1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/proc/base.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * proc base directory handling functions
8 *
9 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10 * Instead of using magical inumbers to determine the kind of object
11 * we allocate and fill in-core inodes upon lookup. They don't even
12 * go into icache. We cache the reference to task_struct upon lookup too.
13 * Eventually it should become a filesystem in its own. We don't use the
14 * rest of procfs anymore.
15 *
16 *
17 * Changelog:
18 * 17-Jan-2005
19 * Allan Bezerra
20 * Bruna Moreira <bruna.moreira@indt.org.br>
21 * Edjard Mota <edjard.mota@indt.org.br>
22 * Ilias Biris <ilias.biris@indt.org.br>
23 * Mauricio Lin <mauricio.lin@indt.org.br>
24 *
25 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 *
27 * A new process specific entry (smaps) included in /proc. It shows the
28 * size of rss for each memory area. The maps entry lacks information
29 * about physical memory size (rss) for each mapped file, i.e.,
30 * rss information for executables and library files.
31 * This additional information is useful for any tools that need to know
32 * about physical memory consumption for a process specific library.
33 *
34 * Changelog:
35 * 21-Feb-2005
36 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37 * Pud inclusion in the page table walking.
38 *
39 * ChangeLog:
40 * 10-Mar-2005
41 * 10LE Instituto Nokia de Tecnologia - INdT:
42 * A better way to walks through the page table as suggested by Hugh Dickins.
43 *
44 * Simo Piiroinen <simo.piiroinen@nokia.com>:
45 * Smaps information related to shared, private, clean and dirty pages.
46 *
47 * Paul Mundt <paul.mundt@nokia.com>:
48 * Overall revision about smaps.
49 */
50
51#include <linux/uaccess.h>
52
53#include <linux/errno.h>
54#include <linux/time.h>
55#include <linux/proc_fs.h>
56#include <linux/stat.h>
57#include <linux/task_io_accounting_ops.h>
58#include <linux/init.h>
59#include <linux/capability.h>
60#include <linux/file.h>
61#include <linux/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/stat.h>
95#include <linux/posix-timers.h>
96#include <linux/time_namespace.h>
97#include <linux/resctrl.h>
98#include <linux/cn_proc.h>
99#include <linux/ksm.h>
100#include <uapi/linux/lsm.h>
101#include <trace/events/oom.h>
102#include "internal.h"
103#include "fd.h"
104
105#include "../../lib/kstrtox.h"
106
107/* NOTE:
108 * Implementing inode permission operations in /proc is almost
109 * certainly an error. Permission checks need to happen during
110 * each system call not at open time. The reason is that most of
111 * what we wish to check for permissions in /proc varies at runtime.
112 *
113 * The classic example of a problem is opening file descriptors
114 * in /proc for a task before it execs a suid executable.
115 */
116
117static u8 nlink_tid __ro_after_init;
118static u8 nlink_tgid __ro_after_init;
119
120enum proc_mem_force {
121 PROC_MEM_FORCE_ALWAYS,
122 PROC_MEM_FORCE_PTRACE,
123 PROC_MEM_FORCE_NEVER
124};
125
126static enum proc_mem_force proc_mem_force_override __ro_after_init =
127 IS_ENABLED(CONFIG_PROC_MEM_NO_FORCE) ? PROC_MEM_FORCE_NEVER :
128 IS_ENABLED(CONFIG_PROC_MEM_FORCE_PTRACE) ? PROC_MEM_FORCE_PTRACE :
129 PROC_MEM_FORCE_ALWAYS;
130
131static const struct constant_table proc_mem_force_table[] __initconst = {
132 { "always", PROC_MEM_FORCE_ALWAYS },
133 { .name: "ptrace", .value: PROC_MEM_FORCE_PTRACE },
134 { .name: "never", .value: PROC_MEM_FORCE_NEVER },
135 { }
136};
137
138static int __init early_proc_mem_force_override(char *buf)
139{
140 if (!buf)
141 return -EINVAL;
142
143 /*
144 * lookup_constant() defaults to proc_mem_force_override to preseve
145 * the initial Kconfig choice in case an invalid param gets passed.
146 */
147 proc_mem_force_override = lookup_constant(tbl: proc_mem_force_table,
148 name: buf, not_found: proc_mem_force_override);
149
150 return 0;
151}
152early_param("proc_mem.force_override", early_proc_mem_force_override);
153
154struct pid_entry {
155 const char *name;
156 unsigned int len;
157 umode_t mode;
158 const struct inode_operations *iop;
159 const struct file_operations *fop;
160 union proc_op op;
161};
162
163#define NOD(NAME, MODE, IOP, FOP, OP) { \
164 .name = (NAME), \
165 .len = sizeof(NAME) - 1, \
166 .mode = MODE, \
167 .iop = IOP, \
168 .fop = FOP, \
169 .op = OP, \
170}
171
172#define DIR(NAME, MODE, iops, fops) \
173 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
174#define LNK(NAME, get_link) \
175 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
176 &proc_pid_link_inode_operations, NULL, \
177 { .proc_get_link = get_link } )
178#define REG(NAME, MODE, fops) \
179 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
180#define ONE(NAME, MODE, show) \
181 NOD(NAME, (S_IFREG|(MODE)), \
182 NULL, &proc_single_file_operations, \
183 { .proc_show = show } )
184#define ATTR(LSMID, NAME, MODE) \
185 NOD(NAME, (S_IFREG|(MODE)), \
186 NULL, &proc_pid_attr_operations, \
187 { .lsmid = LSMID })
188
189/*
190 * Count the number of hardlinks for the pid_entry table, excluding the .
191 * and .. links.
192 */
193static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
194 unsigned int n)
195{
196 unsigned int i;
197 unsigned int count;
198
199 count = 2;
200 for (i = 0; i < n; ++i) {
201 if (S_ISDIR(entries[i].mode))
202 ++count;
203 }
204
205 return count;
206}
207
208static int get_task_root(struct task_struct *task, struct path *root)
209{
210 int result = -ENOENT;
211
212 task_lock(p: task);
213 if (task->fs) {
214 get_fs_root(fs: task->fs, root);
215 result = 0;
216 }
217 task_unlock(p: task);
218 return result;
219}
220
221static int proc_cwd_link(struct dentry *dentry, struct path *path)
222{
223 struct task_struct *task = get_proc_task(inode: d_inode(dentry));
224 int result = -ENOENT;
225
226 if (task) {
227 task_lock(p: task);
228 if (task->fs) {
229 get_fs_pwd(fs: task->fs, pwd: path);
230 result = 0;
231 }
232 task_unlock(p: task);
233 put_task_struct(t: task);
234 }
235 return result;
236}
237
238static int proc_root_link(struct dentry *dentry, struct path *path)
239{
240 struct task_struct *task = get_proc_task(inode: d_inode(dentry));
241 int result = -ENOENT;
242
243 if (task) {
244 result = get_task_root(task, root: path);
245 put_task_struct(t: task);
246 }
247 return result;
248}
249
250/*
251 * If the user used setproctitle(), we just get the string from
252 * user space at arg_start, and limit it to a maximum of one page.
253 */
254static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
255 size_t count, unsigned long pos,
256 unsigned long arg_start)
257{
258 char *page;
259 int ret, got;
260
261 if (pos >= PAGE_SIZE)
262 return 0;
263
264 page = (char *)__get_free_page(GFP_KERNEL);
265 if (!page)
266 return -ENOMEM;
267
268 ret = 0;
269 got = access_remote_vm(mm, addr: arg_start, buf: page, PAGE_SIZE, gup_flags: FOLL_ANON);
270 if (got > 0) {
271 int len = strnlen(page, got);
272
273 /* Include the NUL character if it was found */
274 if (len < got)
275 len++;
276
277 if (len > pos) {
278 len -= pos;
279 if (len > count)
280 len = count;
281 len -= copy_to_user(to: buf, from: page+pos, n: len);
282 if (!len)
283 len = -EFAULT;
284 ret = len;
285 }
286 }
287 free_page((unsigned long)page);
288 return ret;
289}
290
291static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
292 size_t count, loff_t *ppos)
293{
294 unsigned long arg_start, arg_end, env_start, env_end;
295 unsigned long pos, len;
296 char *page, c;
297
298 /* Check if process spawned far enough to have cmdline. */
299 if (!mm->env_end)
300 return 0;
301
302 spin_lock(lock: &mm->arg_lock);
303 arg_start = mm->arg_start;
304 arg_end = mm->arg_end;
305 env_start = mm->env_start;
306 env_end = mm->env_end;
307 spin_unlock(lock: &mm->arg_lock);
308
309 if (arg_start >= arg_end)
310 return 0;
311
312 /*
313 * We allow setproctitle() to overwrite the argument
314 * strings, and overflow past the original end. But
315 * only when it overflows into the environment area.
316 */
317 if (env_start != arg_end || env_end < env_start)
318 env_start = env_end = arg_end;
319 len = env_end - arg_start;
320
321 /* We're not going to care if "*ppos" has high bits set */
322 pos = *ppos;
323 if (pos >= len)
324 return 0;
325 if (count > len - pos)
326 count = len - pos;
327 if (!count)
328 return 0;
329
330 /*
331 * Magical special case: if the argv[] end byte is not
332 * zero, the user has overwritten it with setproctitle(3).
333 *
334 * Possible future enhancement: do this only once when
335 * pos is 0, and set a flag in the 'struct file'.
336 */
337 if (access_remote_vm(mm, addr: arg_end-1, buf: &c, len: 1, gup_flags: FOLL_ANON) == 1 && c)
338 return get_mm_proctitle(mm, buf, count, pos, arg_start);
339
340 /*
341 * For the non-setproctitle() case we limit things strictly
342 * to the [arg_start, arg_end[ range.
343 */
344 pos += arg_start;
345 if (pos < arg_start || pos >= arg_end)
346 return 0;
347 if (count > arg_end - pos)
348 count = arg_end - pos;
349
350 page = (char *)__get_free_page(GFP_KERNEL);
351 if (!page)
352 return -ENOMEM;
353
354 len = 0;
355 while (count) {
356 int got;
357 size_t size = min_t(size_t, PAGE_SIZE, count);
358
359 got = access_remote_vm(mm, addr: pos, buf: page, len: size, gup_flags: FOLL_ANON);
360 if (got <= 0)
361 break;
362 got -= copy_to_user(to: buf, from: page, n: got);
363 if (unlikely(!got)) {
364 if (!len)
365 len = -EFAULT;
366 break;
367 }
368 pos += got;
369 buf += got;
370 len += got;
371 count -= got;
372 }
373
374 free_page((unsigned long)page);
375 return len;
376}
377
378static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
379 size_t count, loff_t *pos)
380{
381 struct mm_struct *mm;
382 ssize_t ret;
383
384 mm = get_task_mm(task: tsk);
385 if (!mm)
386 return 0;
387
388 ret = get_mm_cmdline(mm, buf, count, ppos: pos);
389 mmput(mm);
390 return ret;
391}
392
393static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
394 size_t count, loff_t *pos)
395{
396 struct task_struct *tsk;
397 ssize_t ret;
398
399 BUG_ON(*pos < 0);
400
401 tsk = get_proc_task(inode: file_inode(f: file));
402 if (!tsk)
403 return -ESRCH;
404 ret = get_task_cmdline(tsk, buf, count, pos);
405 put_task_struct(t: tsk);
406 if (ret > 0)
407 *pos += ret;
408 return ret;
409}
410
411static const struct file_operations proc_pid_cmdline_ops = {
412 .read = proc_pid_cmdline_read,
413 .llseek = generic_file_llseek,
414};
415
416#ifdef CONFIG_KALLSYMS
417/*
418 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
419 * Returns the resolved symbol to user space.
420 */
421static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
422 struct pid *pid, struct task_struct *task)
423{
424 unsigned long wchan;
425 char symname[KSYM_NAME_LEN];
426
427 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
428 goto print0;
429
430 wchan = get_wchan(p: task);
431 if (wchan && !lookup_symbol_name(addr: wchan, symname)) {
432 seq_puts(m, s: symname);
433 return 0;
434 }
435
436print0:
437 seq_putc(m, c: '0');
438 return 0;
439}
440#endif /* CONFIG_KALLSYMS */
441
442static int lock_trace(struct task_struct *task)
443{
444 int err = down_read_killable(sem: &task->signal->exec_update_lock);
445 if (err)
446 return err;
447 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
448 up_read(sem: &task->signal->exec_update_lock);
449 return -EPERM;
450 }
451 return 0;
452}
453
454static void unlock_trace(struct task_struct *task)
455{
456 up_read(sem: &task->signal->exec_update_lock);
457}
458
459#ifdef CONFIG_STACKTRACE
460
461#define MAX_STACK_TRACE_DEPTH 64
462
463static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
464 struct pid *pid, struct task_struct *task)
465{
466 unsigned long *entries;
467 int err;
468
469 /*
470 * The ability to racily run the kernel stack unwinder on a running task
471 * and then observe the unwinder output is scary; while it is useful for
472 * debugging kernel issues, it can also allow an attacker to leak kernel
473 * stack contents.
474 * Doing this in a manner that is at least safe from races would require
475 * some work to ensure that the remote task can not be scheduled; and
476 * even then, this would still expose the unwinder as local attack
477 * surface.
478 * Therefore, this interface is restricted to root.
479 */
480 if (!file_ns_capable(file: m->file, ns: &init_user_ns, CAP_SYS_ADMIN))
481 return -EACCES;
482
483 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
484 GFP_KERNEL);
485 if (!entries)
486 return -ENOMEM;
487
488 err = lock_trace(task);
489 if (!err) {
490 unsigned int i, nr_entries;
491
492 nr_entries = stack_trace_save_tsk(task, store: entries,
493 MAX_STACK_TRACE_DEPTH, skipnr: 0);
494
495 for (i = 0; i < nr_entries; i++) {
496 seq_printf(m, fmt: "[<0>] %pB\n", (void *)entries[i]);
497 }
498
499 unlock_trace(task);
500 }
501 kfree(objp: entries);
502
503 return err;
504}
505#endif
506
507#ifdef CONFIG_SCHED_INFO
508/*
509 * Provides /proc/PID/schedstat
510 */
511static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
512 struct pid *pid, struct task_struct *task)
513{
514 if (unlikely(!sched_info_on()))
515 seq_puts(m, s: "0 0 0\n");
516 else
517 seq_printf(m, fmt: "%llu %llu %lu\n",
518 (unsigned long long)task->se.sum_exec_runtime,
519 (unsigned long long)task->sched_info.run_delay,
520 task->sched_info.pcount);
521
522 return 0;
523}
524#endif
525
526#ifdef CONFIG_LATENCYTOP
527static int lstats_show_proc(struct seq_file *m, void *v)
528{
529 int i;
530 struct inode *inode = m->private;
531 struct task_struct *task = get_proc_task(inode);
532
533 if (!task)
534 return -ESRCH;
535 seq_puts(m, "Latency Top version : v0.1\n");
536 for (i = 0; i < LT_SAVECOUNT; i++) {
537 struct latency_record *lr = &task->latency_record[i];
538 if (lr->backtrace[0]) {
539 int q;
540 seq_printf(m, "%i %li %li",
541 lr->count, lr->time, lr->max);
542 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
543 unsigned long bt = lr->backtrace[q];
544
545 if (!bt)
546 break;
547 seq_printf(m, " %ps", (void *)bt);
548 }
549 seq_putc(m, '\n');
550 }
551
552 }
553 put_task_struct(task);
554 return 0;
555}
556
557static int lstats_open(struct inode *inode, struct file *file)
558{
559 return single_open(file, lstats_show_proc, inode);
560}
561
562static ssize_t lstats_write(struct file *file, const char __user *buf,
563 size_t count, loff_t *offs)
564{
565 struct task_struct *task = get_proc_task(file_inode(file));
566
567 if (!task)
568 return -ESRCH;
569 clear_tsk_latency_tracing(task);
570 put_task_struct(task);
571
572 return count;
573}
574
575static const struct file_operations proc_lstats_operations = {
576 .open = lstats_open,
577 .read = seq_read,
578 .write = lstats_write,
579 .llseek = seq_lseek,
580 .release = single_release,
581};
582
583#endif
584
585static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
586 struct pid *pid, struct task_struct *task)
587{
588 unsigned long totalpages = totalram_pages() + total_swap_pages;
589 unsigned long points = 0;
590 long badness;
591
592 badness = oom_badness(p: task, totalpages);
593 /*
594 * Special case OOM_SCORE_ADJ_MIN for all others scale the
595 * badness value into [0, 2000] range which we have been
596 * exporting for a long time so userspace might depend on it.
597 */
598 if (badness != LONG_MIN)
599 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
600
601 seq_printf(m, fmt: "%lu\n", points);
602
603 return 0;
604}
605
606struct limit_names {
607 const char *name;
608 const char *unit;
609};
610
611static const struct limit_names lnames[RLIM_NLIMITS] = {
612 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
613 [RLIMIT_FSIZE] = {.name: "Max file size", .unit: "bytes"},
614 [RLIMIT_DATA] = {.name: "Max data size", .unit: "bytes"},
615 [RLIMIT_STACK] = {.name: "Max stack size", .unit: "bytes"},
616 [RLIMIT_CORE] = {.name: "Max core file size", .unit: "bytes"},
617 [RLIMIT_RSS] = {.name: "Max resident set", .unit: "bytes"},
618 [RLIMIT_NPROC] = {.name: "Max processes", .unit: "processes"},
619 [RLIMIT_NOFILE] = {.name: "Max open files", .unit: "files"},
620 [RLIMIT_MEMLOCK] = {.name: "Max locked memory", .unit: "bytes"},
621 [RLIMIT_AS] = {.name: "Max address space", .unit: "bytes"},
622 [RLIMIT_LOCKS] = {.name: "Max file locks", .unit: "locks"},
623 [RLIMIT_SIGPENDING] = {.name: "Max pending signals", .unit: "signals"},
624 [RLIMIT_MSGQUEUE] = {.name: "Max msgqueue size", .unit: "bytes"},
625 [RLIMIT_NICE] = {.name: "Max nice priority", NULL},
626 [RLIMIT_RTPRIO] = {.name: "Max realtime priority", NULL},
627 [RLIMIT_RTTIME] = {.name: "Max realtime timeout", .unit: "us"},
628};
629
630/* Display limits for a process */
631static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
632 struct pid *pid, struct task_struct *task)
633{
634 unsigned int i;
635 unsigned long flags;
636
637 struct rlimit rlim[RLIM_NLIMITS];
638
639 if (!lock_task_sighand(task, flags: &flags))
640 return 0;
641 memcpy(to: rlim, from: task->signal->rlim, len: sizeof(struct rlimit) * RLIM_NLIMITS);
642 unlock_task_sighand(task, flags: &flags);
643
644 /*
645 * print the file header
646 */
647 seq_puts(m, s: "Limit "
648 "Soft Limit "
649 "Hard Limit "
650 "Units \n");
651
652 for (i = 0; i < RLIM_NLIMITS; i++) {
653 if (rlim[i].rlim_cur == RLIM_INFINITY)
654 seq_printf(m, fmt: "%-25s %-20s ",
655 lnames[i].name, "unlimited");
656 else
657 seq_printf(m, fmt: "%-25s %-20lu ",
658 lnames[i].name, rlim[i].rlim_cur);
659
660 if (rlim[i].rlim_max == RLIM_INFINITY)
661 seq_printf(m, fmt: "%-20s ", "unlimited");
662 else
663 seq_printf(m, fmt: "%-20lu ", rlim[i].rlim_max);
664
665 if (lnames[i].unit)
666 seq_printf(m, fmt: "%-10s\n", lnames[i].unit);
667 else
668 seq_putc(m, c: '\n');
669 }
670
671 return 0;
672}
673
674#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
675static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
676 struct pid *pid, struct task_struct *task)
677{
678 struct syscall_info info;
679 u64 *args = &info.data.args[0];
680 int res;
681
682 res = lock_trace(task);
683 if (res)
684 return res;
685
686 if (task_current_syscall(target: task, info: &info))
687 seq_puts(m, s: "running\n");
688 else if (info.data.nr < 0)
689 seq_printf(m, fmt: "%d 0x%llx 0x%llx\n",
690 info.data.nr, info.sp, info.data.instruction_pointer);
691 else
692 seq_printf(m,
693 fmt: "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
694 info.data.nr,
695 args[0], args[1], args[2], args[3], args[4], args[5],
696 info.sp, info.data.instruction_pointer);
697 unlock_trace(task);
698
699 return 0;
700}
701#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
702
703/************************************************************************/
704/* Here the fs part begins */
705/************************************************************************/
706
707/* permission checks */
708static bool proc_fd_access_allowed(struct inode *inode)
709{
710 struct task_struct *task;
711 bool allowed = false;
712 /* Allow access to a task's file descriptors if it is us or we
713 * may use ptrace attach to the process and find out that
714 * information.
715 */
716 task = get_proc_task(inode);
717 if (task) {
718 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
719 put_task_struct(t: task);
720 }
721 return allowed;
722}
723
724int proc_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
725 struct iattr *attr)
726{
727 int error;
728 struct inode *inode = d_inode(dentry);
729
730 if (attr->ia_valid & ATTR_MODE)
731 return -EPERM;
732
733 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
734 if (error)
735 return error;
736
737 setattr_copy(&nop_mnt_idmap, inode, attr);
738 return 0;
739}
740
741/*
742 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
743 * or euid/egid (for hide_pid_min=2)?
744 */
745static bool has_pid_permissions(struct proc_fs_info *fs_info,
746 struct task_struct *task,
747 enum proc_hidepid hide_pid_min)
748{
749 /*
750 * If 'hidpid' mount option is set force a ptrace check,
751 * we indicate that we are using a filesystem syscall
752 * by passing PTRACE_MODE_READ_FSCREDS
753 */
754 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
755 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
756
757 if (fs_info->hide_pid < hide_pid_min)
758 return true;
759 if (in_group_p(fs_info->pid_gid))
760 return true;
761 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
762}
763
764
765static int proc_pid_permission(struct mnt_idmap *idmap,
766 struct inode *inode, int mask)
767{
768 struct proc_fs_info *fs_info = proc_sb_info(sb: inode->i_sb);
769 struct task_struct *task;
770 bool has_perms;
771
772 task = get_proc_task(inode);
773 if (!task)
774 return -ESRCH;
775 has_perms = has_pid_permissions(fs_info, task, hide_pid_min: HIDEPID_NO_ACCESS);
776 put_task_struct(t: task);
777
778 if (!has_perms) {
779 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
780 /*
781 * Let's make getdents(), stat(), and open()
782 * consistent with each other. If a process
783 * may not stat() a file, it shouldn't be seen
784 * in procfs at all.
785 */
786 return -ENOENT;
787 }
788
789 return -EPERM;
790 }
791 return generic_permission(&nop_mnt_idmap, inode, mask);
792}
793
794
795
796static const struct inode_operations proc_def_inode_operations = {
797 .setattr = proc_setattr,
798};
799
800static int proc_single_show(struct seq_file *m, void *v)
801{
802 struct inode *inode = m->private;
803 struct pid_namespace *ns = proc_pid_ns(sb: inode->i_sb);
804 struct pid *pid = proc_pid(inode);
805 struct task_struct *task;
806 int ret;
807
808 task = get_pid_task(pid, PIDTYPE_PID);
809 if (!task)
810 return -ESRCH;
811
812 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
813
814 put_task_struct(t: task);
815 return ret;
816}
817
818static int proc_single_open(struct inode *inode, struct file *filp)
819{
820 return single_open(filp, proc_single_show, inode);
821}
822
823static const struct file_operations proc_single_file_operations = {
824 .open = proc_single_open,
825 .read = seq_read,
826 .llseek = seq_lseek,
827 .release = single_release,
828};
829
830/*
831 * proc_mem_open() can return errno, NULL or mm_struct*.
832 *
833 * - Returns NULL if the task has no mm (PF_KTHREAD or PF_EXITING)
834 * - Returns mm_struct* on success
835 * - Returns error code on failure
836 */
837struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
838{
839 struct task_struct *task = get_proc_task(inode);
840 struct mm_struct *mm;
841
842 if (!task)
843 return ERR_PTR(error: -ESRCH);
844
845 mm = mm_access(task, mode: mode | PTRACE_MODE_FSCREDS);
846 put_task_struct(t: task);
847
848 if (IS_ERR(ptr: mm))
849 return mm == ERR_PTR(error: -ESRCH) ? NULL : mm;
850
851 /* ensure this mm_struct can't be freed */
852 mmgrab(mm);
853 /* but do not pin its memory */
854 mmput(mm);
855
856 return mm;
857}
858
859static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
860{
861 struct mm_struct *mm = proc_mem_open(inode, mode);
862
863 if (IS_ERR_OR_NULL(ptr: mm))
864 return mm ? PTR_ERR(ptr: mm) : -ESRCH;
865
866 file->private_data = mm;
867 return 0;
868}
869
870static int mem_open(struct inode *inode, struct file *file)
871{
872 if (WARN_ON_ONCE(!(file->f_op->fop_flags & FOP_UNSIGNED_OFFSET)))
873 return -EINVAL;
874 return __mem_open(inode, file, PTRACE_MODE_ATTACH);
875}
876
877static bool proc_mem_foll_force(struct file *file, struct mm_struct *mm)
878{
879 struct task_struct *task;
880 bool ptrace_active = false;
881
882 switch (proc_mem_force_override) {
883 case PROC_MEM_FORCE_NEVER:
884 return false;
885 case PROC_MEM_FORCE_PTRACE:
886 task = get_proc_task(inode: file_inode(f: file));
887 if (task) {
888 ptrace_active = READ_ONCE(task->ptrace) &&
889 READ_ONCE(task->mm) == mm &&
890 READ_ONCE(task->parent) == current;
891 put_task_struct(t: task);
892 }
893 return ptrace_active;
894 default:
895 return true;
896 }
897}
898
899static ssize_t mem_rw(struct file *file, char __user *buf,
900 size_t count, loff_t *ppos, int write)
901{
902 struct mm_struct *mm = file->private_data;
903 unsigned long addr = *ppos;
904 ssize_t copied;
905 char *page;
906 unsigned int flags;
907
908 if (!mm)
909 return 0;
910
911 page = (char *)__get_free_page(GFP_KERNEL);
912 if (!page)
913 return -ENOMEM;
914
915 copied = 0;
916 if (!mmget_not_zero(mm))
917 goto free;
918
919 flags = write ? FOLL_WRITE : 0;
920 if (proc_mem_foll_force(file, mm))
921 flags |= FOLL_FORCE;
922
923 while (count > 0) {
924 size_t this_len = min_t(size_t, count, PAGE_SIZE);
925
926 if (write && copy_from_user(to: page, from: buf, n: this_len)) {
927 copied = -EFAULT;
928 break;
929 }
930
931 this_len = access_remote_vm(mm, addr, buf: page, len: this_len, gup_flags: flags);
932 if (!this_len) {
933 if (!copied)
934 copied = -EIO;
935 break;
936 }
937
938 if (!write && copy_to_user(to: buf, from: page, n: this_len)) {
939 copied = -EFAULT;
940 break;
941 }
942
943 buf += this_len;
944 addr += this_len;
945 copied += this_len;
946 count -= this_len;
947 }
948 *ppos = addr;
949
950 mmput(mm);
951free:
952 free_page((unsigned long) page);
953 return copied;
954}
955
956static ssize_t mem_read(struct file *file, char __user *buf,
957 size_t count, loff_t *ppos)
958{
959 return mem_rw(file, buf, count, ppos, write: 0);
960}
961
962static ssize_t mem_write(struct file *file, const char __user *buf,
963 size_t count, loff_t *ppos)
964{
965 return mem_rw(file, buf: (char __user*)buf, count, ppos, write: 1);
966}
967
968loff_t mem_lseek(struct file *file, loff_t offset, int orig)
969{
970 switch (orig) {
971 case 0:
972 file->f_pos = offset;
973 break;
974 case 1:
975 file->f_pos += offset;
976 break;
977 default:
978 return -EINVAL;
979 }
980 force_successful_syscall_return();
981 return file->f_pos;
982}
983
984static int mem_release(struct inode *inode, struct file *file)
985{
986 struct mm_struct *mm = file->private_data;
987 if (mm)
988 mmdrop(mm);
989 return 0;
990}
991
992static const struct file_operations proc_mem_operations = {
993 .llseek = mem_lseek,
994 .read = mem_read,
995 .write = mem_write,
996 .open = mem_open,
997 .release = mem_release,
998 .fop_flags = FOP_UNSIGNED_OFFSET,
999};
1000
1001static int environ_open(struct inode *inode, struct file *file)
1002{
1003 return __mem_open(inode, file, PTRACE_MODE_READ);
1004}
1005
1006static ssize_t environ_read(struct file *file, char __user *buf,
1007 size_t count, loff_t *ppos)
1008{
1009 char *page;
1010 unsigned long src = *ppos;
1011 int ret = 0;
1012 struct mm_struct *mm = file->private_data;
1013 unsigned long env_start, env_end;
1014
1015 /* Ensure the process spawned far enough to have an environment. */
1016 if (!mm || !mm->env_end)
1017 return 0;
1018
1019 page = (char *)__get_free_page(GFP_KERNEL);
1020 if (!page)
1021 return -ENOMEM;
1022
1023 ret = 0;
1024 if (!mmget_not_zero(mm))
1025 goto free;
1026
1027 spin_lock(lock: &mm->arg_lock);
1028 env_start = mm->env_start;
1029 env_end = mm->env_end;
1030 spin_unlock(lock: &mm->arg_lock);
1031
1032 while (count > 0) {
1033 size_t this_len, max_len;
1034 int retval;
1035
1036 if (src >= (env_end - env_start))
1037 break;
1038
1039 this_len = env_end - (env_start + src);
1040
1041 max_len = min_t(size_t, PAGE_SIZE, count);
1042 this_len = min(max_len, this_len);
1043
1044 retval = access_remote_vm(mm, addr: (env_start + src), buf: page, len: this_len, gup_flags: FOLL_ANON);
1045
1046 if (retval <= 0) {
1047 ret = retval;
1048 break;
1049 }
1050
1051 if (copy_to_user(to: buf, from: page, n: retval)) {
1052 ret = -EFAULT;
1053 break;
1054 }
1055
1056 ret += retval;
1057 src += retval;
1058 buf += retval;
1059 count -= retval;
1060 }
1061 *ppos = src;
1062 mmput(mm);
1063
1064free:
1065 free_page((unsigned long) page);
1066 return ret;
1067}
1068
1069static const struct file_operations proc_environ_operations = {
1070 .open = environ_open,
1071 .read = environ_read,
1072 .llseek = generic_file_llseek,
1073 .release = mem_release,
1074};
1075
1076static int auxv_open(struct inode *inode, struct file *file)
1077{
1078 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1079}
1080
1081static ssize_t auxv_read(struct file *file, char __user *buf,
1082 size_t count, loff_t *ppos)
1083{
1084 struct mm_struct *mm = file->private_data;
1085 unsigned int nwords = 0;
1086
1087 if (!mm)
1088 return 0;
1089 do {
1090 nwords += 2;
1091 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1092 return simple_read_from_buffer(to: buf, count, ppos, from: mm->saved_auxv,
1093 available: nwords * sizeof(mm->saved_auxv[0]));
1094}
1095
1096static const struct file_operations proc_auxv_operations = {
1097 .open = auxv_open,
1098 .read = auxv_read,
1099 .llseek = generic_file_llseek,
1100 .release = mem_release,
1101};
1102
1103static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1104 loff_t *ppos)
1105{
1106 struct task_struct *task = get_proc_task(inode: file_inode(f: file));
1107 char buffer[PROC_NUMBUF];
1108 int oom_adj = OOM_ADJUST_MIN;
1109 size_t len;
1110
1111 if (!task)
1112 return -ESRCH;
1113 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1114 oom_adj = OOM_ADJUST_MAX;
1115 else
1116 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1117 OOM_SCORE_ADJ_MAX;
1118 put_task_struct(t: task);
1119 if (oom_adj > OOM_ADJUST_MAX)
1120 oom_adj = OOM_ADJUST_MAX;
1121 len = snprintf(buf: buffer, size: sizeof(buffer), fmt: "%d\n", oom_adj);
1122 return simple_read_from_buffer(to: buf, count, ppos, from: buffer, available: len);
1123}
1124
1125static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1126{
1127 struct mm_struct *mm = NULL;
1128 struct task_struct *task;
1129 int err = 0;
1130
1131 task = get_proc_task(inode: file_inode(f: file));
1132 if (!task)
1133 return -ESRCH;
1134
1135 mutex_lock(lock: &oom_adj_mutex);
1136 if (legacy) {
1137 if (oom_adj < task->signal->oom_score_adj &&
1138 !capable(CAP_SYS_RESOURCE)) {
1139 err = -EACCES;
1140 goto err_unlock;
1141 }
1142 /*
1143 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1144 * /proc/pid/oom_score_adj instead.
1145 */
1146 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1147 current->comm, task_pid_nr(current), task_pid_nr(task),
1148 task_pid_nr(task));
1149 } else {
1150 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1151 !capable(CAP_SYS_RESOURCE)) {
1152 err = -EACCES;
1153 goto err_unlock;
1154 }
1155 }
1156
1157 /*
1158 * Make sure we will check other processes sharing the mm if this is
1159 * not vfrok which wants its own oom_score_adj.
1160 * pin the mm so it doesn't go away and get reused after task_unlock
1161 */
1162 if (!task->vfork_done) {
1163 struct task_struct *p = find_lock_task_mm(p: task);
1164
1165 if (p) {
1166 if (mm_flags_test(MMF_MULTIPROCESS, mm: p->mm)) {
1167 mm = p->mm;
1168 mmgrab(mm);
1169 }
1170 task_unlock(p);
1171 }
1172 }
1173
1174 task->signal->oom_score_adj = oom_adj;
1175 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1176 task->signal->oom_score_adj_min = (short)oom_adj;
1177 trace_oom_score_adj_update(task);
1178
1179 if (mm) {
1180 struct task_struct *p;
1181
1182 rcu_read_lock();
1183 for_each_process(p) {
1184 if (same_thread_group(p1: task, p2: p))
1185 continue;
1186
1187 /* do not touch kernel threads or the global init */
1188 if (p->flags & PF_KTHREAD || is_global_init(tsk: p))
1189 continue;
1190
1191 task_lock(p);
1192 if (!p->vfork_done && process_shares_mm(p, mm)) {
1193 p->signal->oom_score_adj = oom_adj;
1194 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1195 p->signal->oom_score_adj_min = (short)oom_adj;
1196 }
1197 task_unlock(p);
1198 }
1199 rcu_read_unlock();
1200 mmdrop(mm);
1201 }
1202err_unlock:
1203 mutex_unlock(lock: &oom_adj_mutex);
1204 put_task_struct(t: task);
1205 return err;
1206}
1207
1208/*
1209 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1210 * kernels. The effective policy is defined by oom_score_adj, which has a
1211 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1212 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1213 * Processes that become oom disabled via oom_adj will still be oom disabled
1214 * with this implementation.
1215 *
1216 * oom_adj cannot be removed since existing userspace binaries use it.
1217 */
1218static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1219 size_t count, loff_t *ppos)
1220{
1221 char buffer[PROC_NUMBUF] = {};
1222 int oom_adj;
1223 int err;
1224
1225 if (count > sizeof(buffer) - 1)
1226 count = sizeof(buffer) - 1;
1227 if (copy_from_user(to: buffer, from: buf, n: count)) {
1228 err = -EFAULT;
1229 goto out;
1230 }
1231
1232 err = kstrtoint(s: strstrip(str: buffer), base: 0, res: &oom_adj);
1233 if (err)
1234 goto out;
1235 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1236 oom_adj != OOM_DISABLE) {
1237 err = -EINVAL;
1238 goto out;
1239 }
1240
1241 /*
1242 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1243 * value is always attainable.
1244 */
1245 if (oom_adj == OOM_ADJUST_MAX)
1246 oom_adj = OOM_SCORE_ADJ_MAX;
1247 else
1248 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1249
1250 err = __set_oom_adj(file, oom_adj, legacy: true);
1251out:
1252 return err < 0 ? err : count;
1253}
1254
1255static const struct file_operations proc_oom_adj_operations = {
1256 .read = oom_adj_read,
1257 .write = oom_adj_write,
1258 .llseek = generic_file_llseek,
1259};
1260
1261static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1262 size_t count, loff_t *ppos)
1263{
1264 struct task_struct *task = get_proc_task(inode: file_inode(f: file));
1265 char buffer[PROC_NUMBUF];
1266 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1267 size_t len;
1268
1269 if (!task)
1270 return -ESRCH;
1271 oom_score_adj = task->signal->oom_score_adj;
1272 put_task_struct(t: task);
1273 len = snprintf(buf: buffer, size: sizeof(buffer), fmt: "%hd\n", oom_score_adj);
1274 return simple_read_from_buffer(to: buf, count, ppos, from: buffer, available: len);
1275}
1276
1277static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1278 size_t count, loff_t *ppos)
1279{
1280 char buffer[PROC_NUMBUF] = {};
1281 int oom_score_adj;
1282 int err;
1283
1284 if (count > sizeof(buffer) - 1)
1285 count = sizeof(buffer) - 1;
1286 if (copy_from_user(to: buffer, from: buf, n: count)) {
1287 err = -EFAULT;
1288 goto out;
1289 }
1290
1291 err = kstrtoint(s: strstrip(str: buffer), base: 0, res: &oom_score_adj);
1292 if (err)
1293 goto out;
1294 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1295 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1296 err = -EINVAL;
1297 goto out;
1298 }
1299
1300 err = __set_oom_adj(file, oom_adj: oom_score_adj, legacy: false);
1301out:
1302 return err < 0 ? err : count;
1303}
1304
1305static const struct file_operations proc_oom_score_adj_operations = {
1306 .read = oom_score_adj_read,
1307 .write = oom_score_adj_write,
1308 .llseek = default_llseek,
1309};
1310
1311#ifdef CONFIG_AUDIT
1312#define TMPBUFLEN 11
1313static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1314 size_t count, loff_t *ppos)
1315{
1316 struct inode * inode = file_inode(f: file);
1317 struct task_struct *task = get_proc_task(inode);
1318 ssize_t length;
1319 char tmpbuf[TMPBUFLEN];
1320
1321 if (!task)
1322 return -ESRCH;
1323 length = scnprintf(buf: tmpbuf, TMPBUFLEN, fmt: "%u",
1324 from_kuid(to: file->f_cred->user_ns,
1325 kuid: audit_get_loginuid(tsk: task)));
1326 put_task_struct(t: task);
1327 return simple_read_from_buffer(to: buf, count, ppos, from: tmpbuf, available: length);
1328}
1329
1330static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1331 size_t count, loff_t *ppos)
1332{
1333 struct inode * inode = file_inode(f: file);
1334 uid_t loginuid;
1335 kuid_t kloginuid;
1336 int rv;
1337
1338 /* Don't let kthreads write their own loginuid */
1339 if (current->flags & PF_KTHREAD)
1340 return -EPERM;
1341
1342 rcu_read_lock();
1343 if (current != pid_task(pid: proc_pid(inode), PIDTYPE_PID)) {
1344 rcu_read_unlock();
1345 return -EPERM;
1346 }
1347 rcu_read_unlock();
1348
1349 if (*ppos != 0) {
1350 /* No partial writes. */
1351 return -EINVAL;
1352 }
1353
1354 rv = kstrtou32_from_user(s: buf, count, base: 10, res: &loginuid);
1355 if (rv < 0)
1356 return rv;
1357
1358 /* is userspace tring to explicitly UNSET the loginuid? */
1359 if (loginuid == AUDIT_UID_UNSET) {
1360 kloginuid = INVALID_UID;
1361 } else {
1362 kloginuid = make_kuid(from: file->f_cred->user_ns, uid: loginuid);
1363 if (!uid_valid(uid: kloginuid))
1364 return -EINVAL;
1365 }
1366
1367 rv = audit_set_loginuid(loginuid: kloginuid);
1368 if (rv < 0)
1369 return rv;
1370 return count;
1371}
1372
1373static const struct file_operations proc_loginuid_operations = {
1374 .read = proc_loginuid_read,
1375 .write = proc_loginuid_write,
1376 .llseek = generic_file_llseek,
1377};
1378
1379static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1380 size_t count, loff_t *ppos)
1381{
1382 struct inode * inode = file_inode(f: file);
1383 struct task_struct *task = get_proc_task(inode);
1384 ssize_t length;
1385 char tmpbuf[TMPBUFLEN];
1386
1387 if (!task)
1388 return -ESRCH;
1389 length = scnprintf(buf: tmpbuf, TMPBUFLEN, fmt: "%u",
1390 audit_get_sessionid(tsk: task));
1391 put_task_struct(t: task);
1392 return simple_read_from_buffer(to: buf, count, ppos, from: tmpbuf, available: length);
1393}
1394
1395static const struct file_operations proc_sessionid_operations = {
1396 .read = proc_sessionid_read,
1397 .llseek = generic_file_llseek,
1398};
1399#endif
1400
1401#ifdef CONFIG_FAULT_INJECTION
1402static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1403 size_t count, loff_t *ppos)
1404{
1405 struct task_struct *task = get_proc_task(file_inode(file));
1406 char buffer[PROC_NUMBUF];
1407 size_t len;
1408 int make_it_fail;
1409
1410 if (!task)
1411 return -ESRCH;
1412 make_it_fail = task->make_it_fail;
1413 put_task_struct(task);
1414
1415 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1416
1417 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1418}
1419
1420static ssize_t proc_fault_inject_write(struct file * file,
1421 const char __user * buf, size_t count, loff_t *ppos)
1422{
1423 struct task_struct *task;
1424 char buffer[PROC_NUMBUF] = {};
1425 int make_it_fail;
1426 int rv;
1427
1428 if (!capable(CAP_SYS_RESOURCE))
1429 return -EPERM;
1430
1431 if (count > sizeof(buffer) - 1)
1432 count = sizeof(buffer) - 1;
1433 if (copy_from_user(buffer, buf, count))
1434 return -EFAULT;
1435 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1436 if (rv < 0)
1437 return rv;
1438 if (make_it_fail < 0 || make_it_fail > 1)
1439 return -EINVAL;
1440
1441 task = get_proc_task(file_inode(file));
1442 if (!task)
1443 return -ESRCH;
1444 task->make_it_fail = make_it_fail;
1445 put_task_struct(task);
1446
1447 return count;
1448}
1449
1450static const struct file_operations proc_fault_inject_operations = {
1451 .read = proc_fault_inject_read,
1452 .write = proc_fault_inject_write,
1453 .llseek = generic_file_llseek,
1454};
1455
1456static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1457 size_t count, loff_t *ppos)
1458{
1459 struct task_struct *task;
1460 int err;
1461 unsigned int n;
1462
1463 err = kstrtouint_from_user(buf, count, 0, &n);
1464 if (err)
1465 return err;
1466
1467 task = get_proc_task(file_inode(file));
1468 if (!task)
1469 return -ESRCH;
1470 task->fail_nth = n;
1471 put_task_struct(task);
1472
1473 return count;
1474}
1475
1476static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1477 size_t count, loff_t *ppos)
1478{
1479 struct task_struct *task;
1480 char numbuf[PROC_NUMBUF];
1481 ssize_t len;
1482
1483 task = get_proc_task(file_inode(file));
1484 if (!task)
1485 return -ESRCH;
1486 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1487 put_task_struct(task);
1488 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1489}
1490
1491static const struct file_operations proc_fail_nth_operations = {
1492 .read = proc_fail_nth_read,
1493 .write = proc_fail_nth_write,
1494};
1495#endif
1496
1497
1498/*
1499 * Print out various scheduling related per-task fields:
1500 */
1501static int sched_show(struct seq_file *m, void *v)
1502{
1503 struct inode *inode = m->private;
1504 struct pid_namespace *ns = proc_pid_ns(sb: inode->i_sb);
1505 struct task_struct *p;
1506
1507 p = get_proc_task(inode);
1508 if (!p)
1509 return -ESRCH;
1510 proc_sched_show_task(p, ns, m);
1511
1512 put_task_struct(t: p);
1513
1514 return 0;
1515}
1516
1517static ssize_t
1518sched_write(struct file *file, const char __user *buf,
1519 size_t count, loff_t *offset)
1520{
1521 struct inode *inode = file_inode(f: file);
1522 struct task_struct *p;
1523
1524 p = get_proc_task(inode);
1525 if (!p)
1526 return -ESRCH;
1527 proc_sched_set_task(p);
1528
1529 put_task_struct(t: p);
1530
1531 return count;
1532}
1533
1534static int sched_open(struct inode *inode, struct file *filp)
1535{
1536 return single_open(filp, sched_show, inode);
1537}
1538
1539static const struct file_operations proc_pid_sched_operations = {
1540 .open = sched_open,
1541 .read = seq_read,
1542 .write = sched_write,
1543 .llseek = seq_lseek,
1544 .release = single_release,
1545};
1546
1547#ifdef CONFIG_SCHED_AUTOGROUP
1548/*
1549 * Print out autogroup related information:
1550 */
1551static int sched_autogroup_show(struct seq_file *m, void *v)
1552{
1553 struct inode *inode = m->private;
1554 struct task_struct *p;
1555
1556 p = get_proc_task(inode);
1557 if (!p)
1558 return -ESRCH;
1559 proc_sched_autogroup_show_task(p, m);
1560
1561 put_task_struct(p);
1562
1563 return 0;
1564}
1565
1566static ssize_t
1567sched_autogroup_write(struct file *file, const char __user *buf,
1568 size_t count, loff_t *offset)
1569{
1570 struct inode *inode = file_inode(file);
1571 struct task_struct *p;
1572 char buffer[PROC_NUMBUF] = {};
1573 int nice;
1574 int err;
1575
1576 if (count > sizeof(buffer) - 1)
1577 count = sizeof(buffer) - 1;
1578 if (copy_from_user(buffer, buf, count))
1579 return -EFAULT;
1580
1581 err = kstrtoint(strstrip(buffer), 0, &nice);
1582 if (err < 0)
1583 return err;
1584
1585 p = get_proc_task(inode);
1586 if (!p)
1587 return -ESRCH;
1588
1589 err = proc_sched_autogroup_set_nice(p, nice);
1590 if (err)
1591 count = err;
1592
1593 put_task_struct(p);
1594
1595 return count;
1596}
1597
1598static int sched_autogroup_open(struct inode *inode, struct file *filp)
1599{
1600 int ret;
1601
1602 ret = single_open(filp, sched_autogroup_show, NULL);
1603 if (!ret) {
1604 struct seq_file *m = filp->private_data;
1605
1606 m->private = inode;
1607 }
1608 return ret;
1609}
1610
1611static const struct file_operations proc_pid_sched_autogroup_operations = {
1612 .open = sched_autogroup_open,
1613 .read = seq_read,
1614 .write = sched_autogroup_write,
1615 .llseek = seq_lseek,
1616 .release = single_release,
1617};
1618
1619#endif /* CONFIG_SCHED_AUTOGROUP */
1620
1621#ifdef CONFIG_TIME_NS
1622static int timens_offsets_show(struct seq_file *m, void *v)
1623{
1624 struct task_struct *p;
1625
1626 p = get_proc_task(inode: file_inode(f: m->file));
1627 if (!p)
1628 return -ESRCH;
1629 proc_timens_show_offsets(p, m);
1630
1631 put_task_struct(t: p);
1632
1633 return 0;
1634}
1635
1636static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1637 size_t count, loff_t *ppos)
1638{
1639 struct inode *inode = file_inode(f: file);
1640 struct proc_timens_offset offsets[2];
1641 char *kbuf = NULL, *pos, *next_line;
1642 struct task_struct *p;
1643 int ret, noffsets;
1644
1645 /* Only allow < page size writes at the beginning of the file */
1646 if ((*ppos != 0) || (count >= PAGE_SIZE))
1647 return -EINVAL;
1648
1649 /* Slurp in the user data */
1650 kbuf = memdup_user_nul(buf, count);
1651 if (IS_ERR(ptr: kbuf))
1652 return PTR_ERR(ptr: kbuf);
1653
1654 /* Parse the user data */
1655 ret = -EINVAL;
1656 noffsets = 0;
1657 for (pos = kbuf; pos; pos = next_line) {
1658 struct proc_timens_offset *off = &offsets[noffsets];
1659 char clock[10];
1660 int err;
1661
1662 /* Find the end of line and ensure we don't look past it */
1663 next_line = strchr(pos, '\n');
1664 if (next_line) {
1665 *next_line = '\0';
1666 next_line++;
1667 if (*next_line == '\0')
1668 next_line = NULL;
1669 }
1670
1671 err = sscanf(pos, "%9s %lld %lu", clock,
1672 &off->val.tv_sec, &off->val.tv_nsec);
1673 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1674 goto out;
1675
1676 clock[sizeof(clock) - 1] = 0;
1677 if (strcmp(clock, "monotonic") == 0 ||
1678 strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1679 off->clockid = CLOCK_MONOTONIC;
1680 else if (strcmp(clock, "boottime") == 0 ||
1681 strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1682 off->clockid = CLOCK_BOOTTIME;
1683 else
1684 goto out;
1685
1686 noffsets++;
1687 if (noffsets == ARRAY_SIZE(offsets)) {
1688 if (next_line)
1689 count = next_line - kbuf;
1690 break;
1691 }
1692 }
1693
1694 ret = -ESRCH;
1695 p = get_proc_task(inode);
1696 if (!p)
1697 goto out;
1698 ret = proc_timens_set_offset(file, p, offsets, n: noffsets);
1699 put_task_struct(t: p);
1700 if (ret)
1701 goto out;
1702
1703 ret = count;
1704out:
1705 kfree(objp: kbuf);
1706 return ret;
1707}
1708
1709static int timens_offsets_open(struct inode *inode, struct file *filp)
1710{
1711 return single_open(filp, timens_offsets_show, inode);
1712}
1713
1714static const struct file_operations proc_timens_offsets_operations = {
1715 .open = timens_offsets_open,
1716 .read = seq_read,
1717 .write = timens_offsets_write,
1718 .llseek = seq_lseek,
1719 .release = single_release,
1720};
1721#endif /* CONFIG_TIME_NS */
1722
1723static ssize_t comm_write(struct file *file, const char __user *buf,
1724 size_t count, loff_t *offset)
1725{
1726 struct inode *inode = file_inode(f: file);
1727 struct task_struct *p;
1728 char buffer[TASK_COMM_LEN] = {};
1729 const size_t maxlen = sizeof(buffer) - 1;
1730
1731 if (copy_from_user(to: buffer, from: buf, n: count > maxlen ? maxlen : count))
1732 return -EFAULT;
1733
1734 p = get_proc_task(inode);
1735 if (!p)
1736 return -ESRCH;
1737
1738 if (same_thread_group(current, p2: p)) {
1739 set_task_comm(p, buffer);
1740 proc_comm_connector(task: p);
1741 }
1742 else
1743 count = -EINVAL;
1744
1745 put_task_struct(t: p);
1746
1747 return count;
1748}
1749
1750static int comm_show(struct seq_file *m, void *v)
1751{
1752 struct inode *inode = m->private;
1753 struct task_struct *p;
1754
1755 p = get_proc_task(inode);
1756 if (!p)
1757 return -ESRCH;
1758
1759 proc_task_name(m, p, escape: false);
1760 seq_putc(m, c: '\n');
1761
1762 put_task_struct(t: p);
1763
1764 return 0;
1765}
1766
1767static int comm_open(struct inode *inode, struct file *filp)
1768{
1769 return single_open(filp, comm_show, inode);
1770}
1771
1772static const struct file_operations proc_pid_set_comm_operations = {
1773 .open = comm_open,
1774 .read = seq_read,
1775 .write = comm_write,
1776 .llseek = seq_lseek,
1777 .release = single_release,
1778};
1779
1780static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1781{
1782 struct task_struct *task;
1783 struct file *exe_file;
1784
1785 task = get_proc_task(inode: d_inode(dentry));
1786 if (!task)
1787 return -ENOENT;
1788 exe_file = get_task_exe_file(task);
1789 put_task_struct(t: task);
1790 if (exe_file) {
1791 *exe_path = exe_file->f_path;
1792 path_get(&exe_file->f_path);
1793 fput(exe_file);
1794 return 0;
1795 } else
1796 return -ENOENT;
1797}
1798
1799static const char *proc_pid_get_link(struct dentry *dentry,
1800 struct inode *inode,
1801 struct delayed_call *done)
1802{
1803 struct path path;
1804 int error = -EACCES;
1805
1806 if (!dentry)
1807 return ERR_PTR(error: -ECHILD);
1808
1809 /* Are we allowed to snoop on the tasks file descriptors? */
1810 if (!proc_fd_access_allowed(inode))
1811 goto out;
1812
1813 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1814 if (error)
1815 goto out;
1816
1817 error = nd_jump_link(path: &path);
1818out:
1819 return ERR_PTR(error);
1820}
1821
1822static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1823{
1824 char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1825 char *pathname;
1826 int len;
1827
1828 if (!tmp)
1829 return -ENOMEM;
1830
1831 pathname = d_path(path, tmp, PATH_MAX);
1832 len = PTR_ERR(ptr: pathname);
1833 if (IS_ERR(ptr: pathname))
1834 goto out;
1835 len = tmp + PATH_MAX - 1 - pathname;
1836
1837 if (len > buflen)
1838 len = buflen;
1839 if (copy_to_user(to: buffer, from: pathname, n: len))
1840 len = -EFAULT;
1841 out:
1842 kfree(objp: tmp);
1843 return len;
1844}
1845
1846static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1847{
1848 int error = -EACCES;
1849 struct inode *inode = d_inode(dentry);
1850 struct path path;
1851
1852 /* Are we allowed to snoop on the tasks file descriptors? */
1853 if (!proc_fd_access_allowed(inode))
1854 goto out;
1855
1856 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1857 if (error)
1858 goto out;
1859
1860 error = do_proc_readlink(path: &path, buffer, buflen);
1861 path_put(&path);
1862out:
1863 return error;
1864}
1865
1866const struct inode_operations proc_pid_link_inode_operations = {
1867 .readlink = proc_pid_readlink,
1868 .get_link = proc_pid_get_link,
1869 .setattr = proc_setattr,
1870};
1871
1872
1873/* building an inode */
1874
1875void task_dump_owner(struct task_struct *task, umode_t mode,
1876 kuid_t *ruid, kgid_t *rgid)
1877{
1878 /* Depending on the state of dumpable compute who should own a
1879 * proc file for a task.
1880 */
1881 const struct cred *cred;
1882 kuid_t uid;
1883 kgid_t gid;
1884
1885 if (unlikely(task->flags & PF_KTHREAD)) {
1886 *ruid = GLOBAL_ROOT_UID;
1887 *rgid = GLOBAL_ROOT_GID;
1888 return;
1889 }
1890
1891 /* Default to the tasks effective ownership */
1892 rcu_read_lock();
1893 cred = __task_cred(task);
1894 uid = cred->euid;
1895 gid = cred->egid;
1896 rcu_read_unlock();
1897
1898 /*
1899 * Before the /proc/pid/status file was created the only way to read
1900 * the effective uid of a /process was to stat /proc/pid. Reading
1901 * /proc/pid/status is slow enough that procps and other packages
1902 * kept stating /proc/pid. To keep the rules in /proc simple I have
1903 * made this apply to all per process world readable and executable
1904 * directories.
1905 */
1906 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1907 struct mm_struct *mm;
1908 task_lock(task);
1909 mm = task->mm;
1910 /* Make non-dumpable tasks owned by some root */
1911 if (mm) {
1912 if (get_dumpable(mm) != SUID_DUMP_USER) {
1913 struct user_namespace *user_ns = mm->user_ns;
1914
1915 uid = make_kuid(user_ns, 0);
1916 if (!uid_valid(uid))
1917 uid = GLOBAL_ROOT_UID;
1918
1919 gid = make_kgid(user_ns, 0);
1920 if (!gid_valid(gid))
1921 gid = GLOBAL_ROOT_GID;
1922 }
1923 } else {
1924 uid = GLOBAL_ROOT_UID;
1925 gid = GLOBAL_ROOT_GID;
1926 }
1927 task_unlock(task);
1928 }
1929 *ruid = uid;
1930 *rgid = gid;
1931}
1932
1933void proc_pid_evict_inode(struct proc_inode *ei)
1934{
1935 struct pid *pid = ei->pid;
1936
1937 if (S_ISDIR(ei->vfs_inode.i_mode)) {
1938 spin_lock(lock: &pid->lock);
1939 hlist_del_init_rcu(n: &ei->sibling_inodes);
1940 spin_unlock(lock: &pid->lock);
1941 }
1942}
1943
1944struct inode *proc_pid_make_inode(struct super_block *sb,
1945 struct task_struct *task, umode_t mode)
1946{
1947 struct inode * inode;
1948 struct proc_inode *ei;
1949 struct pid *pid;
1950
1951 /* We need a new inode */
1952
1953 inode = new_inode(sb);
1954 if (!inode)
1955 goto out;
1956
1957 /* Common stuff */
1958 ei = PROC_I(inode);
1959 inode->i_mode = mode;
1960 inode->i_ino = get_next_ino();
1961 simple_inode_init_ts(inode);
1962 inode->i_op = &proc_def_inode_operations;
1963
1964 /*
1965 * grab the reference to task.
1966 */
1967 pid = get_task_pid(task, type: PIDTYPE_PID);
1968 if (!pid)
1969 goto out_unlock;
1970
1971 /* Let the pid remember us for quick removal */
1972 ei->pid = pid;
1973
1974 task_dump_owner(task, mode: 0, ruid: &inode->i_uid, rgid: &inode->i_gid);
1975 security_task_to_inode(p: task, inode);
1976
1977out:
1978 return inode;
1979
1980out_unlock:
1981 iput(inode);
1982 return NULL;
1983}
1984
1985/*
1986 * Generating an inode and adding it into @pid->inodes, so that task will
1987 * invalidate inode's dentry before being released.
1988 *
1989 * This helper is used for creating dir-type entries under '/proc' and
1990 * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1991 * can be released by invalidating '/proc/<tgid>' dentry.
1992 * In theory, dentries under '/proc/<tgid>/task' can also be released by
1993 * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1994 * thread exiting situation: Any one of threads should invalidate its
1995 * '/proc/<tgid>/task/<pid>' dentry before released.
1996 */
1997static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1998 struct task_struct *task, umode_t mode)
1999{
2000 struct inode *inode;
2001 struct proc_inode *ei;
2002 struct pid *pid;
2003
2004 inode = proc_pid_make_inode(sb, task, mode);
2005 if (!inode)
2006 return NULL;
2007
2008 /* Let proc_flush_pid find this directory inode */
2009 ei = PROC_I(inode);
2010 pid = ei->pid;
2011 spin_lock(lock: &pid->lock);
2012 hlist_add_head_rcu(n: &ei->sibling_inodes, h: &pid->inodes);
2013 spin_unlock(lock: &pid->lock);
2014
2015 return inode;
2016}
2017
2018int pid_getattr(struct mnt_idmap *idmap, const struct path *path,
2019 struct kstat *stat, u32 request_mask, unsigned int query_flags)
2020{
2021 struct inode *inode = d_inode(dentry: path->dentry);
2022 struct proc_fs_info *fs_info = proc_sb_info(sb: inode->i_sb);
2023 struct task_struct *task;
2024
2025 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
2026
2027 stat->uid = GLOBAL_ROOT_UID;
2028 stat->gid = GLOBAL_ROOT_GID;
2029 rcu_read_lock();
2030 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2031 if (task) {
2032 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
2033 rcu_read_unlock();
2034 /*
2035 * This doesn't prevent learning whether PID exists,
2036 * it only makes getattr() consistent with readdir().
2037 */
2038 return -ENOENT;
2039 }
2040 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
2041 }
2042 rcu_read_unlock();
2043 return 0;
2044}
2045
2046/* dentry stuff */
2047
2048/*
2049 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
2050 */
2051void pid_update_inode(struct task_struct *task, struct inode *inode)
2052{
2053 task_dump_owner(task, mode: inode->i_mode, ruid: &inode->i_uid, rgid: &inode->i_gid);
2054
2055 inode->i_mode &= ~(S_ISUID | S_ISGID);
2056 security_task_to_inode(p: task, inode);
2057}
2058
2059/*
2060 * Rewrite the inode's ownerships here because the owning task may have
2061 * performed a setuid(), etc.
2062 *
2063 */
2064static int pid_revalidate(struct inode *dir, const struct qstr *name,
2065 struct dentry *dentry, unsigned int flags)
2066{
2067 struct inode *inode;
2068 struct task_struct *task;
2069 int ret = 0;
2070
2071 rcu_read_lock();
2072 inode = d_inode_rcu(dentry);
2073 if (!inode)
2074 goto out;
2075 task = pid_task(pid: proc_pid(inode), PIDTYPE_PID);
2076
2077 if (task) {
2078 pid_update_inode(task, inode);
2079 ret = 1;
2080 }
2081out:
2082 rcu_read_unlock();
2083 return ret;
2084}
2085
2086static inline bool proc_inode_is_dead(struct inode *inode)
2087{
2088 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2089}
2090
2091int pid_delete_dentry(const struct dentry *dentry)
2092{
2093 /* Is the task we represent dead?
2094 * If so, then don't put the dentry on the lru list,
2095 * kill it immediately.
2096 */
2097 return proc_inode_is_dead(inode: d_inode(dentry));
2098}
2099
2100const struct dentry_operations pid_dentry_operations =
2101{
2102 .d_revalidate = pid_revalidate,
2103 .d_delete = pid_delete_dentry,
2104};
2105
2106/* Lookups */
2107
2108/*
2109 * Fill a directory entry.
2110 *
2111 * If possible create the dcache entry and derive our inode number and
2112 * file type from dcache entry.
2113 *
2114 * Since all of the proc inode numbers are dynamically generated, the inode
2115 * numbers do not exist until the inode is cache. This means creating
2116 * the dcache entry in readdir is necessary to keep the inode numbers
2117 * reported by readdir in sync with the inode numbers reported
2118 * by stat.
2119 */
2120bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2121 const char *name, unsigned int len,
2122 instantiate_t instantiate, struct task_struct *task, const void *ptr)
2123{
2124 struct dentry *child, *dir = file->f_path.dentry;
2125 struct qstr qname = QSTR_INIT(name, len);
2126 struct inode *inode;
2127 unsigned type = DT_UNKNOWN;
2128 ino_t ino = 1;
2129
2130 child = try_lookup_noperm(&qname, dir);
2131 if (!child) {
2132 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2133 child = d_alloc_parallel(dir, &qname, &wq);
2134 if (IS_ERR(ptr: child))
2135 goto end_instantiate;
2136 if (d_in_lookup(dentry: child)) {
2137 struct dentry *res;
2138 res = instantiate(child, task, ptr);
2139 d_lookup_done(dentry: child);
2140 if (unlikely(res)) {
2141 dput(child);
2142 child = res;
2143 if (IS_ERR(ptr: child))
2144 goto end_instantiate;
2145 }
2146 }
2147 }
2148 inode = d_inode(dentry: child);
2149 ino = inode->i_ino;
2150 type = inode->i_mode >> 12;
2151 dput(child);
2152end_instantiate:
2153 return dir_emit(ctx, name, namelen: len, ino, type);
2154}
2155
2156/*
2157 * dname_to_vma_addr - maps a dentry name into two unsigned longs
2158 * which represent vma start and end addresses.
2159 */
2160static int dname_to_vma_addr(struct dentry *dentry,
2161 unsigned long *start, unsigned long *end)
2162{
2163 const char *str = dentry->d_name.name;
2164 unsigned long long sval, eval;
2165 unsigned int len;
2166
2167 if (str[0] == '0' && str[1] != '-')
2168 return -EINVAL;
2169 len = _parse_integer(s: str, base: 16, res: &sval);
2170 if (len & KSTRTOX_OVERFLOW)
2171 return -EINVAL;
2172 if (sval != (unsigned long)sval)
2173 return -EINVAL;
2174 str += len;
2175
2176 if (*str != '-')
2177 return -EINVAL;
2178 str++;
2179
2180 if (str[0] == '0' && str[1])
2181 return -EINVAL;
2182 len = _parse_integer(s: str, base: 16, res: &eval);
2183 if (len & KSTRTOX_OVERFLOW)
2184 return -EINVAL;
2185 if (eval != (unsigned long)eval)
2186 return -EINVAL;
2187 str += len;
2188
2189 if (*str != '\0')
2190 return -EINVAL;
2191
2192 *start = sval;
2193 *end = eval;
2194
2195 return 0;
2196}
2197
2198static int map_files_d_revalidate(struct inode *dir, const struct qstr *name,
2199 struct dentry *dentry, unsigned int flags)
2200{
2201 unsigned long vm_start, vm_end;
2202 bool exact_vma_exists = false;
2203 struct mm_struct *mm = NULL;
2204 struct task_struct *task;
2205 struct inode *inode;
2206 int status = 0;
2207
2208 if (flags & LOOKUP_RCU)
2209 return -ECHILD;
2210
2211 inode = d_inode(dentry);
2212 task = get_proc_task(inode);
2213 if (!task)
2214 goto out_notask;
2215
2216 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2217 if (IS_ERR(ptr: mm))
2218 goto out;
2219
2220 if (!dname_to_vma_addr(dentry, start: &vm_start, end: &vm_end)) {
2221 status = mmap_read_lock_killable(mm);
2222 if (!status) {
2223 exact_vma_exists = !!find_exact_vma(mm, vm_start,
2224 vm_end);
2225 mmap_read_unlock(mm);
2226 }
2227 }
2228
2229 mmput(mm);
2230
2231 if (exact_vma_exists) {
2232 task_dump_owner(task, mode: 0, ruid: &inode->i_uid, rgid: &inode->i_gid);
2233
2234 security_task_to_inode(p: task, inode);
2235 status = 1;
2236 }
2237
2238out:
2239 put_task_struct(t: task);
2240
2241out_notask:
2242 return status;
2243}
2244
2245static const struct dentry_operations tid_map_files_dentry_operations = {
2246 .d_revalidate = map_files_d_revalidate,
2247 .d_delete = pid_delete_dentry,
2248};
2249
2250static int map_files_get_link(struct dentry *dentry, struct path *path)
2251{
2252 unsigned long vm_start, vm_end;
2253 struct vm_area_struct *vma;
2254 struct task_struct *task;
2255 struct mm_struct *mm;
2256 int rc;
2257
2258 rc = -ENOENT;
2259 task = get_proc_task(inode: d_inode(dentry));
2260 if (!task)
2261 goto out;
2262
2263 mm = get_task_mm(task);
2264 put_task_struct(t: task);
2265 if (!mm)
2266 goto out;
2267
2268 rc = dname_to_vma_addr(dentry, start: &vm_start, end: &vm_end);
2269 if (rc)
2270 goto out_mmput;
2271
2272 rc = mmap_read_lock_killable(mm);
2273 if (rc)
2274 goto out_mmput;
2275
2276 rc = -ENOENT;
2277 vma = find_exact_vma(mm, vm_start, vm_end);
2278 if (vma && vma->vm_file) {
2279 *path = *file_user_path(f: vma->vm_file);
2280 path_get(path);
2281 rc = 0;
2282 }
2283 mmap_read_unlock(mm);
2284
2285out_mmput:
2286 mmput(mm);
2287out:
2288 return rc;
2289}
2290
2291struct map_files_info {
2292 unsigned long start;
2293 unsigned long end;
2294 fmode_t mode;
2295};
2296
2297/*
2298 * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2299 * to concerns about how the symlinks may be used to bypass permissions on
2300 * ancestor directories in the path to the file in question.
2301 */
2302static const char *
2303proc_map_files_get_link(struct dentry *dentry,
2304 struct inode *inode,
2305 struct delayed_call *done)
2306{
2307 if (!checkpoint_restore_ns_capable(ns: &init_user_ns))
2308 return ERR_PTR(error: -EPERM);
2309
2310 return proc_pid_get_link(dentry, inode, done);
2311}
2312
2313/*
2314 * Identical to proc_pid_link_inode_operations except for get_link()
2315 */
2316static const struct inode_operations proc_map_files_link_inode_operations = {
2317 .readlink = proc_pid_readlink,
2318 .get_link = proc_map_files_get_link,
2319 .setattr = proc_setattr,
2320};
2321
2322static struct dentry *
2323proc_map_files_instantiate(struct dentry *dentry,
2324 struct task_struct *task, const void *ptr)
2325{
2326 fmode_t mode = (fmode_t)(unsigned long)ptr;
2327 struct proc_inode *ei;
2328 struct inode *inode;
2329
2330 inode = proc_pid_make_inode(sb: dentry->d_sb, task, S_IFLNK |
2331 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2332 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2333 if (!inode)
2334 return ERR_PTR(error: -ENOENT);
2335
2336 ei = PROC_I(inode);
2337 ei->op.proc_get_link = map_files_get_link;
2338
2339 inode->i_op = &proc_map_files_link_inode_operations;
2340 inode->i_size = 64;
2341
2342 return proc_splice_unmountable(inode, dentry,
2343 d_ops: &tid_map_files_dentry_operations);
2344}
2345
2346static struct dentry *proc_map_files_lookup(struct inode *dir,
2347 struct dentry *dentry, unsigned int flags)
2348{
2349 unsigned long vm_start, vm_end;
2350 struct vm_area_struct *vma;
2351 struct task_struct *task;
2352 struct dentry *result;
2353 struct mm_struct *mm;
2354
2355 result = ERR_PTR(error: -ENOENT);
2356 task = get_proc_task(inode: dir);
2357 if (!task)
2358 goto out;
2359
2360 result = ERR_PTR(error: -EACCES);
2361 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2362 goto out_put_task;
2363
2364 result = ERR_PTR(error: -ENOENT);
2365 if (dname_to_vma_addr(dentry, start: &vm_start, end: &vm_end))
2366 goto out_put_task;
2367
2368 mm = get_task_mm(task);
2369 if (!mm)
2370 goto out_put_task;
2371
2372 result = ERR_PTR(error: -EINTR);
2373 if (mmap_read_lock_killable(mm))
2374 goto out_put_mm;
2375
2376 result = ERR_PTR(error: -ENOENT);
2377 vma = find_exact_vma(mm, vm_start, vm_end);
2378 if (!vma)
2379 goto out_no_vma;
2380
2381 if (vma->vm_file)
2382 result = proc_map_files_instantiate(dentry, task,
2383 ptr: (void *)(unsigned long)vma->vm_file->f_mode);
2384
2385out_no_vma:
2386 mmap_read_unlock(mm);
2387out_put_mm:
2388 mmput(mm);
2389out_put_task:
2390 put_task_struct(t: task);
2391out:
2392 return result;
2393}
2394
2395static const struct inode_operations proc_map_files_inode_operations = {
2396 .lookup = proc_map_files_lookup,
2397 .permission = proc_fd_permission,
2398 .setattr = proc_setattr,
2399};
2400
2401static int
2402proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2403{
2404 struct vm_area_struct *vma;
2405 struct task_struct *task;
2406 struct mm_struct *mm;
2407 unsigned long nr_files, pos, i;
2408 GENRADIX(struct map_files_info) fa;
2409 struct map_files_info *p;
2410 int ret;
2411 struct vma_iterator vmi;
2412
2413 genradix_init(&fa);
2414
2415 ret = -ENOENT;
2416 task = get_proc_task(inode: file_inode(f: file));
2417 if (!task)
2418 goto out;
2419
2420 ret = -EACCES;
2421 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2422 goto out_put_task;
2423
2424 ret = 0;
2425 if (!dir_emit_dots(file, ctx))
2426 goto out_put_task;
2427
2428 mm = get_task_mm(task);
2429 if (!mm)
2430 goto out_put_task;
2431
2432 ret = mmap_read_lock_killable(mm);
2433 if (ret) {
2434 mmput(mm);
2435 goto out_put_task;
2436 }
2437
2438 nr_files = 0;
2439
2440 /*
2441 * We need two passes here:
2442 *
2443 * 1) Collect vmas of mapped files with mmap_lock taken
2444 * 2) Release mmap_lock and instantiate entries
2445 *
2446 * otherwise we get lockdep complained, since filldir()
2447 * routine might require mmap_lock taken in might_fault().
2448 */
2449
2450 pos = 2;
2451 vma_iter_init(vmi: &vmi, mm, addr: 0);
2452 for_each_vma(vmi, vma) {
2453 if (!vma->vm_file)
2454 continue;
2455 if (++pos <= ctx->pos)
2456 continue;
2457
2458 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2459 if (!p) {
2460 ret = -ENOMEM;
2461 mmap_read_unlock(mm);
2462 mmput(mm);
2463 goto out_put_task;
2464 }
2465
2466 p->start = vma->vm_start;
2467 p->end = vma->vm_end;
2468 p->mode = vma->vm_file->f_mode;
2469 }
2470 mmap_read_unlock(mm);
2471 mmput(mm);
2472
2473 for (i = 0; i < nr_files; i++) {
2474 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2475 unsigned int len;
2476
2477 p = genradix_ptr(&fa, i);
2478 len = snprintf(buf, size: sizeof(buf), fmt: "%lx-%lx", p->start, p->end);
2479 if (!proc_fill_cache(file, ctx,
2480 name: buf, len,
2481 instantiate: proc_map_files_instantiate,
2482 task,
2483 ptr: (void *)(unsigned long)p->mode))
2484 break;
2485 ctx->pos++;
2486 }
2487
2488out_put_task:
2489 put_task_struct(t: task);
2490out:
2491 genradix_free(&fa);
2492 return ret;
2493}
2494
2495static const struct file_operations proc_map_files_operations = {
2496 .read = generic_read_dir,
2497 .iterate_shared = proc_map_files_readdir,
2498 .llseek = generic_file_llseek,
2499};
2500
2501#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2502struct timers_private {
2503 struct pid *pid;
2504 struct task_struct *task;
2505 struct pid_namespace *ns;
2506};
2507
2508static void *timers_start(struct seq_file *m, loff_t *pos)
2509{
2510 struct timers_private *tp = m->private;
2511
2512 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2513 if (!tp->task)
2514 return ERR_PTR(-ESRCH);
2515
2516 rcu_read_lock();
2517 return seq_hlist_start_rcu(&tp->task->signal->posix_timers, *pos);
2518}
2519
2520static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2521{
2522 struct timers_private *tp = m->private;
2523
2524 return seq_hlist_next_rcu(v, &tp->task->signal->posix_timers, pos);
2525}
2526
2527static void timers_stop(struct seq_file *m, void *v)
2528{
2529 struct timers_private *tp = m->private;
2530
2531 if (tp->task) {
2532 put_task_struct(tp->task);
2533 tp->task = NULL;
2534 rcu_read_unlock();
2535 }
2536}
2537
2538static int show_timer(struct seq_file *m, void *v)
2539{
2540 static const char * const nstr[] = {
2541 [SIGEV_SIGNAL] = "signal",
2542 [SIGEV_NONE] = "none",
2543 [SIGEV_THREAD] = "thread",
2544 };
2545
2546 struct k_itimer *timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
2547 struct timers_private *tp = m->private;
2548 int notify = timer->it_sigev_notify;
2549
2550 guard(spinlock_irq)(&timer->it_lock);
2551 if (!posixtimer_valid(timer))
2552 return 0;
2553
2554 seq_printf(m, "ID: %d\n", timer->it_id);
2555 seq_printf(m, "signal: %d/%px\n", timer->sigq.info.si_signo,
2556 timer->sigq.info.si_value.sival_ptr);
2557 seq_printf(m, "notify: %s/%s.%d\n", nstr[notify & ~SIGEV_THREAD_ID],
2558 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2559 pid_nr_ns(timer->it_pid, tp->ns));
2560 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2561
2562 return 0;
2563}
2564
2565static const struct seq_operations proc_timers_seq_ops = {
2566 .start = timers_start,
2567 .next = timers_next,
2568 .stop = timers_stop,
2569 .show = show_timer,
2570};
2571
2572static int proc_timers_open(struct inode *inode, struct file *file)
2573{
2574 struct timers_private *tp;
2575
2576 tp = __seq_open_private(file, &proc_timers_seq_ops,
2577 sizeof(struct timers_private));
2578 if (!tp)
2579 return -ENOMEM;
2580
2581 tp->pid = proc_pid(inode);
2582 tp->ns = proc_pid_ns(inode->i_sb);
2583 return 0;
2584}
2585
2586static const struct file_operations proc_timers_operations = {
2587 .open = proc_timers_open,
2588 .read = seq_read,
2589 .llseek = seq_lseek,
2590 .release = seq_release_private,
2591};
2592#endif
2593
2594static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2595 size_t count, loff_t *offset)
2596{
2597 struct inode *inode = file_inode(f: file);
2598 struct task_struct *p;
2599 u64 slack_ns;
2600 int err;
2601
2602 err = kstrtoull_from_user(s: buf, count, base: 10, res: &slack_ns);
2603 if (err < 0)
2604 return err;
2605
2606 p = get_proc_task(inode);
2607 if (!p)
2608 return -ESRCH;
2609
2610 if (p != current) {
2611 rcu_read_lock();
2612 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2613 rcu_read_unlock();
2614 count = -EPERM;
2615 goto out;
2616 }
2617 rcu_read_unlock();
2618
2619 err = security_task_setscheduler(p);
2620 if (err) {
2621 count = err;
2622 goto out;
2623 }
2624 }
2625
2626 task_lock(p);
2627 if (rt_or_dl_task_policy(tsk: p))
2628 slack_ns = 0;
2629 else if (slack_ns == 0)
2630 slack_ns = p->default_timer_slack_ns;
2631 p->timer_slack_ns = slack_ns;
2632 task_unlock(p);
2633
2634out:
2635 put_task_struct(t: p);
2636
2637 return count;
2638}
2639
2640static int timerslack_ns_show(struct seq_file *m, void *v)
2641{
2642 struct inode *inode = m->private;
2643 struct task_struct *p;
2644 int err = 0;
2645
2646 p = get_proc_task(inode);
2647 if (!p)
2648 return -ESRCH;
2649
2650 if (p != current) {
2651 rcu_read_lock();
2652 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2653 rcu_read_unlock();
2654 err = -EPERM;
2655 goto out;
2656 }
2657 rcu_read_unlock();
2658
2659 err = security_task_getscheduler(p);
2660 if (err)
2661 goto out;
2662 }
2663
2664 task_lock(p);
2665 seq_printf(m, fmt: "%llu\n", p->timer_slack_ns);
2666 task_unlock(p);
2667
2668out:
2669 put_task_struct(t: p);
2670
2671 return err;
2672}
2673
2674static int timerslack_ns_open(struct inode *inode, struct file *filp)
2675{
2676 return single_open(filp, timerslack_ns_show, inode);
2677}
2678
2679static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2680 .open = timerslack_ns_open,
2681 .read = seq_read,
2682 .write = timerslack_ns_write,
2683 .llseek = seq_lseek,
2684 .release = single_release,
2685};
2686
2687static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2688 struct task_struct *task, const void *ptr)
2689{
2690 const struct pid_entry *p = ptr;
2691 struct inode *inode;
2692 struct proc_inode *ei;
2693
2694 inode = proc_pid_make_inode(sb: dentry->d_sb, task, mode: p->mode);
2695 if (!inode)
2696 return ERR_PTR(error: -ENOENT);
2697
2698 ei = PROC_I(inode);
2699 if (S_ISDIR(inode->i_mode))
2700 set_nlink(inode, nlink: 2); /* Use getattr to fix if necessary */
2701 if (p->iop)
2702 inode->i_op = p->iop;
2703 if (p->fop)
2704 inode->i_fop = p->fop;
2705 ei->op = p->op;
2706 pid_update_inode(task, inode);
2707 return d_splice_alias_ops(inode, dentry, &pid_dentry_operations);
2708}
2709
2710static struct dentry *proc_pident_lookup(struct inode *dir,
2711 struct dentry *dentry,
2712 const struct pid_entry *p,
2713 const struct pid_entry *end)
2714{
2715 struct task_struct *task = get_proc_task(inode: dir);
2716 struct dentry *res = ERR_PTR(error: -ENOENT);
2717
2718 if (!task)
2719 goto out_no_task;
2720
2721 /*
2722 * Yes, it does not scale. And it should not. Don't add
2723 * new entries into /proc/<tgid>/ without very good reasons.
2724 */
2725 for (; p < end; p++) {
2726 if (p->len != dentry->d_name.len)
2727 continue;
2728 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2729 res = proc_pident_instantiate(dentry, task, ptr: p);
2730 break;
2731 }
2732 }
2733 put_task_struct(t: task);
2734out_no_task:
2735 return res;
2736}
2737
2738static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2739 const struct pid_entry *ents, unsigned int nents)
2740{
2741 struct task_struct *task = get_proc_task(inode: file_inode(f: file));
2742 const struct pid_entry *p;
2743
2744 if (!task)
2745 return -ENOENT;
2746
2747 if (!dir_emit_dots(file, ctx))
2748 goto out;
2749
2750 if (ctx->pos >= nents + 2)
2751 goto out;
2752
2753 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2754 if (!proc_fill_cache(file, ctx, name: p->name, len: p->len,
2755 instantiate: proc_pident_instantiate, task, ptr: p))
2756 break;
2757 ctx->pos++;
2758 }
2759out:
2760 put_task_struct(t: task);
2761 return 0;
2762}
2763
2764#ifdef CONFIG_SECURITY
2765static int proc_pid_attr_open(struct inode *inode, struct file *file)
2766{
2767 file->private_data = NULL;
2768 __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2769 return 0;
2770}
2771
2772static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2773 size_t count, loff_t *ppos)
2774{
2775 struct inode * inode = file_inode(f: file);
2776 char *p = NULL;
2777 ssize_t length;
2778 struct task_struct *task = get_proc_task(inode);
2779
2780 if (!task)
2781 return -ESRCH;
2782
2783 length = security_getprocattr(p: task, lsmid: PROC_I(inode)->op.lsmid,
2784 name: file->f_path.dentry->d_name.name,
2785 value: &p);
2786 put_task_struct(t: task);
2787 if (length > 0)
2788 length = simple_read_from_buffer(to: buf, count, ppos, from: p, available: length);
2789 kfree(objp: p);
2790 return length;
2791}
2792
2793static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2794 size_t count, loff_t *ppos)
2795{
2796 struct inode * inode = file_inode(f: file);
2797 struct task_struct *task;
2798 void *page;
2799 int rv;
2800
2801 /* A task may only write when it was the opener. */
2802 if (file->private_data != current->mm)
2803 return -EPERM;
2804
2805 rcu_read_lock();
2806 task = pid_task(pid: proc_pid(inode), PIDTYPE_PID);
2807 if (!task) {
2808 rcu_read_unlock();
2809 return -ESRCH;
2810 }
2811 /* A task may only write its own attributes. */
2812 if (current != task) {
2813 rcu_read_unlock();
2814 return -EACCES;
2815 }
2816 /* Prevent changes to overridden credentials. */
2817 if (current_cred() != current_real_cred()) {
2818 rcu_read_unlock();
2819 return -EBUSY;
2820 }
2821 rcu_read_unlock();
2822
2823 if (count > PAGE_SIZE)
2824 count = PAGE_SIZE;
2825
2826 /* No partial writes. */
2827 if (*ppos != 0)
2828 return -EINVAL;
2829
2830 page = memdup_user(buf, count);
2831 if (IS_ERR(ptr: page)) {
2832 rv = PTR_ERR(ptr: page);
2833 goto out;
2834 }
2835
2836 /* Guard against adverse ptrace interaction */
2837 rv = mutex_lock_interruptible(lock: &current->signal->cred_guard_mutex);
2838 if (rv < 0)
2839 goto out_free;
2840
2841 rv = security_setprocattr(lsmid: PROC_I(inode)->op.lsmid,
2842 name: file->f_path.dentry->d_name.name, value: page,
2843 size: count);
2844 mutex_unlock(lock: &current->signal->cred_guard_mutex);
2845out_free:
2846 kfree(objp: page);
2847out:
2848 return rv;
2849}
2850
2851static const struct file_operations proc_pid_attr_operations = {
2852 .open = proc_pid_attr_open,
2853 .read = proc_pid_attr_read,
2854 .write = proc_pid_attr_write,
2855 .llseek = generic_file_llseek,
2856 .release = mem_release,
2857};
2858
2859#define LSM_DIR_OPS(LSM) \
2860static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2861 struct dir_context *ctx) \
2862{ \
2863 return proc_pident_readdir(filp, ctx, \
2864 LSM##_attr_dir_stuff, \
2865 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2866} \
2867\
2868static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2869 .read = generic_read_dir, \
2870 .iterate_shared = proc_##LSM##_attr_dir_iterate, \
2871 .llseek = default_llseek, \
2872}; \
2873\
2874static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2875 struct dentry *dentry, unsigned int flags) \
2876{ \
2877 return proc_pident_lookup(dir, dentry, \
2878 LSM##_attr_dir_stuff, \
2879 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2880} \
2881\
2882static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2883 .lookup = proc_##LSM##_attr_dir_lookup, \
2884 .getattr = pid_getattr, \
2885 .setattr = proc_setattr, \
2886}
2887
2888#ifdef CONFIG_SECURITY_SMACK
2889static const struct pid_entry smack_attr_dir_stuff[] = {
2890 ATTR(LSM_ID_SMACK, "current", 0666),
2891};
2892LSM_DIR_OPS(smack);
2893#endif
2894
2895#ifdef CONFIG_SECURITY_APPARMOR
2896static const struct pid_entry apparmor_attr_dir_stuff[] = {
2897 ATTR(LSM_ID_APPARMOR, "current", 0666),
2898 ATTR(LSM_ID_APPARMOR, "prev", 0444),
2899 ATTR(LSM_ID_APPARMOR, "exec", 0666),
2900};
2901LSM_DIR_OPS(apparmor);
2902#endif
2903
2904static const struct pid_entry attr_dir_stuff[] = {
2905 ATTR(LSM_ID_UNDEF, "current", 0666),
2906 ATTR(LSM_ID_UNDEF, "prev", 0444),
2907 ATTR(LSM_ID_UNDEF, "exec", 0666),
2908 ATTR(LSM_ID_UNDEF, "fscreate", 0666),
2909 ATTR(LSM_ID_UNDEF, "keycreate", 0666),
2910 ATTR(LSM_ID_UNDEF, "sockcreate", 0666),
2911#ifdef CONFIG_SECURITY_SMACK
2912 DIR("smack", 0555,
2913 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2914#endif
2915#ifdef CONFIG_SECURITY_APPARMOR
2916 DIR("apparmor", 0555,
2917 proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2918#endif
2919};
2920
2921static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2922{
2923 return proc_pident_readdir(file, ctx,
2924 ents: attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2925}
2926
2927static const struct file_operations proc_attr_dir_operations = {
2928 .read = generic_read_dir,
2929 .iterate_shared = proc_attr_dir_readdir,
2930 .llseek = generic_file_llseek,
2931};
2932
2933static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2934 struct dentry *dentry, unsigned int flags)
2935{
2936 return proc_pident_lookup(dir, dentry,
2937 p: attr_dir_stuff,
2938 end: attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2939}
2940
2941static const struct inode_operations proc_attr_dir_inode_operations = {
2942 .lookup = proc_attr_dir_lookup,
2943 .getattr = pid_getattr,
2944 .setattr = proc_setattr,
2945};
2946
2947#endif
2948
2949#ifdef CONFIG_ELF_CORE
2950static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2951 size_t count, loff_t *ppos)
2952{
2953 struct task_struct *task = get_proc_task(inode: file_inode(f: file));
2954 struct mm_struct *mm;
2955 char buffer[PROC_NUMBUF];
2956 size_t len;
2957 int ret;
2958
2959 if (!task)
2960 return -ESRCH;
2961
2962 ret = 0;
2963 mm = get_task_mm(task);
2964 if (mm) {
2965 unsigned long flags = __mm_flags_get_dumpable(mm);
2966
2967 len = snprintf(buf: buffer, size: sizeof(buffer), fmt: "%08lx\n",
2968 ((flags & MMF_DUMP_FILTER_MASK) >>
2969 MMF_DUMP_FILTER_SHIFT));
2970 mmput(mm);
2971 ret = simple_read_from_buffer(to: buf, count, ppos, from: buffer, available: len);
2972 }
2973
2974 put_task_struct(t: task);
2975
2976 return ret;
2977}
2978
2979static ssize_t proc_coredump_filter_write(struct file *file,
2980 const char __user *buf,
2981 size_t count,
2982 loff_t *ppos)
2983{
2984 struct task_struct *task;
2985 struct mm_struct *mm;
2986 unsigned int val;
2987 int ret;
2988 int i;
2989 unsigned long mask;
2990
2991 ret = kstrtouint_from_user(s: buf, count, base: 0, res: &val);
2992 if (ret < 0)
2993 return ret;
2994
2995 ret = -ESRCH;
2996 task = get_proc_task(inode: file_inode(f: file));
2997 if (!task)
2998 goto out_no_task;
2999
3000 mm = get_task_mm(task);
3001 if (!mm)
3002 goto out_no_mm;
3003 ret = 0;
3004
3005 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
3006 if (val & mask)
3007 mm_flags_set(flag: i + MMF_DUMP_FILTER_SHIFT, mm);
3008 else
3009 mm_flags_clear(flag: i + MMF_DUMP_FILTER_SHIFT, mm);
3010 }
3011
3012 mmput(mm);
3013 out_no_mm:
3014 put_task_struct(t: task);
3015 out_no_task:
3016 if (ret < 0)
3017 return ret;
3018 return count;
3019}
3020
3021static const struct file_operations proc_coredump_filter_operations = {
3022 .read = proc_coredump_filter_read,
3023 .write = proc_coredump_filter_write,
3024 .llseek = generic_file_llseek,
3025};
3026#endif
3027
3028#ifdef CONFIG_TASK_IO_ACCOUNTING
3029static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
3030{
3031 struct task_io_accounting acct;
3032 int result;
3033
3034 result = down_read_killable(sem: &task->signal->exec_update_lock);
3035 if (result)
3036 return result;
3037
3038 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
3039 result = -EACCES;
3040 goto out_unlock;
3041 }
3042
3043 if (whole) {
3044 struct signal_struct *sig = task->signal;
3045 struct task_struct *t;
3046 unsigned int seq = 1;
3047 unsigned long flags;
3048
3049 rcu_read_lock();
3050 do {
3051 seq++; /* 2 on the 1st/lockless path, otherwise odd */
3052 flags = read_seqbegin_or_lock_irqsave(lock: &sig->stats_lock, seq: &seq);
3053
3054 acct = sig->ioac;
3055 __for_each_thread(sig, t)
3056 task_io_accounting_add(dst: &acct, src: &t->ioac);
3057
3058 } while (need_seqretry(lock: &sig->stats_lock, seq));
3059 done_seqretry_irqrestore(lock: &sig->stats_lock, seq, flags);
3060 rcu_read_unlock();
3061 } else {
3062 acct = task->ioac;
3063 }
3064
3065 seq_printf(m,
3066 fmt: "rchar: %llu\n"
3067 "wchar: %llu\n"
3068 "syscr: %llu\n"
3069 "syscw: %llu\n"
3070 "read_bytes: %llu\n"
3071 "write_bytes: %llu\n"
3072 "cancelled_write_bytes: %llu\n",
3073 (unsigned long long)acct.rchar,
3074 (unsigned long long)acct.wchar,
3075 (unsigned long long)acct.syscr,
3076 (unsigned long long)acct.syscw,
3077 (unsigned long long)acct.read_bytes,
3078 (unsigned long long)acct.write_bytes,
3079 (unsigned long long)acct.cancelled_write_bytes);
3080 result = 0;
3081
3082out_unlock:
3083 up_read(sem: &task->signal->exec_update_lock);
3084 return result;
3085}
3086
3087static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3088 struct pid *pid, struct task_struct *task)
3089{
3090 return do_io_accounting(task, m, whole: 0);
3091}
3092
3093static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3094 struct pid *pid, struct task_struct *task)
3095{
3096 return do_io_accounting(task, m, whole: 1);
3097}
3098#endif /* CONFIG_TASK_IO_ACCOUNTING */
3099
3100#ifdef CONFIG_USER_NS
3101static int proc_id_map_open(struct inode *inode, struct file *file,
3102 const struct seq_operations *seq_ops)
3103{
3104 struct user_namespace *ns = NULL;
3105 struct task_struct *task;
3106 struct seq_file *seq;
3107 int ret = -EINVAL;
3108
3109 task = get_proc_task(inode);
3110 if (task) {
3111 rcu_read_lock();
3112 ns = get_user_ns(task_cred_xxx(task, user_ns));
3113 rcu_read_unlock();
3114 put_task_struct(task);
3115 }
3116 if (!ns)
3117 goto err;
3118
3119 ret = seq_open(file, seq_ops);
3120 if (ret)
3121 goto err_put_ns;
3122
3123 seq = file->private_data;
3124 seq->private = ns;
3125
3126 return 0;
3127err_put_ns:
3128 put_user_ns(ns);
3129err:
3130 return ret;
3131}
3132
3133static int proc_id_map_release(struct inode *inode, struct file *file)
3134{
3135 struct seq_file *seq = file->private_data;
3136 struct user_namespace *ns = seq->private;
3137 put_user_ns(ns);
3138 return seq_release(inode, file);
3139}
3140
3141static int proc_uid_map_open(struct inode *inode, struct file *file)
3142{
3143 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3144}
3145
3146static int proc_gid_map_open(struct inode *inode, struct file *file)
3147{
3148 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3149}
3150
3151static int proc_projid_map_open(struct inode *inode, struct file *file)
3152{
3153 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3154}
3155
3156static const struct file_operations proc_uid_map_operations = {
3157 .open = proc_uid_map_open,
3158 .write = proc_uid_map_write,
3159 .read = seq_read,
3160 .llseek = seq_lseek,
3161 .release = proc_id_map_release,
3162};
3163
3164static const struct file_operations proc_gid_map_operations = {
3165 .open = proc_gid_map_open,
3166 .write = proc_gid_map_write,
3167 .read = seq_read,
3168 .llseek = seq_lseek,
3169 .release = proc_id_map_release,
3170};
3171
3172static const struct file_operations proc_projid_map_operations = {
3173 .open = proc_projid_map_open,
3174 .write = proc_projid_map_write,
3175 .read = seq_read,
3176 .llseek = seq_lseek,
3177 .release = proc_id_map_release,
3178};
3179
3180static int proc_setgroups_open(struct inode *inode, struct file *file)
3181{
3182 struct user_namespace *ns = NULL;
3183 struct task_struct *task;
3184 int ret;
3185
3186 ret = -ESRCH;
3187 task = get_proc_task(inode);
3188 if (task) {
3189 rcu_read_lock();
3190 ns = get_user_ns(task_cred_xxx(task, user_ns));
3191 rcu_read_unlock();
3192 put_task_struct(task);
3193 }
3194 if (!ns)
3195 goto err;
3196
3197 if (file->f_mode & FMODE_WRITE) {
3198 ret = -EACCES;
3199 if (!ns_capable(ns, CAP_SYS_ADMIN))
3200 goto err_put_ns;
3201 }
3202
3203 ret = single_open(file, &proc_setgroups_show, ns);
3204 if (ret)
3205 goto err_put_ns;
3206
3207 return 0;
3208err_put_ns:
3209 put_user_ns(ns);
3210err:
3211 return ret;
3212}
3213
3214static int proc_setgroups_release(struct inode *inode, struct file *file)
3215{
3216 struct seq_file *seq = file->private_data;
3217 struct user_namespace *ns = seq->private;
3218 int ret = single_release(inode, file);
3219 put_user_ns(ns);
3220 return ret;
3221}
3222
3223static const struct file_operations proc_setgroups_operations = {
3224 .open = proc_setgroups_open,
3225 .write = proc_setgroups_write,
3226 .read = seq_read,
3227 .llseek = seq_lseek,
3228 .release = proc_setgroups_release,
3229};
3230#endif /* CONFIG_USER_NS */
3231
3232static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3233 struct pid *pid, struct task_struct *task)
3234{
3235 int err = lock_trace(task);
3236 if (!err) {
3237 seq_printf(m, fmt: "%08x\n", task->personality);
3238 unlock_trace(task);
3239 }
3240 return err;
3241}
3242
3243#ifdef CONFIG_LIVEPATCH
3244static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3245 struct pid *pid, struct task_struct *task)
3246{
3247 seq_printf(m, "%d\n", task->patch_state);
3248 return 0;
3249}
3250#endif /* CONFIG_LIVEPATCH */
3251
3252#ifdef CONFIG_KSM
3253static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3254 struct pid *pid, struct task_struct *task)
3255{
3256 struct mm_struct *mm;
3257
3258 mm = get_task_mm(task);
3259 if (mm) {
3260 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3261 mmput(mm);
3262 }
3263
3264 return 0;
3265}
3266static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3267 struct pid *pid, struct task_struct *task)
3268{
3269 struct mm_struct *mm;
3270 int ret = 0;
3271
3272 mm = get_task_mm(task);
3273 if (mm) {
3274 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3275 seq_printf(m, "ksm_zero_pages %ld\n", mm_ksm_zero_pages(mm));
3276 seq_printf(m, "ksm_merging_pages %lu\n", mm->ksm_merging_pages);
3277 seq_printf(m, "ksm_process_profit %ld\n", ksm_process_profit(mm));
3278 seq_printf(m, "ksm_merge_any: %s\n",
3279 mm_flags_test(MMF_VM_MERGE_ANY, mm) ? "yes" : "no");
3280 ret = mmap_read_lock_killable(mm);
3281 if (ret) {
3282 mmput(mm);
3283 return ret;
3284 }
3285 seq_printf(m, "ksm_mergeable: %s\n",
3286 ksm_process_mergeable(mm) ? "yes" : "no");
3287 mmap_read_unlock(mm);
3288 mmput(mm);
3289 }
3290
3291 return 0;
3292}
3293#endif /* CONFIG_KSM */
3294
3295#ifdef CONFIG_KSTACK_ERASE_METRICS
3296static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3297 struct pid *pid, struct task_struct *task)
3298{
3299 unsigned long prev_depth = THREAD_SIZE -
3300 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3301 unsigned long depth = THREAD_SIZE -
3302 (task->lowest_stack & (THREAD_SIZE - 1));
3303
3304 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3305 prev_depth, depth);
3306 return 0;
3307}
3308#endif /* CONFIG_KSTACK_ERASE_METRICS */
3309
3310/*
3311 * Thread groups
3312 */
3313static const struct file_operations proc_task_operations;
3314static const struct inode_operations proc_task_inode_operations;
3315
3316static const struct pid_entry tgid_base_stuff[] = {
3317 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3318 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3319 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3320 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3321 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3322#ifdef CONFIG_NET
3323 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3324#endif
3325 REG("environ", S_IRUSR, proc_environ_operations),
3326 REG("auxv", S_IRUSR, proc_auxv_operations),
3327 ONE("status", S_IRUGO, proc_pid_status),
3328 ONE("personality", S_IRUSR, proc_pid_personality),
3329 ONE("limits", S_IRUGO, proc_pid_limits),
3330 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3331#ifdef CONFIG_SCHED_AUTOGROUP
3332 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3333#endif
3334#ifdef CONFIG_TIME_NS
3335 REG("timens_offsets", S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3336#endif
3337 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3338#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3339 ONE("syscall", S_IRUSR, proc_pid_syscall),
3340#endif
3341 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3342 ONE("stat", S_IRUGO, proc_tgid_stat),
3343 ONE("statm", S_IRUGO, proc_pid_statm),
3344 REG("maps", S_IRUGO, proc_pid_maps_operations),
3345#ifdef CONFIG_NUMA
3346 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3347#endif
3348 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3349 LNK("cwd", proc_cwd_link),
3350 LNK("root", proc_root_link),
3351 LNK("exe", proc_exe_link),
3352 REG("mounts", S_IRUGO, proc_mounts_operations),
3353 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3354 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3355#ifdef CONFIG_PROC_PAGE_MONITOR
3356 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3357 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3358 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3359 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3360#endif
3361#ifdef CONFIG_SECURITY
3362 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3363#endif
3364#ifdef CONFIG_KALLSYMS
3365 ONE("wchan", S_IRUGO, proc_pid_wchan),
3366#endif
3367#ifdef CONFIG_STACKTRACE
3368 ONE("stack", S_IRUSR, proc_pid_stack),
3369#endif
3370#ifdef CONFIG_SCHED_INFO
3371 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3372#endif
3373#ifdef CONFIG_LATENCYTOP
3374 REG("latency", S_IRUGO, proc_lstats_operations),
3375#endif
3376#ifdef CONFIG_PROC_PID_CPUSET
3377 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3378#endif
3379#ifdef CONFIG_CGROUPS
3380 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3381#endif
3382#ifdef CONFIG_PROC_CPU_RESCTRL
3383 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3384#endif
3385 ONE("oom_score", S_IRUGO, proc_oom_score),
3386 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3387 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3388#ifdef CONFIG_AUDIT
3389 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3390 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3391#endif
3392#ifdef CONFIG_FAULT_INJECTION
3393 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3394 REG("fail-nth", 0644, proc_fail_nth_operations),
3395#endif
3396#ifdef CONFIG_ELF_CORE
3397 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3398#endif
3399#ifdef CONFIG_TASK_IO_ACCOUNTING
3400 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3401#endif
3402#ifdef CONFIG_USER_NS
3403 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3404 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3405 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3406 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3407#endif
3408#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3409 REG("timers", S_IRUGO, proc_timers_operations),
3410#endif
3411 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3412#ifdef CONFIG_LIVEPATCH
3413 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3414#endif
3415#ifdef CONFIG_KSTACK_ERASE_METRICS
3416 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3417#endif
3418#ifdef CONFIG_PROC_PID_ARCH_STATUS
3419 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3420#endif
3421#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3422 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3423#endif
3424#ifdef CONFIG_KSM
3425 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3426 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3427#endif
3428};
3429
3430static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3431{
3432 return proc_pident_readdir(file, ctx,
3433 ents: tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3434}
3435
3436static const struct file_operations proc_tgid_base_operations = {
3437 .read = generic_read_dir,
3438 .iterate_shared = proc_tgid_base_readdir,
3439 .llseek = generic_file_llseek,
3440};
3441
3442struct pid *tgid_pidfd_to_pid(const struct file *file)
3443{
3444 if (file->f_op != &proc_tgid_base_operations)
3445 return ERR_PTR(error: -EBADF);
3446
3447 return proc_pid(inode: file_inode(f: file));
3448}
3449
3450static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3451{
3452 return proc_pident_lookup(dir, dentry,
3453 p: tgid_base_stuff,
3454 end: tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3455}
3456
3457static const struct inode_operations proc_tgid_base_inode_operations = {
3458 .lookup = proc_tgid_base_lookup,
3459 .getattr = pid_getattr,
3460 .setattr = proc_setattr,
3461 .permission = proc_pid_permission,
3462};
3463
3464/**
3465 * proc_flush_pid - Remove dcache entries for @pid from the /proc dcache.
3466 * @pid: pid that should be flushed.
3467 *
3468 * This function walks a list of inodes (that belong to any proc
3469 * filesystem) that are attached to the pid and flushes them from
3470 * the dentry cache.
3471 *
3472 * It is safe and reasonable to cache /proc entries for a task until
3473 * that task exits. After that they just clog up the dcache with
3474 * useless entries, possibly causing useful dcache entries to be
3475 * flushed instead. This routine is provided to flush those useless
3476 * dcache entries when a process is reaped.
3477 *
3478 * NOTE: This routine is just an optimization so it does not guarantee
3479 * that no dcache entries will exist after a process is reaped
3480 * it just makes it very unlikely that any will persist.
3481 */
3482
3483void proc_flush_pid(struct pid *pid)
3484{
3485 proc_invalidate_siblings_dcache(inodes: &pid->inodes, lock: &pid->lock);
3486}
3487
3488static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3489 struct task_struct *task, const void *ptr)
3490{
3491 struct inode *inode;
3492
3493 inode = proc_pid_make_base_inode(sb: dentry->d_sb, task,
3494 S_IFDIR | S_IRUGO | S_IXUGO);
3495 if (!inode)
3496 return ERR_PTR(error: -ENOENT);
3497
3498 inode->i_op = &proc_tgid_base_inode_operations;
3499 inode->i_fop = &proc_tgid_base_operations;
3500 inode->i_flags|=S_IMMUTABLE;
3501
3502 set_nlink(inode, nlink: nlink_tgid);
3503 pid_update_inode(task, inode);
3504
3505 return d_splice_alias_ops(inode, dentry, &pid_dentry_operations);
3506}
3507
3508struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3509{
3510 struct task_struct *task;
3511 unsigned tgid;
3512 struct proc_fs_info *fs_info;
3513 struct pid_namespace *ns;
3514 struct dentry *result = ERR_PTR(error: -ENOENT);
3515
3516 tgid = name_to_int(qstr: &dentry->d_name);
3517 if (tgid == ~0U)
3518 goto out;
3519
3520 fs_info = proc_sb_info(sb: dentry->d_sb);
3521 ns = fs_info->pid_ns;
3522 rcu_read_lock();
3523 task = find_task_by_pid_ns(nr: tgid, ns);
3524 if (task)
3525 get_task_struct(t: task);
3526 rcu_read_unlock();
3527 if (!task)
3528 goto out;
3529
3530 /* Limit procfs to only ptraceable tasks */
3531 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3532 if (!has_pid_permissions(fs_info, task, hide_pid_min: HIDEPID_NO_ACCESS))
3533 goto out_put_task;
3534 }
3535
3536 result = proc_pid_instantiate(dentry, task, NULL);
3537out_put_task:
3538 put_task_struct(t: task);
3539out:
3540 return result;
3541}
3542
3543/*
3544 * Find the first task with tgid >= tgid
3545 *
3546 */
3547struct tgid_iter {
3548 unsigned int tgid;
3549 struct task_struct *task;
3550};
3551static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3552{
3553 struct pid *pid;
3554
3555 if (iter.task)
3556 put_task_struct(t: iter.task);
3557 rcu_read_lock();
3558retry:
3559 iter.task = NULL;
3560 pid = find_ge_pid(nr: iter.tgid, ns);
3561 if (pid) {
3562 iter.tgid = pid_nr_ns(pid, ns);
3563 iter.task = pid_task(pid, PIDTYPE_TGID);
3564 if (!iter.task) {
3565 iter.tgid += 1;
3566 goto retry;
3567 }
3568 get_task_struct(t: iter.task);
3569 }
3570 rcu_read_unlock();
3571 return iter;
3572}
3573
3574#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3575
3576/* for the /proc/ directory itself, after non-process stuff has been done */
3577int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3578{
3579 struct tgid_iter iter;
3580 struct proc_fs_info *fs_info = proc_sb_info(sb: file_inode(f: file)->i_sb);
3581 struct pid_namespace *ns = proc_pid_ns(sb: file_inode(f: file)->i_sb);
3582 loff_t pos = ctx->pos;
3583
3584 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3585 return 0;
3586
3587 if (pos == TGID_OFFSET - 2) {
3588 struct inode *inode = d_inode(dentry: fs_info->proc_self);
3589 if (!dir_emit(ctx, name: "self", namelen: 4, ino: inode->i_ino, DT_LNK))
3590 return 0;
3591 ctx->pos = pos = pos + 1;
3592 }
3593 if (pos == TGID_OFFSET - 1) {
3594 struct inode *inode = d_inode(dentry: fs_info->proc_thread_self);
3595 if (!dir_emit(ctx, name: "thread-self", namelen: 11, ino: inode->i_ino, DT_LNK))
3596 return 0;
3597 ctx->pos = pos = pos + 1;
3598 }
3599 iter.tgid = pos - TGID_OFFSET;
3600 iter.task = NULL;
3601 for (iter = next_tgid(ns, iter);
3602 iter.task;
3603 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3604 char name[10 + 1];
3605 unsigned int len;
3606
3607 cond_resched();
3608 if (!has_pid_permissions(fs_info, task: iter.task, hide_pid_min: HIDEPID_INVISIBLE))
3609 continue;
3610
3611 len = snprintf(buf: name, size: sizeof(name), fmt: "%u", iter.tgid);
3612 ctx->pos = iter.tgid + TGID_OFFSET;
3613 if (!proc_fill_cache(file, ctx, name, len,
3614 instantiate: proc_pid_instantiate, task: iter.task, NULL)) {
3615 put_task_struct(t: iter.task);
3616 return 0;
3617 }
3618 }
3619 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3620 return 0;
3621}
3622
3623/*
3624 * proc_tid_comm_permission is a special permission function exclusively
3625 * used for the node /proc/<pid>/task/<tid>/comm.
3626 * It bypasses generic permission checks in the case where a task of the same
3627 * task group attempts to access the node.
3628 * The rationale behind this is that glibc and bionic access this node for
3629 * cross thread naming (pthread_set/getname_np(!self)). However, if
3630 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3631 * which locks out the cross thread naming implementation.
3632 * This function makes sure that the node is always accessible for members of
3633 * same thread group.
3634 */
3635static int proc_tid_comm_permission(struct mnt_idmap *idmap,
3636 struct inode *inode, int mask)
3637{
3638 bool is_same_tgroup;
3639 struct task_struct *task;
3640
3641 task = get_proc_task(inode);
3642 if (!task)
3643 return -ESRCH;
3644 is_same_tgroup = same_thread_group(current, p2: task);
3645 put_task_struct(t: task);
3646
3647 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3648 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3649 * read or written by the members of the corresponding
3650 * thread group.
3651 */
3652 return 0;
3653 }
3654
3655 return generic_permission(&nop_mnt_idmap, inode, mask);
3656}
3657
3658static const struct inode_operations proc_tid_comm_inode_operations = {
3659 .setattr = proc_setattr,
3660 .permission = proc_tid_comm_permission,
3661};
3662
3663/*
3664 * Tasks
3665 */
3666static const struct pid_entry tid_base_stuff[] = {
3667 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3668 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3669 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3670#ifdef CONFIG_NET
3671 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3672#endif
3673 REG("environ", S_IRUSR, proc_environ_operations),
3674 REG("auxv", S_IRUSR, proc_auxv_operations),
3675 ONE("status", S_IRUGO, proc_pid_status),
3676 ONE("personality", S_IRUSR, proc_pid_personality),
3677 ONE("limits", S_IRUGO, proc_pid_limits),
3678 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3679 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3680 &proc_tid_comm_inode_operations,
3681 &proc_pid_set_comm_operations, {}),
3682#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3683 ONE("syscall", S_IRUSR, proc_pid_syscall),
3684#endif
3685 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3686 ONE("stat", S_IRUGO, proc_tid_stat),
3687 ONE("statm", S_IRUGO, proc_pid_statm),
3688 REG("maps", S_IRUGO, proc_pid_maps_operations),
3689#ifdef CONFIG_PROC_CHILDREN
3690 REG("children", S_IRUGO, proc_tid_children_operations),
3691#endif
3692#ifdef CONFIG_NUMA
3693 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3694#endif
3695 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3696 LNK("cwd", proc_cwd_link),
3697 LNK("root", proc_root_link),
3698 LNK("exe", proc_exe_link),
3699 REG("mounts", S_IRUGO, proc_mounts_operations),
3700 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3701#ifdef CONFIG_PROC_PAGE_MONITOR
3702 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3703 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3704 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3705 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3706#endif
3707#ifdef CONFIG_SECURITY
3708 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3709#endif
3710#ifdef CONFIG_KALLSYMS
3711 ONE("wchan", S_IRUGO, proc_pid_wchan),
3712#endif
3713#ifdef CONFIG_STACKTRACE
3714 ONE("stack", S_IRUSR, proc_pid_stack),
3715#endif
3716#ifdef CONFIG_SCHED_INFO
3717 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3718#endif
3719#ifdef CONFIG_LATENCYTOP
3720 REG("latency", S_IRUGO, proc_lstats_operations),
3721#endif
3722#ifdef CONFIG_PROC_PID_CPUSET
3723 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3724#endif
3725#ifdef CONFIG_CGROUPS
3726 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3727#endif
3728#ifdef CONFIG_PROC_CPU_RESCTRL
3729 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3730#endif
3731 ONE("oom_score", S_IRUGO, proc_oom_score),
3732 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3733 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3734#ifdef CONFIG_AUDIT
3735 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3736 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3737#endif
3738#ifdef CONFIG_FAULT_INJECTION
3739 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3740 REG("fail-nth", 0644, proc_fail_nth_operations),
3741#endif
3742#ifdef CONFIG_TASK_IO_ACCOUNTING
3743 ONE("io", S_IRUSR, proc_tid_io_accounting),
3744#endif
3745#ifdef CONFIG_USER_NS
3746 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3747 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3748 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3749 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3750#endif
3751#ifdef CONFIG_LIVEPATCH
3752 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3753#endif
3754#ifdef CONFIG_PROC_PID_ARCH_STATUS
3755 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3756#endif
3757#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3758 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3759#endif
3760#ifdef CONFIG_KSM
3761 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3762 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3763#endif
3764};
3765
3766static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3767{
3768 return proc_pident_readdir(file, ctx,
3769 ents: tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3770}
3771
3772static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3773{
3774 return proc_pident_lookup(dir, dentry,
3775 p: tid_base_stuff,
3776 end: tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3777}
3778
3779static const struct file_operations proc_tid_base_operations = {
3780 .read = generic_read_dir,
3781 .iterate_shared = proc_tid_base_readdir,
3782 .llseek = generic_file_llseek,
3783};
3784
3785static const struct inode_operations proc_tid_base_inode_operations = {
3786 .lookup = proc_tid_base_lookup,
3787 .getattr = pid_getattr,
3788 .setattr = proc_setattr,
3789};
3790
3791static struct dentry *proc_task_instantiate(struct dentry *dentry,
3792 struct task_struct *task, const void *ptr)
3793{
3794 struct inode *inode;
3795 inode = proc_pid_make_base_inode(sb: dentry->d_sb, task,
3796 S_IFDIR | S_IRUGO | S_IXUGO);
3797 if (!inode)
3798 return ERR_PTR(error: -ENOENT);
3799
3800 inode->i_op = &proc_tid_base_inode_operations;
3801 inode->i_fop = &proc_tid_base_operations;
3802 inode->i_flags |= S_IMMUTABLE;
3803
3804 set_nlink(inode, nlink: nlink_tid);
3805 pid_update_inode(task, inode);
3806
3807 return d_splice_alias_ops(inode, dentry, &pid_dentry_operations);
3808}
3809
3810static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3811{
3812 struct task_struct *task;
3813 struct task_struct *leader = get_proc_task(inode: dir);
3814 unsigned tid;
3815 struct proc_fs_info *fs_info;
3816 struct pid_namespace *ns;
3817 struct dentry *result = ERR_PTR(error: -ENOENT);
3818
3819 if (!leader)
3820 goto out_no_task;
3821
3822 tid = name_to_int(qstr: &dentry->d_name);
3823 if (tid == ~0U)
3824 goto out;
3825
3826 fs_info = proc_sb_info(sb: dentry->d_sb);
3827 ns = fs_info->pid_ns;
3828 rcu_read_lock();
3829 task = find_task_by_pid_ns(nr: tid, ns);
3830 if (task)
3831 get_task_struct(t: task);
3832 rcu_read_unlock();
3833 if (!task)
3834 goto out;
3835 if (!same_thread_group(p1: leader, p2: task))
3836 goto out_drop_task;
3837
3838 result = proc_task_instantiate(dentry, task, NULL);
3839out_drop_task:
3840 put_task_struct(t: task);
3841out:
3842 put_task_struct(t: leader);
3843out_no_task:
3844 return result;
3845}
3846
3847/*
3848 * Find the first tid of a thread group to return to user space.
3849 *
3850 * Usually this is just the thread group leader, but if the users
3851 * buffer was too small or there was a seek into the middle of the
3852 * directory we have more work todo.
3853 *
3854 * In the case of a short read we start with find_task_by_pid.
3855 *
3856 * In the case of a seek we start with the leader and walk nr
3857 * threads past it.
3858 */
3859static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3860 struct pid_namespace *ns)
3861{
3862 struct task_struct *pos, *task;
3863 unsigned long nr = f_pos;
3864
3865 if (nr != f_pos) /* 32bit overflow? */
3866 return NULL;
3867
3868 rcu_read_lock();
3869 task = pid_task(pid, PIDTYPE_PID);
3870 if (!task)
3871 goto fail;
3872
3873 /* Attempt to start with the tid of a thread */
3874 if (tid && nr) {
3875 pos = find_task_by_pid_ns(nr: tid, ns);
3876 if (pos && same_thread_group(p1: pos, p2: task))
3877 goto found;
3878 }
3879
3880 /* If nr exceeds the number of threads there is nothing todo */
3881 if (nr >= get_nr_threads(task))
3882 goto fail;
3883
3884 /* If we haven't found our starting place yet start
3885 * with the leader and walk nr threads forward.
3886 */
3887 for_each_thread(task, pos) {
3888 if (!nr--)
3889 goto found;
3890 }
3891fail:
3892 pos = NULL;
3893 goto out;
3894found:
3895 get_task_struct(t: pos);
3896out:
3897 rcu_read_unlock();
3898 return pos;
3899}
3900
3901/*
3902 * Find the next thread in the thread list.
3903 * Return NULL if there is an error or no next thread.
3904 *
3905 * The reference to the input task_struct is released.
3906 */
3907static struct task_struct *next_tid(struct task_struct *start)
3908{
3909 struct task_struct *pos = NULL;
3910 rcu_read_lock();
3911 if (pid_alive(p: start)) {
3912 pos = __next_thread(p: start);
3913 if (pos)
3914 get_task_struct(t: pos);
3915 }
3916 rcu_read_unlock();
3917 put_task_struct(t: start);
3918 return pos;
3919}
3920
3921/* for the /proc/TGID/task/ directories */
3922static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3923{
3924 struct inode *inode = file_inode(f: file);
3925 struct task_struct *task;
3926 struct pid_namespace *ns;
3927 int tid;
3928
3929 if (proc_inode_is_dead(inode))
3930 return -ENOENT;
3931
3932 if (!dir_emit_dots(file, ctx))
3933 return 0;
3934
3935 /* We cache the tgid value that the last readdir call couldn't
3936 * return and lseek resets it to 0.
3937 */
3938 ns = proc_pid_ns(sb: inode->i_sb);
3939 tid = (int)(intptr_t)file->private_data;
3940 file->private_data = NULL;
3941 for (task = first_tid(pid: proc_pid(inode), tid, f_pos: ctx->pos - 2, ns);
3942 task;
3943 task = next_tid(start: task), ctx->pos++) {
3944 char name[10 + 1];
3945 unsigned int len;
3946
3947 tid = task_pid_nr_ns(tsk: task, ns);
3948 if (!tid)
3949 continue; /* The task has just exited. */
3950 len = snprintf(buf: name, size: sizeof(name), fmt: "%d", tid);
3951 if (!proc_fill_cache(file, ctx, name, len,
3952 instantiate: proc_task_instantiate, task, NULL)) {
3953 /* returning this tgid failed, save it as the first
3954 * pid for the next readir call */
3955 file->private_data = (void *)(intptr_t)tid;
3956 put_task_struct(t: task);
3957 break;
3958 }
3959 }
3960
3961 return 0;
3962}
3963
3964static int proc_task_getattr(struct mnt_idmap *idmap,
3965 const struct path *path, struct kstat *stat,
3966 u32 request_mask, unsigned int query_flags)
3967{
3968 struct inode *inode = d_inode(dentry: path->dentry);
3969 struct task_struct *p = get_proc_task(inode);
3970 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
3971
3972 if (p) {
3973 stat->nlink += get_nr_threads(task: p);
3974 put_task_struct(t: p);
3975 }
3976
3977 return 0;
3978}
3979
3980/*
3981 * proc_task_readdir() set @file->private_data to a positive integer
3982 * value, so casting that to u64 is safe. generic_llseek_cookie() will
3983 * set @cookie to 0, so casting to an int is safe. The WARN_ON_ONCE() is
3984 * here to catch any unexpected change in behavior either in
3985 * proc_task_readdir() or generic_llseek_cookie().
3986 */
3987static loff_t proc_dir_llseek(struct file *file, loff_t offset, int whence)
3988{
3989 u64 cookie = (u64)(intptr_t)file->private_data;
3990 loff_t off;
3991
3992 off = generic_llseek_cookie(file, offset, whence, cookie: &cookie);
3993 WARN_ON_ONCE(cookie > INT_MAX);
3994 file->private_data = (void *)(intptr_t)cookie; /* serialized by f_pos_lock */
3995 return off;
3996}
3997
3998static const struct inode_operations proc_task_inode_operations = {
3999 .lookup = proc_task_lookup,
4000 .getattr = proc_task_getattr,
4001 .setattr = proc_setattr,
4002 .permission = proc_pid_permission,
4003};
4004
4005static const struct file_operations proc_task_operations = {
4006 .read = generic_read_dir,
4007 .iterate_shared = proc_task_readdir,
4008 .llseek = proc_dir_llseek,
4009};
4010
4011void __init set_proc_pid_nlink(void)
4012{
4013 nlink_tid = pid_entry_nlink(entries: tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
4014 nlink_tgid = pid_entry_nlink(entries: tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
4015}
4016