Loading...
1/*
2 * linux/fs/proc/base.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * proc base directory handling functions
7 *
8 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 * Instead of using magical inumbers to determine the kind of object
10 * we allocate and fill in-core inodes upon lookup. They don't even
11 * go into icache. We cache the reference to task_struct upon lookup too.
12 * Eventually it should become a filesystem in its own. We don't use the
13 * rest of procfs anymore.
14 *
15 *
16 * Changelog:
17 * 17-Jan-2005
18 * Allan Bezerra
19 * Bruna Moreira <bruna.moreira@indt.org.br>
20 * Edjard Mota <edjard.mota@indt.org.br>
21 * Ilias Biris <ilias.biris@indt.org.br>
22 * Mauricio Lin <mauricio.lin@indt.org.br>
23 *
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25 *
26 * A new process specific entry (smaps) included in /proc. It shows the
27 * size of rss for each memory area. The maps entry lacks information
28 * about physical memory size (rss) for each mapped file, i.e.,
29 * rss information for executables and library files.
30 * This additional information is useful for any tools that need to know
31 * about physical memory consumption for a process specific library.
32 *
33 * Changelog:
34 * 21-Feb-2005
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
37 *
38 * ChangeLog:
39 * 10-Mar-2005
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
42 *
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
45 *
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
48 */
49
50#include <asm/uaccess.h>
51
52#include <linux/errno.h>
53#include <linux/time.h>
54#include <linux/proc_fs.h>
55#include <linux/stat.h>
56#include <linux/task_io_accounting_ops.h>
57#include <linux/init.h>
58#include <linux/capability.h>
59#include <linux/file.h>
60#include <linux/fdtable.h>
61#include <linux/string.h>
62#include <linux/seq_file.h>
63#include <linux/namei.h>
64#include <linux/mnt_namespace.h>
65#include <linux/mm.h>
66#include <linux/swap.h>
67#include <linux/rcupdate.h>
68#include <linux/kallsyms.h>
69#include <linux/stacktrace.h>
70#include <linux/resource.h>
71#include <linux/module.h>
72#include <linux/mount.h>
73#include <linux/security.h>
74#include <linux/ptrace.h>
75#include <linux/tracehook.h>
76#include <linux/printk.h>
77#include <linux/cgroup.h>
78#include <linux/cpuset.h>
79#include <linux/audit.h>
80#include <linux/poll.h>
81#include <linux/nsproxy.h>
82#include <linux/oom.h>
83#include <linux/elf.h>
84#include <linux/pid_namespace.h>
85#include <linux/user_namespace.h>
86#include <linux/fs_struct.h>
87#include <linux/slab.h>
88#include <linux/flex_array.h>
89#include <linux/posix-timers.h>
90#ifdef CONFIG_HARDWALL
91#include <asm/hardwall.h>
92#endif
93#include <trace/events/oom.h>
94#include "internal.h"
95#include "fd.h"
96
97/* NOTE:
98 * Implementing inode permission operations in /proc is almost
99 * certainly an error. Permission checks need to happen during
100 * each system call not at open time. The reason is that most of
101 * what we wish to check for permissions in /proc varies at runtime.
102 *
103 * The classic example of a problem is opening file descriptors
104 * in /proc for a task before it execs a suid executable.
105 */
106
107struct pid_entry {
108 const char *name;
109 int len;
110 umode_t mode;
111 const struct inode_operations *iop;
112 const struct file_operations *fop;
113 union proc_op op;
114};
115
116#define NOD(NAME, MODE, IOP, FOP, OP) { \
117 .name = (NAME), \
118 .len = sizeof(NAME) - 1, \
119 .mode = MODE, \
120 .iop = IOP, \
121 .fop = FOP, \
122 .op = OP, \
123}
124
125#define DIR(NAME, MODE, iops, fops) \
126 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
127#define LNK(NAME, get_link) \
128 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
129 &proc_pid_link_inode_operations, NULL, \
130 { .proc_get_link = get_link } )
131#define REG(NAME, MODE, fops) \
132 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
133#define ONE(NAME, MODE, show) \
134 NOD(NAME, (S_IFREG|(MODE)), \
135 NULL, &proc_single_file_operations, \
136 { .proc_show = show } )
137
138/*
139 * Count the number of hardlinks for the pid_entry table, excluding the .
140 * and .. links.
141 */
142static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
143 unsigned int n)
144{
145 unsigned int i;
146 unsigned int count;
147
148 count = 0;
149 for (i = 0; i < n; ++i) {
150 if (S_ISDIR(entries[i].mode))
151 ++count;
152 }
153
154 return count;
155}
156
157static int get_task_root(struct task_struct *task, struct path *root)
158{
159 int result = -ENOENT;
160
161 task_lock(task);
162 if (task->fs) {
163 get_fs_root(task->fs, root);
164 result = 0;
165 }
166 task_unlock(task);
167 return result;
168}
169
170static int proc_cwd_link(struct dentry *dentry, struct path *path)
171{
172 struct task_struct *task = get_proc_task(d_inode(dentry));
173 int result = -ENOENT;
174
175 if (task) {
176 task_lock(task);
177 if (task->fs) {
178 get_fs_pwd(task->fs, path);
179 result = 0;
180 }
181 task_unlock(task);
182 put_task_struct(task);
183 }
184 return result;
185}
186
187static int proc_root_link(struct dentry *dentry, struct path *path)
188{
189 struct task_struct *task = get_proc_task(d_inode(dentry));
190 int result = -ENOENT;
191
192 if (task) {
193 result = get_task_root(task, path);
194 put_task_struct(task);
195 }
196 return result;
197}
198
199static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
200 size_t _count, loff_t *pos)
201{
202 struct task_struct *tsk;
203 struct mm_struct *mm;
204 char *page;
205 unsigned long count = _count;
206 unsigned long arg_start, arg_end, env_start, env_end;
207 unsigned long len1, len2, len;
208 unsigned long p;
209 char c;
210 ssize_t rv;
211
212 BUG_ON(*pos < 0);
213
214 tsk = get_proc_task(file_inode(file));
215 if (!tsk)
216 return -ESRCH;
217 mm = get_task_mm(tsk);
218 put_task_struct(tsk);
219 if (!mm)
220 return 0;
221 /* Check if process spawned far enough to have cmdline. */
222 if (!mm->env_end) {
223 rv = 0;
224 goto out_mmput;
225 }
226
227 page = (char *)__get_free_page(GFP_TEMPORARY);
228 if (!page) {
229 rv = -ENOMEM;
230 goto out_mmput;
231 }
232
233 down_read(&mm->mmap_sem);
234 arg_start = mm->arg_start;
235 arg_end = mm->arg_end;
236 env_start = mm->env_start;
237 env_end = mm->env_end;
238 up_read(&mm->mmap_sem);
239
240 BUG_ON(arg_start > arg_end);
241 BUG_ON(env_start > env_end);
242
243 len1 = arg_end - arg_start;
244 len2 = env_end - env_start;
245
246 /* Empty ARGV. */
247 if (len1 == 0) {
248 rv = 0;
249 goto out_free_page;
250 }
251 /*
252 * Inherently racy -- command line shares address space
253 * with code and data.
254 */
255 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
256 if (rv <= 0)
257 goto out_free_page;
258
259 rv = 0;
260
261 if (c == '\0') {
262 /* Command line (set of strings) occupies whole ARGV. */
263 if (len1 <= *pos)
264 goto out_free_page;
265
266 p = arg_start + *pos;
267 len = len1 - *pos;
268 while (count > 0 && len > 0) {
269 unsigned int _count;
270 int nr_read;
271
272 _count = min3(count, len, PAGE_SIZE);
273 nr_read = access_remote_vm(mm, p, page, _count, 0);
274 if (nr_read < 0)
275 rv = nr_read;
276 if (nr_read <= 0)
277 goto out_free_page;
278
279 if (copy_to_user(buf, page, nr_read)) {
280 rv = -EFAULT;
281 goto out_free_page;
282 }
283
284 p += nr_read;
285 len -= nr_read;
286 buf += nr_read;
287 count -= nr_read;
288 rv += nr_read;
289 }
290 } else {
291 /*
292 * Command line (1 string) occupies ARGV and maybe
293 * extends into ENVP.
294 */
295 if (len1 + len2 <= *pos)
296 goto skip_argv_envp;
297 if (len1 <= *pos)
298 goto skip_argv;
299
300 p = arg_start + *pos;
301 len = len1 - *pos;
302 while (count > 0 && len > 0) {
303 unsigned int _count, l;
304 int nr_read;
305 bool final;
306
307 _count = min3(count, len, PAGE_SIZE);
308 nr_read = access_remote_vm(mm, p, page, _count, 0);
309 if (nr_read < 0)
310 rv = nr_read;
311 if (nr_read <= 0)
312 goto out_free_page;
313
314 /*
315 * Command line can be shorter than whole ARGV
316 * even if last "marker" byte says it is not.
317 */
318 final = false;
319 l = strnlen(page, nr_read);
320 if (l < nr_read) {
321 nr_read = l;
322 final = true;
323 }
324
325 if (copy_to_user(buf, page, nr_read)) {
326 rv = -EFAULT;
327 goto out_free_page;
328 }
329
330 p += nr_read;
331 len -= nr_read;
332 buf += nr_read;
333 count -= nr_read;
334 rv += nr_read;
335
336 if (final)
337 goto out_free_page;
338 }
339skip_argv:
340 /*
341 * Command line (1 string) occupies ARGV and
342 * extends into ENVP.
343 */
344 if (len1 <= *pos) {
345 p = env_start + *pos - len1;
346 len = len1 + len2 - *pos;
347 } else {
348 p = env_start;
349 len = len2;
350 }
351 while (count > 0 && len > 0) {
352 unsigned int _count, l;
353 int nr_read;
354 bool final;
355
356 _count = min3(count, len, PAGE_SIZE);
357 nr_read = access_remote_vm(mm, p, page, _count, 0);
358 if (nr_read < 0)
359 rv = nr_read;
360 if (nr_read <= 0)
361 goto out_free_page;
362
363 /* Find EOS. */
364 final = false;
365 l = strnlen(page, nr_read);
366 if (l < nr_read) {
367 nr_read = l;
368 final = true;
369 }
370
371 if (copy_to_user(buf, page, nr_read)) {
372 rv = -EFAULT;
373 goto out_free_page;
374 }
375
376 p += nr_read;
377 len -= nr_read;
378 buf += nr_read;
379 count -= nr_read;
380 rv += nr_read;
381
382 if (final)
383 goto out_free_page;
384 }
385skip_argv_envp:
386 ;
387 }
388
389out_free_page:
390 free_page((unsigned long)page);
391out_mmput:
392 mmput(mm);
393 if (rv > 0)
394 *pos += rv;
395 return rv;
396}
397
398static const struct file_operations proc_pid_cmdline_ops = {
399 .read = proc_pid_cmdline_read,
400 .llseek = generic_file_llseek,
401};
402
403static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
404 struct pid *pid, struct task_struct *task)
405{
406 struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
407 if (mm && !IS_ERR(mm)) {
408 unsigned int nwords = 0;
409 do {
410 nwords += 2;
411 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
412 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
413 mmput(mm);
414 return 0;
415 } else
416 return PTR_ERR(mm);
417}
418
419
420#ifdef CONFIG_KALLSYMS
421/*
422 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
423 * Returns the resolved symbol. If that fails, simply return the address.
424 */
425static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
426 struct pid *pid, struct task_struct *task)
427{
428 unsigned long wchan;
429 char symname[KSYM_NAME_LEN];
430
431 wchan = get_wchan(task);
432
433 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
434 && !lookup_symbol_name(wchan, symname))
435 seq_printf(m, "%s", symname);
436 else
437 seq_putc(m, '0');
438
439 return 0;
440}
441#endif /* CONFIG_KALLSYMS */
442
443static int lock_trace(struct task_struct *task)
444{
445 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
446 if (err)
447 return err;
448 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
449 mutex_unlock(&task->signal->cred_guard_mutex);
450 return -EPERM;
451 }
452 return 0;
453}
454
455static void unlock_trace(struct task_struct *task)
456{
457 mutex_unlock(&task->signal->cred_guard_mutex);
458}
459
460#ifdef CONFIG_STACKTRACE
461
462#define MAX_STACK_TRACE_DEPTH 64
463
464static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
465 struct pid *pid, struct task_struct *task)
466{
467 struct stack_trace trace;
468 unsigned long *entries;
469 int err;
470 int i;
471
472 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
473 if (!entries)
474 return -ENOMEM;
475
476 trace.nr_entries = 0;
477 trace.max_entries = MAX_STACK_TRACE_DEPTH;
478 trace.entries = entries;
479 trace.skip = 0;
480
481 err = lock_trace(task);
482 if (!err) {
483 save_stack_trace_tsk(task, &trace);
484
485 for (i = 0; i < trace.nr_entries; i++) {
486 seq_printf(m, "[<%pK>] %pS\n",
487 (void *)entries[i], (void *)entries[i]);
488 }
489 unlock_trace(task);
490 }
491 kfree(entries);
492
493 return err;
494}
495#endif
496
497#ifdef CONFIG_SCHED_INFO
498/*
499 * Provides /proc/PID/schedstat
500 */
501static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
502 struct pid *pid, struct task_struct *task)
503{
504 if (unlikely(!sched_info_on()))
505 seq_printf(m, "0 0 0\n");
506 else
507 seq_printf(m, "%llu %llu %lu\n",
508 (unsigned long long)task->se.sum_exec_runtime,
509 (unsigned long long)task->sched_info.run_delay,
510 task->sched_info.pcount);
511
512 return 0;
513}
514#endif
515
516#ifdef CONFIG_LATENCYTOP
517static int lstats_show_proc(struct seq_file *m, void *v)
518{
519 int i;
520 struct inode *inode = m->private;
521 struct task_struct *task = get_proc_task(inode);
522
523 if (!task)
524 return -ESRCH;
525 seq_puts(m, "Latency Top version : v0.1\n");
526 for (i = 0; i < 32; i++) {
527 struct latency_record *lr = &task->latency_record[i];
528 if (lr->backtrace[0]) {
529 int q;
530 seq_printf(m, "%i %li %li",
531 lr->count, lr->time, lr->max);
532 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
533 unsigned long bt = lr->backtrace[q];
534 if (!bt)
535 break;
536 if (bt == ULONG_MAX)
537 break;
538 seq_printf(m, " %ps", (void *)bt);
539 }
540 seq_putc(m, '\n');
541 }
542
543 }
544 put_task_struct(task);
545 return 0;
546}
547
548static int lstats_open(struct inode *inode, struct file *file)
549{
550 return single_open(file, lstats_show_proc, inode);
551}
552
553static ssize_t lstats_write(struct file *file, const char __user *buf,
554 size_t count, loff_t *offs)
555{
556 struct task_struct *task = get_proc_task(file_inode(file));
557
558 if (!task)
559 return -ESRCH;
560 clear_all_latency_tracing(task);
561 put_task_struct(task);
562
563 return count;
564}
565
566static const struct file_operations proc_lstats_operations = {
567 .open = lstats_open,
568 .read = seq_read,
569 .write = lstats_write,
570 .llseek = seq_lseek,
571 .release = single_release,
572};
573
574#endif
575
576static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
577 struct pid *pid, struct task_struct *task)
578{
579 unsigned long totalpages = totalram_pages + total_swap_pages;
580 unsigned long points = 0;
581
582 read_lock(&tasklist_lock);
583 if (pid_alive(task))
584 points = oom_badness(task, NULL, NULL, totalpages) *
585 1000 / totalpages;
586 read_unlock(&tasklist_lock);
587 seq_printf(m, "%lu\n", points);
588
589 return 0;
590}
591
592struct limit_names {
593 const char *name;
594 const char *unit;
595};
596
597static const struct limit_names lnames[RLIM_NLIMITS] = {
598 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
599 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
600 [RLIMIT_DATA] = {"Max data size", "bytes"},
601 [RLIMIT_STACK] = {"Max stack size", "bytes"},
602 [RLIMIT_CORE] = {"Max core file size", "bytes"},
603 [RLIMIT_RSS] = {"Max resident set", "bytes"},
604 [RLIMIT_NPROC] = {"Max processes", "processes"},
605 [RLIMIT_NOFILE] = {"Max open files", "files"},
606 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
607 [RLIMIT_AS] = {"Max address space", "bytes"},
608 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
609 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
610 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
611 [RLIMIT_NICE] = {"Max nice priority", NULL},
612 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
613 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
614};
615
616/* Display limits for a process */
617static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
618 struct pid *pid, struct task_struct *task)
619{
620 unsigned int i;
621 unsigned long flags;
622
623 struct rlimit rlim[RLIM_NLIMITS];
624
625 if (!lock_task_sighand(task, &flags))
626 return 0;
627 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
628 unlock_task_sighand(task, &flags);
629
630 /*
631 * print the file header
632 */
633 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
634 "Limit", "Soft Limit", "Hard Limit", "Units");
635
636 for (i = 0; i < RLIM_NLIMITS; i++) {
637 if (rlim[i].rlim_cur == RLIM_INFINITY)
638 seq_printf(m, "%-25s %-20s ",
639 lnames[i].name, "unlimited");
640 else
641 seq_printf(m, "%-25s %-20lu ",
642 lnames[i].name, rlim[i].rlim_cur);
643
644 if (rlim[i].rlim_max == RLIM_INFINITY)
645 seq_printf(m, "%-20s ", "unlimited");
646 else
647 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
648
649 if (lnames[i].unit)
650 seq_printf(m, "%-10s\n", lnames[i].unit);
651 else
652 seq_putc(m, '\n');
653 }
654
655 return 0;
656}
657
658#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
659static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
660 struct pid *pid, struct task_struct *task)
661{
662 long nr;
663 unsigned long args[6], sp, pc;
664 int res;
665
666 res = lock_trace(task);
667 if (res)
668 return res;
669
670 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
671 seq_puts(m, "running\n");
672 else if (nr < 0)
673 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
674 else
675 seq_printf(m,
676 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
677 nr,
678 args[0], args[1], args[2], args[3], args[4], args[5],
679 sp, pc);
680 unlock_trace(task);
681
682 return 0;
683}
684#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
685
686/************************************************************************/
687/* Here the fs part begins */
688/************************************************************************/
689
690/* permission checks */
691static int proc_fd_access_allowed(struct inode *inode)
692{
693 struct task_struct *task;
694 int allowed = 0;
695 /* Allow access to a task's file descriptors if it is us or we
696 * may use ptrace attach to the process and find out that
697 * information.
698 */
699 task = get_proc_task(inode);
700 if (task) {
701 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
702 put_task_struct(task);
703 }
704 return allowed;
705}
706
707int proc_setattr(struct dentry *dentry, struct iattr *attr)
708{
709 int error;
710 struct inode *inode = d_inode(dentry);
711
712 if (attr->ia_valid & ATTR_MODE)
713 return -EPERM;
714
715 error = inode_change_ok(inode, attr);
716 if (error)
717 return error;
718
719 setattr_copy(inode, attr);
720 mark_inode_dirty(inode);
721 return 0;
722}
723
724/*
725 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
726 * or euid/egid (for hide_pid_min=2)?
727 */
728static bool has_pid_permissions(struct pid_namespace *pid,
729 struct task_struct *task,
730 int hide_pid_min)
731{
732 if (pid->hide_pid < hide_pid_min)
733 return true;
734 if (in_group_p(pid->pid_gid))
735 return true;
736 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
737}
738
739
740static int proc_pid_permission(struct inode *inode, int mask)
741{
742 struct pid_namespace *pid = inode->i_sb->s_fs_info;
743 struct task_struct *task;
744 bool has_perms;
745
746 task = get_proc_task(inode);
747 if (!task)
748 return -ESRCH;
749 has_perms = has_pid_permissions(pid, task, 1);
750 put_task_struct(task);
751
752 if (!has_perms) {
753 if (pid->hide_pid == 2) {
754 /*
755 * Let's make getdents(), stat(), and open()
756 * consistent with each other. If a process
757 * may not stat() a file, it shouldn't be seen
758 * in procfs at all.
759 */
760 return -ENOENT;
761 }
762
763 return -EPERM;
764 }
765 return generic_permission(inode, mask);
766}
767
768
769
770static const struct inode_operations proc_def_inode_operations = {
771 .setattr = proc_setattr,
772};
773
774static int proc_single_show(struct seq_file *m, void *v)
775{
776 struct inode *inode = m->private;
777 struct pid_namespace *ns;
778 struct pid *pid;
779 struct task_struct *task;
780 int ret;
781
782 ns = inode->i_sb->s_fs_info;
783 pid = proc_pid(inode);
784 task = get_pid_task(pid, PIDTYPE_PID);
785 if (!task)
786 return -ESRCH;
787
788 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
789
790 put_task_struct(task);
791 return ret;
792}
793
794static int proc_single_open(struct inode *inode, struct file *filp)
795{
796 return single_open(filp, proc_single_show, inode);
797}
798
799static const struct file_operations proc_single_file_operations = {
800 .open = proc_single_open,
801 .read = seq_read,
802 .llseek = seq_lseek,
803 .release = single_release,
804};
805
806
807struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
808{
809 struct task_struct *task = get_proc_task(inode);
810 struct mm_struct *mm = ERR_PTR(-ESRCH);
811
812 if (task) {
813 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
814 put_task_struct(task);
815
816 if (!IS_ERR_OR_NULL(mm)) {
817 /* ensure this mm_struct can't be freed */
818 atomic_inc(&mm->mm_count);
819 /* but do not pin its memory */
820 mmput(mm);
821 }
822 }
823
824 return mm;
825}
826
827static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
828{
829 struct mm_struct *mm = proc_mem_open(inode, mode);
830
831 if (IS_ERR(mm))
832 return PTR_ERR(mm);
833
834 file->private_data = mm;
835 return 0;
836}
837
838static int mem_open(struct inode *inode, struct file *file)
839{
840 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
841
842 /* OK to pass negative loff_t, we can catch out-of-range */
843 file->f_mode |= FMODE_UNSIGNED_OFFSET;
844
845 return ret;
846}
847
848static ssize_t mem_rw(struct file *file, char __user *buf,
849 size_t count, loff_t *ppos, int write)
850{
851 struct mm_struct *mm = file->private_data;
852 unsigned long addr = *ppos;
853 ssize_t copied;
854 char *page;
855
856 if (!mm)
857 return 0;
858
859 page = (char *)__get_free_page(GFP_TEMPORARY);
860 if (!page)
861 return -ENOMEM;
862
863 copied = 0;
864 if (!atomic_inc_not_zero(&mm->mm_users))
865 goto free;
866
867 while (count > 0) {
868 int this_len = min_t(int, count, PAGE_SIZE);
869
870 if (write && copy_from_user(page, buf, this_len)) {
871 copied = -EFAULT;
872 break;
873 }
874
875 this_len = access_remote_vm(mm, addr, page, this_len, write);
876 if (!this_len) {
877 if (!copied)
878 copied = -EIO;
879 break;
880 }
881
882 if (!write && copy_to_user(buf, page, this_len)) {
883 copied = -EFAULT;
884 break;
885 }
886
887 buf += this_len;
888 addr += this_len;
889 copied += this_len;
890 count -= this_len;
891 }
892 *ppos = addr;
893
894 mmput(mm);
895free:
896 free_page((unsigned long) page);
897 return copied;
898}
899
900static ssize_t mem_read(struct file *file, char __user *buf,
901 size_t count, loff_t *ppos)
902{
903 return mem_rw(file, buf, count, ppos, 0);
904}
905
906static ssize_t mem_write(struct file *file, const char __user *buf,
907 size_t count, loff_t *ppos)
908{
909 return mem_rw(file, (char __user*)buf, count, ppos, 1);
910}
911
912loff_t mem_lseek(struct file *file, loff_t offset, int orig)
913{
914 switch (orig) {
915 case 0:
916 file->f_pos = offset;
917 break;
918 case 1:
919 file->f_pos += offset;
920 break;
921 default:
922 return -EINVAL;
923 }
924 force_successful_syscall_return();
925 return file->f_pos;
926}
927
928static int mem_release(struct inode *inode, struct file *file)
929{
930 struct mm_struct *mm = file->private_data;
931 if (mm)
932 mmdrop(mm);
933 return 0;
934}
935
936static const struct file_operations proc_mem_operations = {
937 .llseek = mem_lseek,
938 .read = mem_read,
939 .write = mem_write,
940 .open = mem_open,
941 .release = mem_release,
942};
943
944static int environ_open(struct inode *inode, struct file *file)
945{
946 return __mem_open(inode, file, PTRACE_MODE_READ);
947}
948
949static ssize_t environ_read(struct file *file, char __user *buf,
950 size_t count, loff_t *ppos)
951{
952 char *page;
953 unsigned long src = *ppos;
954 int ret = 0;
955 struct mm_struct *mm = file->private_data;
956 unsigned long env_start, env_end;
957
958 /* Ensure the process spawned far enough to have an environment. */
959 if (!mm || !mm->env_end)
960 return 0;
961
962 page = (char *)__get_free_page(GFP_TEMPORARY);
963 if (!page)
964 return -ENOMEM;
965
966 ret = 0;
967 if (!atomic_inc_not_zero(&mm->mm_users))
968 goto free;
969
970 down_read(&mm->mmap_sem);
971 env_start = mm->env_start;
972 env_end = mm->env_end;
973 up_read(&mm->mmap_sem);
974
975 while (count > 0) {
976 size_t this_len, max_len;
977 int retval;
978
979 if (src >= (env_end - env_start))
980 break;
981
982 this_len = env_end - (env_start + src);
983
984 max_len = min_t(size_t, PAGE_SIZE, count);
985 this_len = min(max_len, this_len);
986
987 retval = access_remote_vm(mm, (env_start + src),
988 page, this_len, 0);
989
990 if (retval <= 0) {
991 ret = retval;
992 break;
993 }
994
995 if (copy_to_user(buf, page, retval)) {
996 ret = -EFAULT;
997 break;
998 }
999
1000 ret += retval;
1001 src += retval;
1002 buf += retval;
1003 count -= retval;
1004 }
1005 *ppos = src;
1006 mmput(mm);
1007
1008free:
1009 free_page((unsigned long) page);
1010 return ret;
1011}
1012
1013static const struct file_operations proc_environ_operations = {
1014 .open = environ_open,
1015 .read = environ_read,
1016 .llseek = generic_file_llseek,
1017 .release = mem_release,
1018};
1019
1020static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1021 loff_t *ppos)
1022{
1023 struct task_struct *task = get_proc_task(file_inode(file));
1024 char buffer[PROC_NUMBUF];
1025 int oom_adj = OOM_ADJUST_MIN;
1026 size_t len;
1027 unsigned long flags;
1028
1029 if (!task)
1030 return -ESRCH;
1031 if (lock_task_sighand(task, &flags)) {
1032 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1033 oom_adj = OOM_ADJUST_MAX;
1034 else
1035 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1036 OOM_SCORE_ADJ_MAX;
1037 unlock_task_sighand(task, &flags);
1038 }
1039 put_task_struct(task);
1040 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1041 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1042}
1043
1044/*
1045 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1046 * kernels. The effective policy is defined by oom_score_adj, which has a
1047 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1048 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1049 * Processes that become oom disabled via oom_adj will still be oom disabled
1050 * with this implementation.
1051 *
1052 * oom_adj cannot be removed since existing userspace binaries use it.
1053 */
1054static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1055 size_t count, loff_t *ppos)
1056{
1057 struct task_struct *task;
1058 char buffer[PROC_NUMBUF];
1059 int oom_adj;
1060 unsigned long flags;
1061 int err;
1062
1063 memset(buffer, 0, sizeof(buffer));
1064 if (count > sizeof(buffer) - 1)
1065 count = sizeof(buffer) - 1;
1066 if (copy_from_user(buffer, buf, count)) {
1067 err = -EFAULT;
1068 goto out;
1069 }
1070
1071 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1072 if (err)
1073 goto out;
1074 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1075 oom_adj != OOM_DISABLE) {
1076 err = -EINVAL;
1077 goto out;
1078 }
1079
1080 task = get_proc_task(file_inode(file));
1081 if (!task) {
1082 err = -ESRCH;
1083 goto out;
1084 }
1085
1086 task_lock(task);
1087 if (!task->mm) {
1088 err = -EINVAL;
1089 goto err_task_lock;
1090 }
1091
1092 if (!lock_task_sighand(task, &flags)) {
1093 err = -ESRCH;
1094 goto err_task_lock;
1095 }
1096
1097 /*
1098 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1099 * value is always attainable.
1100 */
1101 if (oom_adj == OOM_ADJUST_MAX)
1102 oom_adj = OOM_SCORE_ADJ_MAX;
1103 else
1104 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1105
1106 if (oom_adj < task->signal->oom_score_adj &&
1107 !capable(CAP_SYS_RESOURCE)) {
1108 err = -EACCES;
1109 goto err_sighand;
1110 }
1111
1112 /*
1113 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1114 * /proc/pid/oom_score_adj instead.
1115 */
1116 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1117 current->comm, task_pid_nr(current), task_pid_nr(task),
1118 task_pid_nr(task));
1119
1120 task->signal->oom_score_adj = oom_adj;
1121 trace_oom_score_adj_update(task);
1122err_sighand:
1123 unlock_task_sighand(task, &flags);
1124err_task_lock:
1125 task_unlock(task);
1126 put_task_struct(task);
1127out:
1128 return err < 0 ? err : count;
1129}
1130
1131static const struct file_operations proc_oom_adj_operations = {
1132 .read = oom_adj_read,
1133 .write = oom_adj_write,
1134 .llseek = generic_file_llseek,
1135};
1136
1137static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1138 size_t count, loff_t *ppos)
1139{
1140 struct task_struct *task = get_proc_task(file_inode(file));
1141 char buffer[PROC_NUMBUF];
1142 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1143 unsigned long flags;
1144 size_t len;
1145
1146 if (!task)
1147 return -ESRCH;
1148 if (lock_task_sighand(task, &flags)) {
1149 oom_score_adj = task->signal->oom_score_adj;
1150 unlock_task_sighand(task, &flags);
1151 }
1152 put_task_struct(task);
1153 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1154 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1155}
1156
1157static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1158 size_t count, loff_t *ppos)
1159{
1160 struct task_struct *task;
1161 char buffer[PROC_NUMBUF];
1162 unsigned long flags;
1163 int oom_score_adj;
1164 int err;
1165
1166 memset(buffer, 0, sizeof(buffer));
1167 if (count > sizeof(buffer) - 1)
1168 count = sizeof(buffer) - 1;
1169 if (copy_from_user(buffer, buf, count)) {
1170 err = -EFAULT;
1171 goto out;
1172 }
1173
1174 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1175 if (err)
1176 goto out;
1177 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1178 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1179 err = -EINVAL;
1180 goto out;
1181 }
1182
1183 task = get_proc_task(file_inode(file));
1184 if (!task) {
1185 err = -ESRCH;
1186 goto out;
1187 }
1188
1189 task_lock(task);
1190 if (!task->mm) {
1191 err = -EINVAL;
1192 goto err_task_lock;
1193 }
1194
1195 if (!lock_task_sighand(task, &flags)) {
1196 err = -ESRCH;
1197 goto err_task_lock;
1198 }
1199
1200 if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1201 !capable(CAP_SYS_RESOURCE)) {
1202 err = -EACCES;
1203 goto err_sighand;
1204 }
1205
1206 task->signal->oom_score_adj = (short)oom_score_adj;
1207 if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1208 task->signal->oom_score_adj_min = (short)oom_score_adj;
1209 trace_oom_score_adj_update(task);
1210
1211err_sighand:
1212 unlock_task_sighand(task, &flags);
1213err_task_lock:
1214 task_unlock(task);
1215 put_task_struct(task);
1216out:
1217 return err < 0 ? err : count;
1218}
1219
1220static const struct file_operations proc_oom_score_adj_operations = {
1221 .read = oom_score_adj_read,
1222 .write = oom_score_adj_write,
1223 .llseek = default_llseek,
1224};
1225
1226#ifdef CONFIG_AUDITSYSCALL
1227#define TMPBUFLEN 21
1228static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1229 size_t count, loff_t *ppos)
1230{
1231 struct inode * inode = file_inode(file);
1232 struct task_struct *task = get_proc_task(inode);
1233 ssize_t length;
1234 char tmpbuf[TMPBUFLEN];
1235
1236 if (!task)
1237 return -ESRCH;
1238 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1239 from_kuid(file->f_cred->user_ns,
1240 audit_get_loginuid(task)));
1241 put_task_struct(task);
1242 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1243}
1244
1245static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1246 size_t count, loff_t *ppos)
1247{
1248 struct inode * inode = file_inode(file);
1249 uid_t loginuid;
1250 kuid_t kloginuid;
1251 int rv;
1252
1253 rcu_read_lock();
1254 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1255 rcu_read_unlock();
1256 return -EPERM;
1257 }
1258 rcu_read_unlock();
1259
1260 if (*ppos != 0) {
1261 /* No partial writes. */
1262 return -EINVAL;
1263 }
1264
1265 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1266 if (rv < 0)
1267 return rv;
1268
1269 /* is userspace tring to explicitly UNSET the loginuid? */
1270 if (loginuid == AUDIT_UID_UNSET) {
1271 kloginuid = INVALID_UID;
1272 } else {
1273 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1274 if (!uid_valid(kloginuid))
1275 return -EINVAL;
1276 }
1277
1278 rv = audit_set_loginuid(kloginuid);
1279 if (rv < 0)
1280 return rv;
1281 return count;
1282}
1283
1284static const struct file_operations proc_loginuid_operations = {
1285 .read = proc_loginuid_read,
1286 .write = proc_loginuid_write,
1287 .llseek = generic_file_llseek,
1288};
1289
1290static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1291 size_t count, loff_t *ppos)
1292{
1293 struct inode * inode = file_inode(file);
1294 struct task_struct *task = get_proc_task(inode);
1295 ssize_t length;
1296 char tmpbuf[TMPBUFLEN];
1297
1298 if (!task)
1299 return -ESRCH;
1300 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1301 audit_get_sessionid(task));
1302 put_task_struct(task);
1303 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1304}
1305
1306static const struct file_operations proc_sessionid_operations = {
1307 .read = proc_sessionid_read,
1308 .llseek = generic_file_llseek,
1309};
1310#endif
1311
1312#ifdef CONFIG_FAULT_INJECTION
1313static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1314 size_t count, loff_t *ppos)
1315{
1316 struct task_struct *task = get_proc_task(file_inode(file));
1317 char buffer[PROC_NUMBUF];
1318 size_t len;
1319 int make_it_fail;
1320
1321 if (!task)
1322 return -ESRCH;
1323 make_it_fail = task->make_it_fail;
1324 put_task_struct(task);
1325
1326 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1327
1328 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1329}
1330
1331static ssize_t proc_fault_inject_write(struct file * file,
1332 const char __user * buf, size_t count, loff_t *ppos)
1333{
1334 struct task_struct *task;
1335 char buffer[PROC_NUMBUF];
1336 int make_it_fail;
1337 int rv;
1338
1339 if (!capable(CAP_SYS_RESOURCE))
1340 return -EPERM;
1341 memset(buffer, 0, sizeof(buffer));
1342 if (count > sizeof(buffer) - 1)
1343 count = sizeof(buffer) - 1;
1344 if (copy_from_user(buffer, buf, count))
1345 return -EFAULT;
1346 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1347 if (rv < 0)
1348 return rv;
1349 if (make_it_fail < 0 || make_it_fail > 1)
1350 return -EINVAL;
1351
1352 task = get_proc_task(file_inode(file));
1353 if (!task)
1354 return -ESRCH;
1355 task->make_it_fail = make_it_fail;
1356 put_task_struct(task);
1357
1358 return count;
1359}
1360
1361static const struct file_operations proc_fault_inject_operations = {
1362 .read = proc_fault_inject_read,
1363 .write = proc_fault_inject_write,
1364 .llseek = generic_file_llseek,
1365};
1366#endif
1367
1368
1369#ifdef CONFIG_SCHED_DEBUG
1370/*
1371 * Print out various scheduling related per-task fields:
1372 */
1373static int sched_show(struct seq_file *m, void *v)
1374{
1375 struct inode *inode = m->private;
1376 struct task_struct *p;
1377
1378 p = get_proc_task(inode);
1379 if (!p)
1380 return -ESRCH;
1381 proc_sched_show_task(p, m);
1382
1383 put_task_struct(p);
1384
1385 return 0;
1386}
1387
1388static ssize_t
1389sched_write(struct file *file, const char __user *buf,
1390 size_t count, loff_t *offset)
1391{
1392 struct inode *inode = file_inode(file);
1393 struct task_struct *p;
1394
1395 p = get_proc_task(inode);
1396 if (!p)
1397 return -ESRCH;
1398 proc_sched_set_task(p);
1399
1400 put_task_struct(p);
1401
1402 return count;
1403}
1404
1405static int sched_open(struct inode *inode, struct file *filp)
1406{
1407 return single_open(filp, sched_show, inode);
1408}
1409
1410static const struct file_operations proc_pid_sched_operations = {
1411 .open = sched_open,
1412 .read = seq_read,
1413 .write = sched_write,
1414 .llseek = seq_lseek,
1415 .release = single_release,
1416};
1417
1418#endif
1419
1420#ifdef CONFIG_SCHED_AUTOGROUP
1421/*
1422 * Print out autogroup related information:
1423 */
1424static int sched_autogroup_show(struct seq_file *m, void *v)
1425{
1426 struct inode *inode = m->private;
1427 struct task_struct *p;
1428
1429 p = get_proc_task(inode);
1430 if (!p)
1431 return -ESRCH;
1432 proc_sched_autogroup_show_task(p, m);
1433
1434 put_task_struct(p);
1435
1436 return 0;
1437}
1438
1439static ssize_t
1440sched_autogroup_write(struct file *file, const char __user *buf,
1441 size_t count, loff_t *offset)
1442{
1443 struct inode *inode = file_inode(file);
1444 struct task_struct *p;
1445 char buffer[PROC_NUMBUF];
1446 int nice;
1447 int err;
1448
1449 memset(buffer, 0, sizeof(buffer));
1450 if (count > sizeof(buffer) - 1)
1451 count = sizeof(buffer) - 1;
1452 if (copy_from_user(buffer, buf, count))
1453 return -EFAULT;
1454
1455 err = kstrtoint(strstrip(buffer), 0, &nice);
1456 if (err < 0)
1457 return err;
1458
1459 p = get_proc_task(inode);
1460 if (!p)
1461 return -ESRCH;
1462
1463 err = proc_sched_autogroup_set_nice(p, nice);
1464 if (err)
1465 count = err;
1466
1467 put_task_struct(p);
1468
1469 return count;
1470}
1471
1472static int sched_autogroup_open(struct inode *inode, struct file *filp)
1473{
1474 int ret;
1475
1476 ret = single_open(filp, sched_autogroup_show, NULL);
1477 if (!ret) {
1478 struct seq_file *m = filp->private_data;
1479
1480 m->private = inode;
1481 }
1482 return ret;
1483}
1484
1485static const struct file_operations proc_pid_sched_autogroup_operations = {
1486 .open = sched_autogroup_open,
1487 .read = seq_read,
1488 .write = sched_autogroup_write,
1489 .llseek = seq_lseek,
1490 .release = single_release,
1491};
1492
1493#endif /* CONFIG_SCHED_AUTOGROUP */
1494
1495static ssize_t comm_write(struct file *file, const char __user *buf,
1496 size_t count, loff_t *offset)
1497{
1498 struct inode *inode = file_inode(file);
1499 struct task_struct *p;
1500 char buffer[TASK_COMM_LEN];
1501 const size_t maxlen = sizeof(buffer) - 1;
1502
1503 memset(buffer, 0, sizeof(buffer));
1504 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1505 return -EFAULT;
1506
1507 p = get_proc_task(inode);
1508 if (!p)
1509 return -ESRCH;
1510
1511 if (same_thread_group(current, p))
1512 set_task_comm(p, buffer);
1513 else
1514 count = -EINVAL;
1515
1516 put_task_struct(p);
1517
1518 return count;
1519}
1520
1521static int comm_show(struct seq_file *m, void *v)
1522{
1523 struct inode *inode = m->private;
1524 struct task_struct *p;
1525
1526 p = get_proc_task(inode);
1527 if (!p)
1528 return -ESRCH;
1529
1530 task_lock(p);
1531 seq_printf(m, "%s\n", p->comm);
1532 task_unlock(p);
1533
1534 put_task_struct(p);
1535
1536 return 0;
1537}
1538
1539static int comm_open(struct inode *inode, struct file *filp)
1540{
1541 return single_open(filp, comm_show, inode);
1542}
1543
1544static const struct file_operations proc_pid_set_comm_operations = {
1545 .open = comm_open,
1546 .read = seq_read,
1547 .write = comm_write,
1548 .llseek = seq_lseek,
1549 .release = single_release,
1550};
1551
1552static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1553{
1554 struct task_struct *task;
1555 struct mm_struct *mm;
1556 struct file *exe_file;
1557
1558 task = get_proc_task(d_inode(dentry));
1559 if (!task)
1560 return -ENOENT;
1561 mm = get_task_mm(task);
1562 put_task_struct(task);
1563 if (!mm)
1564 return -ENOENT;
1565 exe_file = get_mm_exe_file(mm);
1566 mmput(mm);
1567 if (exe_file) {
1568 *exe_path = exe_file->f_path;
1569 path_get(&exe_file->f_path);
1570 fput(exe_file);
1571 return 0;
1572 } else
1573 return -ENOENT;
1574}
1575
1576static const char *proc_pid_get_link(struct dentry *dentry,
1577 struct inode *inode,
1578 struct delayed_call *done)
1579{
1580 struct path path;
1581 int error = -EACCES;
1582
1583 if (!dentry)
1584 return ERR_PTR(-ECHILD);
1585
1586 /* Are we allowed to snoop on the tasks file descriptors? */
1587 if (!proc_fd_access_allowed(inode))
1588 goto out;
1589
1590 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1591 if (error)
1592 goto out;
1593
1594 nd_jump_link(&path);
1595 return NULL;
1596out:
1597 return ERR_PTR(error);
1598}
1599
1600static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1601{
1602 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1603 char *pathname;
1604 int len;
1605
1606 if (!tmp)
1607 return -ENOMEM;
1608
1609 pathname = d_path(path, tmp, PAGE_SIZE);
1610 len = PTR_ERR(pathname);
1611 if (IS_ERR(pathname))
1612 goto out;
1613 len = tmp + PAGE_SIZE - 1 - pathname;
1614
1615 if (len > buflen)
1616 len = buflen;
1617 if (copy_to_user(buffer, pathname, len))
1618 len = -EFAULT;
1619 out:
1620 free_page((unsigned long)tmp);
1621 return len;
1622}
1623
1624static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1625{
1626 int error = -EACCES;
1627 struct inode *inode = d_inode(dentry);
1628 struct path path;
1629
1630 /* Are we allowed to snoop on the tasks file descriptors? */
1631 if (!proc_fd_access_allowed(inode))
1632 goto out;
1633
1634 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1635 if (error)
1636 goto out;
1637
1638 error = do_proc_readlink(&path, buffer, buflen);
1639 path_put(&path);
1640out:
1641 return error;
1642}
1643
1644const struct inode_operations proc_pid_link_inode_operations = {
1645 .readlink = proc_pid_readlink,
1646 .get_link = proc_pid_get_link,
1647 .setattr = proc_setattr,
1648};
1649
1650
1651/* building an inode */
1652
1653struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1654{
1655 struct inode * inode;
1656 struct proc_inode *ei;
1657 const struct cred *cred;
1658
1659 /* We need a new inode */
1660
1661 inode = new_inode(sb);
1662 if (!inode)
1663 goto out;
1664
1665 /* Common stuff */
1666 ei = PROC_I(inode);
1667 inode->i_ino = get_next_ino();
1668 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1669 inode->i_op = &proc_def_inode_operations;
1670
1671 /*
1672 * grab the reference to task.
1673 */
1674 ei->pid = get_task_pid(task, PIDTYPE_PID);
1675 if (!ei->pid)
1676 goto out_unlock;
1677
1678 if (task_dumpable(task)) {
1679 rcu_read_lock();
1680 cred = __task_cred(task);
1681 inode->i_uid = cred->euid;
1682 inode->i_gid = cred->egid;
1683 rcu_read_unlock();
1684 }
1685 security_task_to_inode(task, inode);
1686
1687out:
1688 return inode;
1689
1690out_unlock:
1691 iput(inode);
1692 return NULL;
1693}
1694
1695int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1696{
1697 struct inode *inode = d_inode(dentry);
1698 struct task_struct *task;
1699 const struct cred *cred;
1700 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1701
1702 generic_fillattr(inode, stat);
1703
1704 rcu_read_lock();
1705 stat->uid = GLOBAL_ROOT_UID;
1706 stat->gid = GLOBAL_ROOT_GID;
1707 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1708 if (task) {
1709 if (!has_pid_permissions(pid, task, 2)) {
1710 rcu_read_unlock();
1711 /*
1712 * This doesn't prevent learning whether PID exists,
1713 * it only makes getattr() consistent with readdir().
1714 */
1715 return -ENOENT;
1716 }
1717 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1718 task_dumpable(task)) {
1719 cred = __task_cred(task);
1720 stat->uid = cred->euid;
1721 stat->gid = cred->egid;
1722 }
1723 }
1724 rcu_read_unlock();
1725 return 0;
1726}
1727
1728/* dentry stuff */
1729
1730/*
1731 * Exceptional case: normally we are not allowed to unhash a busy
1732 * directory. In this case, however, we can do it - no aliasing problems
1733 * due to the way we treat inodes.
1734 *
1735 * Rewrite the inode's ownerships here because the owning task may have
1736 * performed a setuid(), etc.
1737 *
1738 * Before the /proc/pid/status file was created the only way to read
1739 * the effective uid of a /process was to stat /proc/pid. Reading
1740 * /proc/pid/status is slow enough that procps and other packages
1741 * kept stating /proc/pid. To keep the rules in /proc simple I have
1742 * made this apply to all per process world readable and executable
1743 * directories.
1744 */
1745int pid_revalidate(struct dentry *dentry, unsigned int flags)
1746{
1747 struct inode *inode;
1748 struct task_struct *task;
1749 const struct cred *cred;
1750
1751 if (flags & LOOKUP_RCU)
1752 return -ECHILD;
1753
1754 inode = d_inode(dentry);
1755 task = get_proc_task(inode);
1756
1757 if (task) {
1758 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1759 task_dumpable(task)) {
1760 rcu_read_lock();
1761 cred = __task_cred(task);
1762 inode->i_uid = cred->euid;
1763 inode->i_gid = cred->egid;
1764 rcu_read_unlock();
1765 } else {
1766 inode->i_uid = GLOBAL_ROOT_UID;
1767 inode->i_gid = GLOBAL_ROOT_GID;
1768 }
1769 inode->i_mode &= ~(S_ISUID | S_ISGID);
1770 security_task_to_inode(task, inode);
1771 put_task_struct(task);
1772 return 1;
1773 }
1774 return 0;
1775}
1776
1777static inline bool proc_inode_is_dead(struct inode *inode)
1778{
1779 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1780}
1781
1782int pid_delete_dentry(const struct dentry *dentry)
1783{
1784 /* Is the task we represent dead?
1785 * If so, then don't put the dentry on the lru list,
1786 * kill it immediately.
1787 */
1788 return proc_inode_is_dead(d_inode(dentry));
1789}
1790
1791const struct dentry_operations pid_dentry_operations =
1792{
1793 .d_revalidate = pid_revalidate,
1794 .d_delete = pid_delete_dentry,
1795};
1796
1797/* Lookups */
1798
1799/*
1800 * Fill a directory entry.
1801 *
1802 * If possible create the dcache entry and derive our inode number and
1803 * file type from dcache entry.
1804 *
1805 * Since all of the proc inode numbers are dynamically generated, the inode
1806 * numbers do not exist until the inode is cache. This means creating the
1807 * the dcache entry in readdir is necessary to keep the inode numbers
1808 * reported by readdir in sync with the inode numbers reported
1809 * by stat.
1810 */
1811bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1812 const char *name, int len,
1813 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1814{
1815 struct dentry *child, *dir = file->f_path.dentry;
1816 struct qstr qname = QSTR_INIT(name, len);
1817 struct inode *inode;
1818 unsigned type;
1819 ino_t ino;
1820
1821 child = d_hash_and_lookup(dir, &qname);
1822 if (!child) {
1823 child = d_alloc(dir, &qname);
1824 if (!child)
1825 goto end_instantiate;
1826 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1827 dput(child);
1828 goto end_instantiate;
1829 }
1830 }
1831 inode = d_inode(child);
1832 ino = inode->i_ino;
1833 type = inode->i_mode >> 12;
1834 dput(child);
1835 return dir_emit(ctx, name, len, ino, type);
1836
1837end_instantiate:
1838 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1839}
1840
1841/*
1842 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1843 * which represent vma start and end addresses.
1844 */
1845static int dname_to_vma_addr(struct dentry *dentry,
1846 unsigned long *start, unsigned long *end)
1847{
1848 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1849 return -EINVAL;
1850
1851 return 0;
1852}
1853
1854static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1855{
1856 unsigned long vm_start, vm_end;
1857 bool exact_vma_exists = false;
1858 struct mm_struct *mm = NULL;
1859 struct task_struct *task;
1860 const struct cred *cred;
1861 struct inode *inode;
1862 int status = 0;
1863
1864 if (flags & LOOKUP_RCU)
1865 return -ECHILD;
1866
1867 inode = d_inode(dentry);
1868 task = get_proc_task(inode);
1869 if (!task)
1870 goto out_notask;
1871
1872 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1873 if (IS_ERR_OR_NULL(mm))
1874 goto out;
1875
1876 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1877 down_read(&mm->mmap_sem);
1878 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1879 up_read(&mm->mmap_sem);
1880 }
1881
1882 mmput(mm);
1883
1884 if (exact_vma_exists) {
1885 if (task_dumpable(task)) {
1886 rcu_read_lock();
1887 cred = __task_cred(task);
1888 inode->i_uid = cred->euid;
1889 inode->i_gid = cred->egid;
1890 rcu_read_unlock();
1891 } else {
1892 inode->i_uid = GLOBAL_ROOT_UID;
1893 inode->i_gid = GLOBAL_ROOT_GID;
1894 }
1895 security_task_to_inode(task, inode);
1896 status = 1;
1897 }
1898
1899out:
1900 put_task_struct(task);
1901
1902out_notask:
1903 return status;
1904}
1905
1906static const struct dentry_operations tid_map_files_dentry_operations = {
1907 .d_revalidate = map_files_d_revalidate,
1908 .d_delete = pid_delete_dentry,
1909};
1910
1911static int map_files_get_link(struct dentry *dentry, struct path *path)
1912{
1913 unsigned long vm_start, vm_end;
1914 struct vm_area_struct *vma;
1915 struct task_struct *task;
1916 struct mm_struct *mm;
1917 int rc;
1918
1919 rc = -ENOENT;
1920 task = get_proc_task(d_inode(dentry));
1921 if (!task)
1922 goto out;
1923
1924 mm = get_task_mm(task);
1925 put_task_struct(task);
1926 if (!mm)
1927 goto out;
1928
1929 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1930 if (rc)
1931 goto out_mmput;
1932
1933 rc = -ENOENT;
1934 down_read(&mm->mmap_sem);
1935 vma = find_exact_vma(mm, vm_start, vm_end);
1936 if (vma && vma->vm_file) {
1937 *path = vma->vm_file->f_path;
1938 path_get(path);
1939 rc = 0;
1940 }
1941 up_read(&mm->mmap_sem);
1942
1943out_mmput:
1944 mmput(mm);
1945out:
1946 return rc;
1947}
1948
1949struct map_files_info {
1950 fmode_t mode;
1951 unsigned long len;
1952 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1953};
1954
1955/*
1956 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1957 * symlinks may be used to bypass permissions on ancestor directories in the
1958 * path to the file in question.
1959 */
1960static const char *
1961proc_map_files_get_link(struct dentry *dentry,
1962 struct inode *inode,
1963 struct delayed_call *done)
1964{
1965 if (!capable(CAP_SYS_ADMIN))
1966 return ERR_PTR(-EPERM);
1967
1968 return proc_pid_get_link(dentry, inode, done);
1969}
1970
1971/*
1972 * Identical to proc_pid_link_inode_operations except for get_link()
1973 */
1974static const struct inode_operations proc_map_files_link_inode_operations = {
1975 .readlink = proc_pid_readlink,
1976 .get_link = proc_map_files_get_link,
1977 .setattr = proc_setattr,
1978};
1979
1980static int
1981proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1982 struct task_struct *task, const void *ptr)
1983{
1984 fmode_t mode = (fmode_t)(unsigned long)ptr;
1985 struct proc_inode *ei;
1986 struct inode *inode;
1987
1988 inode = proc_pid_make_inode(dir->i_sb, task);
1989 if (!inode)
1990 return -ENOENT;
1991
1992 ei = PROC_I(inode);
1993 ei->op.proc_get_link = map_files_get_link;
1994
1995 inode->i_op = &proc_map_files_link_inode_operations;
1996 inode->i_size = 64;
1997 inode->i_mode = S_IFLNK;
1998
1999 if (mode & FMODE_READ)
2000 inode->i_mode |= S_IRUSR;
2001 if (mode & FMODE_WRITE)
2002 inode->i_mode |= S_IWUSR;
2003
2004 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2005 d_add(dentry, inode);
2006
2007 return 0;
2008}
2009
2010static struct dentry *proc_map_files_lookup(struct inode *dir,
2011 struct dentry *dentry, unsigned int flags)
2012{
2013 unsigned long vm_start, vm_end;
2014 struct vm_area_struct *vma;
2015 struct task_struct *task;
2016 int result;
2017 struct mm_struct *mm;
2018
2019 result = -ENOENT;
2020 task = get_proc_task(dir);
2021 if (!task)
2022 goto out;
2023
2024 result = -EACCES;
2025 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2026 goto out_put_task;
2027
2028 result = -ENOENT;
2029 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2030 goto out_put_task;
2031
2032 mm = get_task_mm(task);
2033 if (!mm)
2034 goto out_put_task;
2035
2036 down_read(&mm->mmap_sem);
2037 vma = find_exact_vma(mm, vm_start, vm_end);
2038 if (!vma)
2039 goto out_no_vma;
2040
2041 if (vma->vm_file)
2042 result = proc_map_files_instantiate(dir, dentry, task,
2043 (void *)(unsigned long)vma->vm_file->f_mode);
2044
2045out_no_vma:
2046 up_read(&mm->mmap_sem);
2047 mmput(mm);
2048out_put_task:
2049 put_task_struct(task);
2050out:
2051 return ERR_PTR(result);
2052}
2053
2054static const struct inode_operations proc_map_files_inode_operations = {
2055 .lookup = proc_map_files_lookup,
2056 .permission = proc_fd_permission,
2057 .setattr = proc_setattr,
2058};
2059
2060static int
2061proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2062{
2063 struct vm_area_struct *vma;
2064 struct task_struct *task;
2065 struct mm_struct *mm;
2066 unsigned long nr_files, pos, i;
2067 struct flex_array *fa = NULL;
2068 struct map_files_info info;
2069 struct map_files_info *p;
2070 int ret;
2071
2072 ret = -ENOENT;
2073 task = get_proc_task(file_inode(file));
2074 if (!task)
2075 goto out;
2076
2077 ret = -EACCES;
2078 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2079 goto out_put_task;
2080
2081 ret = 0;
2082 if (!dir_emit_dots(file, ctx))
2083 goto out_put_task;
2084
2085 mm = get_task_mm(task);
2086 if (!mm)
2087 goto out_put_task;
2088 down_read(&mm->mmap_sem);
2089
2090 nr_files = 0;
2091
2092 /*
2093 * We need two passes here:
2094 *
2095 * 1) Collect vmas of mapped files with mmap_sem taken
2096 * 2) Release mmap_sem and instantiate entries
2097 *
2098 * otherwise we get lockdep complained, since filldir()
2099 * routine might require mmap_sem taken in might_fault().
2100 */
2101
2102 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2103 if (vma->vm_file && ++pos > ctx->pos)
2104 nr_files++;
2105 }
2106
2107 if (nr_files) {
2108 fa = flex_array_alloc(sizeof(info), nr_files,
2109 GFP_KERNEL);
2110 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2111 GFP_KERNEL)) {
2112 ret = -ENOMEM;
2113 if (fa)
2114 flex_array_free(fa);
2115 up_read(&mm->mmap_sem);
2116 mmput(mm);
2117 goto out_put_task;
2118 }
2119 for (i = 0, vma = mm->mmap, pos = 2; vma;
2120 vma = vma->vm_next) {
2121 if (!vma->vm_file)
2122 continue;
2123 if (++pos <= ctx->pos)
2124 continue;
2125
2126 info.mode = vma->vm_file->f_mode;
2127 info.len = snprintf(info.name,
2128 sizeof(info.name), "%lx-%lx",
2129 vma->vm_start, vma->vm_end);
2130 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2131 BUG();
2132 }
2133 }
2134 up_read(&mm->mmap_sem);
2135
2136 for (i = 0; i < nr_files; i++) {
2137 p = flex_array_get(fa, i);
2138 if (!proc_fill_cache(file, ctx,
2139 p->name, p->len,
2140 proc_map_files_instantiate,
2141 task,
2142 (void *)(unsigned long)p->mode))
2143 break;
2144 ctx->pos++;
2145 }
2146 if (fa)
2147 flex_array_free(fa);
2148 mmput(mm);
2149
2150out_put_task:
2151 put_task_struct(task);
2152out:
2153 return ret;
2154}
2155
2156static const struct file_operations proc_map_files_operations = {
2157 .read = generic_read_dir,
2158 .iterate = proc_map_files_readdir,
2159 .llseek = default_llseek,
2160};
2161
2162#ifdef CONFIG_CHECKPOINT_RESTORE
2163struct timers_private {
2164 struct pid *pid;
2165 struct task_struct *task;
2166 struct sighand_struct *sighand;
2167 struct pid_namespace *ns;
2168 unsigned long flags;
2169};
2170
2171static void *timers_start(struct seq_file *m, loff_t *pos)
2172{
2173 struct timers_private *tp = m->private;
2174
2175 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2176 if (!tp->task)
2177 return ERR_PTR(-ESRCH);
2178
2179 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2180 if (!tp->sighand)
2181 return ERR_PTR(-ESRCH);
2182
2183 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2184}
2185
2186static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2187{
2188 struct timers_private *tp = m->private;
2189 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2190}
2191
2192static void timers_stop(struct seq_file *m, void *v)
2193{
2194 struct timers_private *tp = m->private;
2195
2196 if (tp->sighand) {
2197 unlock_task_sighand(tp->task, &tp->flags);
2198 tp->sighand = NULL;
2199 }
2200
2201 if (tp->task) {
2202 put_task_struct(tp->task);
2203 tp->task = NULL;
2204 }
2205}
2206
2207static int show_timer(struct seq_file *m, void *v)
2208{
2209 struct k_itimer *timer;
2210 struct timers_private *tp = m->private;
2211 int notify;
2212 static const char * const nstr[] = {
2213 [SIGEV_SIGNAL] = "signal",
2214 [SIGEV_NONE] = "none",
2215 [SIGEV_THREAD] = "thread",
2216 };
2217
2218 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2219 notify = timer->it_sigev_notify;
2220
2221 seq_printf(m, "ID: %d\n", timer->it_id);
2222 seq_printf(m, "signal: %d/%p\n",
2223 timer->sigq->info.si_signo,
2224 timer->sigq->info.si_value.sival_ptr);
2225 seq_printf(m, "notify: %s/%s.%d\n",
2226 nstr[notify & ~SIGEV_THREAD_ID],
2227 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2228 pid_nr_ns(timer->it_pid, tp->ns));
2229 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2230
2231 return 0;
2232}
2233
2234static const struct seq_operations proc_timers_seq_ops = {
2235 .start = timers_start,
2236 .next = timers_next,
2237 .stop = timers_stop,
2238 .show = show_timer,
2239};
2240
2241static int proc_timers_open(struct inode *inode, struct file *file)
2242{
2243 struct timers_private *tp;
2244
2245 tp = __seq_open_private(file, &proc_timers_seq_ops,
2246 sizeof(struct timers_private));
2247 if (!tp)
2248 return -ENOMEM;
2249
2250 tp->pid = proc_pid(inode);
2251 tp->ns = inode->i_sb->s_fs_info;
2252 return 0;
2253}
2254
2255static const struct file_operations proc_timers_operations = {
2256 .open = proc_timers_open,
2257 .read = seq_read,
2258 .llseek = seq_lseek,
2259 .release = seq_release_private,
2260};
2261#endif
2262
2263static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2264 size_t count, loff_t *offset)
2265{
2266 struct inode *inode = file_inode(file);
2267 struct task_struct *p;
2268 u64 slack_ns;
2269 int err;
2270
2271 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2272 if (err < 0)
2273 return err;
2274
2275 p = get_proc_task(inode);
2276 if (!p)
2277 return -ESRCH;
2278
2279 if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2280 task_lock(p);
2281 if (slack_ns == 0)
2282 p->timer_slack_ns = p->default_timer_slack_ns;
2283 else
2284 p->timer_slack_ns = slack_ns;
2285 task_unlock(p);
2286 } else
2287 count = -EPERM;
2288
2289 put_task_struct(p);
2290
2291 return count;
2292}
2293
2294static int timerslack_ns_show(struct seq_file *m, void *v)
2295{
2296 struct inode *inode = m->private;
2297 struct task_struct *p;
2298 int err = 0;
2299
2300 p = get_proc_task(inode);
2301 if (!p)
2302 return -ESRCH;
2303
2304 if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2305 task_lock(p);
2306 seq_printf(m, "%llu\n", p->timer_slack_ns);
2307 task_unlock(p);
2308 } else
2309 err = -EPERM;
2310
2311 put_task_struct(p);
2312
2313 return err;
2314}
2315
2316static int timerslack_ns_open(struct inode *inode, struct file *filp)
2317{
2318 return single_open(filp, timerslack_ns_show, inode);
2319}
2320
2321static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2322 .open = timerslack_ns_open,
2323 .read = seq_read,
2324 .write = timerslack_ns_write,
2325 .llseek = seq_lseek,
2326 .release = single_release,
2327};
2328
2329static int proc_pident_instantiate(struct inode *dir,
2330 struct dentry *dentry, struct task_struct *task, const void *ptr)
2331{
2332 const struct pid_entry *p = ptr;
2333 struct inode *inode;
2334 struct proc_inode *ei;
2335
2336 inode = proc_pid_make_inode(dir->i_sb, task);
2337 if (!inode)
2338 goto out;
2339
2340 ei = PROC_I(inode);
2341 inode->i_mode = p->mode;
2342 if (S_ISDIR(inode->i_mode))
2343 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2344 if (p->iop)
2345 inode->i_op = p->iop;
2346 if (p->fop)
2347 inode->i_fop = p->fop;
2348 ei->op = p->op;
2349 d_set_d_op(dentry, &pid_dentry_operations);
2350 d_add(dentry, inode);
2351 /* Close the race of the process dying before we return the dentry */
2352 if (pid_revalidate(dentry, 0))
2353 return 0;
2354out:
2355 return -ENOENT;
2356}
2357
2358static struct dentry *proc_pident_lookup(struct inode *dir,
2359 struct dentry *dentry,
2360 const struct pid_entry *ents,
2361 unsigned int nents)
2362{
2363 int error;
2364 struct task_struct *task = get_proc_task(dir);
2365 const struct pid_entry *p, *last;
2366
2367 error = -ENOENT;
2368
2369 if (!task)
2370 goto out_no_task;
2371
2372 /*
2373 * Yes, it does not scale. And it should not. Don't add
2374 * new entries into /proc/<tgid>/ without very good reasons.
2375 */
2376 last = &ents[nents - 1];
2377 for (p = ents; p <= last; p++) {
2378 if (p->len != dentry->d_name.len)
2379 continue;
2380 if (!memcmp(dentry->d_name.name, p->name, p->len))
2381 break;
2382 }
2383 if (p > last)
2384 goto out;
2385
2386 error = proc_pident_instantiate(dir, dentry, task, p);
2387out:
2388 put_task_struct(task);
2389out_no_task:
2390 return ERR_PTR(error);
2391}
2392
2393static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2394 const struct pid_entry *ents, unsigned int nents)
2395{
2396 struct task_struct *task = get_proc_task(file_inode(file));
2397 const struct pid_entry *p;
2398
2399 if (!task)
2400 return -ENOENT;
2401
2402 if (!dir_emit_dots(file, ctx))
2403 goto out;
2404
2405 if (ctx->pos >= nents + 2)
2406 goto out;
2407
2408 for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2409 if (!proc_fill_cache(file, ctx, p->name, p->len,
2410 proc_pident_instantiate, task, p))
2411 break;
2412 ctx->pos++;
2413 }
2414out:
2415 put_task_struct(task);
2416 return 0;
2417}
2418
2419#ifdef CONFIG_SECURITY
2420static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2421 size_t count, loff_t *ppos)
2422{
2423 struct inode * inode = file_inode(file);
2424 char *p = NULL;
2425 ssize_t length;
2426 struct task_struct *task = get_proc_task(inode);
2427
2428 if (!task)
2429 return -ESRCH;
2430
2431 length = security_getprocattr(task,
2432 (char*)file->f_path.dentry->d_name.name,
2433 &p);
2434 put_task_struct(task);
2435 if (length > 0)
2436 length = simple_read_from_buffer(buf, count, ppos, p, length);
2437 kfree(p);
2438 return length;
2439}
2440
2441static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2442 size_t count, loff_t *ppos)
2443{
2444 struct inode * inode = file_inode(file);
2445 void *page;
2446 ssize_t length;
2447 struct task_struct *task = get_proc_task(inode);
2448
2449 length = -ESRCH;
2450 if (!task)
2451 goto out_no_task;
2452 if (count > PAGE_SIZE)
2453 count = PAGE_SIZE;
2454
2455 /* No partial writes. */
2456 length = -EINVAL;
2457 if (*ppos != 0)
2458 goto out;
2459
2460 page = memdup_user(buf, count);
2461 if (IS_ERR(page)) {
2462 length = PTR_ERR(page);
2463 goto out;
2464 }
2465
2466 /* Guard against adverse ptrace interaction */
2467 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2468 if (length < 0)
2469 goto out_free;
2470
2471 length = security_setprocattr(task,
2472 (char*)file->f_path.dentry->d_name.name,
2473 page, count);
2474 mutex_unlock(&task->signal->cred_guard_mutex);
2475out_free:
2476 kfree(page);
2477out:
2478 put_task_struct(task);
2479out_no_task:
2480 return length;
2481}
2482
2483static const struct file_operations proc_pid_attr_operations = {
2484 .read = proc_pid_attr_read,
2485 .write = proc_pid_attr_write,
2486 .llseek = generic_file_llseek,
2487};
2488
2489static const struct pid_entry attr_dir_stuff[] = {
2490 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2491 REG("prev", S_IRUGO, proc_pid_attr_operations),
2492 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2493 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2494 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2495 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2496};
2497
2498static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2499{
2500 return proc_pident_readdir(file, ctx,
2501 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2502}
2503
2504static const struct file_operations proc_attr_dir_operations = {
2505 .read = generic_read_dir,
2506 .iterate = proc_attr_dir_readdir,
2507 .llseek = default_llseek,
2508};
2509
2510static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2511 struct dentry *dentry, unsigned int flags)
2512{
2513 return proc_pident_lookup(dir, dentry,
2514 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2515}
2516
2517static const struct inode_operations proc_attr_dir_inode_operations = {
2518 .lookup = proc_attr_dir_lookup,
2519 .getattr = pid_getattr,
2520 .setattr = proc_setattr,
2521};
2522
2523#endif
2524
2525#ifdef CONFIG_ELF_CORE
2526static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2527 size_t count, loff_t *ppos)
2528{
2529 struct task_struct *task = get_proc_task(file_inode(file));
2530 struct mm_struct *mm;
2531 char buffer[PROC_NUMBUF];
2532 size_t len;
2533 int ret;
2534
2535 if (!task)
2536 return -ESRCH;
2537
2538 ret = 0;
2539 mm = get_task_mm(task);
2540 if (mm) {
2541 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2542 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2543 MMF_DUMP_FILTER_SHIFT));
2544 mmput(mm);
2545 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2546 }
2547
2548 put_task_struct(task);
2549
2550 return ret;
2551}
2552
2553static ssize_t proc_coredump_filter_write(struct file *file,
2554 const char __user *buf,
2555 size_t count,
2556 loff_t *ppos)
2557{
2558 struct task_struct *task;
2559 struct mm_struct *mm;
2560 unsigned int val;
2561 int ret;
2562 int i;
2563 unsigned long mask;
2564
2565 ret = kstrtouint_from_user(buf, count, 0, &val);
2566 if (ret < 0)
2567 return ret;
2568
2569 ret = -ESRCH;
2570 task = get_proc_task(file_inode(file));
2571 if (!task)
2572 goto out_no_task;
2573
2574 mm = get_task_mm(task);
2575 if (!mm)
2576 goto out_no_mm;
2577 ret = 0;
2578
2579 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2580 if (val & mask)
2581 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2582 else
2583 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2584 }
2585
2586 mmput(mm);
2587 out_no_mm:
2588 put_task_struct(task);
2589 out_no_task:
2590 if (ret < 0)
2591 return ret;
2592 return count;
2593}
2594
2595static const struct file_operations proc_coredump_filter_operations = {
2596 .read = proc_coredump_filter_read,
2597 .write = proc_coredump_filter_write,
2598 .llseek = generic_file_llseek,
2599};
2600#endif
2601
2602#ifdef CONFIG_TASK_IO_ACCOUNTING
2603static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2604{
2605 struct task_io_accounting acct = task->ioac;
2606 unsigned long flags;
2607 int result;
2608
2609 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2610 if (result)
2611 return result;
2612
2613 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2614 result = -EACCES;
2615 goto out_unlock;
2616 }
2617
2618 if (whole && lock_task_sighand(task, &flags)) {
2619 struct task_struct *t = task;
2620
2621 task_io_accounting_add(&acct, &task->signal->ioac);
2622 while_each_thread(task, t)
2623 task_io_accounting_add(&acct, &t->ioac);
2624
2625 unlock_task_sighand(task, &flags);
2626 }
2627 seq_printf(m,
2628 "rchar: %llu\n"
2629 "wchar: %llu\n"
2630 "syscr: %llu\n"
2631 "syscw: %llu\n"
2632 "read_bytes: %llu\n"
2633 "write_bytes: %llu\n"
2634 "cancelled_write_bytes: %llu\n",
2635 (unsigned long long)acct.rchar,
2636 (unsigned long long)acct.wchar,
2637 (unsigned long long)acct.syscr,
2638 (unsigned long long)acct.syscw,
2639 (unsigned long long)acct.read_bytes,
2640 (unsigned long long)acct.write_bytes,
2641 (unsigned long long)acct.cancelled_write_bytes);
2642 result = 0;
2643
2644out_unlock:
2645 mutex_unlock(&task->signal->cred_guard_mutex);
2646 return result;
2647}
2648
2649static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2650 struct pid *pid, struct task_struct *task)
2651{
2652 return do_io_accounting(task, m, 0);
2653}
2654
2655static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2656 struct pid *pid, struct task_struct *task)
2657{
2658 return do_io_accounting(task, m, 1);
2659}
2660#endif /* CONFIG_TASK_IO_ACCOUNTING */
2661
2662#ifdef CONFIG_USER_NS
2663static int proc_id_map_open(struct inode *inode, struct file *file,
2664 const struct seq_operations *seq_ops)
2665{
2666 struct user_namespace *ns = NULL;
2667 struct task_struct *task;
2668 struct seq_file *seq;
2669 int ret = -EINVAL;
2670
2671 task = get_proc_task(inode);
2672 if (task) {
2673 rcu_read_lock();
2674 ns = get_user_ns(task_cred_xxx(task, user_ns));
2675 rcu_read_unlock();
2676 put_task_struct(task);
2677 }
2678 if (!ns)
2679 goto err;
2680
2681 ret = seq_open(file, seq_ops);
2682 if (ret)
2683 goto err_put_ns;
2684
2685 seq = file->private_data;
2686 seq->private = ns;
2687
2688 return 0;
2689err_put_ns:
2690 put_user_ns(ns);
2691err:
2692 return ret;
2693}
2694
2695static int proc_id_map_release(struct inode *inode, struct file *file)
2696{
2697 struct seq_file *seq = file->private_data;
2698 struct user_namespace *ns = seq->private;
2699 put_user_ns(ns);
2700 return seq_release(inode, file);
2701}
2702
2703static int proc_uid_map_open(struct inode *inode, struct file *file)
2704{
2705 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2706}
2707
2708static int proc_gid_map_open(struct inode *inode, struct file *file)
2709{
2710 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2711}
2712
2713static int proc_projid_map_open(struct inode *inode, struct file *file)
2714{
2715 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2716}
2717
2718static const struct file_operations proc_uid_map_operations = {
2719 .open = proc_uid_map_open,
2720 .write = proc_uid_map_write,
2721 .read = seq_read,
2722 .llseek = seq_lseek,
2723 .release = proc_id_map_release,
2724};
2725
2726static const struct file_operations proc_gid_map_operations = {
2727 .open = proc_gid_map_open,
2728 .write = proc_gid_map_write,
2729 .read = seq_read,
2730 .llseek = seq_lseek,
2731 .release = proc_id_map_release,
2732};
2733
2734static const struct file_operations proc_projid_map_operations = {
2735 .open = proc_projid_map_open,
2736 .write = proc_projid_map_write,
2737 .read = seq_read,
2738 .llseek = seq_lseek,
2739 .release = proc_id_map_release,
2740};
2741
2742static int proc_setgroups_open(struct inode *inode, struct file *file)
2743{
2744 struct user_namespace *ns = NULL;
2745 struct task_struct *task;
2746 int ret;
2747
2748 ret = -ESRCH;
2749 task = get_proc_task(inode);
2750 if (task) {
2751 rcu_read_lock();
2752 ns = get_user_ns(task_cred_xxx(task, user_ns));
2753 rcu_read_unlock();
2754 put_task_struct(task);
2755 }
2756 if (!ns)
2757 goto err;
2758
2759 if (file->f_mode & FMODE_WRITE) {
2760 ret = -EACCES;
2761 if (!ns_capable(ns, CAP_SYS_ADMIN))
2762 goto err_put_ns;
2763 }
2764
2765 ret = single_open(file, &proc_setgroups_show, ns);
2766 if (ret)
2767 goto err_put_ns;
2768
2769 return 0;
2770err_put_ns:
2771 put_user_ns(ns);
2772err:
2773 return ret;
2774}
2775
2776static int proc_setgroups_release(struct inode *inode, struct file *file)
2777{
2778 struct seq_file *seq = file->private_data;
2779 struct user_namespace *ns = seq->private;
2780 int ret = single_release(inode, file);
2781 put_user_ns(ns);
2782 return ret;
2783}
2784
2785static const struct file_operations proc_setgroups_operations = {
2786 .open = proc_setgroups_open,
2787 .write = proc_setgroups_write,
2788 .read = seq_read,
2789 .llseek = seq_lseek,
2790 .release = proc_setgroups_release,
2791};
2792#endif /* CONFIG_USER_NS */
2793
2794static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2795 struct pid *pid, struct task_struct *task)
2796{
2797 int err = lock_trace(task);
2798 if (!err) {
2799 seq_printf(m, "%08x\n", task->personality);
2800 unlock_trace(task);
2801 }
2802 return err;
2803}
2804
2805/*
2806 * Thread groups
2807 */
2808static const struct file_operations proc_task_operations;
2809static const struct inode_operations proc_task_inode_operations;
2810
2811static const struct pid_entry tgid_base_stuff[] = {
2812 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2813 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2814 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2815 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2816 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2817#ifdef CONFIG_NET
2818 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2819#endif
2820 REG("environ", S_IRUSR, proc_environ_operations),
2821 ONE("auxv", S_IRUSR, proc_pid_auxv),
2822 ONE("status", S_IRUGO, proc_pid_status),
2823 ONE("personality", S_IRUSR, proc_pid_personality),
2824 ONE("limits", S_IRUGO, proc_pid_limits),
2825#ifdef CONFIG_SCHED_DEBUG
2826 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2827#endif
2828#ifdef CONFIG_SCHED_AUTOGROUP
2829 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2830#endif
2831 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2832#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2833 ONE("syscall", S_IRUSR, proc_pid_syscall),
2834#endif
2835 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2836 ONE("stat", S_IRUGO, proc_tgid_stat),
2837 ONE("statm", S_IRUGO, proc_pid_statm),
2838 REG("maps", S_IRUGO, proc_pid_maps_operations),
2839#ifdef CONFIG_NUMA
2840 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2841#endif
2842 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2843 LNK("cwd", proc_cwd_link),
2844 LNK("root", proc_root_link),
2845 LNK("exe", proc_exe_link),
2846 REG("mounts", S_IRUGO, proc_mounts_operations),
2847 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2848 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2849#ifdef CONFIG_PROC_PAGE_MONITOR
2850 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2851 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2852 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2853#endif
2854#ifdef CONFIG_SECURITY
2855 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2856#endif
2857#ifdef CONFIG_KALLSYMS
2858 ONE("wchan", S_IRUGO, proc_pid_wchan),
2859#endif
2860#ifdef CONFIG_STACKTRACE
2861 ONE("stack", S_IRUSR, proc_pid_stack),
2862#endif
2863#ifdef CONFIG_SCHED_INFO
2864 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2865#endif
2866#ifdef CONFIG_LATENCYTOP
2867 REG("latency", S_IRUGO, proc_lstats_operations),
2868#endif
2869#ifdef CONFIG_PROC_PID_CPUSET
2870 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2871#endif
2872#ifdef CONFIG_CGROUPS
2873 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2874#endif
2875 ONE("oom_score", S_IRUGO, proc_oom_score),
2876 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2877 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2878#ifdef CONFIG_AUDITSYSCALL
2879 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2880 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2881#endif
2882#ifdef CONFIG_FAULT_INJECTION
2883 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2884#endif
2885#ifdef CONFIG_ELF_CORE
2886 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2887#endif
2888#ifdef CONFIG_TASK_IO_ACCOUNTING
2889 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2890#endif
2891#ifdef CONFIG_HARDWALL
2892 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2893#endif
2894#ifdef CONFIG_USER_NS
2895 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2896 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2897 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2898 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2899#endif
2900#ifdef CONFIG_CHECKPOINT_RESTORE
2901 REG("timers", S_IRUGO, proc_timers_operations),
2902#endif
2903 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2904};
2905
2906static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2907{
2908 return proc_pident_readdir(file, ctx,
2909 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2910}
2911
2912static const struct file_operations proc_tgid_base_operations = {
2913 .read = generic_read_dir,
2914 .iterate = proc_tgid_base_readdir,
2915 .llseek = default_llseek,
2916};
2917
2918static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2919{
2920 return proc_pident_lookup(dir, dentry,
2921 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2922}
2923
2924static const struct inode_operations proc_tgid_base_inode_operations = {
2925 .lookup = proc_tgid_base_lookup,
2926 .getattr = pid_getattr,
2927 .setattr = proc_setattr,
2928 .permission = proc_pid_permission,
2929};
2930
2931static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2932{
2933 struct dentry *dentry, *leader, *dir;
2934 char buf[PROC_NUMBUF];
2935 struct qstr name;
2936
2937 name.name = buf;
2938 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2939 /* no ->d_hash() rejects on procfs */
2940 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2941 if (dentry) {
2942 d_invalidate(dentry);
2943 dput(dentry);
2944 }
2945
2946 if (pid == tgid)
2947 return;
2948
2949 name.name = buf;
2950 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2951 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2952 if (!leader)
2953 goto out;
2954
2955 name.name = "task";
2956 name.len = strlen(name.name);
2957 dir = d_hash_and_lookup(leader, &name);
2958 if (!dir)
2959 goto out_put_leader;
2960
2961 name.name = buf;
2962 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2963 dentry = d_hash_and_lookup(dir, &name);
2964 if (dentry) {
2965 d_invalidate(dentry);
2966 dput(dentry);
2967 }
2968
2969 dput(dir);
2970out_put_leader:
2971 dput(leader);
2972out:
2973 return;
2974}
2975
2976/**
2977 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
2978 * @task: task that should be flushed.
2979 *
2980 * When flushing dentries from proc, one needs to flush them from global
2981 * proc (proc_mnt) and from all the namespaces' procs this task was seen
2982 * in. This call is supposed to do all of this job.
2983 *
2984 * Looks in the dcache for
2985 * /proc/@pid
2986 * /proc/@tgid/task/@pid
2987 * if either directory is present flushes it and all of it'ts children
2988 * from the dcache.
2989 *
2990 * It is safe and reasonable to cache /proc entries for a task until
2991 * that task exits. After that they just clog up the dcache with
2992 * useless entries, possibly causing useful dcache entries to be
2993 * flushed instead. This routine is proved to flush those useless
2994 * dcache entries at process exit time.
2995 *
2996 * NOTE: This routine is just an optimization so it does not guarantee
2997 * that no dcache entries will exist at process exit time it
2998 * just makes it very unlikely that any will persist.
2999 */
3000
3001void proc_flush_task(struct task_struct *task)
3002{
3003 int i;
3004 struct pid *pid, *tgid;
3005 struct upid *upid;
3006
3007 pid = task_pid(task);
3008 tgid = task_tgid(task);
3009
3010 for (i = 0; i <= pid->level; i++) {
3011 upid = &pid->numbers[i];
3012 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3013 tgid->numbers[i].nr);
3014 }
3015}
3016
3017static int proc_pid_instantiate(struct inode *dir,
3018 struct dentry * dentry,
3019 struct task_struct *task, const void *ptr)
3020{
3021 struct inode *inode;
3022
3023 inode = proc_pid_make_inode(dir->i_sb, task);
3024 if (!inode)
3025 goto out;
3026
3027 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3028 inode->i_op = &proc_tgid_base_inode_operations;
3029 inode->i_fop = &proc_tgid_base_operations;
3030 inode->i_flags|=S_IMMUTABLE;
3031
3032 set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3033 ARRAY_SIZE(tgid_base_stuff)));
3034
3035 d_set_d_op(dentry, &pid_dentry_operations);
3036
3037 d_add(dentry, inode);
3038 /* Close the race of the process dying before we return the dentry */
3039 if (pid_revalidate(dentry, 0))
3040 return 0;
3041out:
3042 return -ENOENT;
3043}
3044
3045struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3046{
3047 int result = -ENOENT;
3048 struct task_struct *task;
3049 unsigned tgid;
3050 struct pid_namespace *ns;
3051
3052 tgid = name_to_int(&dentry->d_name);
3053 if (tgid == ~0U)
3054 goto out;
3055
3056 ns = dentry->d_sb->s_fs_info;
3057 rcu_read_lock();
3058 task = find_task_by_pid_ns(tgid, ns);
3059 if (task)
3060 get_task_struct(task);
3061 rcu_read_unlock();
3062 if (!task)
3063 goto out;
3064
3065 result = proc_pid_instantiate(dir, dentry, task, NULL);
3066 put_task_struct(task);
3067out:
3068 return ERR_PTR(result);
3069}
3070
3071/*
3072 * Find the first task with tgid >= tgid
3073 *
3074 */
3075struct tgid_iter {
3076 unsigned int tgid;
3077 struct task_struct *task;
3078};
3079static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3080{
3081 struct pid *pid;
3082
3083 if (iter.task)
3084 put_task_struct(iter.task);
3085 rcu_read_lock();
3086retry:
3087 iter.task = NULL;
3088 pid = find_ge_pid(iter.tgid, ns);
3089 if (pid) {
3090 iter.tgid = pid_nr_ns(pid, ns);
3091 iter.task = pid_task(pid, PIDTYPE_PID);
3092 /* What we to know is if the pid we have find is the
3093 * pid of a thread_group_leader. Testing for task
3094 * being a thread_group_leader is the obvious thing
3095 * todo but there is a window when it fails, due to
3096 * the pid transfer logic in de_thread.
3097 *
3098 * So we perform the straight forward test of seeing
3099 * if the pid we have found is the pid of a thread
3100 * group leader, and don't worry if the task we have
3101 * found doesn't happen to be a thread group leader.
3102 * As we don't care in the case of readdir.
3103 */
3104 if (!iter.task || !has_group_leader_pid(iter.task)) {
3105 iter.tgid += 1;
3106 goto retry;
3107 }
3108 get_task_struct(iter.task);
3109 }
3110 rcu_read_unlock();
3111 return iter;
3112}
3113
3114#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3115
3116/* for the /proc/ directory itself, after non-process stuff has been done */
3117int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3118{
3119 struct tgid_iter iter;
3120 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3121 loff_t pos = ctx->pos;
3122
3123 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3124 return 0;
3125
3126 if (pos == TGID_OFFSET - 2) {
3127 struct inode *inode = d_inode(ns->proc_self);
3128 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3129 return 0;
3130 ctx->pos = pos = pos + 1;
3131 }
3132 if (pos == TGID_OFFSET - 1) {
3133 struct inode *inode = d_inode(ns->proc_thread_self);
3134 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3135 return 0;
3136 ctx->pos = pos = pos + 1;
3137 }
3138 iter.tgid = pos - TGID_OFFSET;
3139 iter.task = NULL;
3140 for (iter = next_tgid(ns, iter);
3141 iter.task;
3142 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3143 char name[PROC_NUMBUF];
3144 int len;
3145 if (!has_pid_permissions(ns, iter.task, 2))
3146 continue;
3147
3148 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3149 ctx->pos = iter.tgid + TGID_OFFSET;
3150 if (!proc_fill_cache(file, ctx, name, len,
3151 proc_pid_instantiate, iter.task, NULL)) {
3152 put_task_struct(iter.task);
3153 return 0;
3154 }
3155 }
3156 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3157 return 0;
3158}
3159
3160/*
3161 * Tasks
3162 */
3163static const struct pid_entry tid_base_stuff[] = {
3164 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3165 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3166 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3167#ifdef CONFIG_NET
3168 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3169#endif
3170 REG("environ", S_IRUSR, proc_environ_operations),
3171 ONE("auxv", S_IRUSR, proc_pid_auxv),
3172 ONE("status", S_IRUGO, proc_pid_status),
3173 ONE("personality", S_IRUSR, proc_pid_personality),
3174 ONE("limits", S_IRUGO, proc_pid_limits),
3175#ifdef CONFIG_SCHED_DEBUG
3176 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3177#endif
3178 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3179#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3180 ONE("syscall", S_IRUSR, proc_pid_syscall),
3181#endif
3182 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3183 ONE("stat", S_IRUGO, proc_tid_stat),
3184 ONE("statm", S_IRUGO, proc_pid_statm),
3185 REG("maps", S_IRUGO, proc_tid_maps_operations),
3186#ifdef CONFIG_PROC_CHILDREN
3187 REG("children", S_IRUGO, proc_tid_children_operations),
3188#endif
3189#ifdef CONFIG_NUMA
3190 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3191#endif
3192 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3193 LNK("cwd", proc_cwd_link),
3194 LNK("root", proc_root_link),
3195 LNK("exe", proc_exe_link),
3196 REG("mounts", S_IRUGO, proc_mounts_operations),
3197 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3198#ifdef CONFIG_PROC_PAGE_MONITOR
3199 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3200 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3201 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3202#endif
3203#ifdef CONFIG_SECURITY
3204 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3205#endif
3206#ifdef CONFIG_KALLSYMS
3207 ONE("wchan", S_IRUGO, proc_pid_wchan),
3208#endif
3209#ifdef CONFIG_STACKTRACE
3210 ONE("stack", S_IRUSR, proc_pid_stack),
3211#endif
3212#ifdef CONFIG_SCHED_INFO
3213 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3214#endif
3215#ifdef CONFIG_LATENCYTOP
3216 REG("latency", S_IRUGO, proc_lstats_operations),
3217#endif
3218#ifdef CONFIG_PROC_PID_CPUSET
3219 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3220#endif
3221#ifdef CONFIG_CGROUPS
3222 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3223#endif
3224 ONE("oom_score", S_IRUGO, proc_oom_score),
3225 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3226 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3227#ifdef CONFIG_AUDITSYSCALL
3228 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3229 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3230#endif
3231#ifdef CONFIG_FAULT_INJECTION
3232 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3233#endif
3234#ifdef CONFIG_TASK_IO_ACCOUNTING
3235 ONE("io", S_IRUSR, proc_tid_io_accounting),
3236#endif
3237#ifdef CONFIG_HARDWALL
3238 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3239#endif
3240#ifdef CONFIG_USER_NS
3241 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3242 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3243 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3244 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3245#endif
3246};
3247
3248static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3249{
3250 return proc_pident_readdir(file, ctx,
3251 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3252}
3253
3254static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3255{
3256 return proc_pident_lookup(dir, dentry,
3257 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3258}
3259
3260static const struct file_operations proc_tid_base_operations = {
3261 .read = generic_read_dir,
3262 .iterate = proc_tid_base_readdir,
3263 .llseek = default_llseek,
3264};
3265
3266static const struct inode_operations proc_tid_base_inode_operations = {
3267 .lookup = proc_tid_base_lookup,
3268 .getattr = pid_getattr,
3269 .setattr = proc_setattr,
3270};
3271
3272static int proc_task_instantiate(struct inode *dir,
3273 struct dentry *dentry, struct task_struct *task, const void *ptr)
3274{
3275 struct inode *inode;
3276 inode = proc_pid_make_inode(dir->i_sb, task);
3277
3278 if (!inode)
3279 goto out;
3280 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3281 inode->i_op = &proc_tid_base_inode_operations;
3282 inode->i_fop = &proc_tid_base_operations;
3283 inode->i_flags|=S_IMMUTABLE;
3284
3285 set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3286 ARRAY_SIZE(tid_base_stuff)));
3287
3288 d_set_d_op(dentry, &pid_dentry_operations);
3289
3290 d_add(dentry, inode);
3291 /* Close the race of the process dying before we return the dentry */
3292 if (pid_revalidate(dentry, 0))
3293 return 0;
3294out:
3295 return -ENOENT;
3296}
3297
3298static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3299{
3300 int result = -ENOENT;
3301 struct task_struct *task;
3302 struct task_struct *leader = get_proc_task(dir);
3303 unsigned tid;
3304 struct pid_namespace *ns;
3305
3306 if (!leader)
3307 goto out_no_task;
3308
3309 tid = name_to_int(&dentry->d_name);
3310 if (tid == ~0U)
3311 goto out;
3312
3313 ns = dentry->d_sb->s_fs_info;
3314 rcu_read_lock();
3315 task = find_task_by_pid_ns(tid, ns);
3316 if (task)
3317 get_task_struct(task);
3318 rcu_read_unlock();
3319 if (!task)
3320 goto out;
3321 if (!same_thread_group(leader, task))
3322 goto out_drop_task;
3323
3324 result = proc_task_instantiate(dir, dentry, task, NULL);
3325out_drop_task:
3326 put_task_struct(task);
3327out:
3328 put_task_struct(leader);
3329out_no_task:
3330 return ERR_PTR(result);
3331}
3332
3333/*
3334 * Find the first tid of a thread group to return to user space.
3335 *
3336 * Usually this is just the thread group leader, but if the users
3337 * buffer was too small or there was a seek into the middle of the
3338 * directory we have more work todo.
3339 *
3340 * In the case of a short read we start with find_task_by_pid.
3341 *
3342 * In the case of a seek we start with the leader and walk nr
3343 * threads past it.
3344 */
3345static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3346 struct pid_namespace *ns)
3347{
3348 struct task_struct *pos, *task;
3349 unsigned long nr = f_pos;
3350
3351 if (nr != f_pos) /* 32bit overflow? */
3352 return NULL;
3353
3354 rcu_read_lock();
3355 task = pid_task(pid, PIDTYPE_PID);
3356 if (!task)
3357 goto fail;
3358
3359 /* Attempt to start with the tid of a thread */
3360 if (tid && nr) {
3361 pos = find_task_by_pid_ns(tid, ns);
3362 if (pos && same_thread_group(pos, task))
3363 goto found;
3364 }
3365
3366 /* If nr exceeds the number of threads there is nothing todo */
3367 if (nr >= get_nr_threads(task))
3368 goto fail;
3369
3370 /* If we haven't found our starting place yet start
3371 * with the leader and walk nr threads forward.
3372 */
3373 pos = task = task->group_leader;
3374 do {
3375 if (!nr--)
3376 goto found;
3377 } while_each_thread(task, pos);
3378fail:
3379 pos = NULL;
3380 goto out;
3381found:
3382 get_task_struct(pos);
3383out:
3384 rcu_read_unlock();
3385 return pos;
3386}
3387
3388/*
3389 * Find the next thread in the thread list.
3390 * Return NULL if there is an error or no next thread.
3391 *
3392 * The reference to the input task_struct is released.
3393 */
3394static struct task_struct *next_tid(struct task_struct *start)
3395{
3396 struct task_struct *pos = NULL;
3397 rcu_read_lock();
3398 if (pid_alive(start)) {
3399 pos = next_thread(start);
3400 if (thread_group_leader(pos))
3401 pos = NULL;
3402 else
3403 get_task_struct(pos);
3404 }
3405 rcu_read_unlock();
3406 put_task_struct(start);
3407 return pos;
3408}
3409
3410/* for the /proc/TGID/task/ directories */
3411static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3412{
3413 struct inode *inode = file_inode(file);
3414 struct task_struct *task;
3415 struct pid_namespace *ns;
3416 int tid;
3417
3418 if (proc_inode_is_dead(inode))
3419 return -ENOENT;
3420
3421 if (!dir_emit_dots(file, ctx))
3422 return 0;
3423
3424 /* f_version caches the tgid value that the last readdir call couldn't
3425 * return. lseek aka telldir automagically resets f_version to 0.
3426 */
3427 ns = inode->i_sb->s_fs_info;
3428 tid = (int)file->f_version;
3429 file->f_version = 0;
3430 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3431 task;
3432 task = next_tid(task), ctx->pos++) {
3433 char name[PROC_NUMBUF];
3434 int len;
3435 tid = task_pid_nr_ns(task, ns);
3436 len = snprintf(name, sizeof(name), "%d", tid);
3437 if (!proc_fill_cache(file, ctx, name, len,
3438 proc_task_instantiate, task, NULL)) {
3439 /* returning this tgid failed, save it as the first
3440 * pid for the next readir call */
3441 file->f_version = (u64)tid;
3442 put_task_struct(task);
3443 break;
3444 }
3445 }
3446
3447 return 0;
3448}
3449
3450static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3451{
3452 struct inode *inode = d_inode(dentry);
3453 struct task_struct *p = get_proc_task(inode);
3454 generic_fillattr(inode, stat);
3455
3456 if (p) {
3457 stat->nlink += get_nr_threads(p);
3458 put_task_struct(p);
3459 }
3460
3461 return 0;
3462}
3463
3464static const struct inode_operations proc_task_inode_operations = {
3465 .lookup = proc_task_lookup,
3466 .getattr = proc_task_getattr,
3467 .setattr = proc_setattr,
3468 .permission = proc_pid_permission,
3469};
3470
3471static const struct file_operations proc_task_operations = {
3472 .read = generic_read_dir,
3473 .iterate = proc_task_readdir,
3474 .llseek = default_llseek,
3475};
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 { "ptrace", PROC_MEM_FORCE_PTRACE },
134 { "never", 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(proc_mem_force_table,
148 buf, 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(task);
213 if (task->fs) {
214 get_fs_root(task->fs, root);
215 result = 0;
216 }
217 task_unlock(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(d_inode(dentry));
224 int result = -ENOENT;
225
226 if (task) {
227 task_lock(task);
228 if (task->fs) {
229 get_fs_pwd(task->fs, path);
230 result = 0;
231 }
232 task_unlock(task);
233 put_task_struct(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(d_inode(dentry));
241 int result = -ENOENT;
242
243 if (task) {
244 result = get_task_root(task, path);
245 put_task_struct(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, arg_start, page, PAGE_SIZE, 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(buf, page+pos, 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(&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(&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, arg_end-1, &c, 1, 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, pos, page, size, FOLL_ANON);
360 if (got <= 0)
361 break;
362 got -= copy_to_user(buf, page, 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(tsk);
385 if (!mm)
386 return 0;
387
388 ret = get_mm_cmdline(mm, buf, count, 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(file_inode(file));
402 if (!tsk)
403 return -ESRCH;
404 ret = get_task_cmdline(tsk, buf, count, pos);
405 put_task_struct(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. If that fails, simply return the address.
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(task);
431 if (wchan && !lookup_symbol_name(wchan, symname)) {
432 seq_puts(m, symname);
433 return 0;
434 }
435
436print0:
437 seq_putc(m, '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(&task->signal->exec_update_lock);
445 if (err)
446 return err;
447 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
448 up_read(&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(&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(m->file, &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, entries,
493 MAX_STACK_TRACE_DEPTH, 0);
494
495 for (i = 0; i < nr_entries; i++) {
496 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
497 }
498
499 unlock_trace(task);
500 }
501 kfree(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, "0 0 0\n");
516 else
517 seq_printf(m, "%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(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, "%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] = {"Max file size", "bytes"},
614 [RLIMIT_DATA] = {"Max data size", "bytes"},
615 [RLIMIT_STACK] = {"Max stack size", "bytes"},
616 [RLIMIT_CORE] = {"Max core file size", "bytes"},
617 [RLIMIT_RSS] = {"Max resident set", "bytes"},
618 [RLIMIT_NPROC] = {"Max processes", "processes"},
619 [RLIMIT_NOFILE] = {"Max open files", "files"},
620 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
621 [RLIMIT_AS] = {"Max address space", "bytes"},
622 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
623 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
624 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
625 [RLIMIT_NICE] = {"Max nice priority", NULL},
626 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
627 [RLIMIT_RTTIME] = {"Max realtime timeout", "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))
640 return 0;
641 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
642 unlock_task_sighand(task, &flags);
643
644 /*
645 * print the file header
646 */
647 seq_puts(m, "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, "%-25s %-20s ",
655 lnames[i].name, "unlimited");
656 else
657 seq_printf(m, "%-25s %-20lu ",
658 lnames[i].name, rlim[i].rlim_cur);
659
660 if (rlim[i].rlim_max == RLIM_INFINITY)
661 seq_printf(m, "%-20s ", "unlimited");
662 else
663 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
664
665 if (lnames[i].unit)
666 seq_printf(m, "%-10s\n", lnames[i].unit);
667 else
668 seq_putc(m, '\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(task, &info))
687 seq_puts(m, "running\n");
688 else if (info.data.nr < 0)
689 seq_printf(m, "%d 0x%llx 0x%llx\n",
690 info.data.nr, info.sp, info.data.instruction_pointer);
691 else
692 seq_printf(m,
693 "%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(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(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, HIDEPID_NO_ACCESS);
776 put_task_struct(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(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(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
831struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
832{
833 struct task_struct *task = get_proc_task(inode);
834 struct mm_struct *mm;
835
836 if (!task)
837 return ERR_PTR(-ESRCH);
838
839 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
840 put_task_struct(task);
841
842 if (IS_ERR(mm))
843 return mm == ERR_PTR(-ESRCH) ? NULL : mm;
844
845 /* ensure this mm_struct can't be freed */
846 mmgrab(mm);
847 /* but do not pin its memory */
848 mmput(mm);
849
850 return mm;
851}
852
853static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
854{
855 struct mm_struct *mm = proc_mem_open(inode, mode);
856
857 if (IS_ERR(mm))
858 return PTR_ERR(mm);
859
860 file->private_data = mm;
861 return 0;
862}
863
864static int mem_open(struct inode *inode, struct file *file)
865{
866 if (WARN_ON_ONCE(!(file->f_op->fop_flags & FOP_UNSIGNED_OFFSET)))
867 return -EINVAL;
868 return __mem_open(inode, file, PTRACE_MODE_ATTACH);
869}
870
871static bool proc_mem_foll_force(struct file *file, struct mm_struct *mm)
872{
873 struct task_struct *task;
874 bool ptrace_active = false;
875
876 switch (proc_mem_force_override) {
877 case PROC_MEM_FORCE_NEVER:
878 return false;
879 case PROC_MEM_FORCE_PTRACE:
880 task = get_proc_task(file_inode(file));
881 if (task) {
882 ptrace_active = READ_ONCE(task->ptrace) &&
883 READ_ONCE(task->mm) == mm &&
884 READ_ONCE(task->parent) == current;
885 put_task_struct(task);
886 }
887 return ptrace_active;
888 default:
889 return true;
890 }
891}
892
893static ssize_t mem_rw(struct file *file, char __user *buf,
894 size_t count, loff_t *ppos, int write)
895{
896 struct mm_struct *mm = file->private_data;
897 unsigned long addr = *ppos;
898 ssize_t copied;
899 char *page;
900 unsigned int flags;
901
902 if (!mm)
903 return 0;
904
905 page = (char *)__get_free_page(GFP_KERNEL);
906 if (!page)
907 return -ENOMEM;
908
909 copied = 0;
910 if (!mmget_not_zero(mm))
911 goto free;
912
913 flags = write ? FOLL_WRITE : 0;
914 if (proc_mem_foll_force(file, mm))
915 flags |= FOLL_FORCE;
916
917 while (count > 0) {
918 size_t this_len = min_t(size_t, count, PAGE_SIZE);
919
920 if (write && copy_from_user(page, buf, this_len)) {
921 copied = -EFAULT;
922 break;
923 }
924
925 this_len = access_remote_vm(mm, addr, page, this_len, flags);
926 if (!this_len) {
927 if (!copied)
928 copied = -EIO;
929 break;
930 }
931
932 if (!write && copy_to_user(buf, page, this_len)) {
933 copied = -EFAULT;
934 break;
935 }
936
937 buf += this_len;
938 addr += this_len;
939 copied += this_len;
940 count -= this_len;
941 }
942 *ppos = addr;
943
944 mmput(mm);
945free:
946 free_page((unsigned long) page);
947 return copied;
948}
949
950static ssize_t mem_read(struct file *file, char __user *buf,
951 size_t count, loff_t *ppos)
952{
953 return mem_rw(file, buf, count, ppos, 0);
954}
955
956static ssize_t mem_write(struct file *file, const char __user *buf,
957 size_t count, loff_t *ppos)
958{
959 return mem_rw(file, (char __user*)buf, count, ppos, 1);
960}
961
962loff_t mem_lseek(struct file *file, loff_t offset, int orig)
963{
964 switch (orig) {
965 case 0:
966 file->f_pos = offset;
967 break;
968 case 1:
969 file->f_pos += offset;
970 break;
971 default:
972 return -EINVAL;
973 }
974 force_successful_syscall_return();
975 return file->f_pos;
976}
977
978static int mem_release(struct inode *inode, struct file *file)
979{
980 struct mm_struct *mm = file->private_data;
981 if (mm)
982 mmdrop(mm);
983 return 0;
984}
985
986static const struct file_operations proc_mem_operations = {
987 .llseek = mem_lseek,
988 .read = mem_read,
989 .write = mem_write,
990 .open = mem_open,
991 .release = mem_release,
992 .fop_flags = FOP_UNSIGNED_OFFSET,
993};
994
995static int environ_open(struct inode *inode, struct file *file)
996{
997 return __mem_open(inode, file, PTRACE_MODE_READ);
998}
999
1000static ssize_t environ_read(struct file *file, char __user *buf,
1001 size_t count, loff_t *ppos)
1002{
1003 char *page;
1004 unsigned long src = *ppos;
1005 int ret = 0;
1006 struct mm_struct *mm = file->private_data;
1007 unsigned long env_start, env_end;
1008
1009 /* Ensure the process spawned far enough to have an environment. */
1010 if (!mm || !mm->env_end)
1011 return 0;
1012
1013 page = (char *)__get_free_page(GFP_KERNEL);
1014 if (!page)
1015 return -ENOMEM;
1016
1017 ret = 0;
1018 if (!mmget_not_zero(mm))
1019 goto free;
1020
1021 spin_lock(&mm->arg_lock);
1022 env_start = mm->env_start;
1023 env_end = mm->env_end;
1024 spin_unlock(&mm->arg_lock);
1025
1026 while (count > 0) {
1027 size_t this_len, max_len;
1028 int retval;
1029
1030 if (src >= (env_end - env_start))
1031 break;
1032
1033 this_len = env_end - (env_start + src);
1034
1035 max_len = min_t(size_t, PAGE_SIZE, count);
1036 this_len = min(max_len, this_len);
1037
1038 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
1039
1040 if (retval <= 0) {
1041 ret = retval;
1042 break;
1043 }
1044
1045 if (copy_to_user(buf, page, retval)) {
1046 ret = -EFAULT;
1047 break;
1048 }
1049
1050 ret += retval;
1051 src += retval;
1052 buf += retval;
1053 count -= retval;
1054 }
1055 *ppos = src;
1056 mmput(mm);
1057
1058free:
1059 free_page((unsigned long) page);
1060 return ret;
1061}
1062
1063static const struct file_operations proc_environ_operations = {
1064 .open = environ_open,
1065 .read = environ_read,
1066 .llseek = generic_file_llseek,
1067 .release = mem_release,
1068};
1069
1070static int auxv_open(struct inode *inode, struct file *file)
1071{
1072 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1073}
1074
1075static ssize_t auxv_read(struct file *file, char __user *buf,
1076 size_t count, loff_t *ppos)
1077{
1078 struct mm_struct *mm = file->private_data;
1079 unsigned int nwords = 0;
1080
1081 if (!mm)
1082 return 0;
1083 do {
1084 nwords += 2;
1085 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1086 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1087 nwords * sizeof(mm->saved_auxv[0]));
1088}
1089
1090static const struct file_operations proc_auxv_operations = {
1091 .open = auxv_open,
1092 .read = auxv_read,
1093 .llseek = generic_file_llseek,
1094 .release = mem_release,
1095};
1096
1097static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1098 loff_t *ppos)
1099{
1100 struct task_struct *task = get_proc_task(file_inode(file));
1101 char buffer[PROC_NUMBUF];
1102 int oom_adj = OOM_ADJUST_MIN;
1103 size_t len;
1104
1105 if (!task)
1106 return -ESRCH;
1107 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1108 oom_adj = OOM_ADJUST_MAX;
1109 else
1110 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1111 OOM_SCORE_ADJ_MAX;
1112 put_task_struct(task);
1113 if (oom_adj > OOM_ADJUST_MAX)
1114 oom_adj = OOM_ADJUST_MAX;
1115 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1116 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1117}
1118
1119static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1120{
1121 struct mm_struct *mm = NULL;
1122 struct task_struct *task;
1123 int err = 0;
1124
1125 task = get_proc_task(file_inode(file));
1126 if (!task)
1127 return -ESRCH;
1128
1129 mutex_lock(&oom_adj_mutex);
1130 if (legacy) {
1131 if (oom_adj < task->signal->oom_score_adj &&
1132 !capable(CAP_SYS_RESOURCE)) {
1133 err = -EACCES;
1134 goto err_unlock;
1135 }
1136 /*
1137 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1138 * /proc/pid/oom_score_adj instead.
1139 */
1140 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1141 current->comm, task_pid_nr(current), task_pid_nr(task),
1142 task_pid_nr(task));
1143 } else {
1144 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1145 !capable(CAP_SYS_RESOURCE)) {
1146 err = -EACCES;
1147 goto err_unlock;
1148 }
1149 }
1150
1151 /*
1152 * Make sure we will check other processes sharing the mm if this is
1153 * not vfrok which wants its own oom_score_adj.
1154 * pin the mm so it doesn't go away and get reused after task_unlock
1155 */
1156 if (!task->vfork_done) {
1157 struct task_struct *p = find_lock_task_mm(task);
1158
1159 if (p) {
1160 if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1161 mm = p->mm;
1162 mmgrab(mm);
1163 }
1164 task_unlock(p);
1165 }
1166 }
1167
1168 task->signal->oom_score_adj = oom_adj;
1169 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1170 task->signal->oom_score_adj_min = (short)oom_adj;
1171 trace_oom_score_adj_update(task);
1172
1173 if (mm) {
1174 struct task_struct *p;
1175
1176 rcu_read_lock();
1177 for_each_process(p) {
1178 if (same_thread_group(task, p))
1179 continue;
1180
1181 /* do not touch kernel threads or the global init */
1182 if (p->flags & PF_KTHREAD || is_global_init(p))
1183 continue;
1184
1185 task_lock(p);
1186 if (!p->vfork_done && process_shares_mm(p, mm)) {
1187 p->signal->oom_score_adj = oom_adj;
1188 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1189 p->signal->oom_score_adj_min = (short)oom_adj;
1190 }
1191 task_unlock(p);
1192 }
1193 rcu_read_unlock();
1194 mmdrop(mm);
1195 }
1196err_unlock:
1197 mutex_unlock(&oom_adj_mutex);
1198 put_task_struct(task);
1199 return err;
1200}
1201
1202/*
1203 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1204 * kernels. The effective policy is defined by oom_score_adj, which has a
1205 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1206 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1207 * Processes that become oom disabled via oom_adj will still be oom disabled
1208 * with this implementation.
1209 *
1210 * oom_adj cannot be removed since existing userspace binaries use it.
1211 */
1212static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1213 size_t count, loff_t *ppos)
1214{
1215 char buffer[PROC_NUMBUF] = {};
1216 int oom_adj;
1217 int err;
1218
1219 if (count > sizeof(buffer) - 1)
1220 count = sizeof(buffer) - 1;
1221 if (copy_from_user(buffer, buf, count)) {
1222 err = -EFAULT;
1223 goto out;
1224 }
1225
1226 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1227 if (err)
1228 goto out;
1229 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1230 oom_adj != OOM_DISABLE) {
1231 err = -EINVAL;
1232 goto out;
1233 }
1234
1235 /*
1236 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1237 * value is always attainable.
1238 */
1239 if (oom_adj == OOM_ADJUST_MAX)
1240 oom_adj = OOM_SCORE_ADJ_MAX;
1241 else
1242 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1243
1244 err = __set_oom_adj(file, oom_adj, true);
1245out:
1246 return err < 0 ? err : count;
1247}
1248
1249static const struct file_operations proc_oom_adj_operations = {
1250 .read = oom_adj_read,
1251 .write = oom_adj_write,
1252 .llseek = generic_file_llseek,
1253};
1254
1255static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1256 size_t count, loff_t *ppos)
1257{
1258 struct task_struct *task = get_proc_task(file_inode(file));
1259 char buffer[PROC_NUMBUF];
1260 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1261 size_t len;
1262
1263 if (!task)
1264 return -ESRCH;
1265 oom_score_adj = task->signal->oom_score_adj;
1266 put_task_struct(task);
1267 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1268 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1269}
1270
1271static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1272 size_t count, loff_t *ppos)
1273{
1274 char buffer[PROC_NUMBUF] = {};
1275 int oom_score_adj;
1276 int err;
1277
1278 if (count > sizeof(buffer) - 1)
1279 count = sizeof(buffer) - 1;
1280 if (copy_from_user(buffer, buf, count)) {
1281 err = -EFAULT;
1282 goto out;
1283 }
1284
1285 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1286 if (err)
1287 goto out;
1288 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1289 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1290 err = -EINVAL;
1291 goto out;
1292 }
1293
1294 err = __set_oom_adj(file, oom_score_adj, false);
1295out:
1296 return err < 0 ? err : count;
1297}
1298
1299static const struct file_operations proc_oom_score_adj_operations = {
1300 .read = oom_score_adj_read,
1301 .write = oom_score_adj_write,
1302 .llseek = default_llseek,
1303};
1304
1305#ifdef CONFIG_AUDIT
1306#define TMPBUFLEN 11
1307static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1308 size_t count, loff_t *ppos)
1309{
1310 struct inode * inode = file_inode(file);
1311 struct task_struct *task = get_proc_task(inode);
1312 ssize_t length;
1313 char tmpbuf[TMPBUFLEN];
1314
1315 if (!task)
1316 return -ESRCH;
1317 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1318 from_kuid(file->f_cred->user_ns,
1319 audit_get_loginuid(task)));
1320 put_task_struct(task);
1321 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1322}
1323
1324static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1325 size_t count, loff_t *ppos)
1326{
1327 struct inode * inode = file_inode(file);
1328 uid_t loginuid;
1329 kuid_t kloginuid;
1330 int rv;
1331
1332 /* Don't let kthreads write their own loginuid */
1333 if (current->flags & PF_KTHREAD)
1334 return -EPERM;
1335
1336 rcu_read_lock();
1337 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1338 rcu_read_unlock();
1339 return -EPERM;
1340 }
1341 rcu_read_unlock();
1342
1343 if (*ppos != 0) {
1344 /* No partial writes. */
1345 return -EINVAL;
1346 }
1347
1348 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1349 if (rv < 0)
1350 return rv;
1351
1352 /* is userspace tring to explicitly UNSET the loginuid? */
1353 if (loginuid == AUDIT_UID_UNSET) {
1354 kloginuid = INVALID_UID;
1355 } else {
1356 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1357 if (!uid_valid(kloginuid))
1358 return -EINVAL;
1359 }
1360
1361 rv = audit_set_loginuid(kloginuid);
1362 if (rv < 0)
1363 return rv;
1364 return count;
1365}
1366
1367static const struct file_operations proc_loginuid_operations = {
1368 .read = proc_loginuid_read,
1369 .write = proc_loginuid_write,
1370 .llseek = generic_file_llseek,
1371};
1372
1373static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1374 size_t count, loff_t *ppos)
1375{
1376 struct inode * inode = file_inode(file);
1377 struct task_struct *task = get_proc_task(inode);
1378 ssize_t length;
1379 char tmpbuf[TMPBUFLEN];
1380
1381 if (!task)
1382 return -ESRCH;
1383 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1384 audit_get_sessionid(task));
1385 put_task_struct(task);
1386 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1387}
1388
1389static const struct file_operations proc_sessionid_operations = {
1390 .read = proc_sessionid_read,
1391 .llseek = generic_file_llseek,
1392};
1393#endif
1394
1395#ifdef CONFIG_FAULT_INJECTION
1396static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1397 size_t count, loff_t *ppos)
1398{
1399 struct task_struct *task = get_proc_task(file_inode(file));
1400 char buffer[PROC_NUMBUF];
1401 size_t len;
1402 int make_it_fail;
1403
1404 if (!task)
1405 return -ESRCH;
1406 make_it_fail = task->make_it_fail;
1407 put_task_struct(task);
1408
1409 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1410
1411 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1412}
1413
1414static ssize_t proc_fault_inject_write(struct file * file,
1415 const char __user * buf, size_t count, loff_t *ppos)
1416{
1417 struct task_struct *task;
1418 char buffer[PROC_NUMBUF] = {};
1419 int make_it_fail;
1420 int rv;
1421
1422 if (!capable(CAP_SYS_RESOURCE))
1423 return -EPERM;
1424
1425 if (count > sizeof(buffer) - 1)
1426 count = sizeof(buffer) - 1;
1427 if (copy_from_user(buffer, buf, count))
1428 return -EFAULT;
1429 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1430 if (rv < 0)
1431 return rv;
1432 if (make_it_fail < 0 || make_it_fail > 1)
1433 return -EINVAL;
1434
1435 task = get_proc_task(file_inode(file));
1436 if (!task)
1437 return -ESRCH;
1438 task->make_it_fail = make_it_fail;
1439 put_task_struct(task);
1440
1441 return count;
1442}
1443
1444static const struct file_operations proc_fault_inject_operations = {
1445 .read = proc_fault_inject_read,
1446 .write = proc_fault_inject_write,
1447 .llseek = generic_file_llseek,
1448};
1449
1450static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1451 size_t count, loff_t *ppos)
1452{
1453 struct task_struct *task;
1454 int err;
1455 unsigned int n;
1456
1457 err = kstrtouint_from_user(buf, count, 0, &n);
1458 if (err)
1459 return err;
1460
1461 task = get_proc_task(file_inode(file));
1462 if (!task)
1463 return -ESRCH;
1464 task->fail_nth = n;
1465 put_task_struct(task);
1466
1467 return count;
1468}
1469
1470static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1471 size_t count, loff_t *ppos)
1472{
1473 struct task_struct *task;
1474 char numbuf[PROC_NUMBUF];
1475 ssize_t len;
1476
1477 task = get_proc_task(file_inode(file));
1478 if (!task)
1479 return -ESRCH;
1480 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1481 put_task_struct(task);
1482 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1483}
1484
1485static const struct file_operations proc_fail_nth_operations = {
1486 .read = proc_fail_nth_read,
1487 .write = proc_fail_nth_write,
1488};
1489#endif
1490
1491
1492#ifdef CONFIG_SCHED_DEBUG
1493/*
1494 * Print out various scheduling related per-task fields:
1495 */
1496static int sched_show(struct seq_file *m, void *v)
1497{
1498 struct inode *inode = m->private;
1499 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1500 struct task_struct *p;
1501
1502 p = get_proc_task(inode);
1503 if (!p)
1504 return -ESRCH;
1505 proc_sched_show_task(p, ns, m);
1506
1507 put_task_struct(p);
1508
1509 return 0;
1510}
1511
1512static ssize_t
1513sched_write(struct file *file, const char __user *buf,
1514 size_t count, loff_t *offset)
1515{
1516 struct inode *inode = file_inode(file);
1517 struct task_struct *p;
1518
1519 p = get_proc_task(inode);
1520 if (!p)
1521 return -ESRCH;
1522 proc_sched_set_task(p);
1523
1524 put_task_struct(p);
1525
1526 return count;
1527}
1528
1529static int sched_open(struct inode *inode, struct file *filp)
1530{
1531 return single_open(filp, sched_show, inode);
1532}
1533
1534static const struct file_operations proc_pid_sched_operations = {
1535 .open = sched_open,
1536 .read = seq_read,
1537 .write = sched_write,
1538 .llseek = seq_lseek,
1539 .release = single_release,
1540};
1541
1542#endif
1543
1544#ifdef CONFIG_SCHED_AUTOGROUP
1545/*
1546 * Print out autogroup related information:
1547 */
1548static int sched_autogroup_show(struct seq_file *m, void *v)
1549{
1550 struct inode *inode = m->private;
1551 struct task_struct *p;
1552
1553 p = get_proc_task(inode);
1554 if (!p)
1555 return -ESRCH;
1556 proc_sched_autogroup_show_task(p, m);
1557
1558 put_task_struct(p);
1559
1560 return 0;
1561}
1562
1563static ssize_t
1564sched_autogroup_write(struct file *file, const char __user *buf,
1565 size_t count, loff_t *offset)
1566{
1567 struct inode *inode = file_inode(file);
1568 struct task_struct *p;
1569 char buffer[PROC_NUMBUF] = {};
1570 int nice;
1571 int err;
1572
1573 if (count > sizeof(buffer) - 1)
1574 count = sizeof(buffer) - 1;
1575 if (copy_from_user(buffer, buf, count))
1576 return -EFAULT;
1577
1578 err = kstrtoint(strstrip(buffer), 0, &nice);
1579 if (err < 0)
1580 return err;
1581
1582 p = get_proc_task(inode);
1583 if (!p)
1584 return -ESRCH;
1585
1586 err = proc_sched_autogroup_set_nice(p, nice);
1587 if (err)
1588 count = err;
1589
1590 put_task_struct(p);
1591
1592 return count;
1593}
1594
1595static int sched_autogroup_open(struct inode *inode, struct file *filp)
1596{
1597 int ret;
1598
1599 ret = single_open(filp, sched_autogroup_show, NULL);
1600 if (!ret) {
1601 struct seq_file *m = filp->private_data;
1602
1603 m->private = inode;
1604 }
1605 return ret;
1606}
1607
1608static const struct file_operations proc_pid_sched_autogroup_operations = {
1609 .open = sched_autogroup_open,
1610 .read = seq_read,
1611 .write = sched_autogroup_write,
1612 .llseek = seq_lseek,
1613 .release = single_release,
1614};
1615
1616#endif /* CONFIG_SCHED_AUTOGROUP */
1617
1618#ifdef CONFIG_TIME_NS
1619static int timens_offsets_show(struct seq_file *m, void *v)
1620{
1621 struct task_struct *p;
1622
1623 p = get_proc_task(file_inode(m->file));
1624 if (!p)
1625 return -ESRCH;
1626 proc_timens_show_offsets(p, m);
1627
1628 put_task_struct(p);
1629
1630 return 0;
1631}
1632
1633static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1634 size_t count, loff_t *ppos)
1635{
1636 struct inode *inode = file_inode(file);
1637 struct proc_timens_offset offsets[2];
1638 char *kbuf = NULL, *pos, *next_line;
1639 struct task_struct *p;
1640 int ret, noffsets;
1641
1642 /* Only allow < page size writes at the beginning of the file */
1643 if ((*ppos != 0) || (count >= PAGE_SIZE))
1644 return -EINVAL;
1645
1646 /* Slurp in the user data */
1647 kbuf = memdup_user_nul(buf, count);
1648 if (IS_ERR(kbuf))
1649 return PTR_ERR(kbuf);
1650
1651 /* Parse the user data */
1652 ret = -EINVAL;
1653 noffsets = 0;
1654 for (pos = kbuf; pos; pos = next_line) {
1655 struct proc_timens_offset *off = &offsets[noffsets];
1656 char clock[10];
1657 int err;
1658
1659 /* Find the end of line and ensure we don't look past it */
1660 next_line = strchr(pos, '\n');
1661 if (next_line) {
1662 *next_line = '\0';
1663 next_line++;
1664 if (*next_line == '\0')
1665 next_line = NULL;
1666 }
1667
1668 err = sscanf(pos, "%9s %lld %lu", clock,
1669 &off->val.tv_sec, &off->val.tv_nsec);
1670 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1671 goto out;
1672
1673 clock[sizeof(clock) - 1] = 0;
1674 if (strcmp(clock, "monotonic") == 0 ||
1675 strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1676 off->clockid = CLOCK_MONOTONIC;
1677 else if (strcmp(clock, "boottime") == 0 ||
1678 strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1679 off->clockid = CLOCK_BOOTTIME;
1680 else
1681 goto out;
1682
1683 noffsets++;
1684 if (noffsets == ARRAY_SIZE(offsets)) {
1685 if (next_line)
1686 count = next_line - kbuf;
1687 break;
1688 }
1689 }
1690
1691 ret = -ESRCH;
1692 p = get_proc_task(inode);
1693 if (!p)
1694 goto out;
1695 ret = proc_timens_set_offset(file, p, offsets, noffsets);
1696 put_task_struct(p);
1697 if (ret)
1698 goto out;
1699
1700 ret = count;
1701out:
1702 kfree(kbuf);
1703 return ret;
1704}
1705
1706static int timens_offsets_open(struct inode *inode, struct file *filp)
1707{
1708 return single_open(filp, timens_offsets_show, inode);
1709}
1710
1711static const struct file_operations proc_timens_offsets_operations = {
1712 .open = timens_offsets_open,
1713 .read = seq_read,
1714 .write = timens_offsets_write,
1715 .llseek = seq_lseek,
1716 .release = single_release,
1717};
1718#endif /* CONFIG_TIME_NS */
1719
1720static ssize_t comm_write(struct file *file, const char __user *buf,
1721 size_t count, loff_t *offset)
1722{
1723 struct inode *inode = file_inode(file);
1724 struct task_struct *p;
1725 char buffer[TASK_COMM_LEN] = {};
1726 const size_t maxlen = sizeof(buffer) - 1;
1727
1728 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1729 return -EFAULT;
1730
1731 p = get_proc_task(inode);
1732 if (!p)
1733 return -ESRCH;
1734
1735 if (same_thread_group(current, p)) {
1736 set_task_comm(p, buffer);
1737 proc_comm_connector(p);
1738 }
1739 else
1740 count = -EINVAL;
1741
1742 put_task_struct(p);
1743
1744 return count;
1745}
1746
1747static int comm_show(struct seq_file *m, void *v)
1748{
1749 struct inode *inode = m->private;
1750 struct task_struct *p;
1751
1752 p = get_proc_task(inode);
1753 if (!p)
1754 return -ESRCH;
1755
1756 proc_task_name(m, p, false);
1757 seq_putc(m, '\n');
1758
1759 put_task_struct(p);
1760
1761 return 0;
1762}
1763
1764static int comm_open(struct inode *inode, struct file *filp)
1765{
1766 return single_open(filp, comm_show, inode);
1767}
1768
1769static const struct file_operations proc_pid_set_comm_operations = {
1770 .open = comm_open,
1771 .read = seq_read,
1772 .write = comm_write,
1773 .llseek = seq_lseek,
1774 .release = single_release,
1775};
1776
1777static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1778{
1779 struct task_struct *task;
1780 struct file *exe_file;
1781
1782 task = get_proc_task(d_inode(dentry));
1783 if (!task)
1784 return -ENOENT;
1785 exe_file = get_task_exe_file(task);
1786 put_task_struct(task);
1787 if (exe_file) {
1788 *exe_path = exe_file->f_path;
1789 path_get(&exe_file->f_path);
1790 fput(exe_file);
1791 return 0;
1792 } else
1793 return -ENOENT;
1794}
1795
1796static const char *proc_pid_get_link(struct dentry *dentry,
1797 struct inode *inode,
1798 struct delayed_call *done)
1799{
1800 struct path path;
1801 int error = -EACCES;
1802
1803 if (!dentry)
1804 return ERR_PTR(-ECHILD);
1805
1806 /* Are we allowed to snoop on the tasks file descriptors? */
1807 if (!proc_fd_access_allowed(inode))
1808 goto out;
1809
1810 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1811 if (error)
1812 goto out;
1813
1814 error = nd_jump_link(&path);
1815out:
1816 return ERR_PTR(error);
1817}
1818
1819static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1820{
1821 char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1822 char *pathname;
1823 int len;
1824
1825 if (!tmp)
1826 return -ENOMEM;
1827
1828 pathname = d_path(path, tmp, PATH_MAX);
1829 len = PTR_ERR(pathname);
1830 if (IS_ERR(pathname))
1831 goto out;
1832 len = tmp + PATH_MAX - 1 - pathname;
1833
1834 if (len > buflen)
1835 len = buflen;
1836 if (copy_to_user(buffer, pathname, len))
1837 len = -EFAULT;
1838 out:
1839 kfree(tmp);
1840 return len;
1841}
1842
1843static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1844{
1845 int error = -EACCES;
1846 struct inode *inode = d_inode(dentry);
1847 struct path path;
1848
1849 /* Are we allowed to snoop on the tasks file descriptors? */
1850 if (!proc_fd_access_allowed(inode))
1851 goto out;
1852
1853 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1854 if (error)
1855 goto out;
1856
1857 error = do_proc_readlink(&path, buffer, buflen);
1858 path_put(&path);
1859out:
1860 return error;
1861}
1862
1863const struct inode_operations proc_pid_link_inode_operations = {
1864 .readlink = proc_pid_readlink,
1865 .get_link = proc_pid_get_link,
1866 .setattr = proc_setattr,
1867};
1868
1869
1870/* building an inode */
1871
1872void task_dump_owner(struct task_struct *task, umode_t mode,
1873 kuid_t *ruid, kgid_t *rgid)
1874{
1875 /* Depending on the state of dumpable compute who should own a
1876 * proc file for a task.
1877 */
1878 const struct cred *cred;
1879 kuid_t uid;
1880 kgid_t gid;
1881
1882 if (unlikely(task->flags & PF_KTHREAD)) {
1883 *ruid = GLOBAL_ROOT_UID;
1884 *rgid = GLOBAL_ROOT_GID;
1885 return;
1886 }
1887
1888 /* Default to the tasks effective ownership */
1889 rcu_read_lock();
1890 cred = __task_cred(task);
1891 uid = cred->euid;
1892 gid = cred->egid;
1893 rcu_read_unlock();
1894
1895 /*
1896 * Before the /proc/pid/status file was created the only way to read
1897 * the effective uid of a /process was to stat /proc/pid. Reading
1898 * /proc/pid/status is slow enough that procps and other packages
1899 * kept stating /proc/pid. To keep the rules in /proc simple I have
1900 * made this apply to all per process world readable and executable
1901 * directories.
1902 */
1903 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1904 struct mm_struct *mm;
1905 task_lock(task);
1906 mm = task->mm;
1907 /* Make non-dumpable tasks owned by some root */
1908 if (mm) {
1909 if (get_dumpable(mm) != SUID_DUMP_USER) {
1910 struct user_namespace *user_ns = mm->user_ns;
1911
1912 uid = make_kuid(user_ns, 0);
1913 if (!uid_valid(uid))
1914 uid = GLOBAL_ROOT_UID;
1915
1916 gid = make_kgid(user_ns, 0);
1917 if (!gid_valid(gid))
1918 gid = GLOBAL_ROOT_GID;
1919 }
1920 } else {
1921 uid = GLOBAL_ROOT_UID;
1922 gid = GLOBAL_ROOT_GID;
1923 }
1924 task_unlock(task);
1925 }
1926 *ruid = uid;
1927 *rgid = gid;
1928}
1929
1930void proc_pid_evict_inode(struct proc_inode *ei)
1931{
1932 struct pid *pid = ei->pid;
1933
1934 if (S_ISDIR(ei->vfs_inode.i_mode)) {
1935 spin_lock(&pid->lock);
1936 hlist_del_init_rcu(&ei->sibling_inodes);
1937 spin_unlock(&pid->lock);
1938 }
1939}
1940
1941struct inode *proc_pid_make_inode(struct super_block *sb,
1942 struct task_struct *task, umode_t mode)
1943{
1944 struct inode * inode;
1945 struct proc_inode *ei;
1946 struct pid *pid;
1947
1948 /* We need a new inode */
1949
1950 inode = new_inode(sb);
1951 if (!inode)
1952 goto out;
1953
1954 /* Common stuff */
1955 ei = PROC_I(inode);
1956 inode->i_mode = mode;
1957 inode->i_ino = get_next_ino();
1958 simple_inode_init_ts(inode);
1959 inode->i_op = &proc_def_inode_operations;
1960
1961 /*
1962 * grab the reference to task.
1963 */
1964 pid = get_task_pid(task, PIDTYPE_PID);
1965 if (!pid)
1966 goto out_unlock;
1967
1968 /* Let the pid remember us for quick removal */
1969 ei->pid = pid;
1970
1971 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1972 security_task_to_inode(task, inode);
1973
1974out:
1975 return inode;
1976
1977out_unlock:
1978 iput(inode);
1979 return NULL;
1980}
1981
1982/*
1983 * Generating an inode and adding it into @pid->inodes, so that task will
1984 * invalidate inode's dentry before being released.
1985 *
1986 * This helper is used for creating dir-type entries under '/proc' and
1987 * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1988 * can be released by invalidating '/proc/<tgid>' dentry.
1989 * In theory, dentries under '/proc/<tgid>/task' can also be released by
1990 * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1991 * thread exiting situation: Any one of threads should invalidate its
1992 * '/proc/<tgid>/task/<pid>' dentry before released.
1993 */
1994static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1995 struct task_struct *task, umode_t mode)
1996{
1997 struct inode *inode;
1998 struct proc_inode *ei;
1999 struct pid *pid;
2000
2001 inode = proc_pid_make_inode(sb, task, mode);
2002 if (!inode)
2003 return NULL;
2004
2005 /* Let proc_flush_pid find this directory inode */
2006 ei = PROC_I(inode);
2007 pid = ei->pid;
2008 spin_lock(&pid->lock);
2009 hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
2010 spin_unlock(&pid->lock);
2011
2012 return inode;
2013}
2014
2015int pid_getattr(struct mnt_idmap *idmap, const struct path *path,
2016 struct kstat *stat, u32 request_mask, unsigned int query_flags)
2017{
2018 struct inode *inode = d_inode(path->dentry);
2019 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
2020 struct task_struct *task;
2021
2022 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
2023
2024 stat->uid = GLOBAL_ROOT_UID;
2025 stat->gid = GLOBAL_ROOT_GID;
2026 rcu_read_lock();
2027 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2028 if (task) {
2029 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
2030 rcu_read_unlock();
2031 /*
2032 * This doesn't prevent learning whether PID exists,
2033 * it only makes getattr() consistent with readdir().
2034 */
2035 return -ENOENT;
2036 }
2037 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
2038 }
2039 rcu_read_unlock();
2040 return 0;
2041}
2042
2043/* dentry stuff */
2044
2045/*
2046 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
2047 */
2048void pid_update_inode(struct task_struct *task, struct inode *inode)
2049{
2050 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
2051
2052 inode->i_mode &= ~(S_ISUID | S_ISGID);
2053 security_task_to_inode(task, inode);
2054}
2055
2056/*
2057 * Rewrite the inode's ownerships here because the owning task may have
2058 * performed a setuid(), etc.
2059 *
2060 */
2061static int pid_revalidate(struct dentry *dentry, unsigned int flags)
2062{
2063 struct inode *inode;
2064 struct task_struct *task;
2065 int ret = 0;
2066
2067 rcu_read_lock();
2068 inode = d_inode_rcu(dentry);
2069 if (!inode)
2070 goto out;
2071 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2072
2073 if (task) {
2074 pid_update_inode(task, inode);
2075 ret = 1;
2076 }
2077out:
2078 rcu_read_unlock();
2079 return ret;
2080}
2081
2082static inline bool proc_inode_is_dead(struct inode *inode)
2083{
2084 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2085}
2086
2087int pid_delete_dentry(const struct dentry *dentry)
2088{
2089 /* Is the task we represent dead?
2090 * If so, then don't put the dentry on the lru list,
2091 * kill it immediately.
2092 */
2093 return proc_inode_is_dead(d_inode(dentry));
2094}
2095
2096const struct dentry_operations pid_dentry_operations =
2097{
2098 .d_revalidate = pid_revalidate,
2099 .d_delete = pid_delete_dentry,
2100};
2101
2102/* Lookups */
2103
2104/*
2105 * Fill a directory entry.
2106 *
2107 * If possible create the dcache entry and derive our inode number and
2108 * file type from dcache entry.
2109 *
2110 * Since all of the proc inode numbers are dynamically generated, the inode
2111 * numbers do not exist until the inode is cache. This means creating
2112 * the dcache entry in readdir is necessary to keep the inode numbers
2113 * reported by readdir in sync with the inode numbers reported
2114 * by stat.
2115 */
2116bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2117 const char *name, unsigned int len,
2118 instantiate_t instantiate, struct task_struct *task, const void *ptr)
2119{
2120 struct dentry *child, *dir = file->f_path.dentry;
2121 struct qstr qname = QSTR_INIT(name, len);
2122 struct inode *inode;
2123 unsigned type = DT_UNKNOWN;
2124 ino_t ino = 1;
2125
2126 child = d_hash_and_lookup(dir, &qname);
2127 if (!child) {
2128 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2129 child = d_alloc_parallel(dir, &qname, &wq);
2130 if (IS_ERR(child))
2131 goto end_instantiate;
2132 if (d_in_lookup(child)) {
2133 struct dentry *res;
2134 res = instantiate(child, task, ptr);
2135 d_lookup_done(child);
2136 if (unlikely(res)) {
2137 dput(child);
2138 child = res;
2139 if (IS_ERR(child))
2140 goto end_instantiate;
2141 }
2142 }
2143 }
2144 inode = d_inode(child);
2145 ino = inode->i_ino;
2146 type = inode->i_mode >> 12;
2147 dput(child);
2148end_instantiate:
2149 return dir_emit(ctx, name, len, ino, type);
2150}
2151
2152/*
2153 * dname_to_vma_addr - maps a dentry name into two unsigned longs
2154 * which represent vma start and end addresses.
2155 */
2156static int dname_to_vma_addr(struct dentry *dentry,
2157 unsigned long *start, unsigned long *end)
2158{
2159 const char *str = dentry->d_name.name;
2160 unsigned long long sval, eval;
2161 unsigned int len;
2162
2163 if (str[0] == '0' && str[1] != '-')
2164 return -EINVAL;
2165 len = _parse_integer(str, 16, &sval);
2166 if (len & KSTRTOX_OVERFLOW)
2167 return -EINVAL;
2168 if (sval != (unsigned long)sval)
2169 return -EINVAL;
2170 str += len;
2171
2172 if (*str != '-')
2173 return -EINVAL;
2174 str++;
2175
2176 if (str[0] == '0' && str[1])
2177 return -EINVAL;
2178 len = _parse_integer(str, 16, &eval);
2179 if (len & KSTRTOX_OVERFLOW)
2180 return -EINVAL;
2181 if (eval != (unsigned long)eval)
2182 return -EINVAL;
2183 str += len;
2184
2185 if (*str != '\0')
2186 return -EINVAL;
2187
2188 *start = sval;
2189 *end = eval;
2190
2191 return 0;
2192}
2193
2194static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2195{
2196 unsigned long vm_start, vm_end;
2197 bool exact_vma_exists = false;
2198 struct mm_struct *mm = NULL;
2199 struct task_struct *task;
2200 struct inode *inode;
2201 int status = 0;
2202
2203 if (flags & LOOKUP_RCU)
2204 return -ECHILD;
2205
2206 inode = d_inode(dentry);
2207 task = get_proc_task(inode);
2208 if (!task)
2209 goto out_notask;
2210
2211 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2212 if (IS_ERR(mm))
2213 goto out;
2214
2215 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2216 status = mmap_read_lock_killable(mm);
2217 if (!status) {
2218 exact_vma_exists = !!find_exact_vma(mm, vm_start,
2219 vm_end);
2220 mmap_read_unlock(mm);
2221 }
2222 }
2223
2224 mmput(mm);
2225
2226 if (exact_vma_exists) {
2227 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2228
2229 security_task_to_inode(task, inode);
2230 status = 1;
2231 }
2232
2233out:
2234 put_task_struct(task);
2235
2236out_notask:
2237 return status;
2238}
2239
2240static const struct dentry_operations tid_map_files_dentry_operations = {
2241 .d_revalidate = map_files_d_revalidate,
2242 .d_delete = pid_delete_dentry,
2243};
2244
2245static int map_files_get_link(struct dentry *dentry, struct path *path)
2246{
2247 unsigned long vm_start, vm_end;
2248 struct vm_area_struct *vma;
2249 struct task_struct *task;
2250 struct mm_struct *mm;
2251 int rc;
2252
2253 rc = -ENOENT;
2254 task = get_proc_task(d_inode(dentry));
2255 if (!task)
2256 goto out;
2257
2258 mm = get_task_mm(task);
2259 put_task_struct(task);
2260 if (!mm)
2261 goto out;
2262
2263 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2264 if (rc)
2265 goto out_mmput;
2266
2267 rc = mmap_read_lock_killable(mm);
2268 if (rc)
2269 goto out_mmput;
2270
2271 rc = -ENOENT;
2272 vma = find_exact_vma(mm, vm_start, vm_end);
2273 if (vma && vma->vm_file) {
2274 *path = *file_user_path(vma->vm_file);
2275 path_get(path);
2276 rc = 0;
2277 }
2278 mmap_read_unlock(mm);
2279
2280out_mmput:
2281 mmput(mm);
2282out:
2283 return rc;
2284}
2285
2286struct map_files_info {
2287 unsigned long start;
2288 unsigned long end;
2289 fmode_t mode;
2290};
2291
2292/*
2293 * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2294 * to concerns about how the symlinks may be used to bypass permissions on
2295 * ancestor directories in the path to the file in question.
2296 */
2297static const char *
2298proc_map_files_get_link(struct dentry *dentry,
2299 struct inode *inode,
2300 struct delayed_call *done)
2301{
2302 if (!checkpoint_restore_ns_capable(&init_user_ns))
2303 return ERR_PTR(-EPERM);
2304
2305 return proc_pid_get_link(dentry, inode, done);
2306}
2307
2308/*
2309 * Identical to proc_pid_link_inode_operations except for get_link()
2310 */
2311static const struct inode_operations proc_map_files_link_inode_operations = {
2312 .readlink = proc_pid_readlink,
2313 .get_link = proc_map_files_get_link,
2314 .setattr = proc_setattr,
2315};
2316
2317static struct dentry *
2318proc_map_files_instantiate(struct dentry *dentry,
2319 struct task_struct *task, const void *ptr)
2320{
2321 fmode_t mode = (fmode_t)(unsigned long)ptr;
2322 struct proc_inode *ei;
2323 struct inode *inode;
2324
2325 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2326 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2327 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2328 if (!inode)
2329 return ERR_PTR(-ENOENT);
2330
2331 ei = PROC_I(inode);
2332 ei->op.proc_get_link = map_files_get_link;
2333
2334 inode->i_op = &proc_map_files_link_inode_operations;
2335 inode->i_size = 64;
2336
2337 return proc_splice_unmountable(inode, dentry,
2338 &tid_map_files_dentry_operations);
2339}
2340
2341static struct dentry *proc_map_files_lookup(struct inode *dir,
2342 struct dentry *dentry, unsigned int flags)
2343{
2344 unsigned long vm_start, vm_end;
2345 struct vm_area_struct *vma;
2346 struct task_struct *task;
2347 struct dentry *result;
2348 struct mm_struct *mm;
2349
2350 result = ERR_PTR(-ENOENT);
2351 task = get_proc_task(dir);
2352 if (!task)
2353 goto out;
2354
2355 result = ERR_PTR(-EACCES);
2356 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2357 goto out_put_task;
2358
2359 result = ERR_PTR(-ENOENT);
2360 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2361 goto out_put_task;
2362
2363 mm = get_task_mm(task);
2364 if (!mm)
2365 goto out_put_task;
2366
2367 result = ERR_PTR(-EINTR);
2368 if (mmap_read_lock_killable(mm))
2369 goto out_put_mm;
2370
2371 result = ERR_PTR(-ENOENT);
2372 vma = find_exact_vma(mm, vm_start, vm_end);
2373 if (!vma)
2374 goto out_no_vma;
2375
2376 if (vma->vm_file)
2377 result = proc_map_files_instantiate(dentry, task,
2378 (void *)(unsigned long)vma->vm_file->f_mode);
2379
2380out_no_vma:
2381 mmap_read_unlock(mm);
2382out_put_mm:
2383 mmput(mm);
2384out_put_task:
2385 put_task_struct(task);
2386out:
2387 return result;
2388}
2389
2390static const struct inode_operations proc_map_files_inode_operations = {
2391 .lookup = proc_map_files_lookup,
2392 .permission = proc_fd_permission,
2393 .setattr = proc_setattr,
2394};
2395
2396static int
2397proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2398{
2399 struct vm_area_struct *vma;
2400 struct task_struct *task;
2401 struct mm_struct *mm;
2402 unsigned long nr_files, pos, i;
2403 GENRADIX(struct map_files_info) fa;
2404 struct map_files_info *p;
2405 int ret;
2406 struct vma_iterator vmi;
2407
2408 genradix_init(&fa);
2409
2410 ret = -ENOENT;
2411 task = get_proc_task(file_inode(file));
2412 if (!task)
2413 goto out;
2414
2415 ret = -EACCES;
2416 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2417 goto out_put_task;
2418
2419 ret = 0;
2420 if (!dir_emit_dots(file, ctx))
2421 goto out_put_task;
2422
2423 mm = get_task_mm(task);
2424 if (!mm)
2425 goto out_put_task;
2426
2427 ret = mmap_read_lock_killable(mm);
2428 if (ret) {
2429 mmput(mm);
2430 goto out_put_task;
2431 }
2432
2433 nr_files = 0;
2434
2435 /*
2436 * We need two passes here:
2437 *
2438 * 1) Collect vmas of mapped files with mmap_lock taken
2439 * 2) Release mmap_lock and instantiate entries
2440 *
2441 * otherwise we get lockdep complained, since filldir()
2442 * routine might require mmap_lock taken in might_fault().
2443 */
2444
2445 pos = 2;
2446 vma_iter_init(&vmi, mm, 0);
2447 for_each_vma(vmi, vma) {
2448 if (!vma->vm_file)
2449 continue;
2450 if (++pos <= ctx->pos)
2451 continue;
2452
2453 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2454 if (!p) {
2455 ret = -ENOMEM;
2456 mmap_read_unlock(mm);
2457 mmput(mm);
2458 goto out_put_task;
2459 }
2460
2461 p->start = vma->vm_start;
2462 p->end = vma->vm_end;
2463 p->mode = vma->vm_file->f_mode;
2464 }
2465 mmap_read_unlock(mm);
2466 mmput(mm);
2467
2468 for (i = 0; i < nr_files; i++) {
2469 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2470 unsigned int len;
2471
2472 p = genradix_ptr(&fa, i);
2473 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2474 if (!proc_fill_cache(file, ctx,
2475 buf, len,
2476 proc_map_files_instantiate,
2477 task,
2478 (void *)(unsigned long)p->mode))
2479 break;
2480 ctx->pos++;
2481 }
2482
2483out_put_task:
2484 put_task_struct(task);
2485out:
2486 genradix_free(&fa);
2487 return ret;
2488}
2489
2490static const struct file_operations proc_map_files_operations = {
2491 .read = generic_read_dir,
2492 .iterate_shared = proc_map_files_readdir,
2493 .llseek = generic_file_llseek,
2494};
2495
2496#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2497struct timers_private {
2498 struct pid *pid;
2499 struct task_struct *task;
2500 struct sighand_struct *sighand;
2501 struct pid_namespace *ns;
2502 unsigned long flags;
2503};
2504
2505static void *timers_start(struct seq_file *m, loff_t *pos)
2506{
2507 struct timers_private *tp = m->private;
2508
2509 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2510 if (!tp->task)
2511 return ERR_PTR(-ESRCH);
2512
2513 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2514 if (!tp->sighand)
2515 return ERR_PTR(-ESRCH);
2516
2517 return seq_hlist_start(&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 return seq_hlist_next(v, &tp->task->signal->posix_timers, pos);
2524}
2525
2526static void timers_stop(struct seq_file *m, void *v)
2527{
2528 struct timers_private *tp = m->private;
2529
2530 if (tp->sighand) {
2531 unlock_task_sighand(tp->task, &tp->flags);
2532 tp->sighand = NULL;
2533 }
2534
2535 if (tp->task) {
2536 put_task_struct(tp->task);
2537 tp->task = NULL;
2538 }
2539}
2540
2541static int show_timer(struct seq_file *m, void *v)
2542{
2543 struct k_itimer *timer;
2544 struct timers_private *tp = m->private;
2545 int notify;
2546 static const char * const nstr[] = {
2547 [SIGEV_SIGNAL] = "signal",
2548 [SIGEV_NONE] = "none",
2549 [SIGEV_THREAD] = "thread",
2550 };
2551
2552 timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
2553 notify = timer->it_sigev_notify;
2554
2555 seq_printf(m, "ID: %d\n", timer->it_id);
2556 seq_printf(m, "signal: %d/%px\n",
2557 timer->sigq.info.si_signo,
2558 timer->sigq.info.si_value.sival_ptr);
2559 seq_printf(m, "notify: %s/%s.%d\n",
2560 nstr[notify & ~SIGEV_THREAD_ID],
2561 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2562 pid_nr_ns(timer->it_pid, tp->ns));
2563 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2564
2565 return 0;
2566}
2567
2568static const struct seq_operations proc_timers_seq_ops = {
2569 .start = timers_start,
2570 .next = timers_next,
2571 .stop = timers_stop,
2572 .show = show_timer,
2573};
2574
2575static int proc_timers_open(struct inode *inode, struct file *file)
2576{
2577 struct timers_private *tp;
2578
2579 tp = __seq_open_private(file, &proc_timers_seq_ops,
2580 sizeof(struct timers_private));
2581 if (!tp)
2582 return -ENOMEM;
2583
2584 tp->pid = proc_pid(inode);
2585 tp->ns = proc_pid_ns(inode->i_sb);
2586 return 0;
2587}
2588
2589static const struct file_operations proc_timers_operations = {
2590 .open = proc_timers_open,
2591 .read = seq_read,
2592 .llseek = seq_lseek,
2593 .release = seq_release_private,
2594};
2595#endif
2596
2597static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2598 size_t count, loff_t *offset)
2599{
2600 struct inode *inode = file_inode(file);
2601 struct task_struct *p;
2602 u64 slack_ns;
2603 int err;
2604
2605 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2606 if (err < 0)
2607 return err;
2608
2609 p = get_proc_task(inode);
2610 if (!p)
2611 return -ESRCH;
2612
2613 if (p != current) {
2614 rcu_read_lock();
2615 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2616 rcu_read_unlock();
2617 count = -EPERM;
2618 goto out;
2619 }
2620 rcu_read_unlock();
2621
2622 err = security_task_setscheduler(p);
2623 if (err) {
2624 count = err;
2625 goto out;
2626 }
2627 }
2628
2629 task_lock(p);
2630 if (rt_or_dl_task_policy(p))
2631 slack_ns = 0;
2632 else if (slack_ns == 0)
2633 slack_ns = p->default_timer_slack_ns;
2634 p->timer_slack_ns = slack_ns;
2635 task_unlock(p);
2636
2637out:
2638 put_task_struct(p);
2639
2640 return count;
2641}
2642
2643static int timerslack_ns_show(struct seq_file *m, void *v)
2644{
2645 struct inode *inode = m->private;
2646 struct task_struct *p;
2647 int err = 0;
2648
2649 p = get_proc_task(inode);
2650 if (!p)
2651 return -ESRCH;
2652
2653 if (p != current) {
2654 rcu_read_lock();
2655 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2656 rcu_read_unlock();
2657 err = -EPERM;
2658 goto out;
2659 }
2660 rcu_read_unlock();
2661
2662 err = security_task_getscheduler(p);
2663 if (err)
2664 goto out;
2665 }
2666
2667 task_lock(p);
2668 seq_printf(m, "%llu\n", p->timer_slack_ns);
2669 task_unlock(p);
2670
2671out:
2672 put_task_struct(p);
2673
2674 return err;
2675}
2676
2677static int timerslack_ns_open(struct inode *inode, struct file *filp)
2678{
2679 return single_open(filp, timerslack_ns_show, inode);
2680}
2681
2682static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2683 .open = timerslack_ns_open,
2684 .read = seq_read,
2685 .write = timerslack_ns_write,
2686 .llseek = seq_lseek,
2687 .release = single_release,
2688};
2689
2690static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2691 struct task_struct *task, const void *ptr)
2692{
2693 const struct pid_entry *p = ptr;
2694 struct inode *inode;
2695 struct proc_inode *ei;
2696
2697 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2698 if (!inode)
2699 return ERR_PTR(-ENOENT);
2700
2701 ei = PROC_I(inode);
2702 if (S_ISDIR(inode->i_mode))
2703 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2704 if (p->iop)
2705 inode->i_op = p->iop;
2706 if (p->fop)
2707 inode->i_fop = p->fop;
2708 ei->op = p->op;
2709 pid_update_inode(task, inode);
2710 d_set_d_op(dentry, &pid_dentry_operations);
2711 return d_splice_alias(inode, dentry);
2712}
2713
2714static struct dentry *proc_pident_lookup(struct inode *dir,
2715 struct dentry *dentry,
2716 const struct pid_entry *p,
2717 const struct pid_entry *end)
2718{
2719 struct task_struct *task = get_proc_task(dir);
2720 struct dentry *res = ERR_PTR(-ENOENT);
2721
2722 if (!task)
2723 goto out_no_task;
2724
2725 /*
2726 * Yes, it does not scale. And it should not. Don't add
2727 * new entries into /proc/<tgid>/ without very good reasons.
2728 */
2729 for (; p < end; p++) {
2730 if (p->len != dentry->d_name.len)
2731 continue;
2732 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2733 res = proc_pident_instantiate(dentry, task, p);
2734 break;
2735 }
2736 }
2737 put_task_struct(task);
2738out_no_task:
2739 return res;
2740}
2741
2742static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2743 const struct pid_entry *ents, unsigned int nents)
2744{
2745 struct task_struct *task = get_proc_task(file_inode(file));
2746 const struct pid_entry *p;
2747
2748 if (!task)
2749 return -ENOENT;
2750
2751 if (!dir_emit_dots(file, ctx))
2752 goto out;
2753
2754 if (ctx->pos >= nents + 2)
2755 goto out;
2756
2757 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2758 if (!proc_fill_cache(file, ctx, p->name, p->len,
2759 proc_pident_instantiate, task, p))
2760 break;
2761 ctx->pos++;
2762 }
2763out:
2764 put_task_struct(task);
2765 return 0;
2766}
2767
2768#ifdef CONFIG_SECURITY
2769static int proc_pid_attr_open(struct inode *inode, struct file *file)
2770{
2771 file->private_data = NULL;
2772 __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2773 return 0;
2774}
2775
2776static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2777 size_t count, loff_t *ppos)
2778{
2779 struct inode * inode = file_inode(file);
2780 char *p = NULL;
2781 ssize_t length;
2782 struct task_struct *task = get_proc_task(inode);
2783
2784 if (!task)
2785 return -ESRCH;
2786
2787 length = security_getprocattr(task, PROC_I(inode)->op.lsmid,
2788 file->f_path.dentry->d_name.name,
2789 &p);
2790 put_task_struct(task);
2791 if (length > 0)
2792 length = simple_read_from_buffer(buf, count, ppos, p, length);
2793 kfree(p);
2794 return length;
2795}
2796
2797static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2798 size_t count, loff_t *ppos)
2799{
2800 struct inode * inode = file_inode(file);
2801 struct task_struct *task;
2802 void *page;
2803 int rv;
2804
2805 /* A task may only write when it was the opener. */
2806 if (file->private_data != current->mm)
2807 return -EPERM;
2808
2809 rcu_read_lock();
2810 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2811 if (!task) {
2812 rcu_read_unlock();
2813 return -ESRCH;
2814 }
2815 /* A task may only write its own attributes. */
2816 if (current != task) {
2817 rcu_read_unlock();
2818 return -EACCES;
2819 }
2820 /* Prevent changes to overridden credentials. */
2821 if (current_cred() != current_real_cred()) {
2822 rcu_read_unlock();
2823 return -EBUSY;
2824 }
2825 rcu_read_unlock();
2826
2827 if (count > PAGE_SIZE)
2828 count = PAGE_SIZE;
2829
2830 /* No partial writes. */
2831 if (*ppos != 0)
2832 return -EINVAL;
2833
2834 page = memdup_user(buf, count);
2835 if (IS_ERR(page)) {
2836 rv = PTR_ERR(page);
2837 goto out;
2838 }
2839
2840 /* Guard against adverse ptrace interaction */
2841 rv = mutex_lock_interruptible(¤t->signal->cred_guard_mutex);
2842 if (rv < 0)
2843 goto out_free;
2844
2845 rv = security_setprocattr(PROC_I(inode)->op.lsmid,
2846 file->f_path.dentry->d_name.name, page,
2847 count);
2848 mutex_unlock(¤t->signal->cred_guard_mutex);
2849out_free:
2850 kfree(page);
2851out:
2852 return rv;
2853}
2854
2855static const struct file_operations proc_pid_attr_operations = {
2856 .open = proc_pid_attr_open,
2857 .read = proc_pid_attr_read,
2858 .write = proc_pid_attr_write,
2859 .llseek = generic_file_llseek,
2860 .release = mem_release,
2861};
2862
2863#define LSM_DIR_OPS(LSM) \
2864static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2865 struct dir_context *ctx) \
2866{ \
2867 return proc_pident_readdir(filp, ctx, \
2868 LSM##_attr_dir_stuff, \
2869 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2870} \
2871\
2872static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2873 .read = generic_read_dir, \
2874 .iterate_shared = proc_##LSM##_attr_dir_iterate, \
2875 .llseek = default_llseek, \
2876}; \
2877\
2878static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2879 struct dentry *dentry, unsigned int flags) \
2880{ \
2881 return proc_pident_lookup(dir, dentry, \
2882 LSM##_attr_dir_stuff, \
2883 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2884} \
2885\
2886static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2887 .lookup = proc_##LSM##_attr_dir_lookup, \
2888 .getattr = pid_getattr, \
2889 .setattr = proc_setattr, \
2890}
2891
2892#ifdef CONFIG_SECURITY_SMACK
2893static const struct pid_entry smack_attr_dir_stuff[] = {
2894 ATTR(LSM_ID_SMACK, "current", 0666),
2895};
2896LSM_DIR_OPS(smack);
2897#endif
2898
2899#ifdef CONFIG_SECURITY_APPARMOR
2900static const struct pid_entry apparmor_attr_dir_stuff[] = {
2901 ATTR(LSM_ID_APPARMOR, "current", 0666),
2902 ATTR(LSM_ID_APPARMOR, "prev", 0444),
2903 ATTR(LSM_ID_APPARMOR, "exec", 0666),
2904};
2905LSM_DIR_OPS(apparmor);
2906#endif
2907
2908static const struct pid_entry attr_dir_stuff[] = {
2909 ATTR(LSM_ID_UNDEF, "current", 0666),
2910 ATTR(LSM_ID_UNDEF, "prev", 0444),
2911 ATTR(LSM_ID_UNDEF, "exec", 0666),
2912 ATTR(LSM_ID_UNDEF, "fscreate", 0666),
2913 ATTR(LSM_ID_UNDEF, "keycreate", 0666),
2914 ATTR(LSM_ID_UNDEF, "sockcreate", 0666),
2915#ifdef CONFIG_SECURITY_SMACK
2916 DIR("smack", 0555,
2917 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2918#endif
2919#ifdef CONFIG_SECURITY_APPARMOR
2920 DIR("apparmor", 0555,
2921 proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2922#endif
2923};
2924
2925static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2926{
2927 return proc_pident_readdir(file, ctx,
2928 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2929}
2930
2931static const struct file_operations proc_attr_dir_operations = {
2932 .read = generic_read_dir,
2933 .iterate_shared = proc_attr_dir_readdir,
2934 .llseek = generic_file_llseek,
2935};
2936
2937static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2938 struct dentry *dentry, unsigned int flags)
2939{
2940 return proc_pident_lookup(dir, dentry,
2941 attr_dir_stuff,
2942 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2943}
2944
2945static const struct inode_operations proc_attr_dir_inode_operations = {
2946 .lookup = proc_attr_dir_lookup,
2947 .getattr = pid_getattr,
2948 .setattr = proc_setattr,
2949};
2950
2951#endif
2952
2953#ifdef CONFIG_ELF_CORE
2954static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2955 size_t count, loff_t *ppos)
2956{
2957 struct task_struct *task = get_proc_task(file_inode(file));
2958 struct mm_struct *mm;
2959 char buffer[PROC_NUMBUF];
2960 size_t len;
2961 int ret;
2962
2963 if (!task)
2964 return -ESRCH;
2965
2966 ret = 0;
2967 mm = get_task_mm(task);
2968 if (mm) {
2969 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2970 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2971 MMF_DUMP_FILTER_SHIFT));
2972 mmput(mm);
2973 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2974 }
2975
2976 put_task_struct(task);
2977
2978 return ret;
2979}
2980
2981static ssize_t proc_coredump_filter_write(struct file *file,
2982 const char __user *buf,
2983 size_t count,
2984 loff_t *ppos)
2985{
2986 struct task_struct *task;
2987 struct mm_struct *mm;
2988 unsigned int val;
2989 int ret;
2990 int i;
2991 unsigned long mask;
2992
2993 ret = kstrtouint_from_user(buf, count, 0, &val);
2994 if (ret < 0)
2995 return ret;
2996
2997 ret = -ESRCH;
2998 task = get_proc_task(file_inode(file));
2999 if (!task)
3000 goto out_no_task;
3001
3002 mm = get_task_mm(task);
3003 if (!mm)
3004 goto out_no_mm;
3005 ret = 0;
3006
3007 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
3008 if (val & mask)
3009 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3010 else
3011 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3012 }
3013
3014 mmput(mm);
3015 out_no_mm:
3016 put_task_struct(task);
3017 out_no_task:
3018 if (ret < 0)
3019 return ret;
3020 return count;
3021}
3022
3023static const struct file_operations proc_coredump_filter_operations = {
3024 .read = proc_coredump_filter_read,
3025 .write = proc_coredump_filter_write,
3026 .llseek = generic_file_llseek,
3027};
3028#endif
3029
3030#ifdef CONFIG_TASK_IO_ACCOUNTING
3031static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
3032{
3033 struct task_io_accounting acct;
3034 int result;
3035
3036 result = down_read_killable(&task->signal->exec_update_lock);
3037 if (result)
3038 return result;
3039
3040 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
3041 result = -EACCES;
3042 goto out_unlock;
3043 }
3044
3045 if (whole) {
3046 struct signal_struct *sig = task->signal;
3047 struct task_struct *t;
3048 unsigned int seq = 1;
3049 unsigned long flags;
3050
3051 rcu_read_lock();
3052 do {
3053 seq++; /* 2 on the 1st/lockless path, otherwise odd */
3054 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
3055
3056 acct = sig->ioac;
3057 __for_each_thread(sig, t)
3058 task_io_accounting_add(&acct, &t->ioac);
3059
3060 } while (need_seqretry(&sig->stats_lock, seq));
3061 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
3062 rcu_read_unlock();
3063 } else {
3064 acct = task->ioac;
3065 }
3066
3067 seq_printf(m,
3068 "rchar: %llu\n"
3069 "wchar: %llu\n"
3070 "syscr: %llu\n"
3071 "syscw: %llu\n"
3072 "read_bytes: %llu\n"
3073 "write_bytes: %llu\n"
3074 "cancelled_write_bytes: %llu\n",
3075 (unsigned long long)acct.rchar,
3076 (unsigned long long)acct.wchar,
3077 (unsigned long long)acct.syscr,
3078 (unsigned long long)acct.syscw,
3079 (unsigned long long)acct.read_bytes,
3080 (unsigned long long)acct.write_bytes,
3081 (unsigned long long)acct.cancelled_write_bytes);
3082 result = 0;
3083
3084out_unlock:
3085 up_read(&task->signal->exec_update_lock);
3086 return result;
3087}
3088
3089static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3090 struct pid *pid, struct task_struct *task)
3091{
3092 return do_io_accounting(task, m, 0);
3093}
3094
3095static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3096 struct pid *pid, struct task_struct *task)
3097{
3098 return do_io_accounting(task, m, 1);
3099}
3100#endif /* CONFIG_TASK_IO_ACCOUNTING */
3101
3102#ifdef CONFIG_USER_NS
3103static int proc_id_map_open(struct inode *inode, struct file *file,
3104 const struct seq_operations *seq_ops)
3105{
3106 struct user_namespace *ns = NULL;
3107 struct task_struct *task;
3108 struct seq_file *seq;
3109 int ret = -EINVAL;
3110
3111 task = get_proc_task(inode);
3112 if (task) {
3113 rcu_read_lock();
3114 ns = get_user_ns(task_cred_xxx(task, user_ns));
3115 rcu_read_unlock();
3116 put_task_struct(task);
3117 }
3118 if (!ns)
3119 goto err;
3120
3121 ret = seq_open(file, seq_ops);
3122 if (ret)
3123 goto err_put_ns;
3124
3125 seq = file->private_data;
3126 seq->private = ns;
3127
3128 return 0;
3129err_put_ns:
3130 put_user_ns(ns);
3131err:
3132 return ret;
3133}
3134
3135static int proc_id_map_release(struct inode *inode, struct file *file)
3136{
3137 struct seq_file *seq = file->private_data;
3138 struct user_namespace *ns = seq->private;
3139 put_user_ns(ns);
3140 return seq_release(inode, file);
3141}
3142
3143static int proc_uid_map_open(struct inode *inode, struct file *file)
3144{
3145 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3146}
3147
3148static int proc_gid_map_open(struct inode *inode, struct file *file)
3149{
3150 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3151}
3152
3153static int proc_projid_map_open(struct inode *inode, struct file *file)
3154{
3155 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3156}
3157
3158static const struct file_operations proc_uid_map_operations = {
3159 .open = proc_uid_map_open,
3160 .write = proc_uid_map_write,
3161 .read = seq_read,
3162 .llseek = seq_lseek,
3163 .release = proc_id_map_release,
3164};
3165
3166static const struct file_operations proc_gid_map_operations = {
3167 .open = proc_gid_map_open,
3168 .write = proc_gid_map_write,
3169 .read = seq_read,
3170 .llseek = seq_lseek,
3171 .release = proc_id_map_release,
3172};
3173
3174static const struct file_operations proc_projid_map_operations = {
3175 .open = proc_projid_map_open,
3176 .write = proc_projid_map_write,
3177 .read = seq_read,
3178 .llseek = seq_lseek,
3179 .release = proc_id_map_release,
3180};
3181
3182static int proc_setgroups_open(struct inode *inode, struct file *file)
3183{
3184 struct user_namespace *ns = NULL;
3185 struct task_struct *task;
3186 int ret;
3187
3188 ret = -ESRCH;
3189 task = get_proc_task(inode);
3190 if (task) {
3191 rcu_read_lock();
3192 ns = get_user_ns(task_cred_xxx(task, user_ns));
3193 rcu_read_unlock();
3194 put_task_struct(task);
3195 }
3196 if (!ns)
3197 goto err;
3198
3199 if (file->f_mode & FMODE_WRITE) {
3200 ret = -EACCES;
3201 if (!ns_capable(ns, CAP_SYS_ADMIN))
3202 goto err_put_ns;
3203 }
3204
3205 ret = single_open(file, &proc_setgroups_show, ns);
3206 if (ret)
3207 goto err_put_ns;
3208
3209 return 0;
3210err_put_ns:
3211 put_user_ns(ns);
3212err:
3213 return ret;
3214}
3215
3216static int proc_setgroups_release(struct inode *inode, struct file *file)
3217{
3218 struct seq_file *seq = file->private_data;
3219 struct user_namespace *ns = seq->private;
3220 int ret = single_release(inode, file);
3221 put_user_ns(ns);
3222 return ret;
3223}
3224
3225static const struct file_operations proc_setgroups_operations = {
3226 .open = proc_setgroups_open,
3227 .write = proc_setgroups_write,
3228 .read = seq_read,
3229 .llseek = seq_lseek,
3230 .release = proc_setgroups_release,
3231};
3232#endif /* CONFIG_USER_NS */
3233
3234static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3235 struct pid *pid, struct task_struct *task)
3236{
3237 int err = lock_trace(task);
3238 if (!err) {
3239 seq_printf(m, "%08x\n", task->personality);
3240 unlock_trace(task);
3241 }
3242 return err;
3243}
3244
3245#ifdef CONFIG_LIVEPATCH
3246static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3247 struct pid *pid, struct task_struct *task)
3248{
3249 seq_printf(m, "%d\n", task->patch_state);
3250 return 0;
3251}
3252#endif /* CONFIG_LIVEPATCH */
3253
3254#ifdef CONFIG_KSM
3255static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3256 struct pid *pid, struct task_struct *task)
3257{
3258 struct mm_struct *mm;
3259
3260 mm = get_task_mm(task);
3261 if (mm) {
3262 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3263 mmput(mm);
3264 }
3265
3266 return 0;
3267}
3268static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3269 struct pid *pid, struct task_struct *task)
3270{
3271 struct mm_struct *mm;
3272
3273 mm = get_task_mm(task);
3274 if (mm) {
3275 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3276 seq_printf(m, "ksm_zero_pages %ld\n", mm_ksm_zero_pages(mm));
3277 seq_printf(m, "ksm_merging_pages %lu\n", mm->ksm_merging_pages);
3278 seq_printf(m, "ksm_process_profit %ld\n", ksm_process_profit(mm));
3279 mmput(mm);
3280 }
3281
3282 return 0;
3283}
3284#endif /* CONFIG_KSM */
3285
3286#ifdef CONFIG_STACKLEAK_METRICS
3287static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3288 struct pid *pid, struct task_struct *task)
3289{
3290 unsigned long prev_depth = THREAD_SIZE -
3291 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3292 unsigned long depth = THREAD_SIZE -
3293 (task->lowest_stack & (THREAD_SIZE - 1));
3294
3295 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3296 prev_depth, depth);
3297 return 0;
3298}
3299#endif /* CONFIG_STACKLEAK_METRICS */
3300
3301/*
3302 * Thread groups
3303 */
3304static const struct file_operations proc_task_operations;
3305static const struct inode_operations proc_task_inode_operations;
3306
3307static const struct pid_entry tgid_base_stuff[] = {
3308 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3309 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3310 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3311 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3312 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3313#ifdef CONFIG_NET
3314 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3315#endif
3316 REG("environ", S_IRUSR, proc_environ_operations),
3317 REG("auxv", S_IRUSR, proc_auxv_operations),
3318 ONE("status", S_IRUGO, proc_pid_status),
3319 ONE("personality", S_IRUSR, proc_pid_personality),
3320 ONE("limits", S_IRUGO, proc_pid_limits),
3321#ifdef CONFIG_SCHED_DEBUG
3322 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3323#endif
3324#ifdef CONFIG_SCHED_AUTOGROUP
3325 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3326#endif
3327#ifdef CONFIG_TIME_NS
3328 REG("timens_offsets", S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3329#endif
3330 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3331#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3332 ONE("syscall", S_IRUSR, proc_pid_syscall),
3333#endif
3334 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3335 ONE("stat", S_IRUGO, proc_tgid_stat),
3336 ONE("statm", S_IRUGO, proc_pid_statm),
3337 REG("maps", S_IRUGO, proc_pid_maps_operations),
3338#ifdef CONFIG_NUMA
3339 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3340#endif
3341 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3342 LNK("cwd", proc_cwd_link),
3343 LNK("root", proc_root_link),
3344 LNK("exe", proc_exe_link),
3345 REG("mounts", S_IRUGO, proc_mounts_operations),
3346 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3347 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3348#ifdef CONFIG_PROC_PAGE_MONITOR
3349 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3350 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3351 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3352 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3353#endif
3354#ifdef CONFIG_SECURITY
3355 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3356#endif
3357#ifdef CONFIG_KALLSYMS
3358 ONE("wchan", S_IRUGO, proc_pid_wchan),
3359#endif
3360#ifdef CONFIG_STACKTRACE
3361 ONE("stack", S_IRUSR, proc_pid_stack),
3362#endif
3363#ifdef CONFIG_SCHED_INFO
3364 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3365#endif
3366#ifdef CONFIG_LATENCYTOP
3367 REG("latency", S_IRUGO, proc_lstats_operations),
3368#endif
3369#ifdef CONFIG_PROC_PID_CPUSET
3370 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3371#endif
3372#ifdef CONFIG_CGROUPS
3373 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3374#endif
3375#ifdef CONFIG_PROC_CPU_RESCTRL
3376 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3377#endif
3378 ONE("oom_score", S_IRUGO, proc_oom_score),
3379 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3380 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3381#ifdef CONFIG_AUDIT
3382 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3383 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3384#endif
3385#ifdef CONFIG_FAULT_INJECTION
3386 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3387 REG("fail-nth", 0644, proc_fail_nth_operations),
3388#endif
3389#ifdef CONFIG_ELF_CORE
3390 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3391#endif
3392#ifdef CONFIG_TASK_IO_ACCOUNTING
3393 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3394#endif
3395#ifdef CONFIG_USER_NS
3396 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3397 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3398 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3399 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3400#endif
3401#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3402 REG("timers", S_IRUGO, proc_timers_operations),
3403#endif
3404 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3405#ifdef CONFIG_LIVEPATCH
3406 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3407#endif
3408#ifdef CONFIG_STACKLEAK_METRICS
3409 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3410#endif
3411#ifdef CONFIG_PROC_PID_ARCH_STATUS
3412 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3413#endif
3414#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3415 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3416#endif
3417#ifdef CONFIG_KSM
3418 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3419 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3420#endif
3421};
3422
3423static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3424{
3425 return proc_pident_readdir(file, ctx,
3426 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3427}
3428
3429static const struct file_operations proc_tgid_base_operations = {
3430 .read = generic_read_dir,
3431 .iterate_shared = proc_tgid_base_readdir,
3432 .llseek = generic_file_llseek,
3433};
3434
3435struct pid *tgid_pidfd_to_pid(const struct file *file)
3436{
3437 if (file->f_op != &proc_tgid_base_operations)
3438 return ERR_PTR(-EBADF);
3439
3440 return proc_pid(file_inode(file));
3441}
3442
3443static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3444{
3445 return proc_pident_lookup(dir, dentry,
3446 tgid_base_stuff,
3447 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3448}
3449
3450static const struct inode_operations proc_tgid_base_inode_operations = {
3451 .lookup = proc_tgid_base_lookup,
3452 .getattr = pid_getattr,
3453 .setattr = proc_setattr,
3454 .permission = proc_pid_permission,
3455};
3456
3457/**
3458 * proc_flush_pid - Remove dcache entries for @pid from the /proc dcache.
3459 * @pid: pid that should be flushed.
3460 *
3461 * This function walks a list of inodes (that belong to any proc
3462 * filesystem) that are attached to the pid and flushes them from
3463 * the dentry cache.
3464 *
3465 * It is safe and reasonable to cache /proc entries for a task until
3466 * that task exits. After that they just clog up the dcache with
3467 * useless entries, possibly causing useful dcache entries to be
3468 * flushed instead. This routine is provided to flush those useless
3469 * dcache entries when a process is reaped.
3470 *
3471 * NOTE: This routine is just an optimization so it does not guarantee
3472 * that no dcache entries will exist after a process is reaped
3473 * it just makes it very unlikely that any will persist.
3474 */
3475
3476void proc_flush_pid(struct pid *pid)
3477{
3478 proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3479}
3480
3481static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3482 struct task_struct *task, const void *ptr)
3483{
3484 struct inode *inode;
3485
3486 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3487 S_IFDIR | S_IRUGO | S_IXUGO);
3488 if (!inode)
3489 return ERR_PTR(-ENOENT);
3490
3491 inode->i_op = &proc_tgid_base_inode_operations;
3492 inode->i_fop = &proc_tgid_base_operations;
3493 inode->i_flags|=S_IMMUTABLE;
3494
3495 set_nlink(inode, nlink_tgid);
3496 pid_update_inode(task, inode);
3497
3498 d_set_d_op(dentry, &pid_dentry_operations);
3499 return d_splice_alias(inode, dentry);
3500}
3501
3502struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3503{
3504 struct task_struct *task;
3505 unsigned tgid;
3506 struct proc_fs_info *fs_info;
3507 struct pid_namespace *ns;
3508 struct dentry *result = ERR_PTR(-ENOENT);
3509
3510 tgid = name_to_int(&dentry->d_name);
3511 if (tgid == ~0U)
3512 goto out;
3513
3514 fs_info = proc_sb_info(dentry->d_sb);
3515 ns = fs_info->pid_ns;
3516 rcu_read_lock();
3517 task = find_task_by_pid_ns(tgid, ns);
3518 if (task)
3519 get_task_struct(task);
3520 rcu_read_unlock();
3521 if (!task)
3522 goto out;
3523
3524 /* Limit procfs to only ptraceable tasks */
3525 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3526 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3527 goto out_put_task;
3528 }
3529
3530 result = proc_pid_instantiate(dentry, task, NULL);
3531out_put_task:
3532 put_task_struct(task);
3533out:
3534 return result;
3535}
3536
3537/*
3538 * Find the first task with tgid >= tgid
3539 *
3540 */
3541struct tgid_iter {
3542 unsigned int tgid;
3543 struct task_struct *task;
3544};
3545static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3546{
3547 struct pid *pid;
3548
3549 if (iter.task)
3550 put_task_struct(iter.task);
3551 rcu_read_lock();
3552retry:
3553 iter.task = NULL;
3554 pid = find_ge_pid(iter.tgid, ns);
3555 if (pid) {
3556 iter.tgid = pid_nr_ns(pid, ns);
3557 iter.task = pid_task(pid, PIDTYPE_TGID);
3558 if (!iter.task) {
3559 iter.tgid += 1;
3560 goto retry;
3561 }
3562 get_task_struct(iter.task);
3563 }
3564 rcu_read_unlock();
3565 return iter;
3566}
3567
3568#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3569
3570/* for the /proc/ directory itself, after non-process stuff has been done */
3571int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3572{
3573 struct tgid_iter iter;
3574 struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3575 struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3576 loff_t pos = ctx->pos;
3577
3578 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3579 return 0;
3580
3581 if (pos == TGID_OFFSET - 2) {
3582 struct inode *inode = d_inode(fs_info->proc_self);
3583 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3584 return 0;
3585 ctx->pos = pos = pos + 1;
3586 }
3587 if (pos == TGID_OFFSET - 1) {
3588 struct inode *inode = d_inode(fs_info->proc_thread_self);
3589 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3590 return 0;
3591 ctx->pos = pos = pos + 1;
3592 }
3593 iter.tgid = pos - TGID_OFFSET;
3594 iter.task = NULL;
3595 for (iter = next_tgid(ns, iter);
3596 iter.task;
3597 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3598 char name[10 + 1];
3599 unsigned int len;
3600
3601 cond_resched();
3602 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3603 continue;
3604
3605 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3606 ctx->pos = iter.tgid + TGID_OFFSET;
3607 if (!proc_fill_cache(file, ctx, name, len,
3608 proc_pid_instantiate, iter.task, NULL)) {
3609 put_task_struct(iter.task);
3610 return 0;
3611 }
3612 }
3613 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3614 return 0;
3615}
3616
3617/*
3618 * proc_tid_comm_permission is a special permission function exclusively
3619 * used for the node /proc/<pid>/task/<tid>/comm.
3620 * It bypasses generic permission checks in the case where a task of the same
3621 * task group attempts to access the node.
3622 * The rationale behind this is that glibc and bionic access this node for
3623 * cross thread naming (pthread_set/getname_np(!self)). However, if
3624 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3625 * which locks out the cross thread naming implementation.
3626 * This function makes sure that the node is always accessible for members of
3627 * same thread group.
3628 */
3629static int proc_tid_comm_permission(struct mnt_idmap *idmap,
3630 struct inode *inode, int mask)
3631{
3632 bool is_same_tgroup;
3633 struct task_struct *task;
3634
3635 task = get_proc_task(inode);
3636 if (!task)
3637 return -ESRCH;
3638 is_same_tgroup = same_thread_group(current, task);
3639 put_task_struct(task);
3640
3641 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3642 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3643 * read or written by the members of the corresponding
3644 * thread group.
3645 */
3646 return 0;
3647 }
3648
3649 return generic_permission(&nop_mnt_idmap, inode, mask);
3650}
3651
3652static const struct inode_operations proc_tid_comm_inode_operations = {
3653 .setattr = proc_setattr,
3654 .permission = proc_tid_comm_permission,
3655};
3656
3657/*
3658 * Tasks
3659 */
3660static const struct pid_entry tid_base_stuff[] = {
3661 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3662 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3663 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3664#ifdef CONFIG_NET
3665 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3666#endif
3667 REG("environ", S_IRUSR, proc_environ_operations),
3668 REG("auxv", S_IRUSR, proc_auxv_operations),
3669 ONE("status", S_IRUGO, proc_pid_status),
3670 ONE("personality", S_IRUSR, proc_pid_personality),
3671 ONE("limits", S_IRUGO, proc_pid_limits),
3672#ifdef CONFIG_SCHED_DEBUG
3673 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3674#endif
3675 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3676 &proc_tid_comm_inode_operations,
3677 &proc_pid_set_comm_operations, {}),
3678#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3679 ONE("syscall", S_IRUSR, proc_pid_syscall),
3680#endif
3681 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3682 ONE("stat", S_IRUGO, proc_tid_stat),
3683 ONE("statm", S_IRUGO, proc_pid_statm),
3684 REG("maps", S_IRUGO, proc_pid_maps_operations),
3685#ifdef CONFIG_PROC_CHILDREN
3686 REG("children", S_IRUGO, proc_tid_children_operations),
3687#endif
3688#ifdef CONFIG_NUMA
3689 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3690#endif
3691 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3692 LNK("cwd", proc_cwd_link),
3693 LNK("root", proc_root_link),
3694 LNK("exe", proc_exe_link),
3695 REG("mounts", S_IRUGO, proc_mounts_operations),
3696 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3697#ifdef CONFIG_PROC_PAGE_MONITOR
3698 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3699 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3700 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3701 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3702#endif
3703#ifdef CONFIG_SECURITY
3704 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3705#endif
3706#ifdef CONFIG_KALLSYMS
3707 ONE("wchan", S_IRUGO, proc_pid_wchan),
3708#endif
3709#ifdef CONFIG_STACKTRACE
3710 ONE("stack", S_IRUSR, proc_pid_stack),
3711#endif
3712#ifdef CONFIG_SCHED_INFO
3713 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3714#endif
3715#ifdef CONFIG_LATENCYTOP
3716 REG("latency", S_IRUGO, proc_lstats_operations),
3717#endif
3718#ifdef CONFIG_PROC_PID_CPUSET
3719 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3720#endif
3721#ifdef CONFIG_CGROUPS
3722 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3723#endif
3724#ifdef CONFIG_PROC_CPU_RESCTRL
3725 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3726#endif
3727 ONE("oom_score", S_IRUGO, proc_oom_score),
3728 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3729 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3730#ifdef CONFIG_AUDIT
3731 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3732 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3733#endif
3734#ifdef CONFIG_FAULT_INJECTION
3735 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3736 REG("fail-nth", 0644, proc_fail_nth_operations),
3737#endif
3738#ifdef CONFIG_TASK_IO_ACCOUNTING
3739 ONE("io", S_IRUSR, proc_tid_io_accounting),
3740#endif
3741#ifdef CONFIG_USER_NS
3742 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3743 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3744 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3745 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3746#endif
3747#ifdef CONFIG_LIVEPATCH
3748 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3749#endif
3750#ifdef CONFIG_PROC_PID_ARCH_STATUS
3751 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3752#endif
3753#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3754 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3755#endif
3756#ifdef CONFIG_KSM
3757 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3758 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3759#endif
3760};
3761
3762static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3763{
3764 return proc_pident_readdir(file, ctx,
3765 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3766}
3767
3768static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3769{
3770 return proc_pident_lookup(dir, dentry,
3771 tid_base_stuff,
3772 tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3773}
3774
3775static const struct file_operations proc_tid_base_operations = {
3776 .read = generic_read_dir,
3777 .iterate_shared = proc_tid_base_readdir,
3778 .llseek = generic_file_llseek,
3779};
3780
3781static const struct inode_operations proc_tid_base_inode_operations = {
3782 .lookup = proc_tid_base_lookup,
3783 .getattr = pid_getattr,
3784 .setattr = proc_setattr,
3785};
3786
3787static struct dentry *proc_task_instantiate(struct dentry *dentry,
3788 struct task_struct *task, const void *ptr)
3789{
3790 struct inode *inode;
3791 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3792 S_IFDIR | S_IRUGO | S_IXUGO);
3793 if (!inode)
3794 return ERR_PTR(-ENOENT);
3795
3796 inode->i_op = &proc_tid_base_inode_operations;
3797 inode->i_fop = &proc_tid_base_operations;
3798 inode->i_flags |= S_IMMUTABLE;
3799
3800 set_nlink(inode, nlink_tid);
3801 pid_update_inode(task, inode);
3802
3803 d_set_d_op(dentry, &pid_dentry_operations);
3804 return d_splice_alias(inode, dentry);
3805}
3806
3807static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3808{
3809 struct task_struct *task;
3810 struct task_struct *leader = get_proc_task(dir);
3811 unsigned tid;
3812 struct proc_fs_info *fs_info;
3813 struct pid_namespace *ns;
3814 struct dentry *result = ERR_PTR(-ENOENT);
3815
3816 if (!leader)
3817 goto out_no_task;
3818
3819 tid = name_to_int(&dentry->d_name);
3820 if (tid == ~0U)
3821 goto out;
3822
3823 fs_info = proc_sb_info(dentry->d_sb);
3824 ns = fs_info->pid_ns;
3825 rcu_read_lock();
3826 task = find_task_by_pid_ns(tid, ns);
3827 if (task)
3828 get_task_struct(task);
3829 rcu_read_unlock();
3830 if (!task)
3831 goto out;
3832 if (!same_thread_group(leader, task))
3833 goto out_drop_task;
3834
3835 result = proc_task_instantiate(dentry, task, NULL);
3836out_drop_task:
3837 put_task_struct(task);
3838out:
3839 put_task_struct(leader);
3840out_no_task:
3841 return result;
3842}
3843
3844/*
3845 * Find the first tid of a thread group to return to user space.
3846 *
3847 * Usually this is just the thread group leader, but if the users
3848 * buffer was too small or there was a seek into the middle of the
3849 * directory we have more work todo.
3850 *
3851 * In the case of a short read we start with find_task_by_pid.
3852 *
3853 * In the case of a seek we start with the leader and walk nr
3854 * threads past it.
3855 */
3856static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3857 struct pid_namespace *ns)
3858{
3859 struct task_struct *pos, *task;
3860 unsigned long nr = f_pos;
3861
3862 if (nr != f_pos) /* 32bit overflow? */
3863 return NULL;
3864
3865 rcu_read_lock();
3866 task = pid_task(pid, PIDTYPE_PID);
3867 if (!task)
3868 goto fail;
3869
3870 /* Attempt to start with the tid of a thread */
3871 if (tid && nr) {
3872 pos = find_task_by_pid_ns(tid, ns);
3873 if (pos && same_thread_group(pos, task))
3874 goto found;
3875 }
3876
3877 /* If nr exceeds the number of threads there is nothing todo */
3878 if (nr >= get_nr_threads(task))
3879 goto fail;
3880
3881 /* If we haven't found our starting place yet start
3882 * with the leader and walk nr threads forward.
3883 */
3884 for_each_thread(task, pos) {
3885 if (!nr--)
3886 goto found;
3887 }
3888fail:
3889 pos = NULL;
3890 goto out;
3891found:
3892 get_task_struct(pos);
3893out:
3894 rcu_read_unlock();
3895 return pos;
3896}
3897
3898/*
3899 * Find the next thread in the thread list.
3900 * Return NULL if there is an error or no next thread.
3901 *
3902 * The reference to the input task_struct is released.
3903 */
3904static struct task_struct *next_tid(struct task_struct *start)
3905{
3906 struct task_struct *pos = NULL;
3907 rcu_read_lock();
3908 if (pid_alive(start)) {
3909 pos = __next_thread(start);
3910 if (pos)
3911 get_task_struct(pos);
3912 }
3913 rcu_read_unlock();
3914 put_task_struct(start);
3915 return pos;
3916}
3917
3918/* for the /proc/TGID/task/ directories */
3919static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3920{
3921 struct inode *inode = file_inode(file);
3922 struct task_struct *task;
3923 struct pid_namespace *ns;
3924 int tid;
3925
3926 if (proc_inode_is_dead(inode))
3927 return -ENOENT;
3928
3929 if (!dir_emit_dots(file, ctx))
3930 return 0;
3931
3932 /* We cache the tgid value that the last readdir call couldn't
3933 * return and lseek resets it to 0.
3934 */
3935 ns = proc_pid_ns(inode->i_sb);
3936 tid = (int)(intptr_t)file->private_data;
3937 file->private_data = NULL;
3938 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3939 task;
3940 task = next_tid(task), ctx->pos++) {
3941 char name[10 + 1];
3942 unsigned int len;
3943
3944 tid = task_pid_nr_ns(task, ns);
3945 if (!tid)
3946 continue; /* The task has just exited. */
3947 len = snprintf(name, sizeof(name), "%u", tid);
3948 if (!proc_fill_cache(file, ctx, name, len,
3949 proc_task_instantiate, task, NULL)) {
3950 /* returning this tgid failed, save it as the first
3951 * pid for the next readir call */
3952 file->private_data = (void *)(intptr_t)tid;
3953 put_task_struct(task);
3954 break;
3955 }
3956 }
3957
3958 return 0;
3959}
3960
3961static int proc_task_getattr(struct mnt_idmap *idmap,
3962 const struct path *path, struct kstat *stat,
3963 u32 request_mask, unsigned int query_flags)
3964{
3965 struct inode *inode = d_inode(path->dentry);
3966 struct task_struct *p = get_proc_task(inode);
3967 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
3968
3969 if (p) {
3970 stat->nlink += get_nr_threads(p);
3971 put_task_struct(p);
3972 }
3973
3974 return 0;
3975}
3976
3977/*
3978 * proc_task_readdir() set @file->private_data to a positive integer
3979 * value, so casting that to u64 is safe. generic_llseek_cookie() will
3980 * set @cookie to 0, so casting to an int is safe. The WARN_ON_ONCE() is
3981 * here to catch any unexpected change in behavior either in
3982 * proc_task_readdir() or generic_llseek_cookie().
3983 */
3984static loff_t proc_dir_llseek(struct file *file, loff_t offset, int whence)
3985{
3986 u64 cookie = (u64)(intptr_t)file->private_data;
3987 loff_t off;
3988
3989 off = generic_llseek_cookie(file, offset, whence, &cookie);
3990 WARN_ON_ONCE(cookie > INT_MAX);
3991 file->private_data = (void *)(intptr_t)cookie; /* serialized by f_pos_lock */
3992 return off;
3993}
3994
3995static const struct inode_operations proc_task_inode_operations = {
3996 .lookup = proc_task_lookup,
3997 .getattr = proc_task_getattr,
3998 .setattr = proc_setattr,
3999 .permission = proc_pid_permission,
4000};
4001
4002static const struct file_operations proc_task_operations = {
4003 .read = generic_read_dir,
4004 .iterate_shared = proc_task_readdir,
4005 .llseek = proc_dir_llseek,
4006};
4007
4008void __init set_proc_pid_nlink(void)
4009{
4010 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
4011 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
4012}