projects/linux·blob assoc-array-fixes · refs

3990 lines · 95.9 KiB

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 <[email protected]>
21 * Edjard Mota <[email protected]>
22 * Ilias Biris <[email protected]>
23 * Mauricio Lin <[email protected]>
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 <[email protected]>:
45 * Smaps information related to shared, private, clean and dirty pages.
46 *
47 * Paul Mundt <[email protected]>:
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/generic-radix-tree.h>
62#include <linux/string.h>
63#include <linux/seq_file.h>
64#include <linux/namei.h>
65#include <linux/mnt_namespace.h>
66#include <linux/mm.h>
67#include <linux/swap.h>
68#include <linux/rcupdate.h>
69#include <linux/kallsyms.h>
70#include <linux/stacktrace.h>
71#include <linux/resource.h>
72#include <linux/module.h>
73#include <linux/mount.h>
74#include <linux/security.h>
75#include <linux/ptrace.h>
76#include <linux/printk.h>
77#include <linux/cache.h>
78#include <linux/cgroup.h>
79#include <linux/cpuset.h>
80#include <linux/audit.h>
81#include <linux/poll.h>
82#include <linux/nsproxy.h>
83#include <linux/oom.h>
84#include <linux/elf.h>
85#include <linux/pid_namespace.h>
86#include <linux/user_namespace.h>
87#include <linux/fs_parser.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/exec_state.h>
95#include <linux/sched/stat.h>
96#include <linux/posix-timers.h>
97#include <linux/time_namespace.h>
98#include <linux/resctrl.h>
99#include <linux/cn_proc.h>
100#include <linux/ksm.h>
101#include <uapi/linux/lsm.h>
102#include <trace/events/oom.h>
103#include "internal.h"
104#include "fd.h"
105
106#include "../../lib/kstrtox.h"
107
108/* NOTE:
109 * Implementing inode permission operations in /proc is almost
110 * certainly an error. Permission checks need to happen during
111 * each system call not at open time. The reason is that most of
112 * what we wish to check for permissions in /proc varies at runtime.
113 *
114 * The classic example of a problem is opening file descriptors
115 * in /proc for a task before it execs a suid executable.
116 */
117
118static u8 nlink_tid __ro_after_init;
119static u8 nlink_tgid __ro_after_init;
120
121enum proc_mem_force {
122 PROC_MEM_FORCE_ALWAYS,
123 PROC_MEM_FORCE_PTRACE,
124 PROC_MEM_FORCE_NEVER
125};
126
127static enum proc_mem_force proc_mem_force_override __ro_after_init =
128 IS_ENABLED(CONFIG_PROC_MEM_NO_FORCE) ? PROC_MEM_FORCE_NEVER :
129 IS_ENABLED(CONFIG_PROC_MEM_FORCE_PTRACE) ? PROC_MEM_FORCE_PTRACE :
130 PROC_MEM_FORCE_ALWAYS;
131
132static const struct constant_table proc_mem_force_table[] __initconst = {
133 { "always", PROC_MEM_FORCE_ALWAYS },
134 { "ptrace", PROC_MEM_FORCE_PTRACE },
135 { "never", PROC_MEM_FORCE_NEVER },
136 { }
137};
138
139static int __init early_proc_mem_force_override(char *buf)
140{
141 if (!buf)
142 return -EINVAL;
143
144 /*
145 * lookup_constant() defaults to proc_mem_force_override to preseve
146 * the initial Kconfig choice in case an invalid param gets passed.
147 */
148 proc_mem_force_override = lookup_constant(proc_mem_force_table,
149 buf, proc_mem_force_override);
150
151 return 0;
152}
153early_param("proc_mem.force_override", early_proc_mem_force_override);
154
155struct pid_entry {
156 const char *name;
157 unsigned int len;
158 umode_t mode;
159 const struct inode_operations *iop;
160 const struct file_operations *fop;
161 union proc_op op;
162};
163
164#define NOD(NAME, MODE, IOP, FOP, OP) { \
165 .name = (NAME), \
166 .len = sizeof(NAME) - 1, \
167 .mode = MODE, \
168 .iop = IOP, \
169 .fop = FOP, \
170 .op = OP, \
171}
172
173#define DIR(NAME, MODE, iops, fops) \
174 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
175#define LNK(NAME, get_link) \
176 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
177 &proc_pid_link_inode_operations, NULL, \
178 { .proc_get_link = get_link } )
179#define REG(NAME, MODE, fops) \
180 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
181#define ONE(NAME, MODE, show) \
182 NOD(NAME, (S_IFREG|(MODE)), \
183 NULL, &proc_single_file_operations, \
184 { .proc_show = show } )
185#define ATTR(LSMID, NAME, MODE) \
186 NOD(NAME, (S_IFREG|(MODE)), \
187 NULL, &proc_pid_attr_operations, \
188 { .lsmid = LSMID })
189
190/*
191 * Count the number of hardlinks for the pid_entry table, excluding the .
192 * and .. links.
193 */
194static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
195 unsigned int n)
196{
197 unsigned int i;
198 unsigned int count;
199
200 count = 2;
201 for (i = 0; i < n; ++i) {
202 if (S_ISDIR(entries[i].mode))
203 ++count;
204 }
205
206 return count;
207}
208
209static int get_task_root(struct task_struct *task, struct path *root)
210{
211 int result = -ENOENT;
212
213 task_lock(task);
214 if (task->real_fs) {
215 get_fs_root(task->real_fs, root);
216 result = 0;
217 }
218 task_unlock(task);
219 return result;
220}
221
222static int proc_cwd_link(struct dentry *dentry, struct path *path,
223 struct task_struct *task)
224{
225 int result = -ENOENT;
226
227 task_lock(task);
228 if (task->real_fs) {
229 get_fs_pwd(task->real_fs, path);
230 result = 0;
231 }
232 task_unlock(task);
233 return result;
234}
235
236static int proc_root_link(struct dentry *dentry, struct path *path,
237 struct task_struct *task)
238{
239 return get_task_root(task, path);
240}
241
242/*
243 * If the user used setproctitle(), we just get the string from
244 * user space at arg_start, and limit it to a maximum of one page.
245 */
246static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
247 size_t count, unsigned long pos,
248 unsigned long arg_start)
249{
250 char *page;
251 int ret, got;
252
253 if (pos >= PAGE_SIZE)
254 return 0;
255
256 page = kmalloc(PAGE_SIZE, GFP_KERNEL);
257 if (!page)
258 return -ENOMEM;
259
260 ret = 0;
261 got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
262 if (got > 0) {
263 int len = strnlen(page, got);
264
265 /* Include the NUL character if it was found */
266 if (len < got)
267 len++;
268
269 if (len > pos) {
270 len -= pos;
271 if (len > count)
272 len = count;
273 len -= copy_to_user(buf, page+pos, len);
274 if (!len)
275 len = -EFAULT;
276 ret = len;
277 }
278 }
279 kfree(page);
280 return ret;
281}
282
283static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
284 size_t count, loff_t *ppos)
285{
286 unsigned long arg_start, arg_end, env_start, env_end;
287 unsigned long pos, len;
288 char *page, c;
289
290 /* Check if process spawned far enough to have cmdline. */
291 if (!mm->env_end)
292 return 0;
293
294 spin_lock(&mm->arg_lock);
295 arg_start = mm->arg_start;
296 arg_end = mm->arg_end;
297 env_start = mm->env_start;
298 env_end = mm->env_end;
299 spin_unlock(&mm->arg_lock);
300
301 if (arg_start >= arg_end)
302 return 0;
303
304 /*
305 * We allow setproctitle() to overwrite the argument
306 * strings, and overflow past the original end. But
307 * only when it overflows into the environment area.
308 */
309 if (env_start != arg_end || env_end < env_start)
310 env_start = env_end = arg_end;
311 len = env_end - arg_start;
312
313 /* We're not going to care if "*ppos" has high bits set */
314 pos = *ppos;
315 if (pos >= len)
316 return 0;
317 if (count > len - pos)
318 count = len - pos;
319 if (!count)
320 return 0;
321
322 /*
323 * Magical special case: if the argv[] end byte is not
324 * zero, the user has overwritten it with setproctitle(3).
325 *
326 * Possible future enhancement: do this only once when
327 * pos is 0, and set a flag in the 'struct file'.
328 */
329 if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
330 return get_mm_proctitle(mm, buf, count, pos, arg_start);
331
332 /*
333 * For the non-setproctitle() case we limit things strictly
334 * to the [arg_start, arg_end[ range.
335 */
336 pos += arg_start;
337 if (pos < arg_start || pos >= arg_end)
338 return 0;
339 if (count > arg_end - pos)
340 count = arg_end - pos;
341
342 page = kmalloc(PAGE_SIZE, GFP_KERNEL);
343 if (!page)
344 return -ENOMEM;
345
346 len = 0;
347 while (count) {
348 int got;
349 size_t size = min_t(size_t, PAGE_SIZE, count);
350
351 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
352 if (got <= 0)
353 break;
354 got -= copy_to_user(buf, page, got);
355 if (unlikely(!got)) {
356 if (!len)
357 len = -EFAULT;
358 break;
359 }
360 pos += got;
361 buf += got;
362 len += got;
363 count -= got;
364 }
365
366 kfree(page);
367 return len;
368}
369
370static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
371 size_t count, loff_t *pos)
372{
373 struct mm_struct *mm;
374 ssize_t ret;
375
376 mm = get_task_mm(tsk);
377 if (!mm)
378 return 0;
379
380 ret = get_mm_cmdline(mm, buf, count, pos);
381 mmput(mm);
382 return ret;
383}
384
385static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
386 size_t count, loff_t *pos)
387{
388 struct task_struct *tsk;
389 ssize_t ret;
390
391 BUG_ON(*pos < 0);
392
393 tsk = get_proc_task(file_inode(file));
394 if (!tsk)
395 return -ESRCH;
396 ret = get_task_cmdline(tsk, buf, count, pos);
397 put_task_struct(tsk);
398 if (ret > 0)
399 *pos += ret;
400 return ret;
401}
402
403static const struct file_operations proc_pid_cmdline_ops = {
404 .read = proc_pid_cmdline_read,
405 .llseek = generic_file_llseek,
406};
407
408#ifdef CONFIG_KALLSYMS
409/*
410 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
411 * Returns the resolved symbol to user space.
412 */
413static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
414 struct pid *pid, struct task_struct *task)
415{
416 unsigned long wchan;
417 char symname[KSYM_NAME_LEN];
418 int err;
419
420 err = down_read_killable(&task->signal->exec_update_lock);
421 if (err)
422 return err;
423 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
424 goto print0;
425
426 wchan = get_wchan(task);
427 if (wchan && !lookup_symbol_name(wchan, symname)) {
428 seq_puts(m, symname);
429 up_read(&task->signal->exec_update_lock);
430 return 0;
431 }
432
433print0:
434 seq_putc(m, '0');
435 up_read(&task->signal->exec_update_lock);
436 return 0;
437}
438#endif /* CONFIG_KALLSYMS */
439
440static int lock_trace(struct task_struct *task)
441{
442 int err = down_read_killable(&task->signal->exec_update_lock);
443 if (err)
444 return err;
445 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
446 up_read(&task->signal->exec_update_lock);
447 return -EPERM;
448 }
449 return 0;
450}
451
452static void unlock_trace(struct task_struct *task)
453{
454 up_read(&task->signal->exec_update_lock);
455}
456
457#ifdef CONFIG_STACKTRACE
458
459#define MAX_STACK_TRACE_DEPTH 64
460
461static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
462 struct pid *pid, struct task_struct *task)
463{
464 unsigned long *entries;
465 int err;
466
467 /*
468 * The ability to racily run the kernel stack unwinder on a running task
469 * and then observe the unwinder output is scary; while it is useful for
470 * debugging kernel issues, it can also allow an attacker to leak kernel
471 * stack contents.
472 * Doing this in a manner that is at least safe from races would require
473 * some work to ensure that the remote task can not be scheduled; and
474 * even then, this would still expose the unwinder as local attack
475 * surface.
476 * Therefore, this interface is restricted to root.
477 */
478 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
479 return -EACCES;
480
481 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
482 GFP_KERNEL);
483 if (!entries)
484 return -ENOMEM;
485
486 err = lock_trace(task);
487 if (!err) {
488 unsigned int i, nr_entries;
489
490 nr_entries = stack_trace_save_tsk(task, entries,
491 MAX_STACK_TRACE_DEPTH, 0);
492
493 for (i = 0; i < nr_entries; i++) {
494 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
495 }
496
497 unlock_trace(task);
498 }
499 kfree(entries);
500
501 return err;
502}
503#endif
504
505#ifdef CONFIG_SCHED_INFO
506/*
507 * Provides /proc/PID/schedstat
508 */
509static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
510 struct pid *pid, struct task_struct *task)
511{
512 if (unlikely(!sched_info_on()))
513 seq_puts(m, "0 0 0\n");
514 else
515 seq_printf(m, "%llu %llu %lu\n",
516 (unsigned long long)task->se.sum_exec_runtime,
517 (unsigned long long)task->sched_info.run_delay,
518 task->sched_info.pcount);
519
520 return 0;
521}
522#endif
523
524#ifdef CONFIG_LATENCYTOP
525static int lstats_show_proc(struct seq_file *m, void *v)
526{
527 int i;
528 struct inode *inode = m->private;
529 struct task_struct *task = get_proc_task(inode);
530
531 if (!task)
532 return -ESRCH;
533 seq_puts(m, "Latency Top version : v0.1\n");
534 for (i = 0; i < LT_SAVECOUNT; i++) {
535 struct latency_record *lr = &task->latency_record[i];
536 if (lr->backtrace[0]) {
537 int q;
538 seq_printf(m, "%i %li %li",
539 lr->count, lr->time, lr->max);
540 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
541 unsigned long bt = lr->backtrace[q];
542
543 if (!bt)
544 break;
545 seq_printf(m, " %ps", (void *)bt);
546 }
547 seq_putc(m, '\n');
548 }
549
550 }
551 put_task_struct(task);
552 return 0;
553}
554
555static int lstats_open(struct inode *inode, struct file *file)
556{
557 return single_open(file, lstats_show_proc, inode);
558}
559
560static ssize_t lstats_write(struct file *file, const char __user *buf,
561 size_t count, loff_t *offs)
562{
563 struct task_struct *task = get_proc_task(file_inode(file));
564
565 if (!task)
566 return -ESRCH;
567 clear_tsk_latency_tracing(task);
568 put_task_struct(task);
569
570 return count;
571}
572
573static const struct file_operations proc_lstats_operations = {
574 .open = lstats_open,
575 .read = seq_read,
576 .write = lstats_write,
577 .llseek = seq_lseek,
578 .release = single_release,
579};
580
581#endif
582
583static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
584 struct pid *pid, struct task_struct *task)
585{
586 unsigned long totalpages = totalram_pages() + total_swap_pages;
587 unsigned long points = 0;
588 long badness;
589
590 badness = oom_badness(task, totalpages);
591 /*
592 * Special case OOM_SCORE_ADJ_MIN for all others scale the
593 * badness value into [0, 2000] range which we have been
594 * exporting for a long time so userspace might depend on it.
595 */
596 if (badness != LONG_MIN)
597 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
598
599 seq_printf(m, "%lu\n", points);
600
601 return 0;
602}
603
604struct limit_names {
605 const char *name;
606 const char *unit;
607};
608
609static const struct limit_names lnames[RLIM_NLIMITS] = {
610 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
611 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
612 [RLIMIT_DATA] = {"Max data size", "bytes"},
613 [RLIMIT_STACK] = {"Max stack size", "bytes"},
614 [RLIMIT_CORE] = {"Max core file size", "bytes"},
615 [RLIMIT_RSS] = {"Max resident set", "bytes"},
616 [RLIMIT_NPROC] = {"Max processes", "processes"},
617 [RLIMIT_NOFILE] = {"Max open files", "files"},
618 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
619 [RLIMIT_AS] = {"Max address space", "bytes"},
620 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
621 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
622 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
623 [RLIMIT_NICE] = {"Max nice priority", NULL},
624 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
625 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
626};
627
628/* Display limits for a process */
629static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
630 struct pid *pid, struct task_struct *task)
631{
632 unsigned int i;
633 unsigned long flags;
634
635 struct rlimit rlim[RLIM_NLIMITS];
636
637 if (!lock_task_sighand(task, &flags))
638 return 0;
639 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
640 unlock_task_sighand(task, &flags);
641
642 /*
643 * print the file header
644 */
645 seq_puts(m, "Limit "
646 "Soft Limit "
647 "Hard Limit "
648 "Units \n");
649
650 for (i = 0; i < RLIM_NLIMITS; i++) {
651 if (rlim[i].rlim_cur == RLIM_INFINITY)
652 seq_printf(m, "%-25s %-20s ",
653 lnames[i].name, "unlimited");
654 else
655 seq_printf(m, "%-25s %-20lu ",
656 lnames[i].name, rlim[i].rlim_cur);
657
658 if (rlim[i].rlim_max == RLIM_INFINITY)
659 seq_printf(m, "%-20s ", "unlimited");
660 else
661 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
662
663 if (lnames[i].unit)
664 seq_printf(m, "%-10s\n", lnames[i].unit);
665 else
666 seq_putc(m, '\n');
667 }
668
669 return 0;
670}
671
672#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
673static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
674 struct pid *pid, struct task_struct *task)
675{
676 struct syscall_info info;
677 u64 *args = &info.data.args[0];
678 int res;
679
680 res = lock_trace(task);
681 if (res)
682 return res;
683
684 if (task_current_syscall(task, &info))
685 seq_puts(m, "running\n");
686 else if (info.data.nr < 0)
687 seq_printf(m, "%d 0x%llx 0x%llx\n",
688 info.data.nr, info.sp, info.data.instruction_pointer);
689 else
690 seq_printf(m,
691 "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
692 info.data.nr,
693 args[0], args[1], args[2], args[3], args[4], args[5],
694 info.sp, info.data.instruction_pointer);
695 unlock_trace(task);
696
697 return 0;
698}
699#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
700
701/************************************************************************/
702/* Here the fs part begins */
703/************************************************************************/
704
705int proc_nochmod_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
706 struct iattr *attr)
707{
708 int error;
709 struct inode *inode = d_inode(dentry);
710
711 if (attr->ia_valid & ATTR_MODE)
712 return -EPERM;
713
714 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
715 if (error)
716 return error;
717
718 setattr_copy(&nop_mnt_idmap, inode, attr);
719 return 0;
720}
721
722/*
723 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
724 * or euid/egid (for hide_pid_min=2)?
725 */
726static bool has_pid_permissions(struct proc_fs_info *fs_info,
727 struct task_struct *task,
728 enum proc_hidepid hide_pid_min)
729{
730 /*
731 * If 'hidpid' mount option is set force a ptrace check,
732 * we indicate that we are using a filesystem syscall
733 * by passing PTRACE_MODE_READ_FSCREDS
734 */
735 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
736 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
737
738 if (fs_info->hide_pid < hide_pid_min)
739 return true;
740 if (in_group_p(fs_info->pid_gid))
741 return true;
742 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
743}
744
745
746static int proc_pid_permission(struct mnt_idmap *idmap,
747 struct inode *inode, int mask)
748{
749 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
750 struct task_struct *task;
751 bool has_perms;
752
753 task = get_proc_task(inode);
754 if (!task)
755 return -ESRCH;
756 has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
757 put_task_struct(task);
758
759 if (!has_perms) {
760 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
761 /*
762 * Let's make getdents(), stat(), and open()
763 * consistent with each other. If a process
764 * may not stat() a file, it shouldn't be seen
765 * in procfs at all.
766 */
767 return -ENOENT;
768 }
769
770 return -EPERM;
771 }
772 return generic_permission(&nop_mnt_idmap, inode, mask);
773}
774
775
776
777static const struct inode_operations proc_def_inode_operations = {
778 .setattr = proc_nochmod_setattr,
779};
780
781static int proc_single_show(struct seq_file *m, void *v)
782{
783 struct inode *inode = m->private;
784 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
785 struct pid *pid = proc_pid(inode);
786 struct task_struct *task;
787 int ret;
788
789 task = get_pid_task(pid, PIDTYPE_PID);
790 if (!task)
791 return -ESRCH;
792
793 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
794
795 put_task_struct(task);
796 return ret;
797}
798
799static int proc_single_open(struct inode *inode, struct file *filp)
800{
801 return single_open(filp, proc_single_show, inode);
802}
803
804static const struct file_operations proc_single_file_operations = {
805 .open = proc_single_open,
806 .read = seq_read,
807 .llseek = seq_lseek,
808 .release = single_release,
809};
810
811/*
812 * proc_mem_open() can return errno, NULL or mm_struct*.
813 *
814 * - Returns NULL if the task has no mm (PF_KTHREAD or PF_EXITING)
815 * - Returns mm_struct* on success
816 * - Returns error code on failure
817 */
818struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
819{
820 struct task_struct *task = get_proc_task(inode);
821 struct mm_struct *mm;
822
823 if (!task)
824 return ERR_PTR(-ESRCH);
825
826 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
827 put_task_struct(task);
828
829 if (IS_ERR(mm))
830 return mm == ERR_PTR(-ESRCH) ? NULL : mm;
831
832 /* ensure this mm_struct can't be freed */
833 mmgrab(mm);
834 /* but do not pin its memory */
835 mmput(mm);
836
837 return mm;
838}
839
840static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
841{
842 struct mm_struct *mm = proc_mem_open(inode, mode);
843
844 if (IS_ERR_OR_NULL(mm))
845 return mm ? PTR_ERR(mm) : -ESRCH;
846
847 file->private_data = mm;
848 return 0;
849}
850
851static int mem_open(struct inode *inode, struct file *file)
852{
853 if (WARN_ON_ONCE(!(file->f_op->fop_flags & FOP_UNSIGNED_OFFSET)))
854 return -EINVAL;
855 return __mem_open(inode, file, PTRACE_MODE_ATTACH);
856}
857
858static bool proc_mem_foll_force(struct file *file, struct mm_struct *mm)
859{
860 struct task_struct *task;
861 bool ptrace_active = false;
862
863 switch (proc_mem_force_override) {
864 case PROC_MEM_FORCE_NEVER:
865 return false;
866 case PROC_MEM_FORCE_PTRACE:
867 task = get_proc_task(file_inode(file));
868 if (task) {
869 ptrace_active = READ_ONCE(task->ptrace) &&
870 READ_ONCE(task->mm) == mm &&
871 READ_ONCE(task->parent) == current;
872 put_task_struct(task);
873 }
874 return ptrace_active;
875 default:
876 return true;
877 }
878}
879
880static ssize_t mem_rw(struct file *file, char __user *buf,
881 size_t count, loff_t *ppos, int write)
882{
883 struct mm_struct *mm = file->private_data;
884 unsigned long addr = *ppos;
885 ssize_t copied;
886 char *page;
887 unsigned int flags;
888
889 if (!mm)
890 return 0;
891
892 page = kmalloc(PAGE_SIZE, GFP_KERNEL);
893 if (!page)
894 return -ENOMEM;
895
896 copied = 0;
897 if (!mmget_not_zero(mm))
898 goto free;
899
900 flags = write ? FOLL_WRITE : 0;
901 if (proc_mem_foll_force(file, mm))
902 flags |= FOLL_FORCE;
903
904 while (count > 0) {
905 size_t this_len = min_t(size_t, count, PAGE_SIZE);
906
907 if (write && copy_from_user(page, buf, this_len)) {
908 copied = -EFAULT;
909 break;
910 }
911
912 this_len = access_remote_vm(mm, addr, page, this_len, flags);
913 if (!this_len) {
914 if (!copied)
915 copied = -EIO;
916 break;
917 }
918
919 if (!write && copy_to_user(buf, page, this_len)) {
920 copied = -EFAULT;
921 break;
922 }
923
924 buf += this_len;
925 addr += this_len;
926 copied += this_len;
927 count -= this_len;
928 }
929 *ppos = addr;
930
931 mmput(mm);
932free:
933 kfree(page);
934 return copied;
935}
936
937static ssize_t mem_read(struct file *file, char __user *buf,
938 size_t count, loff_t *ppos)
939{
940 return mem_rw(file, buf, count, ppos, 0);
941}
942
943static ssize_t mem_write(struct file *file, const char __user *buf,
944 size_t count, loff_t *ppos)
945{
946 return mem_rw(file, (char __user*)buf, count, ppos, 1);
947}
948
949loff_t mem_lseek(struct file *file, loff_t offset, int orig)
950{
951 switch (orig) {
952 case 0:
953 file->f_pos = offset;
954 break;
955 case 1:
956 file->f_pos += offset;
957 break;
958 default:
959 return -EINVAL;
960 }
961 force_successful_syscall_return();
962 return file->f_pos;
963}
964
965static int mem_release(struct inode *inode, struct file *file)
966{
967 struct mm_struct *mm = file->private_data;
968 if (mm)
969 mmdrop(mm);
970 return 0;
971}
972
973static const struct file_operations proc_mem_operations = {
974 .llseek = mem_lseek,
975 .read = mem_read,
976 .write = mem_write,
977 .open = mem_open,
978 .release = mem_release,
979 .fop_flags = FOP_UNSIGNED_OFFSET,
980};
981
982static int environ_open(struct inode *inode, struct file *file)
983{
984 return __mem_open(inode, file, PTRACE_MODE_READ);
985}
986
987static ssize_t environ_read(struct file *file, char __user *buf,
988 size_t count, loff_t *ppos)
989{
990 char *page;
991 unsigned long src = *ppos;
992 int ret = 0;
993 struct mm_struct *mm = file->private_data;
994 unsigned long env_start, env_end;
995
996 /* Ensure the process spawned far enough to have an environment. */
997 if (!mm || !mm->env_end)
998 return 0;
999
1000 page = kmalloc(PAGE_SIZE, GFP_KERNEL);
1001 if (!page)
1002 return -ENOMEM;
1003
1004 ret = 0;
1005 if (!mmget_not_zero(mm))
1006 goto free;
1007
1008 spin_lock(&mm->arg_lock);
1009 env_start = mm->env_start;
1010 env_end = mm->env_end;
1011 spin_unlock(&mm->arg_lock);
1012
1013 while (count > 0) {
1014 size_t this_len, max_len;
1015 int retval;
1016
1017 if (src >= (env_end - env_start))
1018 break;
1019
1020 this_len = env_end - (env_start + src);
1021
1022 max_len = min_t(size_t, PAGE_SIZE, count);
1023 this_len = min(max_len, this_len);
1024
1025 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
1026
1027 if (retval <= 0) {
1028 ret = retval;
1029 break;
1030 }
1031
1032 if (copy_to_user(buf, page, retval)) {
1033 ret = -EFAULT;
1034 break;
1035 }
1036
1037 ret += retval;
1038 src += retval;
1039 buf += retval;
1040 count -= retval;
1041 }
1042 *ppos = src;
1043 mmput(mm);
1044
1045free:
1046 kfree(page);
1047 return ret;
1048}
1049
1050static const struct file_operations proc_environ_operations = {
1051 .open = environ_open,
1052 .read = environ_read,
1053 .llseek = generic_file_llseek,
1054 .release = mem_release,
1055};
1056
1057static int auxv_open(struct inode *inode, struct file *file)
1058{
1059 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1060}
1061
1062static ssize_t auxv_read(struct file *file, char __user *buf,
1063 size_t count, loff_t *ppos)
1064{
1065 struct mm_struct *mm = file->private_data;
1066 unsigned int nwords = 0;
1067
1068 if (!mm)
1069 return 0;
1070 do {
1071 nwords += 2;
1072 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1073 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1074 nwords * sizeof(mm->saved_auxv[0]));
1075}
1076
1077static const struct file_operations proc_auxv_operations = {
1078 .open = auxv_open,
1079 .read = auxv_read,
1080 .llseek = generic_file_llseek,
1081 .release = mem_release,
1082};
1083
1084static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1085 loff_t *ppos)
1086{
1087 struct task_struct *task = get_proc_task(file_inode(file));
1088 char buffer[PROC_NUMBUF];
1089 int oom_adj = OOM_ADJUST_MIN;
1090 size_t len;
1091
1092 if (!task)
1093 return -ESRCH;
1094 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1095 oom_adj = OOM_ADJUST_MAX;
1096 else
1097 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1098 OOM_SCORE_ADJ_MAX;
1099 put_task_struct(task);
1100 if (oom_adj > OOM_ADJUST_MAX)
1101 oom_adj = OOM_ADJUST_MAX;
1102 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1103 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1104}
1105
1106static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1107{
1108 struct mm_struct *mm = NULL;
1109 struct task_struct *task;
1110 int err = 0;
1111
1112 task = get_proc_task(file_inode(file));
1113 if (!task)
1114 return -ESRCH;
1115
1116 mutex_lock(&oom_adj_mutex);
1117 if (legacy) {
1118 if (oom_adj < task->signal->oom_score_adj &&
1119 !capable(CAP_SYS_RESOURCE)) {
1120 err = -EACCES;
1121 goto err_unlock;
1122 }
1123 /*
1124 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1125 * /proc/pid/oom_score_adj instead.
1126 */
1127 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1128 current->comm, task_pid_nr(current), task_pid_nr(task),
1129 task_pid_nr(task));
1130 } else {
1131 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1132 !capable(CAP_SYS_RESOURCE)) {
1133 err = -EACCES;
1134 goto err_unlock;
1135 }
1136 }
1137
1138 /*
1139 * Make sure we will check other processes sharing the mm if this is
1140 * not vfrok which wants its own oom_score_adj.
1141 * pin the mm so it doesn't go away and get reused after task_unlock
1142 */
1143 if (!task->vfork_done) {
1144 struct task_struct *p = find_lock_task_mm(task);
1145
1146 if (p) {
1147 if (mm_flags_test(MMF_MULTIPROCESS, p->mm)) {
1148 mm = p->mm;
1149 mmgrab(mm);
1150 }
1151 task_unlock(p);
1152 }
1153 }
1154
1155 task->signal->oom_score_adj = oom_adj;
1156 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1157 task->signal->oom_score_adj_min = (short)oom_adj;
1158 trace_oom_score_adj_update(task);
1159
1160 if (mm) {
1161 struct task_struct *p;
1162
1163 rcu_read_lock();
1164 for_each_process(p) {
1165 if (same_thread_group(task, p))
1166 continue;
1167
1168 /* do not touch kernel threads or the global init */
1169 if (p->flags & PF_KTHREAD || is_global_init(p))
1170 continue;
1171
1172 task_lock(p);
1173 if (!p->vfork_done && process_shares_mm(p, mm)) {
1174 p->signal->oom_score_adj = oom_adj;
1175 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1176 p->signal->oom_score_adj_min = (short)oom_adj;
1177 }
1178 task_unlock(p);
1179 }
1180 rcu_read_unlock();
1181 mmdrop(mm);
1182 }
1183err_unlock:
1184 mutex_unlock(&oom_adj_mutex);
1185 put_task_struct(task);
1186 return err;
1187}
1188
1189/*
1190 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1191 * kernels. The effective policy is defined by oom_score_adj, which has a
1192 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1193 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1194 * Processes that become oom disabled via oom_adj will still be oom disabled
1195 * with this implementation.
1196 *
1197 * oom_adj cannot be removed since existing userspace binaries use it.
1198 */
1199static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1200 size_t count, loff_t *ppos)
1201{
1202 char buffer[PROC_NUMBUF] = {};
1203 int oom_adj;
1204 int err;
1205
1206 if (count > sizeof(buffer) - 1)
1207 count = sizeof(buffer) - 1;
1208 if (copy_from_user(buffer, buf, count)) {
1209 err = -EFAULT;
1210 goto out;
1211 }
1212
1213 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1214 if (err)
1215 goto out;
1216 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1217 oom_adj != OOM_DISABLE) {
1218 err = -EINVAL;
1219 goto out;
1220 }
1221
1222 /*
1223 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1224 * value is always attainable.
1225 */
1226 if (oom_adj == OOM_ADJUST_MAX)
1227 oom_adj = OOM_SCORE_ADJ_MAX;
1228 else
1229 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1230
1231 err = __set_oom_adj(file, oom_adj, true);
1232out:
1233 return err < 0 ? err : count;
1234}
1235
1236static const struct file_operations proc_oom_adj_operations = {
1237 .read = oom_adj_read,
1238 .write = oom_adj_write,
1239 .llseek = generic_file_llseek,
1240};
1241
1242static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1243 size_t count, loff_t *ppos)
1244{
1245 struct task_struct *task = get_proc_task(file_inode(file));
1246 char buffer[PROC_NUMBUF];
1247 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1248 size_t len;
1249
1250 if (!task)
1251 return -ESRCH;
1252 oom_score_adj = task->signal->oom_score_adj;
1253 put_task_struct(task);
1254 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1255 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1256}
1257
1258static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1259 size_t count, loff_t *ppos)
1260{
1261 char buffer[PROC_NUMBUF] = {};
1262 int oom_score_adj;
1263 int err;
1264
1265 if (count > sizeof(buffer) - 1)
1266 count = sizeof(buffer) - 1;
1267 if (copy_from_user(buffer, buf, count)) {
1268 err = -EFAULT;
1269 goto out;
1270 }
1271
1272 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1273 if (err)
1274 goto out;
1275 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1276 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1277 err = -EINVAL;
1278 goto out;
1279 }
1280
1281 err = __set_oom_adj(file, oom_score_adj, false);
1282out:
1283 return err < 0 ? err : count;
1284}
1285
1286static const struct file_operations proc_oom_score_adj_operations = {
1287 .read = oom_score_adj_read,
1288 .write = oom_score_adj_write,
1289 .llseek = default_llseek,
1290};
1291
1292#ifdef CONFIG_AUDIT
1293#define TMPBUFLEN 11
1294static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1295 size_t count, loff_t *ppos)
1296{
1297 struct inode * inode = file_inode(file);
1298 struct task_struct *task = get_proc_task(inode);
1299 ssize_t length;
1300 char tmpbuf[TMPBUFLEN];
1301
1302 if (!task)
1303 return -ESRCH;
1304 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1305 from_kuid(file->f_cred->user_ns,
1306 audit_get_loginuid(task)));
1307 put_task_struct(task);
1308 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1309}
1310
1311static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1312 size_t count, loff_t *ppos)
1313{
1314 struct inode * inode = file_inode(file);
1315 uid_t loginuid;
1316 kuid_t kloginuid;
1317 int rv;
1318
1319 /* Don't let kthreads write their own loginuid */
1320 if (current->flags & PF_KTHREAD)
1321 return -EPERM;
1322
1323 rcu_read_lock();
1324 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1325 rcu_read_unlock();
1326 return -EPERM;
1327 }
1328 rcu_read_unlock();
1329
1330 if (*ppos != 0) {
1331 /* No partial writes. */
1332 return -EINVAL;
1333 }
1334
1335 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1336 if (rv < 0)
1337 return rv;
1338
1339 /* is userspace tring to explicitly UNSET the loginuid? */
1340 if (loginuid == AUDIT_UID_UNSET) {
1341 kloginuid = INVALID_UID;
1342 } else {
1343 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1344 if (!uid_valid(kloginuid))
1345 return -EINVAL;
1346 }
1347
1348 rv = audit_set_loginuid(kloginuid);
1349 if (rv < 0)
1350 return rv;
1351 return count;
1352}
1353
1354static const struct file_operations proc_loginuid_operations = {
1355 .read = proc_loginuid_read,
1356 .write = proc_loginuid_write,
1357 .llseek = generic_file_llseek,
1358};
1359
1360static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1361 size_t count, loff_t *ppos)
1362{
1363 struct inode * inode = file_inode(file);
1364 struct task_struct *task = get_proc_task(inode);
1365 ssize_t length;
1366 char tmpbuf[TMPBUFLEN];
1367
1368 if (!task)
1369 return -ESRCH;
1370 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1371 audit_get_sessionid(task));
1372 put_task_struct(task);
1373 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1374}
1375
1376static const struct file_operations proc_sessionid_operations = {
1377 .read = proc_sessionid_read,
1378 .llseek = generic_file_llseek,
1379};
1380#endif
1381
1382#ifdef CONFIG_FAULT_INJECTION
1383static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1384 size_t count, loff_t *ppos)
1385{
1386 struct task_struct *task = get_proc_task(file_inode(file));
1387 char buffer[PROC_NUMBUF];
1388 size_t len;
1389 int make_it_fail;
1390
1391 if (!task)
1392 return -ESRCH;
1393 make_it_fail = task->make_it_fail;
1394 put_task_struct(task);
1395
1396 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1397
1398 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1399}
1400
1401static ssize_t proc_fault_inject_write(struct file * file,
1402 const char __user * buf, size_t count, loff_t *ppos)
1403{
1404 struct task_struct *task;
1405 char buffer[PROC_NUMBUF] = {};
1406 int make_it_fail;
1407 int rv;
1408
1409 if (!capable(CAP_SYS_RESOURCE))
1410 return -EPERM;
1411
1412 if (count > sizeof(buffer) - 1)
1413 count = sizeof(buffer) - 1;
1414 if (copy_from_user(buffer, buf, count))
1415 return -EFAULT;
1416 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1417 if (rv < 0)
1418 return rv;
1419 if (make_it_fail < 0 || make_it_fail > 1)
1420 return -EINVAL;
1421
1422 task = get_proc_task(file_inode(file));
1423 if (!task)
1424 return -ESRCH;
1425 task->make_it_fail = make_it_fail;
1426 put_task_struct(task);
1427
1428 return count;
1429}
1430
1431static const struct file_operations proc_fault_inject_operations = {
1432 .read = proc_fault_inject_read,
1433 .write = proc_fault_inject_write,
1434 .llseek = generic_file_llseek,
1435};
1436
1437static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1438 size_t count, loff_t *ppos)
1439{
1440 struct task_struct *task;
1441 int err;
1442 unsigned int n;
1443
1444 err = kstrtouint_from_user(buf, count, 0, &n);
1445 if (err)
1446 return err;
1447
1448 task = get_proc_task(file_inode(file));
1449 if (!task)
1450 return -ESRCH;
1451 task->fail_nth = n;
1452 put_task_struct(task);
1453
1454 return count;
1455}
1456
1457static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1458 size_t count, loff_t *ppos)
1459{
1460 struct task_struct *task;
1461 char numbuf[PROC_NUMBUF];
1462 ssize_t len;
1463
1464 task = get_proc_task(file_inode(file));
1465 if (!task)
1466 return -ESRCH;
1467 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1468 put_task_struct(task);
1469 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1470}
1471
1472static const struct file_operations proc_fail_nth_operations = {
1473 .read = proc_fail_nth_read,
1474 .write = proc_fail_nth_write,
1475};
1476#endif
1477
1478
1479/*
1480 * Print out various scheduling related per-task fields:
1481 */
1482static int sched_show(struct seq_file *m, void *v)
1483{
1484 struct inode *inode = m->private;
1485 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1486 struct task_struct *p;
1487
1488 p = get_proc_task(inode);
1489 if (!p)
1490 return -ESRCH;
1491 proc_sched_show_task(p, ns, m);
1492
1493 put_task_struct(p);
1494
1495 return 0;
1496}
1497
1498static ssize_t
1499sched_write(struct file *file, const char __user *buf,
1500 size_t count, loff_t *offset)
1501{
1502 struct inode *inode = file_inode(file);
1503 struct task_struct *p;
1504
1505 p = get_proc_task(inode);
1506 if (!p)
1507 return -ESRCH;
1508 proc_sched_set_task(p);
1509
1510 put_task_struct(p);
1511
1512 return count;
1513}
1514
1515static int sched_open(struct inode *inode, struct file *filp)
1516{
1517 return single_open(filp, sched_show, inode);
1518}
1519
1520static const struct file_operations proc_pid_sched_operations = {
1521 .open = sched_open,
1522 .read = seq_read,
1523 .write = sched_write,
1524 .llseek = seq_lseek,
1525 .release = single_release,
1526};
1527
1528#ifdef CONFIG_SCHED_AUTOGROUP
1529/*
1530 * Print out autogroup related information:
1531 */
1532static int sched_autogroup_show(struct seq_file *m, void *v)
1533{
1534 struct inode *inode = m->private;
1535 struct task_struct *p;
1536
1537 p = get_proc_task(inode);
1538 if (!p)
1539 return -ESRCH;
1540 proc_sched_autogroup_show_task(p, m);
1541
1542 put_task_struct(p);
1543
1544 return 0;
1545}
1546
1547static ssize_t
1548sched_autogroup_write(struct file *file, const char __user *buf,
1549 size_t count, loff_t *offset)
1550{
1551 struct inode *inode = file_inode(file);
1552 struct task_struct *p;
1553 char buffer[PROC_NUMBUF] = {};
1554 int nice;
1555 int err;
1556
1557 if (count > sizeof(buffer) - 1)
1558 count = sizeof(buffer) - 1;
1559 if (copy_from_user(buffer, buf, count))
1560 return -EFAULT;
1561
1562 err = kstrtoint(strstrip(buffer), 0, &nice);
1563 if (err < 0)
1564 return err;
1565
1566 p = get_proc_task(inode);
1567 if (!p)
1568 return -ESRCH;
1569
1570 err = proc_sched_autogroup_set_nice(p, nice);
1571 if (err)
1572 count = err;
1573
1574 put_task_struct(p);
1575
1576 return count;
1577}
1578
1579static int sched_autogroup_open(struct inode *inode, struct file *filp)
1580{
1581 int ret;
1582
1583 ret = single_open(filp, sched_autogroup_show, NULL);
1584 if (!ret) {
1585 struct seq_file *m = filp->private_data;
1586
1587 m->private = inode;
1588 }
1589 return ret;
1590}
1591
1592static const struct file_operations proc_pid_sched_autogroup_operations = {
1593 .open = sched_autogroup_open,
1594 .read = seq_read,
1595 .write = sched_autogroup_write,
1596 .llseek = seq_lseek,
1597 .release = single_release,
1598};
1599
1600#endif /* CONFIG_SCHED_AUTOGROUP */
1601
1602#ifdef CONFIG_TIME_NS
1603static int timens_offsets_show(struct seq_file *m, void *v)
1604{
1605 struct task_struct *p;
1606
1607 p = get_proc_task(file_inode(m->file));
1608 if (!p)
1609 return -ESRCH;
1610 proc_timens_show_offsets(p, m);
1611
1612 put_task_struct(p);
1613
1614 return 0;
1615}
1616
1617static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1618 size_t count, loff_t *ppos)
1619{
1620 struct inode *inode = file_inode(file);
1621 struct proc_timens_offset offsets[2];
1622 char *kbuf = NULL, *pos, *next_line;
1623 struct task_struct *p;
1624 int ret, noffsets;
1625
1626 /* Only allow < page size writes at the beginning of the file */
1627 if ((*ppos != 0) || (count >= PAGE_SIZE))
1628 return -EINVAL;
1629
1630 /* Slurp in the user data */
1631 kbuf = memdup_user_nul(buf, count);
1632 if (IS_ERR(kbuf))
1633 return PTR_ERR(kbuf);
1634
1635 /* Parse the user data */
1636 ret = -EINVAL;
1637 noffsets = 0;
1638 for (pos = kbuf; pos; pos = next_line) {
1639 struct proc_timens_offset *off = &offsets[noffsets];
1640 char clock[10];
1641 int err;
1642
1643 /* Find the end of line and ensure we don't look past it */
1644 next_line = strchr(pos, '\n');
1645 if (next_line) {
1646 *next_line = '\0';
1647 next_line++;
1648 if (*next_line == '\0')
1649 next_line = NULL;
1650 }
1651
1652 err = sscanf(pos, "%9s %lld %lu", clock,
1653 &off->val.tv_sec, &off->val.tv_nsec);
1654 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1655 goto out;
1656
1657 clock[sizeof(clock) - 1] = 0;
1658 if (strcmp(clock, "monotonic") == 0 ||
1659 strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1660 off->clockid = CLOCK_MONOTONIC;
1661 else if (strcmp(clock, "boottime") == 0 ||
1662 strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1663 off->clockid = CLOCK_BOOTTIME;
1664 else
1665 goto out;
1666
1667 noffsets++;
1668 if (noffsets == ARRAY_SIZE(offsets)) {
1669 if (next_line)
1670 count = next_line - kbuf;
1671 break;
1672 }
1673 }
1674
1675 ret = -ESRCH;
1676 p = get_proc_task(inode);
1677 if (!p)
1678 goto out;
1679 ret = proc_timens_set_offset(file, p, offsets, noffsets);
1680 put_task_struct(p);
1681 if (ret)
1682 goto out;
1683
1684 ret = count;
1685out:
1686 kfree(kbuf);
1687 return ret;
1688}
1689
1690static int timens_offsets_open(struct inode *inode, struct file *filp)
1691{
1692 return single_open(filp, timens_offsets_show, inode);
1693}
1694
1695static const struct file_operations proc_timens_offsets_operations = {
1696 .open = timens_offsets_open,
1697 .read = seq_read,
1698 .write = timens_offsets_write,
1699 .llseek = seq_lseek,
1700 .release = single_release,
1701};
1702#endif /* CONFIG_TIME_NS */
1703
1704static ssize_t comm_write(struct file *file, const char __user *buf,
1705 size_t count, loff_t *offset)
1706{
1707 struct inode *inode = file_inode(file);
1708 struct task_struct *p;
1709 char buffer[TASK_COMM_LEN] = {};
1710 const size_t maxlen = sizeof(buffer) - 1;
1711
1712 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1713 return -EFAULT;
1714
1715 p = get_proc_task(inode);
1716 if (!p)
1717 return -ESRCH;
1718
1719 if (same_thread_group(current, p)) {
1720 set_task_comm(p, buffer);
1721 proc_comm_connector(p);
1722 }
1723 else
1724 count = -EINVAL;
1725
1726 put_task_struct(p);
1727
1728 return count;
1729}
1730
1731static int comm_show(struct seq_file *m, void *v)
1732{
1733 struct inode *inode = m->private;
1734 struct task_struct *p;
1735
1736 p = get_proc_task(inode);
1737 if (!p)
1738 return -ESRCH;
1739
1740 proc_task_name(m, p, false);
1741 seq_putc(m, '\n');
1742
1743 put_task_struct(p);
1744
1745 return 0;
1746}
1747
1748static int comm_open(struct inode *inode, struct file *filp)
1749{
1750 return single_open(filp, comm_show, inode);
1751}
1752
1753static const struct file_operations proc_pid_set_comm_operations = {
1754 .open = comm_open,
1755 .read = seq_read,
1756 .write = comm_write,
1757 .llseek = seq_lseek,
1758 .release = single_release,
1759};
1760
1761static int proc_exe_link(struct dentry *dentry, struct path *exe_path,
1762 struct task_struct *task)
1763{
1764 struct file *exe_file;
1765
1766 exe_file = get_task_exe_file(task);
1767 if (exe_file) {
1768 *exe_path = exe_file->f_path;
1769 path_get(&exe_file->f_path);
1770 fput(exe_file);
1771 return 0;
1772 } else
1773 return -ENOENT;
1774}
1775
1776static int call_proc_get_link(struct dentry *dentry, struct inode *inode, struct path *path_out)
1777{
1778 struct task_struct *task;
1779 int ret;
1780
1781 task = get_proc_task(inode);
1782 if (!task)
1783 return -ENOENT;
1784 ret = down_read_killable(&task->signal->exec_update_lock);
1785 if (ret)
1786 goto out_put_task;
1787 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
1788 ret = -EACCES;
1789 goto out;
1790 }
1791 ret = PROC_I(inode)->op.proc_get_link(dentry, path_out, task);
1792
1793out:
1794 up_read(&task->signal->exec_update_lock);
1795out_put_task:
1796 put_task_struct(task);
1797 return ret;
1798}
1799
1800static const char *proc_pid_get_link(struct dentry *dentry,
1801 struct inode *inode,
1802 struct delayed_call *done)
1803{
1804 struct path path;
1805 int error;
1806
1807 if (!dentry)
1808 return ERR_PTR(-ECHILD);
1809 error = call_proc_get_link(dentry, inode, &path);
1810 if (!error)
1811 error = nd_jump_link(&path);
1812 return ERR_PTR(error);
1813}
1814
1815static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1816{
1817 char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1818 char *pathname;
1819 int len;
1820
1821 if (!tmp)
1822 return -ENOMEM;
1823
1824 pathname = d_path(path, tmp, PATH_MAX);
1825 len = PTR_ERR(pathname);
1826 if (IS_ERR(pathname))
1827 goto out;
1828 len = tmp + PATH_MAX - 1 - pathname;
1829
1830 if (len > buflen)
1831 len = buflen;
1832 if (copy_to_user(buffer, pathname, len))
1833 len = -EFAULT;
1834 out:
1835 kfree(tmp);
1836 return len;
1837}
1838
1839static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1840{
1841 int error = -EACCES;
1842 struct inode *inode = d_inode(dentry);
1843 struct path path;
1844
1845 error = call_proc_get_link(dentry, inode, &path);
1846 if (!error) {
1847 error = do_proc_readlink(&path, buffer, buflen);
1848 path_put(&path);
1849 }
1850 return error;
1851}
1852
1853const struct inode_operations proc_pid_link_inode_operations = {
1854 .readlink = proc_pid_readlink,
1855 .get_link = proc_pid_get_link,
1856 .setattr = proc_nochmod_setattr,
1857};
1858
1859
1860/* building an inode */
1861
1862void task_dump_owner(struct task_struct *task, umode_t mode,
1863 kuid_t *ruid, kgid_t *rgid)
1864{
1865 /* Depending on the state of dumpable compute who should own a
1866 * proc file for a task.
1867 */
1868 const struct cred *cred;
1869 kuid_t uid;
1870 kgid_t gid;
1871
1872 if (unlikely(task->flags & PF_KTHREAD)) {
1873 *ruid = GLOBAL_ROOT_UID;
1874 *rgid = GLOBAL_ROOT_GID;
1875 return;
1876 }
1877
1878 /* Default to the tasks effective ownership */
1879 rcu_read_lock();
1880 cred = __task_cred(task);
1881 uid = cred->euid;
1882 gid = cred->egid;
1883
1884 /*
1885 * Before the /proc/pid/status file was created the only way to read
1886 * the effective uid of a /process was to stat /proc/pid. Reading
1887 * /proc/pid/status is slow enough that procps and other packages
1888 * kept stating /proc/pid. To keep the rules in /proc simple I have
1889 * made this apply to all per process world readable and executable
1890 * directories.
1891 */
1892 if (mode != (S_IFDIR | S_IRUGO | S_IXUGO)) {
1893 struct task_exec_state *exec_state;
1894
1895 exec_state = task_exec_state_rcu(task);
1896 if (READ_ONCE(exec_state->dumpable) != TASK_DUMPABLE_OWNER) {
1897 uid = make_kuid(exec_state->user_ns, 0);
1898 if (!uid_valid(uid))
1899 uid = GLOBAL_ROOT_UID;
1900
1901 gid = make_kgid(exec_state->user_ns, 0);
1902 if (!gid_valid(gid))
1903 gid = GLOBAL_ROOT_GID;
1904 }
1905 }
1906 rcu_read_unlock();
1907
1908 *ruid = uid;
1909 *rgid = gid;
1910}
1911
1912void proc_pid_evict_inode(struct proc_inode *ei)
1913{
1914 struct pid *pid = ei->pid;
1915
1916 if (S_ISDIR(ei->vfs_inode.i_mode)) {
1917 spin_lock(&pid->lock);
1918 hlist_del_init_rcu(&ei->sibling_inodes);
1919 spin_unlock(&pid->lock);
1920 }
1921}
1922
1923struct inode *proc_pid_make_inode(struct super_block *sb,
1924 struct task_struct *task, umode_t mode)
1925{
1926 struct inode * inode;
1927 struct proc_inode *ei;
1928 struct pid *pid;
1929
1930 /* We need a new inode */
1931
1932 inode = new_inode(sb);
1933 if (!inode)
1934 goto out;
1935
1936 /* Common stuff */
1937 ei = PROC_I(inode);
1938 inode->i_mode = mode;
1939 inode->i_ino = get_next_ino();
1940 simple_inode_init_ts(inode);
1941 inode->i_op = &proc_def_inode_operations;
1942
1943 /*
1944 * grab the reference to task.
1945 */
1946 pid = get_task_pid(task, PIDTYPE_PID);
1947 if (!pid)
1948 goto out_unlock;
1949
1950 /* Let the pid remember us for quick removal */
1951 ei->pid = pid;
1952
1953 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1954 security_task_to_inode(task, inode);
1955
1956out:
1957 return inode;
1958
1959out_unlock:
1960 iput(inode);
1961 return NULL;
1962}
1963
1964/*
1965 * Generating an inode and adding it into @pid->inodes, so that task will
1966 * invalidate inode's dentry before being released.
1967 *
1968 * This helper is used for creating dir-type entries under '/proc' and
1969 * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1970 * can be released by invalidating '/proc/<tgid>' dentry.
1971 * In theory, dentries under '/proc/<tgid>/task' can also be released by
1972 * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1973 * thread exiting situation: Any one of threads should invalidate its
1974 * '/proc/<tgid>/task/<pid>' dentry before released.
1975 */
1976static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1977 struct task_struct *task, umode_t mode)
1978{
1979 struct inode *inode;
1980 struct proc_inode *ei;
1981 struct pid *pid;
1982
1983 inode = proc_pid_make_inode(sb, task, mode);
1984 if (!inode)
1985 return NULL;
1986
1987 /* Let proc_flush_pid find this directory inode */
1988 ei = PROC_I(inode);
1989 pid = ei->pid;
1990 spin_lock(&pid->lock);
1991 hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
1992 spin_unlock(&pid->lock);
1993
1994 return inode;
1995}
1996
1997int pid_getattr(struct mnt_idmap *idmap, const struct path *path,
1998 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1999{
2000 struct inode *inode = d_inode(path->dentry);
2001 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
2002 struct task_struct *task;
2003
2004 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
2005
2006 stat->uid = GLOBAL_ROOT_UID;
2007 stat->gid = GLOBAL_ROOT_GID;
2008 rcu_read_lock();
2009 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2010 if (task) {
2011 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
2012 rcu_read_unlock();
2013 /*
2014 * This doesn't prevent learning whether PID exists,
2015 * it only makes getattr() consistent with readdir().
2016 */
2017 return -ENOENT;
2018 }
2019 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
2020 }
2021 rcu_read_unlock();
2022 return 0;
2023}
2024
2025/* dentry stuff */
2026
2027/*
2028 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
2029 */
2030void pid_update_inode(struct task_struct *task, struct inode *inode)
2031{
2032 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
2033
2034 inode->i_mode &= ~(S_ISUID | S_ISGID);
2035 security_task_to_inode(task, inode);
2036}
2037
2038/*
2039 * Rewrite the inode's ownerships here because the owning task may have
2040 * performed a setuid(), etc.
2041 *
2042 */
2043static int pid_revalidate(struct inode *dir, const struct qstr *name,
2044 struct dentry *dentry, unsigned int flags)
2045{
2046 struct inode *inode;
2047 struct task_struct *task;
2048 int ret = 0;
2049
2050 rcu_read_lock();
2051 inode = d_inode_rcu(dentry);
2052 if (!inode)
2053 goto out;
2054 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2055
2056 if (task) {
2057 pid_update_inode(task, inode);
2058 ret = 1;
2059 }
2060out:
2061 rcu_read_unlock();
2062 return ret;
2063}
2064
2065static inline bool proc_inode_is_dead(struct inode *inode)
2066{
2067 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2068}
2069
2070int pid_delete_dentry(const struct dentry *dentry)
2071{
2072 /* Is the task we represent dead?
2073 * If so, then don't put the dentry on the lru list,
2074 * kill it immediately.
2075 */
2076 return proc_inode_is_dead(d_inode(dentry));
2077}
2078
2079const struct dentry_operations pid_dentry_operations =
2080{
2081 .d_revalidate = pid_revalidate,
2082 .d_delete = pid_delete_dentry,
2083};
2084
2085/* Lookups */
2086
2087/*
2088 * Fill a directory entry.
2089 *
2090 * If possible create the dcache entry and derive our inode number and
2091 * file type from dcache entry.
2092 *
2093 * Since all of the proc inode numbers are dynamically generated, the inode
2094 * numbers do not exist until the inode is cache. This means creating
2095 * the dcache entry in readdir is necessary to keep the inode numbers
2096 * reported by readdir in sync with the inode numbers reported
2097 * by stat.
2098 */
2099bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2100 const char *name, unsigned int len,
2101 instantiate_t instantiate, struct task_struct *task, const void *ptr)
2102{
2103 struct dentry *child, *dir = file->f_path.dentry;
2104 struct qstr qname = QSTR_INIT(name, len);
2105 struct inode *inode;
2106 unsigned type = DT_UNKNOWN;
2107 ino_t ino = 1;
2108
2109 child = try_lookup_noperm(&qname, dir);
2110 if (IS_ERR(child))
2111 goto end_instantiate;
2112
2113 if (!child) {
2114 child = d_alloc_parallel(dir, &qname);
2115 if (IS_ERR(child))
2116 goto end_instantiate;
2117 if (d_in_lookup(child)) {
2118 struct dentry *res;
2119 res = instantiate(child, task, ptr);
2120 d_lookup_done(child);
2121 if (unlikely(res)) {
2122 dput(child);
2123 child = res;
2124 if (IS_ERR(child))
2125 goto end_instantiate;
2126 }
2127 }
2128 }
2129 inode = d_inode(child);
2130 ino = inode->i_ino;
2131 type = inode->i_mode >> 12;
2132 dput(child);
2133end_instantiate:
2134 return dir_emit(ctx, name, len, ino, type);
2135}
2136
2137/*
2138 * dname_to_vma_addr - maps a dentry name into two unsigned longs
2139 * which represent vma start and end addresses.
2140 */
2141static int dname_to_vma_addr(struct dentry *dentry,
2142 unsigned long *start, unsigned long *end)
2143{
2144 const char *str = dentry->d_name.name;
2145 unsigned long long sval, eval;
2146 unsigned int len;
2147
2148 if (str[0] == '0' && str[1] != '-')
2149 return -EINVAL;
2150 len = _parse_integer(str, 16, &sval);
2151 if (len & KSTRTOX_OVERFLOW)
2152 return -EINVAL;
2153 if (sval != (unsigned long)sval)
2154 return -EINVAL;
2155 str += len;
2156
2157 if (*str != '-')
2158 return -EINVAL;
2159 str++;
2160
2161 if (str[0] == '0' && str[1])
2162 return -EINVAL;
2163 len = _parse_integer(str, 16, &eval);
2164 if (len & KSTRTOX_OVERFLOW)
2165 return -EINVAL;
2166 if (eval != (unsigned long)eval)
2167 return -EINVAL;
2168 str += len;
2169
2170 if (*str != '\0')
2171 return -EINVAL;
2172
2173 *start = sval;
2174 *end = eval;
2175
2176 return 0;
2177}
2178
2179static int map_files_d_revalidate(struct inode *dir, const struct qstr *name,
2180 struct dentry *dentry, unsigned int flags)
2181{
2182 unsigned long vm_start, vm_end;
2183 bool exact_vma_exists = false;
2184 struct mm_struct *mm = NULL;
2185 struct task_struct *task;
2186 struct inode *inode;
2187 int status = 0;
2188
2189 if (flags & LOOKUP_RCU)
2190 return -ECHILD;
2191
2192 inode = d_inode(dentry);
2193 task = get_proc_task(inode);
2194 if (!task)
2195 goto out_notask;
2196
2197 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2198 if (IS_ERR(mm))
2199 goto out;
2200
2201 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2202 status = mmap_read_lock_killable(mm);
2203 if (!status) {
2204 exact_vma_exists = !!find_exact_vma(mm, vm_start,
2205 vm_end);
2206 mmap_read_unlock(mm);
2207 }
2208 }
2209
2210 mmput(mm);
2211
2212 if (exact_vma_exists) {
2213 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2214
2215 security_task_to_inode(task, inode);
2216 status = 1;
2217 }
2218
2219out:
2220 put_task_struct(task);
2221
2222out_notask:
2223 return status;
2224}
2225
2226static const struct dentry_operations tid_map_files_dentry_operations = {
2227 .d_revalidate = map_files_d_revalidate,
2228 .d_delete = pid_delete_dentry,
2229};
2230
2231static int map_files_get_link(struct dentry *dentry, struct path *path,
2232 struct task_struct *task)
2233{
2234 unsigned long vm_start, vm_end;
2235 struct vm_area_struct *vma;
2236 struct mm_struct *mm;
2237 int rc;
2238
2239 rc = -ENOENT;
2240 mm = get_task_mm(task);
2241 if (!mm)
2242 goto out;
2243
2244 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2245 if (rc)
2246 goto out_mmput;
2247
2248 rc = mmap_read_lock_killable(mm);
2249 if (rc)
2250 goto out_mmput;
2251
2252 rc = -ENOENT;
2253 vma = find_exact_vma(mm, vm_start, vm_end);
2254 if (vma && vma->vm_file) {
2255 *path = *file_user_path(vma->vm_file);
2256 path_get(path);
2257 rc = 0;
2258 }
2259 mmap_read_unlock(mm);
2260
2261out_mmput:
2262 mmput(mm);
2263out:
2264 return rc;
2265}
2266
2267struct map_files_info {
2268 unsigned long start;
2269 unsigned long end;
2270 fmode_t mode;
2271};
2272
2273/*
2274 * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2275 * to concerns about how the symlinks may be used to bypass permissions on
2276 * ancestor directories in the path to the file in question.
2277 */
2278static const char *
2279proc_map_files_get_link(struct dentry *dentry,
2280 struct inode *inode,
2281 struct delayed_call *done)
2282{
2283 if (!checkpoint_restore_ns_capable(&init_user_ns))
2284 return ERR_PTR(-EPERM);
2285
2286 return proc_pid_get_link(dentry, inode, done);
2287}
2288
2289/*
2290 * Identical to proc_pid_link_inode_operations except for get_link()
2291 */
2292static const struct inode_operations proc_map_files_link_inode_operations = {
2293 .readlink = proc_pid_readlink,
2294 .get_link = proc_map_files_get_link,
2295 .setattr = proc_nochmod_setattr,
2296};
2297
2298static struct dentry *
2299proc_map_files_instantiate(struct dentry *dentry,
2300 struct task_struct *task, const void *ptr)
2301{
2302 fmode_t mode = (fmode_t)(unsigned long)ptr;
2303 struct proc_inode *ei;
2304 struct inode *inode;
2305
2306 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2307 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2308 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2309 if (!inode)
2310 return ERR_PTR(-ENOENT);
2311
2312 ei = PROC_I(inode);
2313 ei->op.proc_get_link = map_files_get_link;
2314
2315 inode->i_op = &proc_map_files_link_inode_operations;
2316 inode->i_size = 64;
2317
2318 return proc_splice_unmountable(inode, dentry,
2319 &tid_map_files_dentry_operations);
2320}
2321
2322static struct dentry *proc_map_files_lookup(struct inode *dir,
2323 struct dentry *dentry, unsigned int flags)
2324{
2325 unsigned long vm_start, vm_end;
2326 struct vm_area_struct *vma;
2327 struct task_struct *task;
2328 struct dentry *result;
2329 struct mm_struct *mm;
2330
2331 result = ERR_PTR(-ENOENT);
2332 task = get_proc_task(dir);
2333 if (!task)
2334 goto out;
2335
2336 result = ERR_PTR(-ENOENT);
2337 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2338 goto out_put_task;
2339
2340 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2341 if (IS_ERR(mm)) {
2342 result = ERR_CAST(mm);
2343 goto out_put_task;
2344 }
2345
2346 result = ERR_PTR(-EINTR);
2347 if (mmap_read_lock_killable(mm))
2348 goto out_put_mm;
2349
2350 result = ERR_PTR(-ENOENT);
2351 vma = find_exact_vma(mm, vm_start, vm_end);
2352 if (!vma)
2353 goto out_no_vma;
2354
2355 if (vma->vm_file)
2356 result = proc_map_files_instantiate(dentry, task,
2357 (void *)(unsigned long)vma->vm_file->f_mode);
2358
2359out_no_vma:
2360 mmap_read_unlock(mm);
2361out_put_mm:
2362 mmput(mm);
2363out_put_task:
2364 put_task_struct(task);
2365out:
2366 return result;
2367}
2368
2369static const struct inode_operations proc_map_files_inode_operations = {
2370 .lookup = proc_map_files_lookup,
2371 .permission = proc_fd_permission,
2372 .setattr = proc_nochmod_setattr,
2373};
2374
2375static int
2376proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2377{
2378 struct vm_area_struct *vma;
2379 struct task_struct *task;
2380 struct mm_struct *mm;
2381 unsigned long nr_files, pos, i;
2382 GENRADIX(struct map_files_info) fa;
2383 struct map_files_info *p;
2384 int ret;
2385 struct vma_iterator vmi;
2386
2387 genradix_init(&fa);
2388
2389 ret = -ENOENT;
2390 task = get_proc_task(file_inode(file));
2391 if (!task)
2392 goto out;
2393
2394 ret = 0;
2395 if (!dir_emit_dots(file, ctx))
2396 goto out_put_task;
2397
2398 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2399 if (IS_ERR(mm)) {
2400 ret = PTR_ERR(mm);
2401 /* if the task has no mm, the directory should just be empty */
2402 if (ret == -ESRCH)
2403 ret = 0;
2404 goto out_put_task;
2405 }
2406
2407 ret = mmap_read_lock_killable(mm);
2408 if (ret)
2409 goto out_put_mm;
2410
2411 nr_files = 0;
2412
2413 /*
2414 * We need two passes here:
2415 *
2416 * 1) Collect vmas of mapped files with mmap_lock taken
2417 * 2) Release mmap_lock and instantiate entries
2418 *
2419 * otherwise we get lockdep complained, since filldir()
2420 * routine might require mmap_lock taken in might_fault().
2421 */
2422
2423 pos = 2;
2424 vma_iter_init(&vmi, mm, 0);
2425 for_each_vma(vmi, vma) {
2426 if (!vma->vm_file)
2427 continue;
2428 if (++pos <= ctx->pos)
2429 continue;
2430
2431 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2432 if (!p) {
2433 ret = -ENOMEM;
2434 mmap_read_unlock(mm);
2435 goto out_put_mm;
2436 }
2437
2438 p->start = vma->vm_start;
2439 p->end = vma->vm_end;
2440 p->mode = vma->vm_file->f_mode;
2441 }
2442 mmap_read_unlock(mm);
2443
2444 for (i = 0; i < nr_files; i++) {
2445 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2446 unsigned int len;
2447
2448 p = genradix_ptr(&fa, i);
2449 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2450 if (!proc_fill_cache(file, ctx,
2451 buf, len,
2452 proc_map_files_instantiate,
2453 task,
2454 (void *)(unsigned long)p->mode))
2455 break;
2456 ctx->pos++;
2457 }
2458
2459out_put_mm:
2460 mmput(mm);
2461out_put_task:
2462 put_task_struct(task);
2463out:
2464 genradix_free(&fa);
2465 return ret;
2466}
2467
2468static const struct file_operations proc_map_files_operations = {
2469 .read = generic_read_dir,
2470 .iterate_shared = proc_map_files_readdir,
2471 .llseek = generic_file_llseek,
2472};
2473
2474#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2475struct timers_private {
2476 struct pid *pid;
2477 struct task_struct *task;
2478 struct pid_namespace *ns;
2479};
2480
2481static void *timers_start(struct seq_file *m, loff_t *pos)
2482{
2483 struct timers_private *tp = m->private;
2484
2485 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2486 if (!tp->task)
2487 return ERR_PTR(-ESRCH);
2488
2489 rcu_read_lock();
2490 return seq_hlist_start_rcu(&tp->task->signal->posix_timers, *pos);
2491}
2492
2493static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2494{
2495 struct timers_private *tp = m->private;
2496
2497 return seq_hlist_next_rcu(v, &tp->task->signal->posix_timers, pos);
2498}
2499
2500static void timers_stop(struct seq_file *m, void *v)
2501{
2502 struct timers_private *tp = m->private;
2503
2504 if (tp->task) {
2505 put_task_struct(tp->task);
2506 tp->task = NULL;
2507 rcu_read_unlock();
2508 }
2509}
2510
2511static int show_timer(struct seq_file *m, void *v)
2512{
2513 static const char * const nstr[] = {
2514 [SIGEV_SIGNAL] = "signal",
2515 [SIGEV_NONE] = "none",
2516 [SIGEV_THREAD] = "thread",
2517 };
2518
2519 struct k_itimer *timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
2520 struct timers_private *tp = m->private;
2521 int notify = timer->it_sigev_notify;
2522
2523 guard(spinlock_irq)(&timer->it_lock);
2524 if (!posixtimer_valid(timer))
2525 return 0;
2526
2527 seq_printf(m, "ID: %d\n", timer->it_id);
2528 seq_printf(m, "signal: %d/%px\n", timer->sigq.info.si_signo,
2529 timer->sigq.info.si_value.sival_ptr);
2530 seq_printf(m, "notify: %s/%s.%d\n", nstr[notify & ~SIGEV_THREAD_ID],
2531 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2532 pid_nr_ns(timer->it_pid, tp->ns));
2533 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2534
2535 return 0;
2536}
2537
2538static const struct seq_operations proc_timers_seq_ops = {
2539 .start = timers_start,
2540 .next = timers_next,
2541 .stop = timers_stop,
2542 .show = show_timer,
2543};
2544
2545static int proc_timers_open(struct inode *inode, struct file *file)
2546{
2547 struct timers_private *tp;
2548
2549 tp = __seq_open_private(file, &proc_timers_seq_ops,
2550 sizeof(struct timers_private));
2551 if (!tp)
2552 return -ENOMEM;
2553
2554 tp->pid = proc_pid(inode);
2555 tp->ns = proc_pid_ns(inode->i_sb);
2556 return 0;
2557}
2558
2559static const struct file_operations proc_timers_operations = {
2560 .open = proc_timers_open,
2561 .read = seq_read,
2562 .llseek = seq_lseek,
2563 .release = seq_release_private,
2564};
2565#endif
2566
2567static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2568 size_t count, loff_t *offset)
2569{
2570 struct inode *inode = file_inode(file);
2571 struct task_struct *p;
2572 u64 slack_ns;
2573 int err;
2574
2575 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2576 if (err < 0)
2577 return err;
2578
2579 p = get_proc_task(inode);
2580 if (!p)
2581 return -ESRCH;
2582
2583 if (p != current) {
2584 rcu_read_lock();
2585 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2586 rcu_read_unlock();
2587 count = -EPERM;
2588 goto out;
2589 }
2590 rcu_read_unlock();
2591
2592 err = security_task_setscheduler(p);
2593 if (err) {
2594 count = err;
2595 goto out;
2596 }
2597 }
2598
2599 task_lock(p);
2600 if (rt_or_dl_task_policy(p))
2601 slack_ns = 0;
2602 else if (slack_ns == 0)
2603 slack_ns = p->default_timer_slack_ns;
2604 p->timer_slack_ns = slack_ns;
2605 task_unlock(p);
2606
2607out:
2608 put_task_struct(p);
2609
2610 return count;
2611}
2612
2613static int timerslack_ns_show(struct seq_file *m, void *v)
2614{
2615 struct inode *inode = m->private;
2616 struct task_struct *p;
2617 int err = 0;
2618
2619 p = get_proc_task(inode);
2620 if (!p)
2621 return -ESRCH;
2622
2623 if (p != current) {
2624 rcu_read_lock();
2625 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2626 rcu_read_unlock();
2627 err = -EPERM;
2628 goto out;
2629 }
2630 rcu_read_unlock();
2631
2632 err = security_task_getscheduler(p);
2633 if (err)
2634 goto out;
2635 }
2636
2637 task_lock(p);
2638 seq_printf(m, "%llu\n", p->timer_slack_ns);
2639 task_unlock(p);
2640
2641out:
2642 put_task_struct(p);
2643
2644 return err;
2645}
2646
2647static int timerslack_ns_open(struct inode *inode, struct file *filp)
2648{
2649 return single_open(filp, timerslack_ns_show, inode);
2650}
2651
2652static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2653 .open = timerslack_ns_open,
2654 .read = seq_read,
2655 .write = timerslack_ns_write,
2656 .llseek = seq_lseek,
2657 .release = single_release,
2658};
2659
2660static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2661 struct task_struct *task, const void *ptr)
2662{
2663 const struct pid_entry *p = ptr;
2664 struct inode *inode;
2665 struct proc_inode *ei;
2666
2667 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2668 if (!inode)
2669 return ERR_PTR(-ENOENT);
2670
2671 ei = PROC_I(inode);
2672 if (S_ISDIR(inode->i_mode))
2673 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2674 if (p->iop)
2675 inode->i_op = p->iop;
2676 if (p->fop)
2677 inode->i_fop = p->fop;
2678 ei->op = p->op;
2679 pid_update_inode(task, inode);
2680 return d_splice_alias_ops(inode, dentry, &pid_dentry_operations);
2681}
2682
2683static struct dentry *proc_pident_lookup(struct inode *dir,
2684 struct dentry *dentry,
2685 const struct pid_entry *p,
2686 const struct pid_entry *end)
2687{
2688 struct task_struct *task = get_proc_task(dir);
2689 struct dentry *res = ERR_PTR(-ENOENT);
2690
2691 if (!task)
2692 goto out_no_task;
2693
2694 /*
2695 * Yes, it does not scale. And it should not. Don't add
2696 * new entries into /proc/<tgid>/ without very good reasons.
2697 */
2698 for (; p < end; p++) {
2699 if (p->len != dentry->d_name.len)
2700 continue;
2701 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2702 res = proc_pident_instantiate(dentry, task, p);
2703 break;
2704 }
2705 }
2706 put_task_struct(task);
2707out_no_task:
2708 return res;
2709}
2710
2711static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2712 const struct pid_entry *ents, unsigned int nents)
2713{
2714 struct task_struct *task = get_proc_task(file_inode(file));
2715 const struct pid_entry *p;
2716
2717 if (!task)
2718 return -ENOENT;
2719
2720 if (!dir_emit_dots(file, ctx))
2721 goto out;
2722
2723 if (ctx->pos >= nents + 2)
2724 goto out;
2725
2726 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2727 if (!proc_fill_cache(file, ctx, p->name, p->len,
2728 proc_pident_instantiate, task, p))
2729 break;
2730 ctx->pos++;
2731 }
2732out:
2733 put_task_struct(task);
2734 return 0;
2735}
2736
2737#ifdef CONFIG_SECURITY
2738static int proc_pid_attr_open(struct inode *inode, struct file *file)
2739{
2740 file->private_data = NULL;
2741 __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2742 return 0;
2743}
2744
2745static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2746 size_t count, loff_t *ppos)
2747{
2748 struct inode * inode = file_inode(file);
2749 char *p = NULL;
2750 ssize_t length;
2751 struct task_struct *task = get_proc_task(inode);
2752
2753 if (!task)
2754 return -ESRCH;
2755
2756 length = security_getprocattr(task, PROC_I(inode)->op.lsmid,
2757 file->f_path.dentry->d_name.name,
2758 &p);
2759 put_task_struct(task);
2760 if (length > 0)
2761 length = simple_read_from_buffer(buf, count, ppos, p, length);
2762 kfree(p);
2763 return length;
2764}
2765
2766static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2767 size_t count, loff_t *ppos)
2768{
2769 struct inode * inode = file_inode(file);
2770 struct task_struct *task;
2771 void *page;
2772 int rv;
2773
2774 /* A task may only write when it was the opener. */
2775 if (file->private_data != current->mm)
2776 return -EPERM;
2777
2778 rcu_read_lock();
2779 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2780 if (!task) {
2781 rcu_read_unlock();
2782 return -ESRCH;
2783 }
2784 /* A task may only write its own attributes. */
2785 if (current != task) {
2786 rcu_read_unlock();
2787 return -EACCES;
2788 }
2789 /* Prevent changes to overridden credentials. */
2790 if (current_cred() != current_real_cred()) {
2791 rcu_read_unlock();
2792 return -EBUSY;
2793 }
2794 rcu_read_unlock();
2795
2796 if (count > PAGE_SIZE)
2797 count = PAGE_SIZE;
2798
2799 /* No partial writes. */
2800 if (*ppos != 0)
2801 return -EINVAL;
2802
2803 page = memdup_user(buf, count);
2804 if (IS_ERR(page)) {
2805 rv = PTR_ERR(page);
2806 goto out;
2807 }
2808
2809 /* Guard against adverse ptrace interaction */
2810 rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2811 if (rv < 0)
2812 goto out_free;
2813
2814 rv = security_setprocattr(PROC_I(inode)->op.lsmid,
2815 file->f_path.dentry->d_name.name, page,
2816 count);
2817 mutex_unlock(&current->signal->cred_guard_mutex);
2818out_free:
2819 kfree(page);
2820out:
2821 return rv;
2822}
2823
2824static const struct file_operations proc_pid_attr_operations = {
2825 .open = proc_pid_attr_open,
2826 .read = proc_pid_attr_read,
2827 .write = proc_pid_attr_write,
2828 .llseek = generic_file_llseek,
2829 .release = mem_release,
2830};
2831
2832#define LSM_DIR_OPS(LSM) \
2833static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2834 struct dir_context *ctx) \
2835{ \
2836 return proc_pident_readdir(filp, ctx, \
2837 LSM##_attr_dir_stuff, \
2838 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2839} \
2840\
2841static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2842 .read = generic_read_dir, \
2843 .iterate_shared = proc_##LSM##_attr_dir_iterate, \
2844 .llseek = default_llseek, \
2845}; \
2846\
2847static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2848 struct dentry *dentry, unsigned int flags) \
2849{ \
2850 return proc_pident_lookup(dir, dentry, \
2851 LSM##_attr_dir_stuff, \
2852 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2853} \
2854\
2855static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2856 .lookup = proc_##LSM##_attr_dir_lookup, \
2857 .getattr = pid_getattr, \
2858 .setattr = proc_nochmod_setattr, \
2859}
2860
2861#ifdef CONFIG_SECURITY_SMACK
2862static const struct pid_entry smack_attr_dir_stuff[] = {
2863 ATTR(LSM_ID_SMACK, "current", 0666),
2864};
2865LSM_DIR_OPS(smack);
2866#endif
2867
2868#ifdef CONFIG_SECURITY_APPARMOR
2869static const struct pid_entry apparmor_attr_dir_stuff[] = {
2870 ATTR(LSM_ID_APPARMOR, "current", 0666),
2871 ATTR(LSM_ID_APPARMOR, "prev", 0444),
2872 ATTR(LSM_ID_APPARMOR, "exec", 0666),
2873};
2874LSM_DIR_OPS(apparmor);
2875#endif
2876
2877static const struct pid_entry attr_dir_stuff[] = {
2878 ATTR(LSM_ID_UNDEF, "current", 0666),
2879 ATTR(LSM_ID_UNDEF, "prev", 0444),
2880 ATTR(LSM_ID_UNDEF, "exec", 0666),
2881 ATTR(LSM_ID_UNDEF, "fscreate", 0666),
2882 ATTR(LSM_ID_UNDEF, "keycreate", 0666),
2883 ATTR(LSM_ID_UNDEF, "sockcreate", 0666),
2884#ifdef CONFIG_SECURITY_SMACK
2885 DIR("smack", 0555,
2886 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2887#endif
2888#ifdef CONFIG_SECURITY_APPARMOR
2889 DIR("apparmor", 0555,
2890 proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2891#endif
2892};
2893
2894static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2895{
2896 return proc_pident_readdir(file, ctx,
2897 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2898}
2899
2900static const struct file_operations proc_attr_dir_operations = {
2901 .read = generic_read_dir,
2902 .iterate_shared = proc_attr_dir_readdir,
2903 .llseek = generic_file_llseek,
2904};
2905
2906static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2907 struct dentry *dentry, unsigned int flags)
2908{
2909 return proc_pident_lookup(dir, dentry,
2910 attr_dir_stuff,
2911 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2912}
2913
2914static const struct inode_operations proc_attr_dir_inode_operations = {
2915 .lookup = proc_attr_dir_lookup,
2916 .getattr = pid_getattr,
2917 .setattr = proc_nochmod_setattr,
2918};
2919
2920#endif
2921
2922#ifdef CONFIG_ELF_CORE
2923static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2924 size_t count, loff_t *ppos)
2925{
2926 struct task_struct *task = get_proc_task(file_inode(file));
2927 struct mm_struct *mm;
2928 char buffer[PROC_NUMBUF];
2929 size_t len;
2930 int ret;
2931
2932 if (!task)
2933 return -ESRCH;
2934
2935 ret = 0;
2936 mm = get_task_mm(task);
2937 if (mm) {
2938 unsigned long flags = __mm_flags_get_word(mm);
2939
2940 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2941 ((flags & MMF_DUMP_FILTER_MASK) >>
2942 MMF_DUMP_FILTER_SHIFT));
2943 mmput(mm);
2944 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2945 }
2946
2947 put_task_struct(task);
2948
2949 return ret;
2950}
2951
2952static ssize_t proc_coredump_filter_write(struct file *file,
2953 const char __user *buf,
2954 size_t count,
2955 loff_t *ppos)
2956{
2957 struct task_struct *task;
2958 struct mm_struct *mm;
2959 unsigned int val;
2960 int ret;
2961 int i;
2962 unsigned long mask;
2963
2964 ret = kstrtouint_from_user(buf, count, 0, &val);
2965 if (ret < 0)
2966 return ret;
2967
2968 ret = -ESRCH;
2969 task = get_proc_task(file_inode(file));
2970 if (!task)
2971 goto out_no_task;
2972
2973 mm = get_task_mm(task);
2974 if (!mm)
2975 goto out_no_mm;
2976 ret = 0;
2977
2978 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2979 if (val & mask)
2980 mm_flags_set(i + MMF_DUMP_FILTER_SHIFT, mm);
2981 else
2982 mm_flags_clear(i + MMF_DUMP_FILTER_SHIFT, mm);
2983 }
2984
2985 mmput(mm);
2986 out_no_mm:
2987 put_task_struct(task);
2988 out_no_task:
2989 if (ret < 0)
2990 return ret;
2991 return count;
2992}
2993
2994static const struct file_operations proc_coredump_filter_operations = {
2995 .read = proc_coredump_filter_read,
2996 .write = proc_coredump_filter_write,
2997 .llseek = generic_file_llseek,
2998};
2999#endif
3000
3001#ifdef CONFIG_TASK_IO_ACCOUNTING
3002static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
3003{
3004 struct task_io_accounting acct;
3005 int result;
3006
3007 result = down_read_killable(&task->signal->exec_update_lock);
3008 if (result)
3009 return result;
3010
3011 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
3012 result = -EACCES;
3013 goto out_unlock;
3014 }
3015
3016 if (whole) {
3017 struct signal_struct *sig = task->signal;
3018 struct task_struct *t;
3019
3020 guard(rcu)();
3021 scoped_seqlock_read (&sig->stats_lock, ss_lock_irqsave) {
3022 acct = sig->ioac;
3023 __for_each_thread(sig, t)
3024 task_io_accounting_add(&acct, &t->ioac);
3025
3026 }
3027 } else {
3028 acct = task->ioac;
3029 }
3030
3031 seq_printf(m,
3032 "rchar: %llu\n"
3033 "wchar: %llu\n"
3034 "syscr: %llu\n"
3035 "syscw: %llu\n"
3036 "read_bytes: %llu\n"
3037 "write_bytes: %llu\n"
3038 "cancelled_write_bytes: %llu\n",
3039 (unsigned long long)acct.rchar,
3040 (unsigned long long)acct.wchar,
3041 (unsigned long long)acct.syscr,
3042 (unsigned long long)acct.syscw,
3043 (unsigned long long)acct.read_bytes,
3044 (unsigned long long)acct.write_bytes,
3045 (unsigned long long)acct.cancelled_write_bytes);
3046 result = 0;
3047
3048out_unlock:
3049 up_read(&task->signal->exec_update_lock);
3050 return result;
3051}
3052
3053static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3054 struct pid *pid, struct task_struct *task)
3055{
3056 return do_io_accounting(task, m, 0);
3057}
3058
3059static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3060 struct pid *pid, struct task_struct *task)
3061{
3062 return do_io_accounting(task, m, 1);
3063}
3064#endif /* CONFIG_TASK_IO_ACCOUNTING */
3065
3066#ifdef CONFIG_USER_NS
3067static int proc_id_map_open(struct inode *inode, struct file *file,
3068 const struct seq_operations *seq_ops)
3069{
3070 struct user_namespace *ns = NULL;
3071 struct task_struct *task;
3072 struct seq_file *seq;
3073 int ret = -EINVAL;
3074
3075 task = get_proc_task(inode);
3076 if (task) {
3077 rcu_read_lock();
3078 ns = get_user_ns(task_cred_xxx(task, user_ns));
3079 rcu_read_unlock();
3080 put_task_struct(task);
3081 }
3082 if (!ns)
3083 goto err;
3084
3085 ret = seq_open(file, seq_ops);
3086 if (ret)
3087 goto err_put_ns;
3088
3089 seq = file->private_data;
3090 seq->private = ns;
3091
3092 return 0;
3093err_put_ns:
3094 put_user_ns(ns);
3095err:
3096 return ret;
3097}
3098
3099static int proc_id_map_release(struct inode *inode, struct file *file)
3100{
3101 struct seq_file *seq = file->private_data;
3102 struct user_namespace *ns = seq->private;
3103 put_user_ns(ns);
3104 return seq_release(inode, file);
3105}
3106
3107static int proc_uid_map_open(struct inode *inode, struct file *file)
3108{
3109 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3110}
3111
3112static int proc_gid_map_open(struct inode *inode, struct file *file)
3113{
3114 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3115}
3116
3117static int proc_projid_map_open(struct inode *inode, struct file *file)
3118{
3119 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3120}
3121
3122static const struct file_operations proc_uid_map_operations = {
3123 .open = proc_uid_map_open,
3124 .write = proc_uid_map_write,
3125 .read = seq_read,
3126 .llseek = seq_lseek,
3127 .release = proc_id_map_release,
3128};
3129
3130static const struct file_operations proc_gid_map_operations = {
3131 .open = proc_gid_map_open,
3132 .write = proc_gid_map_write,
3133 .read = seq_read,
3134 .llseek = seq_lseek,
3135 .release = proc_id_map_release,
3136};
3137
3138static const struct file_operations proc_projid_map_operations = {
3139 .open = proc_projid_map_open,
3140 .write = proc_projid_map_write,
3141 .read = seq_read,
3142 .llseek = seq_lseek,
3143 .release = proc_id_map_release,
3144};
3145
3146static int proc_setgroups_open(struct inode *inode, struct file *file)
3147{
3148 struct user_namespace *ns = NULL;
3149 struct task_struct *task;
3150 int ret;
3151
3152 ret = -ESRCH;
3153 task = get_proc_task(inode);
3154 if (task) {
3155 rcu_read_lock();
3156 ns = get_user_ns(task_cred_xxx(task, user_ns));
3157 rcu_read_unlock();
3158 put_task_struct(task);
3159 }
3160 if (!ns)
3161 goto err;
3162
3163 if (file->f_mode & FMODE_WRITE) {
3164 ret = -EACCES;
3165 if (!ns_capable(ns, CAP_SYS_ADMIN))
3166 goto err_put_ns;
3167 }
3168
3169 ret = single_open(file, &proc_setgroups_show, ns);
3170 if (ret)
3171 goto err_put_ns;
3172
3173 return 0;
3174err_put_ns:
3175 put_user_ns(ns);
3176err:
3177 return ret;
3178}
3179
3180static int proc_setgroups_release(struct inode *inode, struct file *file)
3181{
3182 struct seq_file *seq = file->private_data;
3183 struct user_namespace *ns = seq->private;
3184 int ret = single_release(inode, file);
3185 put_user_ns(ns);
3186 return ret;
3187}
3188
3189static const struct file_operations proc_setgroups_operations = {
3190 .open = proc_setgroups_open,
3191 .write = proc_setgroups_write,
3192 .read = seq_read,
3193 .llseek = seq_lseek,
3194 .release = proc_setgroups_release,
3195};
3196#endif /* CONFIG_USER_NS */
3197
3198static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3199 struct pid *pid, struct task_struct *task)
3200{
3201 int err = lock_trace(task);
3202 if (!err) {
3203 seq_printf(m, "%08x\n", task->personality);
3204 unlock_trace(task);
3205 }
3206 return err;
3207}
3208
3209#ifdef CONFIG_LIVEPATCH
3210static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3211 struct pid *pid, struct task_struct *task)
3212{
3213 seq_printf(m, "%d\n", task->patch_state);
3214 return 0;
3215}
3216#endif /* CONFIG_LIVEPATCH */
3217
3218#ifdef CONFIG_KSM
3219static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3220 struct pid *pid, struct task_struct *task)
3221{
3222 struct mm_struct *mm;
3223
3224 mm = get_task_mm(task);
3225 if (mm) {
3226 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3227 mmput(mm);
3228 }
3229
3230 return 0;
3231}
3232static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3233 struct pid *pid, struct task_struct *task)
3234{
3235 struct mm_struct *mm;
3236 int ret = 0;
3237
3238 mm = get_task_mm(task);
3239 if (mm) {
3240 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3241 seq_printf(m, "ksm_zero_pages %ld\n", mm_ksm_zero_pages(mm));
3242 seq_printf(m, "ksm_merging_pages %lu\n", mm->ksm_merging_pages);
3243 seq_printf(m, "ksm_process_profit %ld\n", ksm_process_profit(mm));
3244 seq_printf(m, "ksm_merge_any: %s\n",
3245 mm_flags_test(MMF_VM_MERGE_ANY, mm) ? "yes" : "no");
3246 ret = mmap_read_lock_killable(mm);
3247 if (ret) {
3248 mmput(mm);
3249 return ret;
3250 }
3251 seq_printf(m, "ksm_mergeable: %s\n",
3252 ksm_process_mergeable(mm) ? "yes" : "no");
3253 mmap_read_unlock(mm);
3254 mmput(mm);
3255 }
3256
3257 return 0;
3258}
3259#endif /* CONFIG_KSM */
3260
3261#ifdef CONFIG_KSTACK_ERASE_METRICS
3262static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3263 struct pid *pid, struct task_struct *task)
3264{
3265 unsigned long prev_depth = THREAD_SIZE -
3266 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3267 unsigned long depth = THREAD_SIZE -
3268 (task->lowest_stack & (THREAD_SIZE - 1));
3269
3270 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3271 prev_depth, depth);
3272 return 0;
3273}
3274#endif /* CONFIG_KSTACK_ERASE_METRICS */
3275
3276/*
3277 * Thread groups
3278 */
3279static const struct file_operations proc_task_operations;
3280static const struct inode_operations proc_task_inode_operations;
3281
3282static const struct pid_entry tgid_base_stuff[] = {
3283 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3284 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3285 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3286 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3287 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3288#ifdef CONFIG_NET
3289 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3290#endif
3291 REG("environ", S_IRUSR, proc_environ_operations),
3292 REG("auxv", S_IRUSR, proc_auxv_operations),
3293 ONE("status", S_IRUGO, proc_pid_status),
3294 ONE("personality", S_IRUSR, proc_pid_personality),
3295 ONE("limits", S_IRUGO, proc_pid_limits),
3296 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3297#ifdef CONFIG_SCHED_AUTOGROUP
3298 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3299#endif
3300#ifdef CONFIG_TIME_NS
3301 REG("timens_offsets", S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3302#endif
3303 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3304#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3305 ONE("syscall", S_IRUSR, proc_pid_syscall),
3306#endif
3307 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3308 ONE("stat", S_IRUGO, proc_tgid_stat),
3309 ONE("statm", S_IRUGO, proc_pid_statm),
3310 REG("maps", S_IRUGO, proc_pid_maps_operations),
3311#ifdef CONFIG_NUMA
3312 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3313#endif
3314 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3315 LNK("cwd", proc_cwd_link),
3316 LNK("root", proc_root_link),
3317 LNK("exe", proc_exe_link),
3318 REG("mounts", S_IRUGO, proc_mounts_operations),
3319 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3320 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3321#ifdef CONFIG_PROC_PAGE_MONITOR
3322 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3323 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3324 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3325 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3326#endif
3327#ifdef CONFIG_SECURITY
3328 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3329#endif
3330#ifdef CONFIG_KALLSYMS
3331 ONE("wchan", S_IRUGO, proc_pid_wchan),
3332#endif
3333#ifdef CONFIG_STACKTRACE
3334 ONE("stack", S_IRUSR, proc_pid_stack),
3335#endif
3336#ifdef CONFIG_SCHED_INFO
3337 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3338#endif
3339#ifdef CONFIG_LATENCYTOP
3340 REG("latency", S_IRUGO, proc_lstats_operations),
3341#endif
3342#ifdef CONFIG_PROC_PID_CPUSET
3343 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3344#endif
3345#ifdef CONFIG_CGROUPS
3346 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3347#endif
3348#ifdef CONFIG_PROC_CPU_RESCTRL
3349 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3350#endif
3351 ONE("oom_score", S_IRUGO, proc_oom_score),
3352 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3353 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3354#ifdef CONFIG_AUDIT
3355 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3356 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3357#endif
3358#ifdef CONFIG_FAULT_INJECTION
3359 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3360 REG("fail-nth", 0644, proc_fail_nth_operations),
3361#endif
3362#ifdef CONFIG_ELF_CORE
3363 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3364#endif
3365#ifdef CONFIG_TASK_IO_ACCOUNTING
3366 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3367#endif
3368#ifdef CONFIG_USER_NS
3369 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3370 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3371 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3372 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3373#endif
3374#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3375 REG("timers", S_IRUGO, proc_timers_operations),
3376#endif
3377 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3378#ifdef CONFIG_LIVEPATCH
3379 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3380#endif
3381#ifdef CONFIG_KSTACK_ERASE_METRICS
3382 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3383#endif
3384#ifdef CONFIG_PROC_PID_ARCH_STATUS
3385 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3386#endif
3387#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3388 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3389#endif
3390#ifdef CONFIG_KSM
3391 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3392 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3393#endif
3394};
3395
3396static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3397{
3398 return proc_pident_readdir(file, ctx,
3399 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3400}
3401
3402static const struct file_operations proc_tgid_base_operations = {
3403 .read = generic_read_dir,
3404 .iterate_shared = proc_tgid_base_readdir,
3405 .llseek = generic_file_llseek,
3406};
3407
3408struct pid *tgid_pidfd_to_pid(const struct file *file)
3409{
3410 if (file->f_op != &proc_tgid_base_operations)
3411 return ERR_PTR(-EBADF);
3412
3413 return proc_pid(file_inode(file));
3414}
3415
3416static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3417{
3418 return proc_pident_lookup(dir, dentry,
3419 tgid_base_stuff,
3420 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3421}
3422
3423static const struct inode_operations proc_tgid_base_inode_operations = {
3424 .lookup = proc_tgid_base_lookup,
3425 .getattr = pid_getattr,
3426 .setattr = proc_nochmod_setattr,
3427 .permission = proc_pid_permission,
3428};
3429
3430/**
3431 * proc_flush_pid - Remove dcache entries for @pid from the /proc dcache.
3432 * @pid: pid that should be flushed.
3433 *
3434 * This function walks a list of inodes (that belong to any proc
3435 * filesystem) that are attached to the pid and flushes them from
3436 * the dentry cache.
3437 *
3438 * It is safe and reasonable to cache /proc entries for a task until
3439 * that task exits. After that they just clog up the dcache with
3440 * useless entries, possibly causing useful dcache entries to be
3441 * flushed instead. This routine is provided to flush those useless
3442 * dcache entries when a process is reaped.
3443 *
3444 * NOTE: This routine is just an optimization so it does not guarantee
3445 * that no dcache entries will exist after a process is reaped
3446 * it just makes it very unlikely that any will persist.
3447 */
3448
3449void proc_flush_pid(struct pid *pid)
3450{
3451 proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3452}
3453
3454static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3455 struct task_struct *task, const void *ptr)
3456{
3457 struct inode *inode;
3458
3459 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3460 S_IFDIR | S_IRUGO | S_IXUGO);
3461 if (!inode)
3462 return ERR_PTR(-ENOENT);
3463
3464 inode->i_op = &proc_tgid_base_inode_operations;
3465 inode->i_fop = &proc_tgid_base_operations;
3466 inode->i_flags|=S_IMMUTABLE;
3467
3468 set_nlink(inode, nlink_tgid);
3469 pid_update_inode(task, inode);
3470
3471 return d_splice_alias_ops(inode, dentry, &pid_dentry_operations);
3472}
3473
3474struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3475{
3476 struct task_struct *task;
3477 unsigned tgid;
3478 struct proc_fs_info *fs_info;
3479 struct pid_namespace *ns;
3480 struct dentry *result = ERR_PTR(-ENOENT);
3481
3482 tgid = name_to_int(&dentry->d_name);
3483 if (tgid == ~0U)
3484 goto out;
3485
3486 fs_info = proc_sb_info(dentry->d_sb);
3487 ns = fs_info->pid_ns;
3488 rcu_read_lock();
3489 task = find_task_by_pid_ns(tgid, ns);
3490 if (task)
3491 get_task_struct(task);
3492 rcu_read_unlock();
3493 if (!task)
3494 goto out;
3495
3496 /* Limit procfs to only ptraceable tasks */
3497 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3498 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3499 goto out_put_task;
3500 }
3501
3502 result = proc_pid_instantiate(dentry, task, NULL);
3503out_put_task:
3504 put_task_struct(task);
3505out:
3506 return result;
3507}
3508
3509/*
3510 * Find the first task with tgid >= tgid
3511 *
3512 */
3513struct tgid_iter {
3514 unsigned int tgid;
3515 struct task_struct *task;
3516 struct pid_namespace *const pid_ns;
3517};
3518
3519static struct tgid_iter
3520make_tgid_iter(unsigned int init_tgid, struct pid_namespace *pid_ns)
3521{
3522 return (struct tgid_iter){
3523 .tgid = init_tgid - 1,
3524 .pid_ns = pid_ns,
3525 };
3526}
3527
3528static bool next_tgid(struct tgid_iter *it)
3529{
3530 if (it->task) {
3531 put_task_struct(it->task);
3532 it->task = NULL;
3533 }
3534
3535 rcu_read_lock();
3536 while (1) {
3537 it->tgid += 1;
3538 const auto pid = find_ge_pid(it->tgid, it->pid_ns);
3539 if (pid) {
3540 it->tgid = pid_nr_ns(pid, it->pid_ns);
3541 it->task = pid_task(pid, PIDTYPE_TGID);
3542 if (it->task) {
3543 get_task_struct(it->task);
3544 rcu_read_unlock();
3545 return true;
3546 }
3547 } else {
3548 rcu_read_unlock();
3549 return false;
3550 }
3551 }
3552}
3553
3554#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3555
3556/* for the /proc/ directory itself, after non-process stuff has been done */
3557int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3558{
3559 struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3560 struct pid_namespace *pid_ns = proc_pid_ns(file_inode(file)->i_sb);
3561 loff_t pos = ctx->pos;
3562
3563 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3564 return 0;
3565
3566 if (pos == TGID_OFFSET - 2) {
3567 if (!dir_emit(ctx, "self", 4, self_inum, DT_LNK))
3568 return 0;
3569 ctx->pos = pos = pos + 1;
3570 }
3571 if (pos == TGID_OFFSET - 1) {
3572 if (!dir_emit(ctx, "thread-self", 11, thread_self_inum, DT_LNK))
3573 return 0;
3574 ctx->pos = pos = pos + 1;
3575 }
3576
3577 auto iter = make_tgid_iter(pos - TGID_OFFSET, pid_ns);
3578 while (next_tgid(&iter)) {
3579 char name[10 + 1];
3580 unsigned int len;
3581
3582 cond_resched();
3583 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3584 continue;
3585
3586 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3587 ctx->pos = iter.tgid + TGID_OFFSET;
3588 if (!proc_fill_cache(file, ctx, name, len,
3589 proc_pid_instantiate, iter.task, NULL)) {
3590 put_task_struct(iter.task);
3591 return 0;
3592 }
3593 }
3594 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3595 return 0;
3596}
3597
3598/*
3599 * proc_tid_comm_permission is a special permission function exclusively
3600 * used for the node /proc/<pid>/task/<tid>/comm.
3601 * It bypasses generic permission checks in the case where a task of the same
3602 * task group attempts to access the node.
3603 * The rationale behind this is that glibc and bionic access this node for
3604 * cross thread naming (pthread_set/getname_np(!self)). However, if
3605 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3606 * which locks out the cross thread naming implementation.
3607 * This function makes sure that the node is always accessible for members of
3608 * same thread group.
3609 */
3610static int proc_tid_comm_permission(struct mnt_idmap *idmap,
3611 struct inode *inode, int mask)
3612{
3613 bool is_same_tgroup;
3614 struct task_struct *task;
3615
3616 task = get_proc_task(inode);
3617 if (!task)
3618 return -ESRCH;
3619 is_same_tgroup = same_thread_group(current, task);
3620 put_task_struct(task);
3621
3622 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3623 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3624 * read or written by the members of the corresponding
3625 * thread group.
3626 */
3627 return 0;
3628 }
3629
3630 return generic_permission(&nop_mnt_idmap, inode, mask);
3631}
3632
3633static const struct inode_operations proc_tid_comm_inode_operations = {
3634 .setattr = proc_nochmod_setattr,
3635 .permission = proc_tid_comm_permission,
3636};
3637
3638/*
3639 * Tasks
3640 */
3641static const struct pid_entry tid_base_stuff[] = {
3642 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3643 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3644 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3645#ifdef CONFIG_NET
3646 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3647#endif
3648 REG("environ", S_IRUSR, proc_environ_operations),
3649 REG("auxv", S_IRUSR, proc_auxv_operations),
3650 ONE("status", S_IRUGO, proc_pid_status),
3651 ONE("personality", S_IRUSR, proc_pid_personality),
3652 ONE("limits", S_IRUGO, proc_pid_limits),
3653 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3654 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3655 &proc_tid_comm_inode_operations,
3656 &proc_pid_set_comm_operations, {}),
3657#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3658 ONE("syscall", S_IRUSR, proc_pid_syscall),
3659#endif
3660 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3661 ONE("stat", S_IRUGO, proc_tid_stat),
3662 ONE("statm", S_IRUGO, proc_pid_statm),
3663 REG("maps", S_IRUGO, proc_pid_maps_operations),
3664#ifdef CONFIG_PROC_CHILDREN
3665 REG("children", S_IRUGO, proc_tid_children_operations),
3666#endif
3667#ifdef CONFIG_NUMA
3668 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3669#endif
3670 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3671 LNK("cwd", proc_cwd_link),
3672 LNK("root", proc_root_link),
3673 LNK("exe", proc_exe_link),
3674 REG("mounts", S_IRUGO, proc_mounts_operations),
3675 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3676#ifdef CONFIG_PROC_PAGE_MONITOR
3677 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3678 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3679 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3680 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3681#endif
3682#ifdef CONFIG_SECURITY
3683 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3684#endif
3685#ifdef CONFIG_KALLSYMS
3686 ONE("wchan", S_IRUGO, proc_pid_wchan),
3687#endif
3688#ifdef CONFIG_STACKTRACE
3689 ONE("stack", S_IRUSR, proc_pid_stack),
3690#endif
3691#ifdef CONFIG_SCHED_INFO
3692 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3693#endif
3694#ifdef CONFIG_LATENCYTOP
3695 REG("latency", S_IRUGO, proc_lstats_operations),
3696#endif
3697#ifdef CONFIG_PROC_PID_CPUSET
3698 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3699#endif
3700#ifdef CONFIG_CGROUPS
3701 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3702#endif
3703#ifdef CONFIG_PROC_CPU_RESCTRL
3704 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3705#endif
3706 ONE("oom_score", S_IRUGO, proc_oom_score),
3707 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3708 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3709#ifdef CONFIG_AUDIT
3710 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3711 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3712#endif
3713#ifdef CONFIG_FAULT_INJECTION
3714 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3715 REG("fail-nth", 0644, proc_fail_nth_operations),
3716#endif
3717#ifdef CONFIG_TASK_IO_ACCOUNTING
3718 ONE("io", S_IRUSR, proc_tid_io_accounting),
3719#endif
3720#ifdef CONFIG_USER_NS
3721 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3722 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3723 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3724 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3725#endif
3726#ifdef CONFIG_LIVEPATCH
3727 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3728#endif
3729#ifdef CONFIG_PROC_PID_ARCH_STATUS
3730 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3731#endif
3732#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3733 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3734#endif
3735#ifdef CONFIG_KSM
3736 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3737 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3738#endif
3739};
3740
3741static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3742{
3743 return proc_pident_readdir(file, ctx,
3744 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3745}
3746
3747static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3748{
3749 return proc_pident_lookup(dir, dentry,
3750 tid_base_stuff,
3751 tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3752}
3753
3754static const struct file_operations proc_tid_base_operations = {
3755 .read = generic_read_dir,
3756 .iterate_shared = proc_tid_base_readdir,
3757 .llseek = generic_file_llseek,
3758};
3759
3760static const struct inode_operations proc_tid_base_inode_operations = {
3761 .lookup = proc_tid_base_lookup,
3762 .getattr = pid_getattr,
3763 .setattr = proc_nochmod_setattr,
3764};
3765
3766static struct dentry *proc_task_instantiate(struct dentry *dentry,
3767 struct task_struct *task, const void *ptr)
3768{
3769 struct inode *inode;
3770 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3771 S_IFDIR | S_IRUGO | S_IXUGO);
3772 if (!inode)
3773 return ERR_PTR(-ENOENT);
3774
3775 inode->i_op = &proc_tid_base_inode_operations;
3776 inode->i_fop = &proc_tid_base_operations;
3777 inode->i_flags |= S_IMMUTABLE;
3778
3779 set_nlink(inode, nlink_tid);
3780 pid_update_inode(task, inode);
3781
3782 return d_splice_alias_ops(inode, dentry, &pid_dentry_operations);
3783}
3784
3785static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3786{
3787 struct task_struct *task;
3788 struct task_struct *leader = get_proc_task(dir);
3789 unsigned tid;
3790 struct proc_fs_info *fs_info;
3791 struct pid_namespace *ns;
3792 struct dentry *result = ERR_PTR(-ENOENT);
3793
3794 if (!leader)
3795 goto out_no_task;
3796
3797 tid = name_to_int(&dentry->d_name);
3798 if (tid == ~0U)
3799 goto out;
3800
3801 fs_info = proc_sb_info(dentry->d_sb);
3802 ns = fs_info->pid_ns;
3803 rcu_read_lock();
3804 task = find_task_by_pid_ns(tid, ns);
3805 if (task)
3806 get_task_struct(task);
3807 rcu_read_unlock();
3808 if (!task)
3809 goto out;
3810 if (!same_thread_group(leader, task))
3811 goto out_drop_task;
3812
3813 result = proc_task_instantiate(dentry, task, NULL);
3814out_drop_task:
3815 put_task_struct(task);
3816out:
3817 put_task_struct(leader);
3818out_no_task:
3819 return result;
3820}
3821
3822/*
3823 * Find the first tid of a thread group to return to user space.
3824 *
3825 * Usually this is just the thread group leader, but if the users
3826 * buffer was too small or there was a seek into the middle of the
3827 * directory we have more work todo.
3828 *
3829 * In the case of a short read we start with find_task_by_pid.
3830 *
3831 * In the case of a seek we start with the leader and walk nr
3832 * threads past it.
3833 */
3834static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3835 struct pid_namespace *ns)
3836{
3837 struct task_struct *pos, *task;
3838 unsigned long nr = f_pos;
3839
3840 if (nr != f_pos) /* 32bit overflow? */
3841 return NULL;
3842
3843 rcu_read_lock();
3844 task = pid_task(pid, PIDTYPE_PID);
3845 if (!task)
3846 goto fail;
3847
3848 /* Attempt to start with the tid of a thread */
3849 if (tid && nr) {
3850 pos = find_task_by_pid_ns(tid, ns);
3851 if (pos && same_thread_group(pos, task))
3852 goto found;
3853 }
3854
3855 /* If nr exceeds the number of threads there is nothing todo */
3856 if (nr >= get_nr_threads(task))
3857 goto fail;
3858
3859 /* If we haven't found our starting place yet start
3860 * with the leader and walk nr threads forward.
3861 */
3862 for_each_thread(task, pos) {
3863 if (!nr--)
3864 goto found;
3865 }
3866fail:
3867 pos = NULL;
3868 goto out;
3869found:
3870 get_task_struct(pos);
3871out:
3872 rcu_read_unlock();
3873 return pos;
3874}
3875
3876/*
3877 * Find the next thread in the thread list.
3878 * Return NULL if there is an error or no next thread.
3879 *
3880 * The reference to the input task_struct is released.
3881 */
3882static struct task_struct *next_tid(struct task_struct *start)
3883{
3884 struct task_struct *pos = NULL;
3885 rcu_read_lock();
3886 if (pid_alive(start)) {
3887 pos = __next_thread(start);
3888 if (pos)
3889 get_task_struct(pos);
3890 }
3891 rcu_read_unlock();
3892 put_task_struct(start);
3893 return pos;
3894}
3895
3896/* for the /proc/TGID/task/ directories */
3897static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3898{
3899 struct inode *inode = file_inode(file);
3900 struct task_struct *task;
3901 struct pid_namespace *ns;
3902 int tid;
3903
3904 if (proc_inode_is_dead(inode))
3905 return -ENOENT;
3906
3907 if (!dir_emit_dots(file, ctx))
3908 return 0;
3909
3910 /* We cache the tgid value that the last readdir call couldn't
3911 * return and lseek resets it to 0.
3912 */
3913 ns = proc_pid_ns(inode->i_sb);
3914 tid = (int)(intptr_t)file->private_data;
3915 file->private_data = NULL;
3916 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3917 task;
3918 task = next_tid(task), ctx->pos++) {
3919 char name[10 + 1];
3920 unsigned int len;
3921
3922 tid = task_pid_nr_ns(task, ns);
3923 if (!tid)
3924 continue; /* The task has just exited. */
3925 len = snprintf(name, sizeof(name), "%d", tid);
3926 if (!proc_fill_cache(file, ctx, name, len,
3927 proc_task_instantiate, task, NULL)) {
3928 /* returning this tgid failed, save it as the first
3929 * pid for the next readir call */
3930 file->private_data = (void *)(intptr_t)tid;
3931 put_task_struct(task);
3932 break;
3933 }
3934 }
3935
3936 return 0;
3937}
3938
3939static int proc_task_getattr(struct mnt_idmap *idmap,
3940 const struct path *path, struct kstat *stat,
3941 u32 request_mask, unsigned int query_flags)
3942{
3943 struct inode *inode = d_inode(path->dentry);
3944 struct task_struct *p = get_proc_task(inode);
3945 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
3946
3947 if (p) {
3948 stat->nlink += get_nr_threads(p);
3949 put_task_struct(p);
3950 }
3951
3952 return 0;
3953}
3954
3955/*
3956 * proc_task_readdir() set @file->private_data to a positive integer
3957 * value, so casting that to u64 is safe. generic_llseek_cookie() will
3958 * set @cookie to 0, so casting to an int is safe. The WARN_ON_ONCE() is
3959 * here to catch any unexpected change in behavior either in
3960 * proc_task_readdir() or generic_llseek_cookie().
3961 */
3962static loff_t proc_dir_llseek(struct file *file, loff_t offset, int whence)
3963{
3964 u64 cookie = (u64)(intptr_t)file->private_data;
3965 loff_t off;
3966
3967 off = generic_llseek_cookie(file, offset, whence, &cookie);
3968 WARN_ON_ONCE(cookie > INT_MAX);
3969 file->private_data = (void *)(intptr_t)cookie; /* serialized by f_pos_lock */
3970 return off;
3971}
3972
3973static const struct inode_operations proc_task_inode_operations = {
3974 .lookup = proc_task_lookup,
3975 .getattr = proc_task_getattr,
3976 .setattr = proc_nochmod_setattr,
3977 .permission = proc_pid_permission,
3978};
3979
3980static const struct file_operations proc_task_operations = {
3981 .read = generic_read_dir,
3982 .iterate_shared = proc_task_readdir,
3983 .llseek = proc_dir_llseek,
3984};
3985
3986void __init set_proc_pid_nlink(void)
3987{
3988 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3989 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3990}