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/kallsyms.h>
71#include <linux/stacktrace.h>
72#include <linux/resource.h>
73#include <linux/module.h>
74#include <linux/mount.h>
75#include <linux/security.h>
76#include <linux/ptrace.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 <linux/cn_proc.h>
99#include <trace/events/oom.h>
100#include "internal.h"
101#include "fd.h"
102
103#include "../../lib/kstrtox.h"
104
105/* NOTE:
106 * Implementing inode permission operations in /proc is almost
107 * certainly an error. Permission checks need to happen during
108 * each system call not at open time. The reason is that most of
109 * what we wish to check for permissions in /proc varies at runtime.
110 *
111 * The classic example of a problem is opening file descriptors
112 * in /proc for a task before it execs a suid executable.
113 */
114
115static u8 nlink_tid __ro_after_init;
116static u8 nlink_tgid __ro_after_init;
117
118struct pid_entry {
119 const char *name;
120 unsigned int len;
121 umode_t mode;
122 const struct inode_operations *iop;
123 const struct file_operations *fop;
124 union proc_op op;
125};
126
127#define NOD(NAME, MODE, IOP, FOP, OP) { \
128 .name = (NAME), \
129 .len = sizeof(NAME) - 1, \
130 .mode = MODE, \
131 .iop = IOP, \
132 .fop = FOP, \
133 .op = OP, \
134}
135
136#define DIR(NAME, MODE, iops, fops) \
137 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
138#define LNK(NAME, get_link) \
139 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
140 &proc_pid_link_inode_operations, NULL, \
141 { .proc_get_link = get_link } )
142#define REG(NAME, MODE, fops) \
143 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
144#define ONE(NAME, MODE, show) \
145 NOD(NAME, (S_IFREG|(MODE)), \
146 NULL, &proc_single_file_operations, \
147 { .proc_show = show } )
148#define ATTR(LSM, NAME, MODE) \
149 NOD(NAME, (S_IFREG|(MODE)), \
150 NULL, &proc_pid_attr_operations, \
151 { .lsm = LSM })
152
153/*
154 * Count the number of hardlinks for the pid_entry table, excluding the .
155 * and .. links.
156 */
157static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
158 unsigned int n)
159{
160 unsigned int i;
161 unsigned int count;
162
163 count = 2;
164 for (i = 0; i < n; ++i) {
165 if (S_ISDIR(entries[i].mode))
166 ++count;
167 }
168
169 return count;
170}
171
172static int get_task_root(struct task_struct *task, struct path *root)
173{
174 int result = -ENOENT;
175
176 task_lock(task);
177 if (task->fs) {
178 get_fs_root(task->fs, root);
179 result = 0;
180 }
181 task_unlock(task);
182 return result;
183}
184
185static int proc_cwd_link(struct dentry *dentry, struct path *path)
186{
187 struct task_struct *task = get_proc_task(d_inode(dentry));
188 int result = -ENOENT;
189
190 if (task) {
191 task_lock(task);
192 if (task->fs) {
193 get_fs_pwd(task->fs, path);
194 result = 0;
195 }
196 task_unlock(task);
197 put_task_struct(task);
198 }
199 return result;
200}
201
202static int proc_root_link(struct dentry *dentry, struct path *path)
203{
204 struct task_struct *task = get_proc_task(d_inode(dentry));
205 int result = -ENOENT;
206
207 if (task) {
208 result = get_task_root(task, path);
209 put_task_struct(task);
210 }
211 return result;
212}
213
214/*
215 * If the user used setproctitle(), we just get the string from
216 * user space at arg_start, and limit it to a maximum of one page.
217 */
218static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
219 size_t count, unsigned long pos,
220 unsigned long arg_start)
221{
222 char *page;
223 int ret, got;
224
225 if (pos >= PAGE_SIZE)
226 return 0;
227
228 page = (char *)__get_free_page(GFP_KERNEL);
229 if (!page)
230 return -ENOMEM;
231
232 ret = 0;
233 got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
234 if (got > 0) {
235 int len = strnlen(page, got);
236
237 /* Include the NUL character if it was found */
238 if (len < got)
239 len++;
240
241 if (len > pos) {
242 len -= pos;
243 if (len > count)
244 len = count;
245 len -= copy_to_user(buf, page+pos, len);
246 if (!len)
247 len = -EFAULT;
248 ret = len;
249 }
250 }
251 free_page((unsigned long)page);
252 return ret;
253}
254
255static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
256 size_t count, loff_t *ppos)
257{
258 unsigned long arg_start, arg_end, env_start, env_end;
259 unsigned long pos, len;
260 char *page, c;
261
262 /* Check if process spawned far enough to have cmdline. */
263 if (!mm->env_end)
264 return 0;
265
266 spin_lock(&mm->arg_lock);
267 arg_start = mm->arg_start;
268 arg_end = mm->arg_end;
269 env_start = mm->env_start;
270 env_end = mm->env_end;
271 spin_unlock(&mm->arg_lock);
272
273 if (arg_start >= arg_end)
274 return 0;
275
276 /*
277 * We allow setproctitle() to overwrite the argument
278 * strings, and overflow past the original end. But
279 * only when it overflows into the environment area.
280 */
281 if (env_start != arg_end || env_end < env_start)
282 env_start = env_end = arg_end;
283 len = env_end - arg_start;
284
285 /* We're not going to care if "*ppos" has high bits set */
286 pos = *ppos;
287 if (pos >= len)
288 return 0;
289 if (count > len - pos)
290 count = len - pos;
291 if (!count)
292 return 0;
293
294 /*
295 * Magical special case: if the argv[] end byte is not
296 * zero, the user has overwritten it with setproctitle(3).
297 *
298 * Possible future enhancement: do this only once when
299 * pos is 0, and set a flag in the 'struct file'.
300 */
301 if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
302 return get_mm_proctitle(mm, buf, count, pos, arg_start);
303
304 /*
305 * For the non-setproctitle() case we limit things strictly
306 * to the [arg_start, arg_end[ range.
307 */
308 pos += arg_start;
309 if (pos < arg_start || pos >= arg_end)
310 return 0;
311 if (count > arg_end - pos)
312 count = arg_end - pos;
313
314 page = (char *)__get_free_page(GFP_KERNEL);
315 if (!page)
316 return -ENOMEM;
317
318 len = 0;
319 while (count) {
320 int got;
321 size_t size = min_t(size_t, PAGE_SIZE, count);
322
323 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
324 if (got <= 0)
325 break;
326 got -= copy_to_user(buf, page, got);
327 if (unlikely(!got)) {
328 if (!len)
329 len = -EFAULT;
330 break;
331 }
332 pos += got;
333 buf += got;
334 len += got;
335 count -= got;
336 }
337
338 free_page((unsigned long)page);
339 return len;
340}
341
342static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
343 size_t count, loff_t *pos)
344{
345 struct mm_struct *mm;
346 ssize_t ret;
347
348 mm = get_task_mm(tsk);
349 if (!mm)
350 return 0;
351
352 ret = get_mm_cmdline(mm, buf, count, pos);
353 mmput(mm);
354 return ret;
355}
356
357static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
358 size_t count, loff_t *pos)
359{
360 struct task_struct *tsk;
361 ssize_t ret;
362
363 BUG_ON(*pos < 0);
364
365 tsk = get_proc_task(file_inode(file));
366 if (!tsk)
367 return -ESRCH;
368 ret = get_task_cmdline(tsk, buf, count, pos);
369 put_task_struct(tsk);
370 if (ret > 0)
371 *pos += ret;
372 return ret;
373}
374
375static const struct file_operations proc_pid_cmdline_ops = {
376 .read = proc_pid_cmdline_read,
377 .llseek = generic_file_llseek,
378};
379
380#ifdef CONFIG_KALLSYMS
381/*
382 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
383 * Returns the resolved symbol. If that fails, simply return the address.
384 */
385static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
386 struct pid *pid, struct task_struct *task)
387{
388 unsigned long wchan;
389 char symname[KSYM_NAME_LEN];
390
391 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
392 goto print0;
393
394 wchan = get_wchan(task);
395 if (wchan && !lookup_symbol_name(wchan, symname)) {
396 seq_puts(m, symname);
397 return 0;
398 }
399
400print0:
401 seq_putc(m, '0');
402 return 0;
403}
404#endif /* CONFIG_KALLSYMS */
405
406static int lock_trace(struct task_struct *task)
407{
408 int err = down_read_killable(&task->signal->exec_update_lock);
409 if (err)
410 return err;
411 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
412 up_read(&task->signal->exec_update_lock);
413 return -EPERM;
414 }
415 return 0;
416}
417
418static void unlock_trace(struct task_struct *task)
419{
420 up_read(&task->signal->exec_update_lock);
421}
422
423#ifdef CONFIG_STACKTRACE
424
425#define MAX_STACK_TRACE_DEPTH 64
426
427static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
428 struct pid *pid, struct task_struct *task)
429{
430 unsigned long *entries;
431 int err;
432
433 /*
434 * The ability to racily run the kernel stack unwinder on a running task
435 * and then observe the unwinder output is scary; while it is useful for
436 * debugging kernel issues, it can also allow an attacker to leak kernel
437 * stack contents.
438 * Doing this in a manner that is at least safe from races would require
439 * some work to ensure that the remote task can not be scheduled; and
440 * even then, this would still expose the unwinder as local attack
441 * surface.
442 * Therefore, this interface is restricted to root.
443 */
444 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
445 return -EACCES;
446
447 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
448 GFP_KERNEL);
449 if (!entries)
450 return -ENOMEM;
451
452 err = lock_trace(task);
453 if (!err) {
454 unsigned int i, nr_entries;
455
456 nr_entries = stack_trace_save_tsk(task, entries,
457 MAX_STACK_TRACE_DEPTH, 0);
458
459 for (i = 0; i < nr_entries; i++) {
460 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
461 }
462
463 unlock_trace(task);
464 }
465 kfree(entries);
466
467 return err;
468}
469#endif
470
471#ifdef CONFIG_SCHED_INFO
472/*
473 * Provides /proc/PID/schedstat
474 */
475static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
476 struct pid *pid, struct task_struct *task)
477{
478 if (unlikely(!sched_info_on()))
479 seq_puts(m, "0 0 0\n");
480 else
481 seq_printf(m, "%llu %llu %lu\n",
482 (unsigned long long)task->se.sum_exec_runtime,
483 (unsigned long long)task->sched_info.run_delay,
484 task->sched_info.pcount);
485
486 return 0;
487}
488#endif
489
490#ifdef CONFIG_LATENCYTOP
491static int lstats_show_proc(struct seq_file *m, void *v)
492{
493 int i;
494 struct inode *inode = m->private;
495 struct task_struct *task = get_proc_task(inode);
496
497 if (!task)
498 return -ESRCH;
499 seq_puts(m, "Latency Top version : v0.1\n");
500 for (i = 0; i < LT_SAVECOUNT; i++) {
501 struct latency_record *lr = &task->latency_record[i];
502 if (lr->backtrace[0]) {
503 int q;
504 seq_printf(m, "%i %li %li",
505 lr->count, lr->time, lr->max);
506 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
507 unsigned long bt = lr->backtrace[q];
508
509 if (!bt)
510 break;
511 seq_printf(m, " %ps", (void *)bt);
512 }
513 seq_putc(m, '\n');
514 }
515
516 }
517 put_task_struct(task);
518 return 0;
519}
520
521static int lstats_open(struct inode *inode, struct file *file)
522{
523 return single_open(file, lstats_show_proc, inode);
524}
525
526static ssize_t lstats_write(struct file *file, const char __user *buf,
527 size_t count, loff_t *offs)
528{
529 struct task_struct *task = get_proc_task(file_inode(file));
530
531 if (!task)
532 return -ESRCH;
533 clear_tsk_latency_tracing(task);
534 put_task_struct(task);
535
536 return count;
537}
538
539static const struct file_operations proc_lstats_operations = {
540 .open = lstats_open,
541 .read = seq_read,
542 .write = lstats_write,
543 .llseek = seq_lseek,
544 .release = single_release,
545};
546
547#endif
548
549static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
550 struct pid *pid, struct task_struct *task)
551{
552 unsigned long totalpages = totalram_pages() + total_swap_pages;
553 unsigned long points = 0;
554 long badness;
555
556 badness = oom_badness(task, totalpages);
557 /*
558 * Special case OOM_SCORE_ADJ_MIN for all others scale the
559 * badness value into [0, 2000] range which we have been
560 * exporting for a long time so userspace might depend on it.
561 */
562 if (badness != LONG_MIN)
563 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
564
565 seq_printf(m, "%lu\n", points);
566
567 return 0;
568}
569
570struct limit_names {
571 const char *name;
572 const char *unit;
573};
574
575static const struct limit_names lnames[RLIM_NLIMITS] = {
576 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
577 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
578 [RLIMIT_DATA] = {"Max data size", "bytes"},
579 [RLIMIT_STACK] = {"Max stack size", "bytes"},
580 [RLIMIT_CORE] = {"Max core file size", "bytes"},
581 [RLIMIT_RSS] = {"Max resident set", "bytes"},
582 [RLIMIT_NPROC] = {"Max processes", "processes"},
583 [RLIMIT_NOFILE] = {"Max open files", "files"},
584 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
585 [RLIMIT_AS] = {"Max address space", "bytes"},
586 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
587 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
588 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
589 [RLIMIT_NICE] = {"Max nice priority", NULL},
590 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
591 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
592};
593
594/* Display limits for a process */
595static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
596 struct pid *pid, struct task_struct *task)
597{
598 unsigned int i;
599 unsigned long flags;
600
601 struct rlimit rlim[RLIM_NLIMITS];
602
603 if (!lock_task_sighand(task, &flags))
604 return 0;
605 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
606 unlock_task_sighand(task, &flags);
607
608 /*
609 * print the file header
610 */
611 seq_puts(m, "Limit "
612 "Soft Limit "
613 "Hard Limit "
614 "Units \n");
615
616 for (i = 0; i < RLIM_NLIMITS; i++) {
617 if (rlim[i].rlim_cur == RLIM_INFINITY)
618 seq_printf(m, "%-25s %-20s ",
619 lnames[i].name, "unlimited");
620 else
621 seq_printf(m, "%-25s %-20lu ",
622 lnames[i].name, rlim[i].rlim_cur);
623
624 if (rlim[i].rlim_max == RLIM_INFINITY)
625 seq_printf(m, "%-20s ", "unlimited");
626 else
627 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
628
629 if (lnames[i].unit)
630 seq_printf(m, "%-10s\n", lnames[i].unit);
631 else
632 seq_putc(m, '\n');
633 }
634
635 return 0;
636}
637
638#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
639static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
640 struct pid *pid, struct task_struct *task)
641{
642 struct syscall_info info;
643 u64 *args = &info.data.args[0];
644 int res;
645
646 res = lock_trace(task);
647 if (res)
648 return res;
649
650 if (task_current_syscall(task, &info))
651 seq_puts(m, "running\n");
652 else if (info.data.nr < 0)
653 seq_printf(m, "%d 0x%llx 0x%llx\n",
654 info.data.nr, info.sp, info.data.instruction_pointer);
655 else
656 seq_printf(m,
657 "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
658 info.data.nr,
659 args[0], args[1], args[2], args[3], args[4], args[5],
660 info.sp, info.data.instruction_pointer);
661 unlock_trace(task);
662
663 return 0;
664}
665#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
666
667/************************************************************************/
668/* Here the fs part begins */
669/************************************************************************/
670
671/* permission checks */
672static bool proc_fd_access_allowed(struct inode *inode)
673{
674 struct task_struct *task;
675 bool allowed = false;
676 /* Allow access to a task's file descriptors if it is us or we
677 * may use ptrace attach to the process and find out that
678 * information.
679 */
680 task = get_proc_task(inode);
681 if (task) {
682 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
683 put_task_struct(task);
684 }
685 return allowed;
686}
687
688int proc_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
689 struct iattr *attr)
690{
691 int error;
692 struct inode *inode = d_inode(dentry);
693
694 if (attr->ia_valid & ATTR_MODE)
695 return -EPERM;
696
697 error = setattr_prepare(&init_user_ns, dentry, attr);
698 if (error)
699 return error;
700
701 setattr_copy(&init_user_ns, inode, attr);
702 mark_inode_dirty(inode);
703 return 0;
704}
705
706/*
707 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
708 * or euid/egid (for hide_pid_min=2)?
709 */
710static bool has_pid_permissions(struct proc_fs_info *fs_info,
711 struct task_struct *task,
712 enum proc_hidepid hide_pid_min)
713{
714 /*
715 * If 'hidpid' mount option is set force a ptrace check,
716 * we indicate that we are using a filesystem syscall
717 * by passing PTRACE_MODE_READ_FSCREDS
718 */
719 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
720 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
721
722 if (fs_info->hide_pid < hide_pid_min)
723 return true;
724 if (in_group_p(fs_info->pid_gid))
725 return true;
726 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
727}
728
729
730static int proc_pid_permission(struct user_namespace *mnt_userns,
731 struct inode *inode, int mask)
732{
733 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
734 struct task_struct *task;
735 bool has_perms;
736
737 task = get_proc_task(inode);
738 if (!task)
739 return -ESRCH;
740 has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
741 put_task_struct(task);
742
743 if (!has_perms) {
744 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
745 /*
746 * Let's make getdents(), stat(), and open()
747 * consistent with each other. If a process
748 * may not stat() a file, it shouldn't be seen
749 * in procfs at all.
750 */
751 return -ENOENT;
752 }
753
754 return -EPERM;
755 }
756 return generic_permission(&init_user_ns, inode, mask);
757}
758
759
760
761static const struct inode_operations proc_def_inode_operations = {
762 .setattr = proc_setattr,
763};
764
765static int proc_single_show(struct seq_file *m, void *v)
766{
767 struct inode *inode = m->private;
768 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
769 struct pid *pid = proc_pid(inode);
770 struct task_struct *task;
771 int ret;
772
773 task = get_pid_task(pid, PIDTYPE_PID);
774 if (!task)
775 return -ESRCH;
776
777 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
778
779 put_task_struct(task);
780 return ret;
781}
782
783static int proc_single_open(struct inode *inode, struct file *filp)
784{
785 return single_open(filp, proc_single_show, inode);
786}
787
788static const struct file_operations proc_single_file_operations = {
789 .open = proc_single_open,
790 .read = seq_read,
791 .llseek = seq_lseek,
792 .release = single_release,
793};
794
795
796struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
797{
798 struct task_struct *task = get_proc_task(inode);
799 struct mm_struct *mm = ERR_PTR(-ESRCH);
800
801 if (task) {
802 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
803 put_task_struct(task);
804
805 if (!IS_ERR_OR_NULL(mm)) {
806 /* ensure this mm_struct can't be freed */
807 mmgrab(mm);
808 /* but do not pin its memory */
809 mmput(mm);
810 }
811 }
812
813 return mm;
814}
815
816static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
817{
818 struct mm_struct *mm = proc_mem_open(inode, mode);
819
820 if (IS_ERR(mm))
821 return PTR_ERR(mm);
822
823 file->private_data = mm;
824 return 0;
825}
826
827static int mem_open(struct inode *inode, struct file *file)
828{
829 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
830
831 /* OK to pass negative loff_t, we can catch out-of-range */
832 file->f_mode |= FMODE_UNSIGNED_OFFSET;
833
834 return ret;
835}
836
837static ssize_t mem_rw(struct file *file, char __user *buf,
838 size_t count, loff_t *ppos, int write)
839{
840 struct mm_struct *mm = file->private_data;
841 unsigned long addr = *ppos;
842 ssize_t copied;
843 char *page;
844 unsigned int flags;
845
846 if (!mm)
847 return 0;
848
849 page = (char *)__get_free_page(GFP_KERNEL);
850 if (!page)
851 return -ENOMEM;
852
853 copied = 0;
854 if (!mmget_not_zero(mm))
855 goto free;
856
857 flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
858
859 while (count > 0) {
860 size_t this_len = min_t(size_t, count, PAGE_SIZE);
861
862 if (write && copy_from_user(page, buf, this_len)) {
863 copied = -EFAULT;
864 break;
865 }
866
867 this_len = access_remote_vm(mm, addr, page, this_len, flags);
868 if (!this_len) {
869 if (!copied)
870 copied = -EIO;
871 break;
872 }
873
874 if (!write && copy_to_user(buf, page, this_len)) {
875 copied = -EFAULT;
876 break;
877 }
878
879 buf += this_len;
880 addr += this_len;
881 copied += this_len;
882 count -= this_len;
883 }
884 *ppos = addr;
885
886 mmput(mm);
887free:
888 free_page((unsigned long) page);
889 return copied;
890}
891
892static ssize_t mem_read(struct file *file, char __user *buf,
893 size_t count, loff_t *ppos)
894{
895 return mem_rw(file, buf, count, ppos, 0);
896}
897
898static ssize_t mem_write(struct file *file, const char __user *buf,
899 size_t count, loff_t *ppos)
900{
901 return mem_rw(file, (char __user*)buf, count, ppos, 1);
902}
903
904loff_t mem_lseek(struct file *file, loff_t offset, int orig)
905{
906 switch (orig) {
907 case 0:
908 file->f_pos = offset;
909 break;
910 case 1:
911 file->f_pos += offset;
912 break;
913 default:
914 return -EINVAL;
915 }
916 force_successful_syscall_return();
917 return file->f_pos;
918}
919
920static int mem_release(struct inode *inode, struct file *file)
921{
922 struct mm_struct *mm = file->private_data;
923 if (mm)
924 mmdrop(mm);
925 return 0;
926}
927
928static const struct file_operations proc_mem_operations = {
929 .llseek = mem_lseek,
930 .read = mem_read,
931 .write = mem_write,
932 .open = mem_open,
933 .release = mem_release,
934};
935
936static int environ_open(struct inode *inode, struct file *file)
937{
938 return __mem_open(inode, file, PTRACE_MODE_READ);
939}
940
941static ssize_t environ_read(struct file *file, char __user *buf,
942 size_t count, loff_t *ppos)
943{
944 char *page;
945 unsigned long src = *ppos;
946 int ret = 0;
947 struct mm_struct *mm = file->private_data;
948 unsigned long env_start, env_end;
949
950 /* Ensure the process spawned far enough to have an environment. */
951 if (!mm || !mm->env_end)
952 return 0;
953
954 page = (char *)__get_free_page(GFP_KERNEL);
955 if (!page)
956 return -ENOMEM;
957
958 ret = 0;
959 if (!mmget_not_zero(mm))
960 goto free;
961
962 spin_lock(&mm->arg_lock);
963 env_start = mm->env_start;
964 env_end = mm->env_end;
965 spin_unlock(&mm->arg_lock);
966
967 while (count > 0) {
968 size_t this_len, max_len;
969 int retval;
970
971 if (src >= (env_end - env_start))
972 break;
973
974 this_len = env_end - (env_start + src);
975
976 max_len = min_t(size_t, PAGE_SIZE, count);
977 this_len = min(max_len, this_len);
978
979 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
980
981 if (retval <= 0) {
982 ret = retval;
983 break;
984 }
985
986 if (copy_to_user(buf, page, retval)) {
987 ret = -EFAULT;
988 break;
989 }
990
991 ret += retval;
992 src += retval;
993 buf += retval;
994 count -= retval;
995 }
996 *ppos = src;
997 mmput(mm);
998
999free:
1000 free_page((unsigned long) page);
1001 return ret;
1002}
1003
1004static const struct file_operations proc_environ_operations = {
1005 .open = environ_open,
1006 .read = environ_read,
1007 .llseek = generic_file_llseek,
1008 .release = mem_release,
1009};
1010
1011static int auxv_open(struct inode *inode, struct file *file)
1012{
1013 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1014}
1015
1016static ssize_t auxv_read(struct file *file, char __user *buf,
1017 size_t count, loff_t *ppos)
1018{
1019 struct mm_struct *mm = file->private_data;
1020 unsigned int nwords = 0;
1021
1022 if (!mm)
1023 return 0;
1024 do {
1025 nwords += 2;
1026 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1027 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1028 nwords * sizeof(mm->saved_auxv[0]));
1029}
1030
1031static const struct file_operations proc_auxv_operations = {
1032 .open = auxv_open,
1033 .read = auxv_read,
1034 .llseek = generic_file_llseek,
1035 .release = mem_release,
1036};
1037
1038static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1039 loff_t *ppos)
1040{
1041 struct task_struct *task = get_proc_task(file_inode(file));
1042 char buffer[PROC_NUMBUF];
1043 int oom_adj = OOM_ADJUST_MIN;
1044 size_t len;
1045
1046 if (!task)
1047 return -ESRCH;
1048 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1049 oom_adj = OOM_ADJUST_MAX;
1050 else
1051 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1052 OOM_SCORE_ADJ_MAX;
1053 put_task_struct(task);
1054 if (oom_adj > OOM_ADJUST_MAX)
1055 oom_adj = OOM_ADJUST_MAX;
1056 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1057 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1058}
1059
1060static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1061{
1062 struct mm_struct *mm = NULL;
1063 struct task_struct *task;
1064 int err = 0;
1065
1066 task = get_proc_task(file_inode(file));
1067 if (!task)
1068 return -ESRCH;
1069
1070 mutex_lock(&oom_adj_mutex);
1071 if (legacy) {
1072 if (oom_adj < task->signal->oom_score_adj &&
1073 !capable(CAP_SYS_RESOURCE)) {
1074 err = -EACCES;
1075 goto err_unlock;
1076 }
1077 /*
1078 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1079 * /proc/pid/oom_score_adj instead.
1080 */
1081 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1082 current->comm, task_pid_nr(current), task_pid_nr(task),
1083 task_pid_nr(task));
1084 } else {
1085 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1086 !capable(CAP_SYS_RESOURCE)) {
1087 err = -EACCES;
1088 goto err_unlock;
1089 }
1090 }
1091
1092 /*
1093 * Make sure we will check other processes sharing the mm if this is
1094 * not vfrok which wants its own oom_score_adj.
1095 * pin the mm so it doesn't go away and get reused after task_unlock
1096 */
1097 if (!task->vfork_done) {
1098 struct task_struct *p = find_lock_task_mm(task);
1099
1100 if (p) {
1101 if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1102 mm = p->mm;
1103 mmgrab(mm);
1104 }
1105 task_unlock(p);
1106 }
1107 }
1108
1109 task->signal->oom_score_adj = oom_adj;
1110 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1111 task->signal->oom_score_adj_min = (short)oom_adj;
1112 trace_oom_score_adj_update(task);
1113
1114 if (mm) {
1115 struct task_struct *p;
1116
1117 rcu_read_lock();
1118 for_each_process(p) {
1119 if (same_thread_group(task, p))
1120 continue;
1121
1122 /* do not touch kernel threads or the global init */
1123 if (p->flags & PF_KTHREAD || is_global_init(p))
1124 continue;
1125
1126 task_lock(p);
1127 if (!p->vfork_done && process_shares_mm(p, mm)) {
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_AUDIT
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 /* Don't let kthreads write their own loginuid */
1276 if (current->flags & PF_KTHREAD)
1277 return -EPERM;
1278
1279 rcu_read_lock();
1280 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1281 rcu_read_unlock();
1282 return -EPERM;
1283 }
1284 rcu_read_unlock();
1285
1286 if (*ppos != 0) {
1287 /* No partial writes. */
1288 return -EINVAL;
1289 }
1290
1291 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1292 if (rv < 0)
1293 return rv;
1294
1295 /* is userspace tring to explicitly UNSET the loginuid? */
1296 if (loginuid == AUDIT_UID_UNSET) {
1297 kloginuid = INVALID_UID;
1298 } else {
1299 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1300 if (!uid_valid(kloginuid))
1301 return -EINVAL;
1302 }
1303
1304 rv = audit_set_loginuid(kloginuid);
1305 if (rv < 0)
1306 return rv;
1307 return count;
1308}
1309
1310static const struct file_operations proc_loginuid_operations = {
1311 .read = proc_loginuid_read,
1312 .write = proc_loginuid_write,
1313 .llseek = generic_file_llseek,
1314};
1315
1316static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1317 size_t count, loff_t *ppos)
1318{
1319 struct inode * inode = file_inode(file);
1320 struct task_struct *task = get_proc_task(inode);
1321 ssize_t length;
1322 char tmpbuf[TMPBUFLEN];
1323
1324 if (!task)
1325 return -ESRCH;
1326 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1327 audit_get_sessionid(task));
1328 put_task_struct(task);
1329 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1330}
1331
1332static const struct file_operations proc_sessionid_operations = {
1333 .read = proc_sessionid_read,
1334 .llseek = generic_file_llseek,
1335};
1336#endif
1337
1338#ifdef CONFIG_FAULT_INJECTION
1339static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1340 size_t count, loff_t *ppos)
1341{
1342 struct task_struct *task = get_proc_task(file_inode(file));
1343 char buffer[PROC_NUMBUF];
1344 size_t len;
1345 int make_it_fail;
1346
1347 if (!task)
1348 return -ESRCH;
1349 make_it_fail = task->make_it_fail;
1350 put_task_struct(task);
1351
1352 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1353
1354 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1355}
1356
1357static ssize_t proc_fault_inject_write(struct file * file,
1358 const char __user * buf, size_t count, loff_t *ppos)
1359{
1360 struct task_struct *task;
1361 char buffer[PROC_NUMBUF];
1362 int make_it_fail;
1363 int rv;
1364
1365 if (!capable(CAP_SYS_RESOURCE))
1366 return -EPERM;
1367 memset(buffer, 0, sizeof(buffer));
1368 if (count > sizeof(buffer) - 1)
1369 count = sizeof(buffer) - 1;
1370 if (copy_from_user(buffer, buf, count))
1371 return -EFAULT;
1372 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1373 if (rv < 0)
1374 return rv;
1375 if (make_it_fail < 0 || make_it_fail > 1)
1376 return -EINVAL;
1377
1378 task = get_proc_task(file_inode(file));
1379 if (!task)
1380 return -ESRCH;
1381 task->make_it_fail = make_it_fail;
1382 put_task_struct(task);
1383
1384 return count;
1385}
1386
1387static const struct file_operations proc_fault_inject_operations = {
1388 .read = proc_fault_inject_read,
1389 .write = proc_fault_inject_write,
1390 .llseek = generic_file_llseek,
1391};
1392
1393static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1394 size_t count, loff_t *ppos)
1395{
1396 struct task_struct *task;
1397 int err;
1398 unsigned int n;
1399
1400 err = kstrtouint_from_user(buf, count, 0, &n);
1401 if (err)
1402 return err;
1403
1404 task = get_proc_task(file_inode(file));
1405 if (!task)
1406 return -ESRCH;
1407 task->fail_nth = n;
1408 put_task_struct(task);
1409
1410 return count;
1411}
1412
1413static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1414 size_t count, loff_t *ppos)
1415{
1416 struct task_struct *task;
1417 char numbuf[PROC_NUMBUF];
1418 ssize_t len;
1419
1420 task = get_proc_task(file_inode(file));
1421 if (!task)
1422 return -ESRCH;
1423 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1424 put_task_struct(task);
1425 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1426}
1427
1428static const struct file_operations proc_fail_nth_operations = {
1429 .read = proc_fail_nth_read,
1430 .write = proc_fail_nth_write,
1431};
1432#endif
1433
1434
1435#ifdef CONFIG_SCHED_DEBUG
1436/*
1437 * Print out various scheduling related per-task fields:
1438 */
1439static int sched_show(struct seq_file *m, void *v)
1440{
1441 struct inode *inode = m->private;
1442 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1443 struct task_struct *p;
1444
1445 p = get_proc_task(inode);
1446 if (!p)
1447 return -ESRCH;
1448 proc_sched_show_task(p, ns, m);
1449
1450 put_task_struct(p);
1451
1452 return 0;
1453}
1454
1455static ssize_t
1456sched_write(struct file *file, const char __user *buf,
1457 size_t count, loff_t *offset)
1458{
1459 struct inode *inode = file_inode(file);
1460 struct task_struct *p;
1461
1462 p = get_proc_task(inode);
1463 if (!p)
1464 return -ESRCH;
1465 proc_sched_set_task(p);
1466
1467 put_task_struct(p);
1468
1469 return count;
1470}
1471
1472static int sched_open(struct inode *inode, struct file *filp)
1473{
1474 return single_open(filp, sched_show, inode);
1475}
1476
1477static const struct file_operations proc_pid_sched_operations = {
1478 .open = sched_open,
1479 .read = seq_read,
1480 .write = sched_write,
1481 .llseek = seq_lseek,
1482 .release = single_release,
1483};
1484
1485#endif
1486
1487#ifdef CONFIG_SCHED_AUTOGROUP
1488/*
1489 * Print out autogroup related information:
1490 */
1491static int sched_autogroup_show(struct seq_file *m, void *v)
1492{
1493 struct inode *inode = m->private;
1494 struct task_struct *p;
1495
1496 p = get_proc_task(inode);
1497 if (!p)
1498 return -ESRCH;
1499 proc_sched_autogroup_show_task(p, m);
1500
1501 put_task_struct(p);
1502
1503 return 0;
1504}
1505
1506static ssize_t
1507sched_autogroup_write(struct file *file, const char __user *buf,
1508 size_t count, loff_t *offset)
1509{
1510 struct inode *inode = file_inode(file);
1511 struct task_struct *p;
1512 char buffer[PROC_NUMBUF];
1513 int nice;
1514 int err;
1515
1516 memset(buffer, 0, sizeof(buffer));
1517 if (count > sizeof(buffer) - 1)
1518 count = sizeof(buffer) - 1;
1519 if (copy_from_user(buffer, buf, count))
1520 return -EFAULT;
1521
1522 err = kstrtoint(strstrip(buffer), 0, &nice);
1523 if (err < 0)
1524 return err;
1525
1526 p = get_proc_task(inode);
1527 if (!p)
1528 return -ESRCH;
1529
1530 err = proc_sched_autogroup_set_nice(p, nice);
1531 if (err)
1532 count = err;
1533
1534 put_task_struct(p);
1535
1536 return count;
1537}
1538
1539static int sched_autogroup_open(struct inode *inode, struct file *filp)
1540{
1541 int ret;
1542
1543 ret = single_open(filp, sched_autogroup_show, NULL);
1544 if (!ret) {
1545 struct seq_file *m = filp->private_data;
1546
1547 m->private = inode;
1548 }
1549 return ret;
1550}
1551
1552static const struct file_operations proc_pid_sched_autogroup_operations = {
1553 .open = sched_autogroup_open,
1554 .read = seq_read,
1555 .write = sched_autogroup_write,
1556 .llseek = seq_lseek,
1557 .release = single_release,
1558};
1559
1560#endif /* CONFIG_SCHED_AUTOGROUP */
1561
1562#ifdef CONFIG_TIME_NS
1563static int timens_offsets_show(struct seq_file *m, void *v)
1564{
1565 struct task_struct *p;
1566
1567 p = get_proc_task(file_inode(m->file));
1568 if (!p)
1569 return -ESRCH;
1570 proc_timens_show_offsets(p, m);
1571
1572 put_task_struct(p);
1573
1574 return 0;
1575}
1576
1577static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1578 size_t count, loff_t *ppos)
1579{
1580 struct inode *inode = file_inode(file);
1581 struct proc_timens_offset offsets[2];
1582 char *kbuf = NULL, *pos, *next_line;
1583 struct task_struct *p;
1584 int ret, noffsets;
1585
1586 /* Only allow < page size writes at the beginning of the file */
1587 if ((*ppos != 0) || (count >= PAGE_SIZE))
1588 return -EINVAL;
1589
1590 /* Slurp in the user data */
1591 kbuf = memdup_user_nul(buf, count);
1592 if (IS_ERR(kbuf))
1593 return PTR_ERR(kbuf);
1594
1595 /* Parse the user data */
1596 ret = -EINVAL;
1597 noffsets = 0;
1598 for (pos = kbuf; pos; pos = next_line) {
1599 struct proc_timens_offset *off = &offsets[noffsets];
1600 char clock[10];
1601 int err;
1602
1603 /* Find the end of line and ensure we don't look past it */
1604 next_line = strchr(pos, '\n');
1605 if (next_line) {
1606 *next_line = '\0';
1607 next_line++;
1608 if (*next_line == '\0')
1609 next_line = NULL;
1610 }
1611
1612 err = sscanf(pos, "%9s %lld %lu", clock,
1613 &off->val.tv_sec, &off->val.tv_nsec);
1614 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1615 goto out;
1616
1617 clock[sizeof(clock) - 1] = 0;
1618 if (strcmp(clock, "monotonic") == 0 ||
1619 strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1620 off->clockid = CLOCK_MONOTONIC;
1621 else if (strcmp(clock, "boottime") == 0 ||
1622 strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1623 off->clockid = CLOCK_BOOTTIME;
1624 else
1625 goto out;
1626
1627 noffsets++;
1628 if (noffsets == ARRAY_SIZE(offsets)) {
1629 if (next_line)
1630 count = next_line - kbuf;
1631 break;
1632 }
1633 }
1634
1635 ret = -ESRCH;
1636 p = get_proc_task(inode);
1637 if (!p)
1638 goto out;
1639 ret = proc_timens_set_offset(file, p, offsets, noffsets);
1640 put_task_struct(p);
1641 if (ret)
1642 goto out;
1643
1644 ret = count;
1645out:
1646 kfree(kbuf);
1647 return ret;
1648}
1649
1650static int timens_offsets_open(struct inode *inode, struct file *filp)
1651{
1652 return single_open(filp, timens_offsets_show, inode);
1653}
1654
1655static const struct file_operations proc_timens_offsets_operations = {
1656 .open = timens_offsets_open,
1657 .read = seq_read,
1658 .write = timens_offsets_write,
1659 .llseek = seq_lseek,
1660 .release = single_release,
1661};
1662#endif /* CONFIG_TIME_NS */
1663
1664static ssize_t comm_write(struct file *file, const char __user *buf,
1665 size_t count, loff_t *offset)
1666{
1667 struct inode *inode = file_inode(file);
1668 struct task_struct *p;
1669 char buffer[TASK_COMM_LEN];
1670 const size_t maxlen = sizeof(buffer) - 1;
1671
1672 memset(buffer, 0, sizeof(buffer));
1673 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1674 return -EFAULT;
1675
1676 p = get_proc_task(inode);
1677 if (!p)
1678 return -ESRCH;
1679
1680 if (same_thread_group(current, p)) {
1681 set_task_comm(p, buffer);
1682 proc_comm_connector(p);
1683 }
1684 else
1685 count = -EINVAL;
1686
1687 put_task_struct(p);
1688
1689 return count;
1690}
1691
1692static int comm_show(struct seq_file *m, void *v)
1693{
1694 struct inode *inode = m->private;
1695 struct task_struct *p;
1696
1697 p = get_proc_task(inode);
1698 if (!p)
1699 return -ESRCH;
1700
1701 proc_task_name(m, p, false);
1702 seq_putc(m, '\n');
1703
1704 put_task_struct(p);
1705
1706 return 0;
1707}
1708
1709static int comm_open(struct inode *inode, struct file *filp)
1710{
1711 return single_open(filp, comm_show, inode);
1712}
1713
1714static const struct file_operations proc_pid_set_comm_operations = {
1715 .open = comm_open,
1716 .read = seq_read,
1717 .write = comm_write,
1718 .llseek = seq_lseek,
1719 .release = single_release,
1720};
1721
1722static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1723{
1724 struct task_struct *task;
1725 struct file *exe_file;
1726
1727 task = get_proc_task(d_inode(dentry));
1728 if (!task)
1729 return -ENOENT;
1730 exe_file = get_task_exe_file(task);
1731 put_task_struct(task);
1732 if (exe_file) {
1733 *exe_path = exe_file->f_path;
1734 path_get(&exe_file->f_path);
1735 fput(exe_file);
1736 return 0;
1737 } else
1738 return -ENOENT;
1739}
1740
1741static const char *proc_pid_get_link(struct dentry *dentry,
1742 struct inode *inode,
1743 struct delayed_call *done)
1744{
1745 struct path path;
1746 int error = -EACCES;
1747
1748 if (!dentry)
1749 return ERR_PTR(-ECHILD);
1750
1751 /* Are we allowed to snoop on the tasks file descriptors? */
1752 if (!proc_fd_access_allowed(inode))
1753 goto out;
1754
1755 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1756 if (error)
1757 goto out;
1758
1759 error = nd_jump_link(&path);
1760out:
1761 return ERR_PTR(error);
1762}
1763
1764static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1765{
1766 char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1767 char *pathname;
1768 int len;
1769
1770 if (!tmp)
1771 return -ENOMEM;
1772
1773 pathname = d_path(path, tmp, PATH_MAX);
1774 len = PTR_ERR(pathname);
1775 if (IS_ERR(pathname))
1776 goto out;
1777 len = tmp + PATH_MAX - 1 - pathname;
1778
1779 if (len > buflen)
1780 len = buflen;
1781 if (copy_to_user(buffer, pathname, len))
1782 len = -EFAULT;
1783 out:
1784 kfree(tmp);
1785 return len;
1786}
1787
1788static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1789{
1790 int error = -EACCES;
1791 struct inode *inode = d_inode(dentry);
1792 struct path path;
1793
1794 /* Are we allowed to snoop on the tasks file descriptors? */
1795 if (!proc_fd_access_allowed(inode))
1796 goto out;
1797
1798 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1799 if (error)
1800 goto out;
1801
1802 error = do_proc_readlink(&path, buffer, buflen);
1803 path_put(&path);
1804out:
1805 return error;
1806}
1807
1808const struct inode_operations proc_pid_link_inode_operations = {
1809 .readlink = proc_pid_readlink,
1810 .get_link = proc_pid_get_link,
1811 .setattr = proc_setattr,
1812};
1813
1814
1815/* building an inode */
1816
1817void task_dump_owner(struct task_struct *task, umode_t mode,
1818 kuid_t *ruid, kgid_t *rgid)
1819{
1820 /* Depending on the state of dumpable compute who should own a
1821 * proc file for a task.
1822 */
1823 const struct cred *cred;
1824 kuid_t uid;
1825 kgid_t gid;
1826
1827 if (unlikely(task->flags & PF_KTHREAD)) {
1828 *ruid = GLOBAL_ROOT_UID;
1829 *rgid = GLOBAL_ROOT_GID;
1830 return;
1831 }
1832
1833 /* Default to the tasks effective ownership */
1834 rcu_read_lock();
1835 cred = __task_cred(task);
1836 uid = cred->euid;
1837 gid = cred->egid;
1838 rcu_read_unlock();
1839
1840 /*
1841 * Before the /proc/pid/status file was created the only way to read
1842 * the effective uid of a /process was to stat /proc/pid. Reading
1843 * /proc/pid/status is slow enough that procps and other packages
1844 * kept stating /proc/pid. To keep the rules in /proc simple I have
1845 * made this apply to all per process world readable and executable
1846 * directories.
1847 */
1848 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1849 struct mm_struct *mm;
1850 task_lock(task);
1851 mm = task->mm;
1852 /* Make non-dumpable tasks owned by some root */
1853 if (mm) {
1854 if (get_dumpable(mm) != SUID_DUMP_USER) {
1855 struct user_namespace *user_ns = mm->user_ns;
1856
1857 uid = make_kuid(user_ns, 0);
1858 if (!uid_valid(uid))
1859 uid = GLOBAL_ROOT_UID;
1860
1861 gid = make_kgid(user_ns, 0);
1862 if (!gid_valid(gid))
1863 gid = GLOBAL_ROOT_GID;
1864 }
1865 } else {
1866 uid = GLOBAL_ROOT_UID;
1867 gid = GLOBAL_ROOT_GID;
1868 }
1869 task_unlock(task);
1870 }
1871 *ruid = uid;
1872 *rgid = gid;
1873}
1874
1875void proc_pid_evict_inode(struct proc_inode *ei)
1876{
1877 struct pid *pid = ei->pid;
1878
1879 if (S_ISDIR(ei->vfs_inode.i_mode)) {
1880 spin_lock(&pid->lock);
1881 hlist_del_init_rcu(&ei->sibling_inodes);
1882 spin_unlock(&pid->lock);
1883 }
1884
1885 put_pid(pid);
1886}
1887
1888struct inode *proc_pid_make_inode(struct super_block *sb,
1889 struct task_struct *task, umode_t mode)
1890{
1891 struct inode * inode;
1892 struct proc_inode *ei;
1893 struct pid *pid;
1894
1895 /* We need a new inode */
1896
1897 inode = new_inode(sb);
1898 if (!inode)
1899 goto out;
1900
1901 /* Common stuff */
1902 ei = PROC_I(inode);
1903 inode->i_mode = mode;
1904 inode->i_ino = get_next_ino();
1905 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1906 inode->i_op = &proc_def_inode_operations;
1907
1908 /*
1909 * grab the reference to task.
1910 */
1911 pid = get_task_pid(task, PIDTYPE_PID);
1912 if (!pid)
1913 goto out_unlock;
1914
1915 /* Let the pid remember us for quick removal */
1916 ei->pid = pid;
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
1929/*
1930 * Generating an inode and adding it into @pid->inodes, so that task will
1931 * invalidate inode's dentry before being released.
1932 *
1933 * This helper is used for creating dir-type entries under '/proc' and
1934 * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1935 * can be released by invalidating '/proc/<tgid>' dentry.
1936 * In theory, dentries under '/proc/<tgid>/task' can also be released by
1937 * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1938 * thread exiting situation: Any one of threads should invalidate its
1939 * '/proc/<tgid>/task/<pid>' dentry before released.
1940 */
1941static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1942 struct task_struct *task, umode_t mode)
1943{
1944 struct inode *inode;
1945 struct proc_inode *ei;
1946 struct pid *pid;
1947
1948 inode = proc_pid_make_inode(sb, task, mode);
1949 if (!inode)
1950 return NULL;
1951
1952 /* Let proc_flush_pid find this directory inode */
1953 ei = PROC_I(inode);
1954 pid = ei->pid;
1955 spin_lock(&pid->lock);
1956 hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
1957 spin_unlock(&pid->lock);
1958
1959 return inode;
1960}
1961
1962int pid_getattr(struct user_namespace *mnt_userns, const struct path *path,
1963 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1964{
1965 struct inode *inode = d_inode(path->dentry);
1966 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
1967 struct task_struct *task;
1968
1969 generic_fillattr(&init_user_ns, inode, stat);
1970
1971 stat->uid = GLOBAL_ROOT_UID;
1972 stat->gid = GLOBAL_ROOT_GID;
1973 rcu_read_lock();
1974 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1975 if (task) {
1976 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
1977 rcu_read_unlock();
1978 /*
1979 * This doesn't prevent learning whether PID exists,
1980 * it only makes getattr() consistent with readdir().
1981 */
1982 return -ENOENT;
1983 }
1984 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1985 }
1986 rcu_read_unlock();
1987 return 0;
1988}
1989
1990/* dentry stuff */
1991
1992/*
1993 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1994 */
1995void pid_update_inode(struct task_struct *task, struct inode *inode)
1996{
1997 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1998
1999 inode->i_mode &= ~(S_ISUID | S_ISGID);
2000 security_task_to_inode(task, inode);
2001}
2002
2003/*
2004 * Rewrite the inode's ownerships here because the owning task may have
2005 * performed a setuid(), etc.
2006 *
2007 */
2008static int pid_revalidate(struct dentry *dentry, unsigned int flags)
2009{
2010 struct inode *inode;
2011 struct task_struct *task;
2012 int ret = 0;
2013
2014 rcu_read_lock();
2015 inode = d_inode_rcu(dentry);
2016 if (!inode)
2017 goto out;
2018 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2019
2020 if (task) {
2021 pid_update_inode(task, inode);
2022 ret = 1;
2023 }
2024out:
2025 rcu_read_unlock();
2026 return ret;
2027}
2028
2029static inline bool proc_inode_is_dead(struct inode *inode)
2030{
2031 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2032}
2033
2034int pid_delete_dentry(const struct dentry *dentry)
2035{
2036 /* Is the task we represent dead?
2037 * If so, then don't put the dentry on the lru list,
2038 * kill it immediately.
2039 */
2040 return proc_inode_is_dead(d_inode(dentry));
2041}
2042
2043const struct dentry_operations pid_dentry_operations =
2044{
2045 .d_revalidate = pid_revalidate,
2046 .d_delete = pid_delete_dentry,
2047};
2048
2049/* Lookups */
2050
2051/*
2052 * Fill a directory entry.
2053 *
2054 * If possible create the dcache entry and derive our inode number and
2055 * file type from dcache entry.
2056 *
2057 * Since all of the proc inode numbers are dynamically generated, the inode
2058 * numbers do not exist until the inode is cache. This means creating
2059 * the dcache entry in readdir is necessary to keep the inode numbers
2060 * reported by readdir in sync with the inode numbers reported
2061 * by stat.
2062 */
2063bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2064 const char *name, unsigned int len,
2065 instantiate_t instantiate, struct task_struct *task, const void *ptr)
2066{
2067 struct dentry *child, *dir = file->f_path.dentry;
2068 struct qstr qname = QSTR_INIT(name, len);
2069 struct inode *inode;
2070 unsigned type = DT_UNKNOWN;
2071 ino_t ino = 1;
2072
2073 child = d_hash_and_lookup(dir, &qname);
2074 if (!child) {
2075 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2076 child = d_alloc_parallel(dir, &qname, &wq);
2077 if (IS_ERR(child))
2078 goto end_instantiate;
2079 if (d_in_lookup(child)) {
2080 struct dentry *res;
2081 res = instantiate(child, task, ptr);
2082 d_lookup_done(child);
2083 if (unlikely(res)) {
2084 dput(child);
2085 child = res;
2086 if (IS_ERR(child))
2087 goto end_instantiate;
2088 }
2089 }
2090 }
2091 inode = d_inode(child);
2092 ino = inode->i_ino;
2093 type = inode->i_mode >> 12;
2094 dput(child);
2095end_instantiate:
2096 return dir_emit(ctx, name, len, ino, type);
2097}
2098
2099/*
2100 * dname_to_vma_addr - maps a dentry name into two unsigned longs
2101 * which represent vma start and end addresses.
2102 */
2103static int dname_to_vma_addr(struct dentry *dentry,
2104 unsigned long *start, unsigned long *end)
2105{
2106 const char *str = dentry->d_name.name;
2107 unsigned long long sval, eval;
2108 unsigned int len;
2109
2110 if (str[0] == '0' && str[1] != '-')
2111 return -EINVAL;
2112 len = _parse_integer(str, 16, &sval);
2113 if (len & KSTRTOX_OVERFLOW)
2114 return -EINVAL;
2115 if (sval != (unsigned long)sval)
2116 return -EINVAL;
2117 str += len;
2118
2119 if (*str != '-')
2120 return -EINVAL;
2121 str++;
2122
2123 if (str[0] == '0' && str[1])
2124 return -EINVAL;
2125 len = _parse_integer(str, 16, &eval);
2126 if (len & KSTRTOX_OVERFLOW)
2127 return -EINVAL;
2128 if (eval != (unsigned long)eval)
2129 return -EINVAL;
2130 str += len;
2131
2132 if (*str != '\0')
2133 return -EINVAL;
2134
2135 *start = sval;
2136 *end = eval;
2137
2138 return 0;
2139}
2140
2141static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2142{
2143 unsigned long vm_start, vm_end;
2144 bool exact_vma_exists = false;
2145 struct mm_struct *mm = NULL;
2146 struct task_struct *task;
2147 struct inode *inode;
2148 int status = 0;
2149
2150 if (flags & LOOKUP_RCU)
2151 return -ECHILD;
2152
2153 inode = d_inode(dentry);
2154 task = get_proc_task(inode);
2155 if (!task)
2156 goto out_notask;
2157
2158 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2159 if (IS_ERR_OR_NULL(mm))
2160 goto out;
2161
2162 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2163 status = mmap_read_lock_killable(mm);
2164 if (!status) {
2165 exact_vma_exists = !!find_exact_vma(mm, vm_start,
2166 vm_end);
2167 mmap_read_unlock(mm);
2168 }
2169 }
2170
2171 mmput(mm);
2172
2173 if (exact_vma_exists) {
2174 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2175
2176 security_task_to_inode(task, inode);
2177 status = 1;
2178 }
2179
2180out:
2181 put_task_struct(task);
2182
2183out_notask:
2184 return status;
2185}
2186
2187static const struct dentry_operations tid_map_files_dentry_operations = {
2188 .d_revalidate = map_files_d_revalidate,
2189 .d_delete = pid_delete_dentry,
2190};
2191
2192static int map_files_get_link(struct dentry *dentry, struct path *path)
2193{
2194 unsigned long vm_start, vm_end;
2195 struct vm_area_struct *vma;
2196 struct task_struct *task;
2197 struct mm_struct *mm;
2198 int rc;
2199
2200 rc = -ENOENT;
2201 task = get_proc_task(d_inode(dentry));
2202 if (!task)
2203 goto out;
2204
2205 mm = get_task_mm(task);
2206 put_task_struct(task);
2207 if (!mm)
2208 goto out;
2209
2210 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2211 if (rc)
2212 goto out_mmput;
2213
2214 rc = mmap_read_lock_killable(mm);
2215 if (rc)
2216 goto out_mmput;
2217
2218 rc = -ENOENT;
2219 vma = find_exact_vma(mm, vm_start, vm_end);
2220 if (vma && vma->vm_file) {
2221 *path = vma->vm_file->f_path;
2222 path_get(path);
2223 rc = 0;
2224 }
2225 mmap_read_unlock(mm);
2226
2227out_mmput:
2228 mmput(mm);
2229out:
2230 return rc;
2231}
2232
2233struct map_files_info {
2234 unsigned long start;
2235 unsigned long end;
2236 fmode_t mode;
2237};
2238
2239/*
2240 * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2241 * to concerns about how the symlinks may be used to bypass permissions on
2242 * ancestor directories in the path to the file in question.
2243 */
2244static const char *
2245proc_map_files_get_link(struct dentry *dentry,
2246 struct inode *inode,
2247 struct delayed_call *done)
2248{
2249 if (!checkpoint_restore_ns_capable(&init_user_ns))
2250 return ERR_PTR(-EPERM);
2251
2252 return proc_pid_get_link(dentry, inode, done);
2253}
2254
2255/*
2256 * Identical to proc_pid_link_inode_operations except for get_link()
2257 */
2258static const struct inode_operations proc_map_files_link_inode_operations = {
2259 .readlink = proc_pid_readlink,
2260 .get_link = proc_map_files_get_link,
2261 .setattr = proc_setattr,
2262};
2263
2264static struct dentry *
2265proc_map_files_instantiate(struct dentry *dentry,
2266 struct task_struct *task, const void *ptr)
2267{
2268 fmode_t mode = (fmode_t)(unsigned long)ptr;
2269 struct proc_inode *ei;
2270 struct inode *inode;
2271
2272 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2273 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2274 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2275 if (!inode)
2276 return ERR_PTR(-ENOENT);
2277
2278 ei = PROC_I(inode);
2279 ei->op.proc_get_link = map_files_get_link;
2280
2281 inode->i_op = &proc_map_files_link_inode_operations;
2282 inode->i_size = 64;
2283
2284 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2285 return d_splice_alias(inode, dentry);
2286}
2287
2288static struct dentry *proc_map_files_lookup(struct inode *dir,
2289 struct dentry *dentry, unsigned int flags)
2290{
2291 unsigned long vm_start, vm_end;
2292 struct vm_area_struct *vma;
2293 struct task_struct *task;
2294 struct dentry *result;
2295 struct mm_struct *mm;
2296
2297 result = ERR_PTR(-ENOENT);
2298 task = get_proc_task(dir);
2299 if (!task)
2300 goto out;
2301
2302 result = ERR_PTR(-EACCES);
2303 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2304 goto out_put_task;
2305
2306 result = ERR_PTR(-ENOENT);
2307 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2308 goto out_put_task;
2309
2310 mm = get_task_mm(task);
2311 if (!mm)
2312 goto out_put_task;
2313
2314 result = ERR_PTR(-EINTR);
2315 if (mmap_read_lock_killable(mm))
2316 goto out_put_mm;
2317
2318 result = ERR_PTR(-ENOENT);
2319 vma = find_exact_vma(mm, vm_start, vm_end);
2320 if (!vma)
2321 goto out_no_vma;
2322
2323 if (vma->vm_file)
2324 result = proc_map_files_instantiate(dentry, task,
2325 (void *)(unsigned long)vma->vm_file->f_mode);
2326
2327out_no_vma:
2328 mmap_read_unlock(mm);
2329out_put_mm:
2330 mmput(mm);
2331out_put_task:
2332 put_task_struct(task);
2333out:
2334 return result;
2335}
2336
2337static const struct inode_operations proc_map_files_inode_operations = {
2338 .lookup = proc_map_files_lookup,
2339 .permission = proc_fd_permission,
2340 .setattr = proc_setattr,
2341};
2342
2343static int
2344proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2345{
2346 struct vm_area_struct *vma;
2347 struct task_struct *task;
2348 struct mm_struct *mm;
2349 unsigned long nr_files, pos, i;
2350 GENRADIX(struct map_files_info) fa;
2351 struct map_files_info *p;
2352 int ret;
2353 struct vma_iterator vmi;
2354
2355 genradix_init(&fa);
2356
2357 ret = -ENOENT;
2358 task = get_proc_task(file_inode(file));
2359 if (!task)
2360 goto out;
2361
2362 ret = -EACCES;
2363 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2364 goto out_put_task;
2365
2366 ret = 0;
2367 if (!dir_emit_dots(file, ctx))
2368 goto out_put_task;
2369
2370 mm = get_task_mm(task);
2371 if (!mm)
2372 goto out_put_task;
2373
2374 ret = mmap_read_lock_killable(mm);
2375 if (ret) {
2376 mmput(mm);
2377 goto out_put_task;
2378 }
2379
2380 nr_files = 0;
2381
2382 /*
2383 * We need two passes here:
2384 *
2385 * 1) Collect vmas of mapped files with mmap_lock taken
2386 * 2) Release mmap_lock and instantiate entries
2387 *
2388 * otherwise we get lockdep complained, since filldir()
2389 * routine might require mmap_lock taken in might_fault().
2390 */
2391
2392 pos = 2;
2393 vma_iter_init(&vmi, mm, 0);
2394 for_each_vma(vmi, vma) {
2395 if (!vma->vm_file)
2396 continue;
2397 if (++pos <= ctx->pos)
2398 continue;
2399
2400 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2401 if (!p) {
2402 ret = -ENOMEM;
2403 mmap_read_unlock(mm);
2404 mmput(mm);
2405 goto out_put_task;
2406 }
2407
2408 p->start = vma->vm_start;
2409 p->end = vma->vm_end;
2410 p->mode = vma->vm_file->f_mode;
2411 }
2412 mmap_read_unlock(mm);
2413 mmput(mm);
2414
2415 for (i = 0; i < nr_files; i++) {
2416 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2417 unsigned int len;
2418
2419 p = genradix_ptr(&fa, i);
2420 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2421 if (!proc_fill_cache(file, ctx,
2422 buf, len,
2423 proc_map_files_instantiate,
2424 task,
2425 (void *)(unsigned long)p->mode))
2426 break;
2427 ctx->pos++;
2428 }
2429
2430out_put_task:
2431 put_task_struct(task);
2432out:
2433 genradix_free(&fa);
2434 return ret;
2435}
2436
2437static const struct file_operations proc_map_files_operations = {
2438 .read = generic_read_dir,
2439 .iterate_shared = proc_map_files_readdir,
2440 .llseek = generic_file_llseek,
2441};
2442
2443#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2444struct timers_private {
2445 struct pid *pid;
2446 struct task_struct *task;
2447 struct sighand_struct *sighand;
2448 struct pid_namespace *ns;
2449 unsigned long flags;
2450};
2451
2452static void *timers_start(struct seq_file *m, loff_t *pos)
2453{
2454 struct timers_private *tp = m->private;
2455
2456 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2457 if (!tp->task)
2458 return ERR_PTR(-ESRCH);
2459
2460 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2461 if (!tp->sighand)
2462 return ERR_PTR(-ESRCH);
2463
2464 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2465}
2466
2467static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2468{
2469 struct timers_private *tp = m->private;
2470 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2471}
2472
2473static void timers_stop(struct seq_file *m, void *v)
2474{
2475 struct timers_private *tp = m->private;
2476
2477 if (tp->sighand) {
2478 unlock_task_sighand(tp->task, &tp->flags);
2479 tp->sighand = NULL;
2480 }
2481
2482 if (tp->task) {
2483 put_task_struct(tp->task);
2484 tp->task = NULL;
2485 }
2486}
2487
2488static int show_timer(struct seq_file *m, void *v)
2489{
2490 struct k_itimer *timer;
2491 struct timers_private *tp = m->private;
2492 int notify;
2493 static const char * const nstr[] = {
2494 [SIGEV_SIGNAL] = "signal",
2495 [SIGEV_NONE] = "none",
2496 [SIGEV_THREAD] = "thread",
2497 };
2498
2499 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2500 notify = timer->it_sigev_notify;
2501
2502 seq_printf(m, "ID: %d\n", timer->it_id);
2503 seq_printf(m, "signal: %d/%px\n",
2504 timer->sigq->info.si_signo,
2505 timer->sigq->info.si_value.sival_ptr);
2506 seq_printf(m, "notify: %s/%s.%d\n",
2507 nstr[notify & ~SIGEV_THREAD_ID],
2508 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2509 pid_nr_ns(timer->it_pid, tp->ns));
2510 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2511
2512 return 0;
2513}
2514
2515static const struct seq_operations proc_timers_seq_ops = {
2516 .start = timers_start,
2517 .next = timers_next,
2518 .stop = timers_stop,
2519 .show = show_timer,
2520};
2521
2522static int proc_timers_open(struct inode *inode, struct file *file)
2523{
2524 struct timers_private *tp;
2525
2526 tp = __seq_open_private(file, &proc_timers_seq_ops,
2527 sizeof(struct timers_private));
2528 if (!tp)
2529 return -ENOMEM;
2530
2531 tp->pid = proc_pid(inode);
2532 tp->ns = proc_pid_ns(inode->i_sb);
2533 return 0;
2534}
2535
2536static const struct file_operations proc_timers_operations = {
2537 .open = proc_timers_open,
2538 .read = seq_read,
2539 .llseek = seq_lseek,
2540 .release = seq_release_private,
2541};
2542#endif
2543
2544static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2545 size_t count, loff_t *offset)
2546{
2547 struct inode *inode = file_inode(file);
2548 struct task_struct *p;
2549 u64 slack_ns;
2550 int err;
2551
2552 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2553 if (err < 0)
2554 return err;
2555
2556 p = get_proc_task(inode);
2557 if (!p)
2558 return -ESRCH;
2559
2560 if (p != current) {
2561 rcu_read_lock();
2562 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2563 rcu_read_unlock();
2564 count = -EPERM;
2565 goto out;
2566 }
2567 rcu_read_unlock();
2568
2569 err = security_task_setscheduler(p);
2570 if (err) {
2571 count = err;
2572 goto out;
2573 }
2574 }
2575
2576 task_lock(p);
2577 if (slack_ns == 0)
2578 p->timer_slack_ns = p->default_timer_slack_ns;
2579 else
2580 p->timer_slack_ns = slack_ns;
2581 task_unlock(p);
2582
2583out:
2584 put_task_struct(p);
2585
2586 return count;
2587}
2588
2589static int timerslack_ns_show(struct seq_file *m, void *v)
2590{
2591 struct inode *inode = m->private;
2592 struct task_struct *p;
2593 int err = 0;
2594
2595 p = get_proc_task(inode);
2596 if (!p)
2597 return -ESRCH;
2598
2599 if (p != current) {
2600 rcu_read_lock();
2601 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2602 rcu_read_unlock();
2603 err = -EPERM;
2604 goto out;
2605 }
2606 rcu_read_unlock();
2607
2608 err = security_task_getscheduler(p);
2609 if (err)
2610 goto out;
2611 }
2612
2613 task_lock(p);
2614 seq_printf(m, "%llu\n", p->timer_slack_ns);
2615 task_unlock(p);
2616
2617out:
2618 put_task_struct(p);
2619
2620 return err;
2621}
2622
2623static int timerslack_ns_open(struct inode *inode, struct file *filp)
2624{
2625 return single_open(filp, timerslack_ns_show, inode);
2626}
2627
2628static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2629 .open = timerslack_ns_open,
2630 .read = seq_read,
2631 .write = timerslack_ns_write,
2632 .llseek = seq_lseek,
2633 .release = single_release,
2634};
2635
2636static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2637 struct task_struct *task, const void *ptr)
2638{
2639 const struct pid_entry *p = ptr;
2640 struct inode *inode;
2641 struct proc_inode *ei;
2642
2643 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2644 if (!inode)
2645 return ERR_PTR(-ENOENT);
2646
2647 ei = PROC_I(inode);
2648 if (S_ISDIR(inode->i_mode))
2649 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2650 if (p->iop)
2651 inode->i_op = p->iop;
2652 if (p->fop)
2653 inode->i_fop = p->fop;
2654 ei->op = p->op;
2655 pid_update_inode(task, inode);
2656 d_set_d_op(dentry, &pid_dentry_operations);
2657 return d_splice_alias(inode, dentry);
2658}
2659
2660static struct dentry *proc_pident_lookup(struct inode *dir,
2661 struct dentry *dentry,
2662 const struct pid_entry *p,
2663 const struct pid_entry *end)
2664{
2665 struct task_struct *task = get_proc_task(dir);
2666 struct dentry *res = ERR_PTR(-ENOENT);
2667
2668 if (!task)
2669 goto out_no_task;
2670
2671 /*
2672 * Yes, it does not scale. And it should not. Don't add
2673 * new entries into /proc/<tgid>/ without very good reasons.
2674 */
2675 for (; p < end; p++) {
2676 if (p->len != dentry->d_name.len)
2677 continue;
2678 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2679 res = proc_pident_instantiate(dentry, task, p);
2680 break;
2681 }
2682 }
2683 put_task_struct(task);
2684out_no_task:
2685 return res;
2686}
2687
2688static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2689 const struct pid_entry *ents, unsigned int nents)
2690{
2691 struct task_struct *task = get_proc_task(file_inode(file));
2692 const struct pid_entry *p;
2693
2694 if (!task)
2695 return -ENOENT;
2696
2697 if (!dir_emit_dots(file, ctx))
2698 goto out;
2699
2700 if (ctx->pos >= nents + 2)
2701 goto out;
2702
2703 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2704 if (!proc_fill_cache(file, ctx, p->name, p->len,
2705 proc_pident_instantiate, task, p))
2706 break;
2707 ctx->pos++;
2708 }
2709out:
2710 put_task_struct(task);
2711 return 0;
2712}
2713
2714#ifdef CONFIG_SECURITY
2715static int proc_pid_attr_open(struct inode *inode, struct file *file)
2716{
2717 file->private_data = NULL;
2718 __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2719 return 0;
2720}
2721
2722static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2723 size_t count, loff_t *ppos)
2724{
2725 struct inode * inode = file_inode(file);
2726 char *p = NULL;
2727 ssize_t length;
2728 struct task_struct *task = get_proc_task(inode);
2729
2730 if (!task)
2731 return -ESRCH;
2732
2733 length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2734 file->f_path.dentry->d_name.name,
2735 &p);
2736 put_task_struct(task);
2737 if (length > 0)
2738 length = simple_read_from_buffer(buf, count, ppos, p, length);
2739 kfree(p);
2740 return length;
2741}
2742
2743static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2744 size_t count, loff_t *ppos)
2745{
2746 struct inode * inode = file_inode(file);
2747 struct task_struct *task;
2748 void *page;
2749 int rv;
2750
2751 /* A task may only write when it was the opener. */
2752 if (file->private_data != current->mm)
2753 return -EPERM;
2754
2755 rcu_read_lock();
2756 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2757 if (!task) {
2758 rcu_read_unlock();
2759 return -ESRCH;
2760 }
2761 /* A task may only write its own attributes. */
2762 if (current != task) {
2763 rcu_read_unlock();
2764 return -EACCES;
2765 }
2766 /* Prevent changes to overridden credentials. */
2767 if (current_cred() != current_real_cred()) {
2768 rcu_read_unlock();
2769 return -EBUSY;
2770 }
2771 rcu_read_unlock();
2772
2773 if (count > PAGE_SIZE)
2774 count = PAGE_SIZE;
2775
2776 /* No partial writes. */
2777 if (*ppos != 0)
2778 return -EINVAL;
2779
2780 page = memdup_user(buf, count);
2781 if (IS_ERR(page)) {
2782 rv = PTR_ERR(page);
2783 goto out;
2784 }
2785
2786 /* Guard against adverse ptrace interaction */
2787 rv = mutex_lock_interruptible(¤t->signal->cred_guard_mutex);
2788 if (rv < 0)
2789 goto out_free;
2790
2791 rv = security_setprocattr(PROC_I(inode)->op.lsm,
2792 file->f_path.dentry->d_name.name, page,
2793 count);
2794 mutex_unlock(¤t->signal->cred_guard_mutex);
2795out_free:
2796 kfree(page);
2797out:
2798 return rv;
2799}
2800
2801static const struct file_operations proc_pid_attr_operations = {
2802 .open = proc_pid_attr_open,
2803 .read = proc_pid_attr_read,
2804 .write = proc_pid_attr_write,
2805 .llseek = generic_file_llseek,
2806 .release = mem_release,
2807};
2808
2809#define LSM_DIR_OPS(LSM) \
2810static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2811 struct dir_context *ctx) \
2812{ \
2813 return proc_pident_readdir(filp, ctx, \
2814 LSM##_attr_dir_stuff, \
2815 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2816} \
2817\
2818static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2819 .read = generic_read_dir, \
2820 .iterate = proc_##LSM##_attr_dir_iterate, \
2821 .llseek = default_llseek, \
2822}; \
2823\
2824static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2825 struct dentry *dentry, unsigned int flags) \
2826{ \
2827 return proc_pident_lookup(dir, dentry, \
2828 LSM##_attr_dir_stuff, \
2829 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2830} \
2831\
2832static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2833 .lookup = proc_##LSM##_attr_dir_lookup, \
2834 .getattr = pid_getattr, \
2835 .setattr = proc_setattr, \
2836}
2837
2838#ifdef CONFIG_SECURITY_SMACK
2839static const struct pid_entry smack_attr_dir_stuff[] = {
2840 ATTR("smack", "current", 0666),
2841};
2842LSM_DIR_OPS(smack);
2843#endif
2844
2845#ifdef CONFIG_SECURITY_APPARMOR
2846static const struct pid_entry apparmor_attr_dir_stuff[] = {
2847 ATTR("apparmor", "current", 0666),
2848 ATTR("apparmor", "prev", 0444),
2849 ATTR("apparmor", "exec", 0666),
2850};
2851LSM_DIR_OPS(apparmor);
2852#endif
2853
2854static const struct pid_entry attr_dir_stuff[] = {
2855 ATTR(NULL, "current", 0666),
2856 ATTR(NULL, "prev", 0444),
2857 ATTR(NULL, "exec", 0666),
2858 ATTR(NULL, "fscreate", 0666),
2859 ATTR(NULL, "keycreate", 0666),
2860 ATTR(NULL, "sockcreate", 0666),
2861#ifdef CONFIG_SECURITY_SMACK
2862 DIR("smack", 0555,
2863 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2864#endif
2865#ifdef CONFIG_SECURITY_APPARMOR
2866 DIR("apparmor", 0555,
2867 proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2868#endif
2869};
2870
2871static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2872{
2873 return proc_pident_readdir(file, ctx,
2874 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2875}
2876
2877static const struct file_operations proc_attr_dir_operations = {
2878 .read = generic_read_dir,
2879 .iterate_shared = proc_attr_dir_readdir,
2880 .llseek = generic_file_llseek,
2881};
2882
2883static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2884 struct dentry *dentry, unsigned int flags)
2885{
2886 return proc_pident_lookup(dir, dentry,
2887 attr_dir_stuff,
2888 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2889}
2890
2891static const struct inode_operations proc_attr_dir_inode_operations = {
2892 .lookup = proc_attr_dir_lookup,
2893 .getattr = pid_getattr,
2894 .setattr = proc_setattr,
2895};
2896
2897#endif
2898
2899#ifdef CONFIG_ELF_CORE
2900static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2901 size_t count, loff_t *ppos)
2902{
2903 struct task_struct *task = get_proc_task(file_inode(file));
2904 struct mm_struct *mm;
2905 char buffer[PROC_NUMBUF];
2906 size_t len;
2907 int ret;
2908
2909 if (!task)
2910 return -ESRCH;
2911
2912 ret = 0;
2913 mm = get_task_mm(task);
2914 if (mm) {
2915 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2916 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2917 MMF_DUMP_FILTER_SHIFT));
2918 mmput(mm);
2919 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2920 }
2921
2922 put_task_struct(task);
2923
2924 return ret;
2925}
2926
2927static ssize_t proc_coredump_filter_write(struct file *file,
2928 const char __user *buf,
2929 size_t count,
2930 loff_t *ppos)
2931{
2932 struct task_struct *task;
2933 struct mm_struct *mm;
2934 unsigned int val;
2935 int ret;
2936 int i;
2937 unsigned long mask;
2938
2939 ret = kstrtouint_from_user(buf, count, 0, &val);
2940 if (ret < 0)
2941 return ret;
2942
2943 ret = -ESRCH;
2944 task = get_proc_task(file_inode(file));
2945 if (!task)
2946 goto out_no_task;
2947
2948 mm = get_task_mm(task);
2949 if (!mm)
2950 goto out_no_mm;
2951 ret = 0;
2952
2953 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2954 if (val & mask)
2955 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2956 else
2957 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2958 }
2959
2960 mmput(mm);
2961 out_no_mm:
2962 put_task_struct(task);
2963 out_no_task:
2964 if (ret < 0)
2965 return ret;
2966 return count;
2967}
2968
2969static const struct file_operations proc_coredump_filter_operations = {
2970 .read = proc_coredump_filter_read,
2971 .write = proc_coredump_filter_write,
2972 .llseek = generic_file_llseek,
2973};
2974#endif
2975
2976#ifdef CONFIG_TASK_IO_ACCOUNTING
2977static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2978{
2979 struct task_io_accounting acct = task->ioac;
2980 unsigned long flags;
2981 int result;
2982
2983 result = down_read_killable(&task->signal->exec_update_lock);
2984 if (result)
2985 return result;
2986
2987 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2988 result = -EACCES;
2989 goto out_unlock;
2990 }
2991
2992 if (whole && lock_task_sighand(task, &flags)) {
2993 struct task_struct *t = task;
2994
2995 task_io_accounting_add(&acct, &task->signal->ioac);
2996 while_each_thread(task, t)
2997 task_io_accounting_add(&acct, &t->ioac);
2998
2999 unlock_task_sighand(task, &flags);
3000 }
3001 seq_printf(m,
3002 "rchar: %llu\n"
3003 "wchar: %llu\n"
3004 "syscr: %llu\n"
3005 "syscw: %llu\n"
3006 "read_bytes: %llu\n"
3007 "write_bytes: %llu\n"
3008 "cancelled_write_bytes: %llu\n",
3009 (unsigned long long)acct.rchar,
3010 (unsigned long long)acct.wchar,
3011 (unsigned long long)acct.syscr,
3012 (unsigned long long)acct.syscw,
3013 (unsigned long long)acct.read_bytes,
3014 (unsigned long long)acct.write_bytes,
3015 (unsigned long long)acct.cancelled_write_bytes);
3016 result = 0;
3017
3018out_unlock:
3019 up_read(&task->signal->exec_update_lock);
3020 return result;
3021}
3022
3023static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3024 struct pid *pid, struct task_struct *task)
3025{
3026 return do_io_accounting(task, m, 0);
3027}
3028
3029static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3030 struct pid *pid, struct task_struct *task)
3031{
3032 return do_io_accounting(task, m, 1);
3033}
3034#endif /* CONFIG_TASK_IO_ACCOUNTING */
3035
3036#ifdef CONFIG_USER_NS
3037static int proc_id_map_open(struct inode *inode, struct file *file,
3038 const struct seq_operations *seq_ops)
3039{
3040 struct user_namespace *ns = NULL;
3041 struct task_struct *task;
3042 struct seq_file *seq;
3043 int ret = -EINVAL;
3044
3045 task = get_proc_task(inode);
3046 if (task) {
3047 rcu_read_lock();
3048 ns = get_user_ns(task_cred_xxx(task, user_ns));
3049 rcu_read_unlock();
3050 put_task_struct(task);
3051 }
3052 if (!ns)
3053 goto err;
3054
3055 ret = seq_open(file, seq_ops);
3056 if (ret)
3057 goto err_put_ns;
3058
3059 seq = file->private_data;
3060 seq->private = ns;
3061
3062 return 0;
3063err_put_ns:
3064 put_user_ns(ns);
3065err:
3066 return ret;
3067}
3068
3069static int proc_id_map_release(struct inode *inode, struct file *file)
3070{
3071 struct seq_file *seq = file->private_data;
3072 struct user_namespace *ns = seq->private;
3073 put_user_ns(ns);
3074 return seq_release(inode, file);
3075}
3076
3077static int proc_uid_map_open(struct inode *inode, struct file *file)
3078{
3079 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3080}
3081
3082static int proc_gid_map_open(struct inode *inode, struct file *file)
3083{
3084 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3085}
3086
3087static int proc_projid_map_open(struct inode *inode, struct file *file)
3088{
3089 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3090}
3091
3092static const struct file_operations proc_uid_map_operations = {
3093 .open = proc_uid_map_open,
3094 .write = proc_uid_map_write,
3095 .read = seq_read,
3096 .llseek = seq_lseek,
3097 .release = proc_id_map_release,
3098};
3099
3100static const struct file_operations proc_gid_map_operations = {
3101 .open = proc_gid_map_open,
3102 .write = proc_gid_map_write,
3103 .read = seq_read,
3104 .llseek = seq_lseek,
3105 .release = proc_id_map_release,
3106};
3107
3108static const struct file_operations proc_projid_map_operations = {
3109 .open = proc_projid_map_open,
3110 .write = proc_projid_map_write,
3111 .read = seq_read,
3112 .llseek = seq_lseek,
3113 .release = proc_id_map_release,
3114};
3115
3116static int proc_setgroups_open(struct inode *inode, struct file *file)
3117{
3118 struct user_namespace *ns = NULL;
3119 struct task_struct *task;
3120 int ret;
3121
3122 ret = -ESRCH;
3123 task = get_proc_task(inode);
3124 if (task) {
3125 rcu_read_lock();
3126 ns = get_user_ns(task_cred_xxx(task, user_ns));
3127 rcu_read_unlock();
3128 put_task_struct(task);
3129 }
3130 if (!ns)
3131 goto err;
3132
3133 if (file->f_mode & FMODE_WRITE) {
3134 ret = -EACCES;
3135 if (!ns_capable(ns, CAP_SYS_ADMIN))
3136 goto err_put_ns;
3137 }
3138
3139 ret = single_open(file, &proc_setgroups_show, ns);
3140 if (ret)
3141 goto err_put_ns;
3142
3143 return 0;
3144err_put_ns:
3145 put_user_ns(ns);
3146err:
3147 return ret;
3148}
3149
3150static int proc_setgroups_release(struct inode *inode, struct file *file)
3151{
3152 struct seq_file *seq = file->private_data;
3153 struct user_namespace *ns = seq->private;
3154 int ret = single_release(inode, file);
3155 put_user_ns(ns);
3156 return ret;
3157}
3158
3159static const struct file_operations proc_setgroups_operations = {
3160 .open = proc_setgroups_open,
3161 .write = proc_setgroups_write,
3162 .read = seq_read,
3163 .llseek = seq_lseek,
3164 .release = proc_setgroups_release,
3165};
3166#endif /* CONFIG_USER_NS */
3167
3168static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3169 struct pid *pid, struct task_struct *task)
3170{
3171 int err = lock_trace(task);
3172 if (!err) {
3173 seq_printf(m, "%08x\n", task->personality);
3174 unlock_trace(task);
3175 }
3176 return err;
3177}
3178
3179#ifdef CONFIG_LIVEPATCH
3180static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3181 struct pid *pid, struct task_struct *task)
3182{
3183 seq_printf(m, "%d\n", task->patch_state);
3184 return 0;
3185}
3186#endif /* CONFIG_LIVEPATCH */
3187
3188#ifdef CONFIG_KSM
3189static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3190 struct pid *pid, struct task_struct *task)
3191{
3192 struct mm_struct *mm;
3193
3194 mm = get_task_mm(task);
3195 if (mm) {
3196 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3197 mmput(mm);
3198 }
3199
3200 return 0;
3201}
3202static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3203 struct pid *pid, struct task_struct *task)
3204{
3205 struct mm_struct *mm;
3206
3207 mm = get_task_mm(task);
3208 if (mm) {
3209 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3210 mmput(mm);
3211 }
3212
3213 return 0;
3214}
3215#endif /* CONFIG_KSM */
3216
3217#ifdef CONFIG_STACKLEAK_METRICS
3218static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3219 struct pid *pid, struct task_struct *task)
3220{
3221 unsigned long prev_depth = THREAD_SIZE -
3222 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3223 unsigned long depth = THREAD_SIZE -
3224 (task->lowest_stack & (THREAD_SIZE - 1));
3225
3226 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3227 prev_depth, depth);
3228 return 0;
3229}
3230#endif /* CONFIG_STACKLEAK_METRICS */
3231
3232/*
3233 * Thread groups
3234 */
3235static const struct file_operations proc_task_operations;
3236static const struct inode_operations proc_task_inode_operations;
3237
3238static const struct pid_entry tgid_base_stuff[] = {
3239 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3240 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3241 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3242 DIR("fdinfo", S_IRUGO|S_IXUGO, 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#ifdef CONFIG_SCHED_AUTOGROUP
3256 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3257#endif
3258#ifdef CONFIG_TIME_NS
3259 REG("timens_offsets", S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3260#endif
3261 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3262#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3263 ONE("syscall", S_IRUSR, proc_pid_syscall),
3264#endif
3265 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3266 ONE("stat", S_IRUGO, proc_tgid_stat),
3267 ONE("statm", S_IRUGO, proc_pid_statm),
3268 REG("maps", S_IRUGO, proc_pid_maps_operations),
3269#ifdef CONFIG_NUMA
3270 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3271#endif
3272 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3273 LNK("cwd", proc_cwd_link),
3274 LNK("root", proc_root_link),
3275 LNK("exe", proc_exe_link),
3276 REG("mounts", S_IRUGO, proc_mounts_operations),
3277 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3278 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3279#ifdef CONFIG_PROC_PAGE_MONITOR
3280 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3281 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3282 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3283 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3284#endif
3285#ifdef CONFIG_SECURITY
3286 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3287#endif
3288#ifdef CONFIG_KALLSYMS
3289 ONE("wchan", S_IRUGO, proc_pid_wchan),
3290#endif
3291#ifdef CONFIG_STACKTRACE
3292 ONE("stack", S_IRUSR, proc_pid_stack),
3293#endif
3294#ifdef CONFIG_SCHED_INFO
3295 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3296#endif
3297#ifdef CONFIG_LATENCYTOP
3298 REG("latency", S_IRUGO, proc_lstats_operations),
3299#endif
3300#ifdef CONFIG_PROC_PID_CPUSET
3301 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3302#endif
3303#ifdef CONFIG_CGROUPS
3304 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3305#endif
3306#ifdef CONFIG_PROC_CPU_RESCTRL
3307 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3308#endif
3309 ONE("oom_score", S_IRUGO, proc_oom_score),
3310 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3311 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3312#ifdef CONFIG_AUDIT
3313 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3314 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3315#endif
3316#ifdef CONFIG_FAULT_INJECTION
3317 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3318 REG("fail-nth", 0644, proc_fail_nth_operations),
3319#endif
3320#ifdef CONFIG_ELF_CORE
3321 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3322#endif
3323#ifdef CONFIG_TASK_IO_ACCOUNTING
3324 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3325#endif
3326#ifdef CONFIG_USER_NS
3327 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3328 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3329 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3330 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3331#endif
3332#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3333 REG("timers", S_IRUGO, proc_timers_operations),
3334#endif
3335 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3336#ifdef CONFIG_LIVEPATCH
3337 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3338#endif
3339#ifdef CONFIG_STACKLEAK_METRICS
3340 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3341#endif
3342#ifdef CONFIG_PROC_PID_ARCH_STATUS
3343 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3344#endif
3345#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3346 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3347#endif
3348#ifdef CONFIG_KSM
3349 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3350 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3351#endif
3352};
3353
3354static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3355{
3356 return proc_pident_readdir(file, ctx,
3357 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3358}
3359
3360static const struct file_operations proc_tgid_base_operations = {
3361 .read = generic_read_dir,
3362 .iterate_shared = proc_tgid_base_readdir,
3363 .llseek = generic_file_llseek,
3364};
3365
3366struct pid *tgid_pidfd_to_pid(const struct file *file)
3367{
3368 if (file->f_op != &proc_tgid_base_operations)
3369 return ERR_PTR(-EBADF);
3370
3371 return proc_pid(file_inode(file));
3372}
3373
3374static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3375{
3376 return proc_pident_lookup(dir, dentry,
3377 tgid_base_stuff,
3378 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3379}
3380
3381static const struct inode_operations proc_tgid_base_inode_operations = {
3382 .lookup = proc_tgid_base_lookup,
3383 .getattr = pid_getattr,
3384 .setattr = proc_setattr,
3385 .permission = proc_pid_permission,
3386};
3387
3388/**
3389 * proc_flush_pid - Remove dcache entries for @pid from the /proc dcache.
3390 * @pid: pid that should be flushed.
3391 *
3392 * This function walks a list of inodes (that belong to any proc
3393 * filesystem) that are attached to the pid and flushes them from
3394 * the dentry cache.
3395 *
3396 * It is safe and reasonable to cache /proc entries for a task until
3397 * that task exits. After that they just clog up the dcache with
3398 * useless entries, possibly causing useful dcache entries to be
3399 * flushed instead. This routine is provided to flush those useless
3400 * dcache entries when a process is reaped.
3401 *
3402 * NOTE: This routine is just an optimization so it does not guarantee
3403 * that no dcache entries will exist after a process is reaped
3404 * it just makes it very unlikely that any will persist.
3405 */
3406
3407void proc_flush_pid(struct pid *pid)
3408{
3409 proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3410}
3411
3412static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3413 struct task_struct *task, const void *ptr)
3414{
3415 struct inode *inode;
3416
3417 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3418 S_IFDIR | S_IRUGO | S_IXUGO);
3419 if (!inode)
3420 return ERR_PTR(-ENOENT);
3421
3422 inode->i_op = &proc_tgid_base_inode_operations;
3423 inode->i_fop = &proc_tgid_base_operations;
3424 inode->i_flags|=S_IMMUTABLE;
3425
3426 set_nlink(inode, nlink_tgid);
3427 pid_update_inode(task, inode);
3428
3429 d_set_d_op(dentry, &pid_dentry_operations);
3430 return d_splice_alias(inode, dentry);
3431}
3432
3433struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3434{
3435 struct task_struct *task;
3436 unsigned tgid;
3437 struct proc_fs_info *fs_info;
3438 struct pid_namespace *ns;
3439 struct dentry *result = ERR_PTR(-ENOENT);
3440
3441 tgid = name_to_int(&dentry->d_name);
3442 if (tgid == ~0U)
3443 goto out;
3444
3445 fs_info = proc_sb_info(dentry->d_sb);
3446 ns = fs_info->pid_ns;
3447 rcu_read_lock();
3448 task = find_task_by_pid_ns(tgid, ns);
3449 if (task)
3450 get_task_struct(task);
3451 rcu_read_unlock();
3452 if (!task)
3453 goto out;
3454
3455 /* Limit procfs to only ptraceable tasks */
3456 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3457 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3458 goto out_put_task;
3459 }
3460
3461 result = proc_pid_instantiate(dentry, task, NULL);
3462out_put_task:
3463 put_task_struct(task);
3464out:
3465 return result;
3466}
3467
3468/*
3469 * Find the first task with tgid >= tgid
3470 *
3471 */
3472struct tgid_iter {
3473 unsigned int tgid;
3474 struct task_struct *task;
3475};
3476static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3477{
3478 struct pid *pid;
3479
3480 if (iter.task)
3481 put_task_struct(iter.task);
3482 rcu_read_lock();
3483retry:
3484 iter.task = NULL;
3485 pid = find_ge_pid(iter.tgid, ns);
3486 if (pid) {
3487 iter.tgid = pid_nr_ns(pid, ns);
3488 iter.task = pid_task(pid, PIDTYPE_TGID);
3489 if (!iter.task) {
3490 iter.tgid += 1;
3491 goto retry;
3492 }
3493 get_task_struct(iter.task);
3494 }
3495 rcu_read_unlock();
3496 return iter;
3497}
3498
3499#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3500
3501/* for the /proc/ directory itself, after non-process stuff has been done */
3502int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3503{
3504 struct tgid_iter iter;
3505 struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3506 struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3507 loff_t pos = ctx->pos;
3508
3509 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3510 return 0;
3511
3512 if (pos == TGID_OFFSET - 2) {
3513 struct inode *inode = d_inode(fs_info->proc_self);
3514 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3515 return 0;
3516 ctx->pos = pos = pos + 1;
3517 }
3518 if (pos == TGID_OFFSET - 1) {
3519 struct inode *inode = d_inode(fs_info->proc_thread_self);
3520 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3521 return 0;
3522 ctx->pos = pos = pos + 1;
3523 }
3524 iter.tgid = pos - TGID_OFFSET;
3525 iter.task = NULL;
3526 for (iter = next_tgid(ns, iter);
3527 iter.task;
3528 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3529 char name[10 + 1];
3530 unsigned int len;
3531
3532 cond_resched();
3533 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3534 continue;
3535
3536 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3537 ctx->pos = iter.tgid + TGID_OFFSET;
3538 if (!proc_fill_cache(file, ctx, name, len,
3539 proc_pid_instantiate, iter.task, NULL)) {
3540 put_task_struct(iter.task);
3541 return 0;
3542 }
3543 }
3544 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3545 return 0;
3546}
3547
3548/*
3549 * proc_tid_comm_permission is a special permission function exclusively
3550 * used for the node /proc/<pid>/task/<tid>/comm.
3551 * It bypasses generic permission checks in the case where a task of the same
3552 * task group attempts to access the node.
3553 * The rationale behind this is that glibc and bionic access this node for
3554 * cross thread naming (pthread_set/getname_np(!self)). However, if
3555 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3556 * which locks out the cross thread naming implementation.
3557 * This function makes sure that the node is always accessible for members of
3558 * same thread group.
3559 */
3560static int proc_tid_comm_permission(struct user_namespace *mnt_userns,
3561 struct inode *inode, int mask)
3562{
3563 bool is_same_tgroup;
3564 struct task_struct *task;
3565
3566 task = get_proc_task(inode);
3567 if (!task)
3568 return -ESRCH;
3569 is_same_tgroup = same_thread_group(current, task);
3570 put_task_struct(task);
3571
3572 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3573 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3574 * read or written by the members of the corresponding
3575 * thread group.
3576 */
3577 return 0;
3578 }
3579
3580 return generic_permission(&init_user_ns, inode, mask);
3581}
3582
3583static const struct inode_operations proc_tid_comm_inode_operations = {
3584 .permission = proc_tid_comm_permission,
3585};
3586
3587/*
3588 * Tasks
3589 */
3590static const struct pid_entry tid_base_stuff[] = {
3591 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3592 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3593 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3594#ifdef CONFIG_NET
3595 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3596#endif
3597 REG("environ", S_IRUSR, proc_environ_operations),
3598 REG("auxv", S_IRUSR, proc_auxv_operations),
3599 ONE("status", S_IRUGO, proc_pid_status),
3600 ONE("personality", S_IRUSR, proc_pid_personality),
3601 ONE("limits", S_IRUGO, proc_pid_limits),
3602#ifdef CONFIG_SCHED_DEBUG
3603 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3604#endif
3605 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3606 &proc_tid_comm_inode_operations,
3607 &proc_pid_set_comm_operations, {}),
3608#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3609 ONE("syscall", S_IRUSR, proc_pid_syscall),
3610#endif
3611 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3612 ONE("stat", S_IRUGO, proc_tid_stat),
3613 ONE("statm", S_IRUGO, proc_pid_statm),
3614 REG("maps", S_IRUGO, proc_pid_maps_operations),
3615#ifdef CONFIG_PROC_CHILDREN
3616 REG("children", S_IRUGO, proc_tid_children_operations),
3617#endif
3618#ifdef CONFIG_NUMA
3619 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3620#endif
3621 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3622 LNK("cwd", proc_cwd_link),
3623 LNK("root", proc_root_link),
3624 LNK("exe", proc_exe_link),
3625 REG("mounts", S_IRUGO, proc_mounts_operations),
3626 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3627#ifdef CONFIG_PROC_PAGE_MONITOR
3628 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3629 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3630 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3631 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3632#endif
3633#ifdef CONFIG_SECURITY
3634 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3635#endif
3636#ifdef CONFIG_KALLSYMS
3637 ONE("wchan", S_IRUGO, proc_pid_wchan),
3638#endif
3639#ifdef CONFIG_STACKTRACE
3640 ONE("stack", S_IRUSR, proc_pid_stack),
3641#endif
3642#ifdef CONFIG_SCHED_INFO
3643 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3644#endif
3645#ifdef CONFIG_LATENCYTOP
3646 REG("latency", S_IRUGO, proc_lstats_operations),
3647#endif
3648#ifdef CONFIG_PROC_PID_CPUSET
3649 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3650#endif
3651#ifdef CONFIG_CGROUPS
3652 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3653#endif
3654#ifdef CONFIG_PROC_CPU_RESCTRL
3655 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3656#endif
3657 ONE("oom_score", S_IRUGO, proc_oom_score),
3658 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3659 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3660#ifdef CONFIG_AUDIT
3661 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3662 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3663#endif
3664#ifdef CONFIG_FAULT_INJECTION
3665 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3666 REG("fail-nth", 0644, proc_fail_nth_operations),
3667#endif
3668#ifdef CONFIG_TASK_IO_ACCOUNTING
3669 ONE("io", S_IRUSR, proc_tid_io_accounting),
3670#endif
3671#ifdef CONFIG_USER_NS
3672 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3673 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3674 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3675 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3676#endif
3677#ifdef CONFIG_LIVEPATCH
3678 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3679#endif
3680#ifdef CONFIG_PROC_PID_ARCH_STATUS
3681 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3682#endif
3683#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3684 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3685#endif
3686#ifdef CONFIG_KSM
3687 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3688 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3689#endif
3690};
3691
3692static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3693{
3694 return proc_pident_readdir(file, ctx,
3695 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3696}
3697
3698static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3699{
3700 return proc_pident_lookup(dir, dentry,
3701 tid_base_stuff,
3702 tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3703}
3704
3705static const struct file_operations proc_tid_base_operations = {
3706 .read = generic_read_dir,
3707 .iterate_shared = proc_tid_base_readdir,
3708 .llseek = generic_file_llseek,
3709};
3710
3711static const struct inode_operations proc_tid_base_inode_operations = {
3712 .lookup = proc_tid_base_lookup,
3713 .getattr = pid_getattr,
3714 .setattr = proc_setattr,
3715};
3716
3717static struct dentry *proc_task_instantiate(struct dentry *dentry,
3718 struct task_struct *task, const void *ptr)
3719{
3720 struct inode *inode;
3721 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3722 S_IFDIR | S_IRUGO | S_IXUGO);
3723 if (!inode)
3724 return ERR_PTR(-ENOENT);
3725
3726 inode->i_op = &proc_tid_base_inode_operations;
3727 inode->i_fop = &proc_tid_base_operations;
3728 inode->i_flags |= S_IMMUTABLE;
3729
3730 set_nlink(inode, nlink_tid);
3731 pid_update_inode(task, inode);
3732
3733 d_set_d_op(dentry, &pid_dentry_operations);
3734 return d_splice_alias(inode, dentry);
3735}
3736
3737static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3738{
3739 struct task_struct *task;
3740 struct task_struct *leader = get_proc_task(dir);
3741 unsigned tid;
3742 struct proc_fs_info *fs_info;
3743 struct pid_namespace *ns;
3744 struct dentry *result = ERR_PTR(-ENOENT);
3745
3746 if (!leader)
3747 goto out_no_task;
3748
3749 tid = name_to_int(&dentry->d_name);
3750 if (tid == ~0U)
3751 goto out;
3752
3753 fs_info = proc_sb_info(dentry->d_sb);
3754 ns = fs_info->pid_ns;
3755 rcu_read_lock();
3756 task = find_task_by_pid_ns(tid, ns);
3757 if (task)
3758 get_task_struct(task);
3759 rcu_read_unlock();
3760 if (!task)
3761 goto out;
3762 if (!same_thread_group(leader, task))
3763 goto out_drop_task;
3764
3765 result = proc_task_instantiate(dentry, task, NULL);
3766out_drop_task:
3767 put_task_struct(task);
3768out:
3769 put_task_struct(leader);
3770out_no_task:
3771 return result;
3772}
3773
3774/*
3775 * Find the first tid of a thread group to return to user space.
3776 *
3777 * Usually this is just the thread group leader, but if the users
3778 * buffer was too small or there was a seek into the middle of the
3779 * directory we have more work todo.
3780 *
3781 * In the case of a short read we start with find_task_by_pid.
3782 *
3783 * In the case of a seek we start with the leader and walk nr
3784 * threads past it.
3785 */
3786static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3787 struct pid_namespace *ns)
3788{
3789 struct task_struct *pos, *task;
3790 unsigned long nr = f_pos;
3791
3792 if (nr != f_pos) /* 32bit overflow? */
3793 return NULL;
3794
3795 rcu_read_lock();
3796 task = pid_task(pid, PIDTYPE_PID);
3797 if (!task)
3798 goto fail;
3799
3800 /* Attempt to start with the tid of a thread */
3801 if (tid && nr) {
3802 pos = find_task_by_pid_ns(tid, ns);
3803 if (pos && same_thread_group(pos, task))
3804 goto found;
3805 }
3806
3807 /* If nr exceeds the number of threads there is nothing todo */
3808 if (nr >= get_nr_threads(task))
3809 goto fail;
3810
3811 /* If we haven't found our starting place yet start
3812 * with the leader and walk nr threads forward.
3813 */
3814 pos = task = task->group_leader;
3815 do {
3816 if (!nr--)
3817 goto found;
3818 } while_each_thread(task, pos);
3819fail:
3820 pos = NULL;
3821 goto out;
3822found:
3823 get_task_struct(pos);
3824out:
3825 rcu_read_unlock();
3826 return pos;
3827}
3828
3829/*
3830 * Find the next thread in the thread list.
3831 * Return NULL if there is an error or no next thread.
3832 *
3833 * The reference to the input task_struct is released.
3834 */
3835static struct task_struct *next_tid(struct task_struct *start)
3836{
3837 struct task_struct *pos = NULL;
3838 rcu_read_lock();
3839 if (pid_alive(start)) {
3840 pos = next_thread(start);
3841 if (thread_group_leader(pos))
3842 pos = NULL;
3843 else
3844 get_task_struct(pos);
3845 }
3846 rcu_read_unlock();
3847 put_task_struct(start);
3848 return pos;
3849}
3850
3851/* for the /proc/TGID/task/ directories */
3852static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3853{
3854 struct inode *inode = file_inode(file);
3855 struct task_struct *task;
3856 struct pid_namespace *ns;
3857 int tid;
3858
3859 if (proc_inode_is_dead(inode))
3860 return -ENOENT;
3861
3862 if (!dir_emit_dots(file, ctx))
3863 return 0;
3864
3865 /* f_version caches the tgid value that the last readdir call couldn't
3866 * return. lseek aka telldir automagically resets f_version to 0.
3867 */
3868 ns = proc_pid_ns(inode->i_sb);
3869 tid = (int)file->f_version;
3870 file->f_version = 0;
3871 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3872 task;
3873 task = next_tid(task), ctx->pos++) {
3874 char name[10 + 1];
3875 unsigned int len;
3876
3877 tid = task_pid_nr_ns(task, ns);
3878 if (!tid)
3879 continue; /* The task has just exited. */
3880 len = snprintf(name, sizeof(name), "%u", tid);
3881 if (!proc_fill_cache(file, ctx, name, len,
3882 proc_task_instantiate, task, NULL)) {
3883 /* returning this tgid failed, save it as the first
3884 * pid for the next readir call */
3885 file->f_version = (u64)tid;
3886 put_task_struct(task);
3887 break;
3888 }
3889 }
3890
3891 return 0;
3892}
3893
3894static int proc_task_getattr(struct user_namespace *mnt_userns,
3895 const struct path *path, struct kstat *stat,
3896 u32 request_mask, unsigned int query_flags)
3897{
3898 struct inode *inode = d_inode(path->dentry);
3899 struct task_struct *p = get_proc_task(inode);
3900 generic_fillattr(&init_user_ns, inode, stat);
3901
3902 if (p) {
3903 stat->nlink += get_nr_threads(p);
3904 put_task_struct(p);
3905 }
3906
3907 return 0;
3908}
3909
3910static const struct inode_operations proc_task_inode_operations = {
3911 .lookup = proc_task_lookup,
3912 .getattr = proc_task_getattr,
3913 .setattr = proc_setattr,
3914 .permission = proc_pid_permission,
3915};
3916
3917static const struct file_operations proc_task_operations = {
3918 .read = generic_read_dir,
3919 .iterate_shared = proc_task_readdir,
3920 .llseek = generic_file_llseek,
3921};
3922
3923void __init set_proc_pid_nlink(void)
3924{
3925 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3926 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3927}
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}