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 <linux/uaccess.h>
51
52#include <linux/errno.h>
53#include <linux/time.h>
54#include <linux/proc_fs.h>
55#include <linux/stat.h>
56#include <linux/task_io_accounting_ops.h>
57#include <linux/init.h>
58#include <linux/capability.h>
59#include <linux/file.h>
60#include <linux/fdtable.h>
61#include <linux/string.h>
62#include <linux/seq_file.h>
63#include <linux/namei.h>
64#include <linux/mnt_namespace.h>
65#include <linux/mm.h>
66#include <linux/swap.h>
67#include <linux/rcupdate.h>
68#include <linux/kallsyms.h>
69#include <linux/stacktrace.h>
70#include <linux/resource.h>
71#include <linux/module.h>
72#include <linux/mount.h>
73#include <linux/security.h>
74#include <linux/ptrace.h>
75#include <linux/tracehook.h>
76#include <linux/printk.h>
77#include <linux/cgroup.h>
78#include <linux/cpuset.h>
79#include <linux/audit.h>
80#include <linux/poll.h>
81#include <linux/nsproxy.h>
82#include <linux/oom.h>
83#include <linux/elf.h>
84#include <linux/pid_namespace.h>
85#include <linux/user_namespace.h>
86#include <linux/fs_struct.h>
87#include <linux/slab.h>
88#include <linux/flex_array.h>
89#include <linux/posix-timers.h>
90#ifdef CONFIG_HARDWALL
91#include <asm/hardwall.h>
92#endif
93#include <trace/events/oom.h>
94#include "internal.h"
95#include "fd.h"
96
97/* NOTE:
98 * Implementing inode permission operations in /proc is almost
99 * certainly an error. Permission checks need to happen during
100 * each system call not at open time. The reason is that most of
101 * what we wish to check for permissions in /proc varies at runtime.
102 *
103 * The classic example of a problem is opening file descriptors
104 * in /proc for a task before it execs a suid executable.
105 */
106
107static u8 nlink_tid;
108static u8 nlink_tgid;
109
110struct pid_entry {
111 const char *name;
112 unsigned int len;
113 umode_t mode;
114 const struct inode_operations *iop;
115 const struct file_operations *fop;
116 union proc_op op;
117};
118
119#define NOD(NAME, MODE, IOP, FOP, OP) { \
120 .name = (NAME), \
121 .len = sizeof(NAME) - 1, \
122 .mode = MODE, \
123 .iop = IOP, \
124 .fop = FOP, \
125 .op = OP, \
126}
127
128#define DIR(NAME, MODE, iops, fops) \
129 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
130#define LNK(NAME, get_link) \
131 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
132 &proc_pid_link_inode_operations, NULL, \
133 { .proc_get_link = get_link } )
134#define REG(NAME, MODE, fops) \
135 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
136#define ONE(NAME, MODE, show) \
137 NOD(NAME, (S_IFREG|(MODE)), \
138 NULL, &proc_single_file_operations, \
139 { .proc_show = show } )
140
141/*
142 * Count the number of hardlinks for the pid_entry table, excluding the .
143 * and .. links.
144 */
145static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
146 unsigned int n)
147{
148 unsigned int i;
149 unsigned int count;
150
151 count = 2;
152 for (i = 0; i < n; ++i) {
153 if (S_ISDIR(entries[i].mode))
154 ++count;
155 }
156
157 return count;
158}
159
160static int get_task_root(struct task_struct *task, struct path *root)
161{
162 int result = -ENOENT;
163
164 task_lock(task);
165 if (task->fs) {
166 get_fs_root(task->fs, root);
167 result = 0;
168 }
169 task_unlock(task);
170 return result;
171}
172
173static int proc_cwd_link(struct dentry *dentry, struct path *path)
174{
175 struct task_struct *task = get_proc_task(d_inode(dentry));
176 int result = -ENOENT;
177
178 if (task) {
179 task_lock(task);
180 if (task->fs) {
181 get_fs_pwd(task->fs, path);
182 result = 0;
183 }
184 task_unlock(task);
185 put_task_struct(task);
186 }
187 return result;
188}
189
190static int proc_root_link(struct dentry *dentry, struct path *path)
191{
192 struct task_struct *task = get_proc_task(d_inode(dentry));
193 int result = -ENOENT;
194
195 if (task) {
196 result = get_task_root(task, path);
197 put_task_struct(task);
198 }
199 return result;
200}
201
202static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
203 size_t _count, loff_t *pos)
204{
205 struct task_struct *tsk;
206 struct mm_struct *mm;
207 char *page;
208 unsigned long count = _count;
209 unsigned long arg_start, arg_end, env_start, env_end;
210 unsigned long len1, len2, len;
211 unsigned long p;
212 char c;
213 ssize_t rv;
214
215 BUG_ON(*pos < 0);
216
217 tsk = get_proc_task(file_inode(file));
218 if (!tsk)
219 return -ESRCH;
220 mm = get_task_mm(tsk);
221 put_task_struct(tsk);
222 if (!mm)
223 return 0;
224 /* Check if process spawned far enough to have cmdline. */
225 if (!mm->env_end) {
226 rv = 0;
227 goto out_mmput;
228 }
229
230 page = (char *)__get_free_page(GFP_TEMPORARY);
231 if (!page) {
232 rv = -ENOMEM;
233 goto out_mmput;
234 }
235
236 down_read(&mm->mmap_sem);
237 arg_start = mm->arg_start;
238 arg_end = mm->arg_end;
239 env_start = mm->env_start;
240 env_end = mm->env_end;
241 up_read(&mm->mmap_sem);
242
243 BUG_ON(arg_start > arg_end);
244 BUG_ON(env_start > env_end);
245
246 len1 = arg_end - arg_start;
247 len2 = env_end - env_start;
248
249 /* Empty ARGV. */
250 if (len1 == 0) {
251 rv = 0;
252 goto out_free_page;
253 }
254 /*
255 * Inherently racy -- command line shares address space
256 * with code and data.
257 */
258 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
259 if (rv <= 0)
260 goto out_free_page;
261
262 rv = 0;
263
264 if (c == '\0') {
265 /* Command line (set of strings) occupies whole ARGV. */
266 if (len1 <= *pos)
267 goto out_free_page;
268
269 p = arg_start + *pos;
270 len = len1 - *pos;
271 while (count > 0 && len > 0) {
272 unsigned int _count;
273 int nr_read;
274
275 _count = min3(count, len, PAGE_SIZE);
276 nr_read = access_remote_vm(mm, p, page, _count, 0);
277 if (nr_read < 0)
278 rv = nr_read;
279 if (nr_read <= 0)
280 goto out_free_page;
281
282 if (copy_to_user(buf, page, nr_read)) {
283 rv = -EFAULT;
284 goto out_free_page;
285 }
286
287 p += nr_read;
288 len -= nr_read;
289 buf += nr_read;
290 count -= nr_read;
291 rv += nr_read;
292 }
293 } else {
294 /*
295 * Command line (1 string) occupies ARGV and maybe
296 * extends into ENVP.
297 */
298 if (len1 + len2 <= *pos)
299 goto skip_argv_envp;
300 if (len1 <= *pos)
301 goto skip_argv;
302
303 p = arg_start + *pos;
304 len = len1 - *pos;
305 while (count > 0 && len > 0) {
306 unsigned int _count, l;
307 int nr_read;
308 bool final;
309
310 _count = min3(count, len, PAGE_SIZE);
311 nr_read = access_remote_vm(mm, p, page, _count, 0);
312 if (nr_read < 0)
313 rv = nr_read;
314 if (nr_read <= 0)
315 goto out_free_page;
316
317 /*
318 * Command line can be shorter than whole ARGV
319 * even if last "marker" byte says it is not.
320 */
321 final = false;
322 l = strnlen(page, nr_read);
323 if (l < nr_read) {
324 nr_read = l;
325 final = true;
326 }
327
328 if (copy_to_user(buf, page, nr_read)) {
329 rv = -EFAULT;
330 goto out_free_page;
331 }
332
333 p += nr_read;
334 len -= nr_read;
335 buf += nr_read;
336 count -= nr_read;
337 rv += nr_read;
338
339 if (final)
340 goto out_free_page;
341 }
342skip_argv:
343 /*
344 * Command line (1 string) occupies ARGV and
345 * extends into ENVP.
346 */
347 if (len1 <= *pos) {
348 p = env_start + *pos - len1;
349 len = len1 + len2 - *pos;
350 } else {
351 p = env_start;
352 len = len2;
353 }
354 while (count > 0 && len > 0) {
355 unsigned int _count, l;
356 int nr_read;
357 bool final;
358
359 _count = min3(count, len, PAGE_SIZE);
360 nr_read = access_remote_vm(mm, p, page, _count, 0);
361 if (nr_read < 0)
362 rv = nr_read;
363 if (nr_read <= 0)
364 goto out_free_page;
365
366 /* Find EOS. */
367 final = false;
368 l = strnlen(page, nr_read);
369 if (l < nr_read) {
370 nr_read = l;
371 final = true;
372 }
373
374 if (copy_to_user(buf, page, nr_read)) {
375 rv = -EFAULT;
376 goto out_free_page;
377 }
378
379 p += nr_read;
380 len -= nr_read;
381 buf += nr_read;
382 count -= nr_read;
383 rv += nr_read;
384
385 if (final)
386 goto out_free_page;
387 }
388skip_argv_envp:
389 ;
390 }
391
392out_free_page:
393 free_page((unsigned long)page);
394out_mmput:
395 mmput(mm);
396 if (rv > 0)
397 *pos += rv;
398 return rv;
399}
400
401static const struct file_operations proc_pid_cmdline_ops = {
402 .read = proc_pid_cmdline_read,
403 .llseek = generic_file_llseek,
404};
405
406#ifdef CONFIG_KALLSYMS
407/*
408 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
409 * Returns the resolved symbol. If that fails, simply return the address.
410 */
411static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
412 struct pid *pid, struct task_struct *task)
413{
414 unsigned long wchan;
415 char symname[KSYM_NAME_LEN];
416
417 wchan = get_wchan(task);
418
419 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
420 && !lookup_symbol_name(wchan, symname))
421 seq_printf(m, "%s", symname);
422 else
423 seq_putc(m, '0');
424
425 return 0;
426}
427#endif /* CONFIG_KALLSYMS */
428
429static int lock_trace(struct task_struct *task)
430{
431 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
432 if (err)
433 return err;
434 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
435 mutex_unlock(&task->signal->cred_guard_mutex);
436 return -EPERM;
437 }
438 return 0;
439}
440
441static void unlock_trace(struct task_struct *task)
442{
443 mutex_unlock(&task->signal->cred_guard_mutex);
444}
445
446#ifdef CONFIG_STACKTRACE
447
448#define MAX_STACK_TRACE_DEPTH 64
449
450static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
451 struct pid *pid, struct task_struct *task)
452{
453 struct stack_trace trace;
454 unsigned long *entries;
455 int err;
456 int i;
457
458 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
459 if (!entries)
460 return -ENOMEM;
461
462 trace.nr_entries = 0;
463 trace.max_entries = MAX_STACK_TRACE_DEPTH;
464 trace.entries = entries;
465 trace.skip = 0;
466
467 err = lock_trace(task);
468 if (!err) {
469 save_stack_trace_tsk(task, &trace);
470
471 for (i = 0; i < trace.nr_entries; i++) {
472 seq_printf(m, "[<%pK>] %pB\n",
473 (void *)entries[i], (void *)entries[i]);
474 }
475 unlock_trace(task);
476 }
477 kfree(entries);
478
479 return err;
480}
481#endif
482
483#ifdef CONFIG_SCHED_INFO
484/*
485 * Provides /proc/PID/schedstat
486 */
487static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
488 struct pid *pid, struct task_struct *task)
489{
490 if (unlikely(!sched_info_on()))
491 seq_printf(m, "0 0 0\n");
492 else
493 seq_printf(m, "%llu %llu %lu\n",
494 (unsigned long long)task->se.sum_exec_runtime,
495 (unsigned long long)task->sched_info.run_delay,
496 task->sched_info.pcount);
497
498 return 0;
499}
500#endif
501
502#ifdef CONFIG_LATENCYTOP
503static int lstats_show_proc(struct seq_file *m, void *v)
504{
505 int i;
506 struct inode *inode = m->private;
507 struct task_struct *task = get_proc_task(inode);
508
509 if (!task)
510 return -ESRCH;
511 seq_puts(m, "Latency Top version : v0.1\n");
512 for (i = 0; i < 32; i++) {
513 struct latency_record *lr = &task->latency_record[i];
514 if (lr->backtrace[0]) {
515 int q;
516 seq_printf(m, "%i %li %li",
517 lr->count, lr->time, lr->max);
518 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
519 unsigned long bt = lr->backtrace[q];
520 if (!bt)
521 break;
522 if (bt == ULONG_MAX)
523 break;
524 seq_printf(m, " %ps", (void *)bt);
525 }
526 seq_putc(m, '\n');
527 }
528
529 }
530 put_task_struct(task);
531 return 0;
532}
533
534static int lstats_open(struct inode *inode, struct file *file)
535{
536 return single_open(file, lstats_show_proc, inode);
537}
538
539static ssize_t lstats_write(struct file *file, const char __user *buf,
540 size_t count, loff_t *offs)
541{
542 struct task_struct *task = get_proc_task(file_inode(file));
543
544 if (!task)
545 return -ESRCH;
546 clear_all_latency_tracing(task);
547 put_task_struct(task);
548
549 return count;
550}
551
552static const struct file_operations proc_lstats_operations = {
553 .open = lstats_open,
554 .read = seq_read,
555 .write = lstats_write,
556 .llseek = seq_lseek,
557 .release = single_release,
558};
559
560#endif
561
562static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
563 struct pid *pid, struct task_struct *task)
564{
565 unsigned long totalpages = totalram_pages + total_swap_pages;
566 unsigned long points = 0;
567
568 points = oom_badness(task, NULL, NULL, totalpages) *
569 1000 / totalpages;
570 seq_printf(m, "%lu\n", points);
571
572 return 0;
573}
574
575struct limit_names {
576 const char *name;
577 const char *unit;
578};
579
580static const struct limit_names lnames[RLIM_NLIMITS] = {
581 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
582 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
583 [RLIMIT_DATA] = {"Max data size", "bytes"},
584 [RLIMIT_STACK] = {"Max stack size", "bytes"},
585 [RLIMIT_CORE] = {"Max core file size", "bytes"},
586 [RLIMIT_RSS] = {"Max resident set", "bytes"},
587 [RLIMIT_NPROC] = {"Max processes", "processes"},
588 [RLIMIT_NOFILE] = {"Max open files", "files"},
589 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
590 [RLIMIT_AS] = {"Max address space", "bytes"},
591 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
592 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
593 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
594 [RLIMIT_NICE] = {"Max nice priority", NULL},
595 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
596 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
597};
598
599/* Display limits for a process */
600static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
601 struct pid *pid, struct task_struct *task)
602{
603 unsigned int i;
604 unsigned long flags;
605
606 struct rlimit rlim[RLIM_NLIMITS];
607
608 if (!lock_task_sighand(task, &flags))
609 return 0;
610 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
611 unlock_task_sighand(task, &flags);
612
613 /*
614 * print the file header
615 */
616 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
617 "Limit", "Soft Limit", "Hard Limit", "Units");
618
619 for (i = 0; i < RLIM_NLIMITS; i++) {
620 if (rlim[i].rlim_cur == RLIM_INFINITY)
621 seq_printf(m, "%-25s %-20s ",
622 lnames[i].name, "unlimited");
623 else
624 seq_printf(m, "%-25s %-20lu ",
625 lnames[i].name, rlim[i].rlim_cur);
626
627 if (rlim[i].rlim_max == RLIM_INFINITY)
628 seq_printf(m, "%-20s ", "unlimited");
629 else
630 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
631
632 if (lnames[i].unit)
633 seq_printf(m, "%-10s\n", lnames[i].unit);
634 else
635 seq_putc(m, '\n');
636 }
637
638 return 0;
639}
640
641#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
642static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
643 struct pid *pid, struct task_struct *task)
644{
645 long nr;
646 unsigned long args[6], sp, pc;
647 int res;
648
649 res = lock_trace(task);
650 if (res)
651 return res;
652
653 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
654 seq_puts(m, "running\n");
655 else if (nr < 0)
656 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
657 else
658 seq_printf(m,
659 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
660 nr,
661 args[0], args[1], args[2], args[3], args[4], args[5],
662 sp, pc);
663 unlock_trace(task);
664
665 return 0;
666}
667#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
668
669/************************************************************************/
670/* Here the fs part begins */
671/************************************************************************/
672
673/* permission checks */
674static int proc_fd_access_allowed(struct inode *inode)
675{
676 struct task_struct *task;
677 int allowed = 0;
678 /* Allow access to a task's file descriptors if it is us or we
679 * may use ptrace attach to the process and find out that
680 * information.
681 */
682 task = get_proc_task(inode);
683 if (task) {
684 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
685 put_task_struct(task);
686 }
687 return allowed;
688}
689
690int proc_setattr(struct dentry *dentry, struct iattr *attr)
691{
692 int error;
693 struct inode *inode = d_inode(dentry);
694
695 if (attr->ia_valid & ATTR_MODE)
696 return -EPERM;
697
698 error = setattr_prepare(dentry, attr);
699 if (error)
700 return error;
701
702 setattr_copy(inode, attr);
703 mark_inode_dirty(inode);
704 return 0;
705}
706
707/*
708 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
709 * or euid/egid (for hide_pid_min=2)?
710 */
711static bool has_pid_permissions(struct pid_namespace *pid,
712 struct task_struct *task,
713 int hide_pid_min)
714{
715 if (pid->hide_pid < hide_pid_min)
716 return true;
717 if (in_group_p(pid->pid_gid))
718 return true;
719 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
720}
721
722
723static int proc_pid_permission(struct inode *inode, int mask)
724{
725 struct pid_namespace *pid = inode->i_sb->s_fs_info;
726 struct task_struct *task;
727 bool has_perms;
728
729 task = get_proc_task(inode);
730 if (!task)
731 return -ESRCH;
732 has_perms = has_pid_permissions(pid, task, 1);
733 put_task_struct(task);
734
735 if (!has_perms) {
736 if (pid->hide_pid == 2) {
737 /*
738 * Let's make getdents(), stat(), and open()
739 * consistent with each other. If a process
740 * may not stat() a file, it shouldn't be seen
741 * in procfs at all.
742 */
743 return -ENOENT;
744 }
745
746 return -EPERM;
747 }
748 return generic_permission(inode, mask);
749}
750
751
752
753static const struct inode_operations proc_def_inode_operations = {
754 .setattr = proc_setattr,
755};
756
757static int proc_single_show(struct seq_file *m, void *v)
758{
759 struct inode *inode = m->private;
760 struct pid_namespace *ns;
761 struct pid *pid;
762 struct task_struct *task;
763 int ret;
764
765 ns = inode->i_sb->s_fs_info;
766 pid = proc_pid(inode);
767 task = get_pid_task(pid, PIDTYPE_PID);
768 if (!task)
769 return -ESRCH;
770
771 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
772
773 put_task_struct(task);
774 return ret;
775}
776
777static int proc_single_open(struct inode *inode, struct file *filp)
778{
779 return single_open(filp, proc_single_show, inode);
780}
781
782static const struct file_operations proc_single_file_operations = {
783 .open = proc_single_open,
784 .read = seq_read,
785 .llseek = seq_lseek,
786 .release = single_release,
787};
788
789
790struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
791{
792 struct task_struct *task = get_proc_task(inode);
793 struct mm_struct *mm = ERR_PTR(-ESRCH);
794
795 if (task) {
796 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
797 put_task_struct(task);
798
799 if (!IS_ERR_OR_NULL(mm)) {
800 /* ensure this mm_struct can't be freed */
801 atomic_inc(&mm->mm_count);
802 /* but do not pin its memory */
803 mmput(mm);
804 }
805 }
806
807 return mm;
808}
809
810static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
811{
812 struct mm_struct *mm = proc_mem_open(inode, mode);
813
814 if (IS_ERR(mm))
815 return PTR_ERR(mm);
816
817 file->private_data = mm;
818 return 0;
819}
820
821static int mem_open(struct inode *inode, struct file *file)
822{
823 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
824
825 /* OK to pass negative loff_t, we can catch out-of-range */
826 file->f_mode |= FMODE_UNSIGNED_OFFSET;
827
828 return ret;
829}
830
831static ssize_t mem_rw(struct file *file, char __user *buf,
832 size_t count, loff_t *ppos, int write)
833{
834 struct mm_struct *mm = file->private_data;
835 unsigned long addr = *ppos;
836 ssize_t copied;
837 char *page;
838 unsigned int flags;
839
840 if (!mm)
841 return 0;
842
843 page = (char *)__get_free_page(GFP_TEMPORARY);
844 if (!page)
845 return -ENOMEM;
846
847 copied = 0;
848 if (!atomic_inc_not_zero(&mm->mm_users))
849 goto free;
850
851 /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
852 flags = FOLL_FORCE;
853 if (write)
854 flags |= FOLL_WRITE;
855
856 while (count > 0) {
857 int this_len = min_t(int, count, PAGE_SIZE);
858
859 if (write && copy_from_user(page, buf, this_len)) {
860 copied = -EFAULT;
861 break;
862 }
863
864 this_len = access_remote_vm(mm, addr, page, this_len, flags);
865 if (!this_len) {
866 if (!copied)
867 copied = -EIO;
868 break;
869 }
870
871 if (!write && copy_to_user(buf, page, this_len)) {
872 copied = -EFAULT;
873 break;
874 }
875
876 buf += this_len;
877 addr += this_len;
878 copied += this_len;
879 count -= this_len;
880 }
881 *ppos = addr;
882
883 mmput(mm);
884free:
885 free_page((unsigned long) page);
886 return copied;
887}
888
889static ssize_t mem_read(struct file *file, char __user *buf,
890 size_t count, loff_t *ppos)
891{
892 return mem_rw(file, buf, count, ppos, 0);
893}
894
895static ssize_t mem_write(struct file *file, const char __user *buf,
896 size_t count, loff_t *ppos)
897{
898 return mem_rw(file, (char __user*)buf, count, ppos, 1);
899}
900
901loff_t mem_lseek(struct file *file, loff_t offset, int orig)
902{
903 switch (orig) {
904 case 0:
905 file->f_pos = offset;
906 break;
907 case 1:
908 file->f_pos += offset;
909 break;
910 default:
911 return -EINVAL;
912 }
913 force_successful_syscall_return();
914 return file->f_pos;
915}
916
917static int mem_release(struct inode *inode, struct file *file)
918{
919 struct mm_struct *mm = file->private_data;
920 if (mm)
921 mmdrop(mm);
922 return 0;
923}
924
925static const struct file_operations proc_mem_operations = {
926 .llseek = mem_lseek,
927 .read = mem_read,
928 .write = mem_write,
929 .open = mem_open,
930 .release = mem_release,
931};
932
933static int environ_open(struct inode *inode, struct file *file)
934{
935 return __mem_open(inode, file, PTRACE_MODE_READ);
936}
937
938static ssize_t environ_read(struct file *file, char __user *buf,
939 size_t count, loff_t *ppos)
940{
941 char *page;
942 unsigned long src = *ppos;
943 int ret = 0;
944 struct mm_struct *mm = file->private_data;
945 unsigned long env_start, env_end;
946
947 /* Ensure the process spawned far enough to have an environment. */
948 if (!mm || !mm->env_end)
949 return 0;
950
951 page = (char *)__get_free_page(GFP_TEMPORARY);
952 if (!page)
953 return -ENOMEM;
954
955 ret = 0;
956 if (!atomic_inc_not_zero(&mm->mm_users))
957 goto free;
958
959 down_read(&mm->mmap_sem);
960 env_start = mm->env_start;
961 env_end = mm->env_end;
962 up_read(&mm->mmap_sem);
963
964 while (count > 0) {
965 size_t this_len, max_len;
966 int retval;
967
968 if (src >= (env_end - env_start))
969 break;
970
971 this_len = env_end - (env_start + src);
972
973 max_len = min_t(size_t, PAGE_SIZE, count);
974 this_len = min(max_len, this_len);
975
976 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
977
978 if (retval <= 0) {
979 ret = retval;
980 break;
981 }
982
983 if (copy_to_user(buf, page, retval)) {
984 ret = -EFAULT;
985 break;
986 }
987
988 ret += retval;
989 src += retval;
990 buf += retval;
991 count -= retval;
992 }
993 *ppos = src;
994 mmput(mm);
995
996free:
997 free_page((unsigned long) page);
998 return ret;
999}
1000
1001static const struct file_operations proc_environ_operations = {
1002 .open = environ_open,
1003 .read = environ_read,
1004 .llseek = generic_file_llseek,
1005 .release = mem_release,
1006};
1007
1008static int auxv_open(struct inode *inode, struct file *file)
1009{
1010 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1011}
1012
1013static ssize_t auxv_read(struct file *file, char __user *buf,
1014 size_t count, loff_t *ppos)
1015{
1016 struct mm_struct *mm = file->private_data;
1017 unsigned int nwords = 0;
1018
1019 if (!mm)
1020 return 0;
1021 do {
1022 nwords += 2;
1023 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1024 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1025 nwords * sizeof(mm->saved_auxv[0]));
1026}
1027
1028static const struct file_operations proc_auxv_operations = {
1029 .open = auxv_open,
1030 .read = auxv_read,
1031 .llseek = generic_file_llseek,
1032 .release = mem_release,
1033};
1034
1035static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1036 loff_t *ppos)
1037{
1038 struct task_struct *task = get_proc_task(file_inode(file));
1039 char buffer[PROC_NUMBUF];
1040 int oom_adj = OOM_ADJUST_MIN;
1041 size_t len;
1042
1043 if (!task)
1044 return -ESRCH;
1045 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1046 oom_adj = OOM_ADJUST_MAX;
1047 else
1048 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1049 OOM_SCORE_ADJ_MAX;
1050 put_task_struct(task);
1051 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1052 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1053}
1054
1055static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1056{
1057 static DEFINE_MUTEX(oom_adj_mutex);
1058 struct mm_struct *mm = NULL;
1059 struct task_struct *task;
1060 int err = 0;
1061
1062 task = get_proc_task(file_inode(file));
1063 if (!task)
1064 return -ESRCH;
1065
1066 mutex_lock(&oom_adj_mutex);
1067 if (legacy) {
1068 if (oom_adj < task->signal->oom_score_adj &&
1069 !capable(CAP_SYS_RESOURCE)) {
1070 err = -EACCES;
1071 goto err_unlock;
1072 }
1073 /*
1074 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1075 * /proc/pid/oom_score_adj instead.
1076 */
1077 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1078 current->comm, task_pid_nr(current), task_pid_nr(task),
1079 task_pid_nr(task));
1080 } else {
1081 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1082 !capable(CAP_SYS_RESOURCE)) {
1083 err = -EACCES;
1084 goto err_unlock;
1085 }
1086 }
1087
1088 /*
1089 * Make sure we will check other processes sharing the mm if this is
1090 * not vfrok which wants its own oom_score_adj.
1091 * pin the mm so it doesn't go away and get reused after task_unlock
1092 */
1093 if (!task->vfork_done) {
1094 struct task_struct *p = find_lock_task_mm(task);
1095
1096 if (p) {
1097 if (atomic_read(&p->mm->mm_users) > 1) {
1098 mm = p->mm;
1099 atomic_inc(&mm->mm_count);
1100 }
1101 task_unlock(p);
1102 }
1103 }
1104
1105 task->signal->oom_score_adj = oom_adj;
1106 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1107 task->signal->oom_score_adj_min = (short)oom_adj;
1108 trace_oom_score_adj_update(task);
1109
1110 if (mm) {
1111 struct task_struct *p;
1112
1113 rcu_read_lock();
1114 for_each_process(p) {
1115 if (same_thread_group(task, p))
1116 continue;
1117
1118 /* do not touch kernel threads or the global init */
1119 if (p->flags & PF_KTHREAD || is_global_init(p))
1120 continue;
1121
1122 task_lock(p);
1123 if (!p->vfork_done && process_shares_mm(p, mm)) {
1124 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1125 task_pid_nr(p), p->comm,
1126 p->signal->oom_score_adj, oom_adj,
1127 task_pid_nr(task), task->comm);
1128 p->signal->oom_score_adj = oom_adj;
1129 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1130 p->signal->oom_score_adj_min = (short)oom_adj;
1131 }
1132 task_unlock(p);
1133 }
1134 rcu_read_unlock();
1135 mmdrop(mm);
1136 }
1137err_unlock:
1138 mutex_unlock(&oom_adj_mutex);
1139 put_task_struct(task);
1140 return err;
1141}
1142
1143/*
1144 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1145 * kernels. The effective policy is defined by oom_score_adj, which has a
1146 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1147 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1148 * Processes that become oom disabled via oom_adj will still be oom disabled
1149 * with this implementation.
1150 *
1151 * oom_adj cannot be removed since existing userspace binaries use it.
1152 */
1153static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1154 size_t count, loff_t *ppos)
1155{
1156 char buffer[PROC_NUMBUF];
1157 int oom_adj;
1158 int err;
1159
1160 memset(buffer, 0, sizeof(buffer));
1161 if (count > sizeof(buffer) - 1)
1162 count = sizeof(buffer) - 1;
1163 if (copy_from_user(buffer, buf, count)) {
1164 err = -EFAULT;
1165 goto out;
1166 }
1167
1168 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1169 if (err)
1170 goto out;
1171 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1172 oom_adj != OOM_DISABLE) {
1173 err = -EINVAL;
1174 goto out;
1175 }
1176
1177 /*
1178 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1179 * value is always attainable.
1180 */
1181 if (oom_adj == OOM_ADJUST_MAX)
1182 oom_adj = OOM_SCORE_ADJ_MAX;
1183 else
1184 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1185
1186 err = __set_oom_adj(file, oom_adj, true);
1187out:
1188 return err < 0 ? err : count;
1189}
1190
1191static const struct file_operations proc_oom_adj_operations = {
1192 .read = oom_adj_read,
1193 .write = oom_adj_write,
1194 .llseek = generic_file_llseek,
1195};
1196
1197static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1198 size_t count, loff_t *ppos)
1199{
1200 struct task_struct *task = get_proc_task(file_inode(file));
1201 char buffer[PROC_NUMBUF];
1202 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1203 size_t len;
1204
1205 if (!task)
1206 return -ESRCH;
1207 oom_score_adj = task->signal->oom_score_adj;
1208 put_task_struct(task);
1209 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1210 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1211}
1212
1213static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1214 size_t count, loff_t *ppos)
1215{
1216 char buffer[PROC_NUMBUF];
1217 int oom_score_adj;
1218 int err;
1219
1220 memset(buffer, 0, sizeof(buffer));
1221 if (count > sizeof(buffer) - 1)
1222 count = sizeof(buffer) - 1;
1223 if (copy_from_user(buffer, buf, count)) {
1224 err = -EFAULT;
1225 goto out;
1226 }
1227
1228 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1229 if (err)
1230 goto out;
1231 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1232 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1233 err = -EINVAL;
1234 goto out;
1235 }
1236
1237 err = __set_oom_adj(file, oom_score_adj, false);
1238out:
1239 return err < 0 ? err : count;
1240}
1241
1242static const struct file_operations proc_oom_score_adj_operations = {
1243 .read = oom_score_adj_read,
1244 .write = oom_score_adj_write,
1245 .llseek = default_llseek,
1246};
1247
1248#ifdef CONFIG_AUDITSYSCALL
1249#define TMPBUFLEN 11
1250static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1251 size_t count, loff_t *ppos)
1252{
1253 struct inode * inode = file_inode(file);
1254 struct task_struct *task = get_proc_task(inode);
1255 ssize_t length;
1256 char tmpbuf[TMPBUFLEN];
1257
1258 if (!task)
1259 return -ESRCH;
1260 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1261 from_kuid(file->f_cred->user_ns,
1262 audit_get_loginuid(task)));
1263 put_task_struct(task);
1264 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1265}
1266
1267static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1268 size_t count, loff_t *ppos)
1269{
1270 struct inode * inode = file_inode(file);
1271 uid_t loginuid;
1272 kuid_t kloginuid;
1273 int rv;
1274
1275 rcu_read_lock();
1276 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1277 rcu_read_unlock();
1278 return -EPERM;
1279 }
1280 rcu_read_unlock();
1281
1282 if (*ppos != 0) {
1283 /* No partial writes. */
1284 return -EINVAL;
1285 }
1286
1287 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1288 if (rv < 0)
1289 return rv;
1290
1291 /* is userspace tring to explicitly UNSET the loginuid? */
1292 if (loginuid == AUDIT_UID_UNSET) {
1293 kloginuid = INVALID_UID;
1294 } else {
1295 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1296 if (!uid_valid(kloginuid))
1297 return -EINVAL;
1298 }
1299
1300 rv = audit_set_loginuid(kloginuid);
1301 if (rv < 0)
1302 return rv;
1303 return count;
1304}
1305
1306static const struct file_operations proc_loginuid_operations = {
1307 .read = proc_loginuid_read,
1308 .write = proc_loginuid_write,
1309 .llseek = generic_file_llseek,
1310};
1311
1312static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1313 size_t count, loff_t *ppos)
1314{
1315 struct inode * inode = file_inode(file);
1316 struct task_struct *task = get_proc_task(inode);
1317 ssize_t length;
1318 char tmpbuf[TMPBUFLEN];
1319
1320 if (!task)
1321 return -ESRCH;
1322 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1323 audit_get_sessionid(task));
1324 put_task_struct(task);
1325 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1326}
1327
1328static const struct file_operations proc_sessionid_operations = {
1329 .read = proc_sessionid_read,
1330 .llseek = generic_file_llseek,
1331};
1332#endif
1333
1334#ifdef CONFIG_FAULT_INJECTION
1335static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1336 size_t count, loff_t *ppos)
1337{
1338 struct task_struct *task = get_proc_task(file_inode(file));
1339 char buffer[PROC_NUMBUF];
1340 size_t len;
1341 int make_it_fail;
1342
1343 if (!task)
1344 return -ESRCH;
1345 make_it_fail = task->make_it_fail;
1346 put_task_struct(task);
1347
1348 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1349
1350 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1351}
1352
1353static ssize_t proc_fault_inject_write(struct file * file,
1354 const char __user * buf, size_t count, loff_t *ppos)
1355{
1356 struct task_struct *task;
1357 char buffer[PROC_NUMBUF];
1358 int make_it_fail;
1359 int rv;
1360
1361 if (!capable(CAP_SYS_RESOURCE))
1362 return -EPERM;
1363 memset(buffer, 0, sizeof(buffer));
1364 if (count > sizeof(buffer) - 1)
1365 count = sizeof(buffer) - 1;
1366 if (copy_from_user(buffer, buf, count))
1367 return -EFAULT;
1368 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1369 if (rv < 0)
1370 return rv;
1371 if (make_it_fail < 0 || make_it_fail > 1)
1372 return -EINVAL;
1373
1374 task = get_proc_task(file_inode(file));
1375 if (!task)
1376 return -ESRCH;
1377 task->make_it_fail = make_it_fail;
1378 put_task_struct(task);
1379
1380 return count;
1381}
1382
1383static const struct file_operations proc_fault_inject_operations = {
1384 .read = proc_fault_inject_read,
1385 .write = proc_fault_inject_write,
1386 .llseek = generic_file_llseek,
1387};
1388#endif
1389
1390
1391#ifdef CONFIG_SCHED_DEBUG
1392/*
1393 * Print out various scheduling related per-task fields:
1394 */
1395static int sched_show(struct seq_file *m, void *v)
1396{
1397 struct inode *inode = m->private;
1398 struct task_struct *p;
1399
1400 p = get_proc_task(inode);
1401 if (!p)
1402 return -ESRCH;
1403 proc_sched_show_task(p, m);
1404
1405 put_task_struct(p);
1406
1407 return 0;
1408}
1409
1410static ssize_t
1411sched_write(struct file *file, const char __user *buf,
1412 size_t count, loff_t *offset)
1413{
1414 struct inode *inode = file_inode(file);
1415 struct task_struct *p;
1416
1417 p = get_proc_task(inode);
1418 if (!p)
1419 return -ESRCH;
1420 proc_sched_set_task(p);
1421
1422 put_task_struct(p);
1423
1424 return count;
1425}
1426
1427static int sched_open(struct inode *inode, struct file *filp)
1428{
1429 return single_open(filp, sched_show, inode);
1430}
1431
1432static const struct file_operations proc_pid_sched_operations = {
1433 .open = sched_open,
1434 .read = seq_read,
1435 .write = sched_write,
1436 .llseek = seq_lseek,
1437 .release = single_release,
1438};
1439
1440#endif
1441
1442#ifdef CONFIG_SCHED_AUTOGROUP
1443/*
1444 * Print out autogroup related information:
1445 */
1446static int sched_autogroup_show(struct seq_file *m, void *v)
1447{
1448 struct inode *inode = m->private;
1449 struct task_struct *p;
1450
1451 p = get_proc_task(inode);
1452 if (!p)
1453 return -ESRCH;
1454 proc_sched_autogroup_show_task(p, m);
1455
1456 put_task_struct(p);
1457
1458 return 0;
1459}
1460
1461static ssize_t
1462sched_autogroup_write(struct file *file, const char __user *buf,
1463 size_t count, loff_t *offset)
1464{
1465 struct inode *inode = file_inode(file);
1466 struct task_struct *p;
1467 char buffer[PROC_NUMBUF];
1468 int nice;
1469 int err;
1470
1471 memset(buffer, 0, sizeof(buffer));
1472 if (count > sizeof(buffer) - 1)
1473 count = sizeof(buffer) - 1;
1474 if (copy_from_user(buffer, buf, count))
1475 return -EFAULT;
1476
1477 err = kstrtoint(strstrip(buffer), 0, &nice);
1478 if (err < 0)
1479 return err;
1480
1481 p = get_proc_task(inode);
1482 if (!p)
1483 return -ESRCH;
1484
1485 err = proc_sched_autogroup_set_nice(p, nice);
1486 if (err)
1487 count = err;
1488
1489 put_task_struct(p);
1490
1491 return count;
1492}
1493
1494static int sched_autogroup_open(struct inode *inode, struct file *filp)
1495{
1496 int ret;
1497
1498 ret = single_open(filp, sched_autogroup_show, NULL);
1499 if (!ret) {
1500 struct seq_file *m = filp->private_data;
1501
1502 m->private = inode;
1503 }
1504 return ret;
1505}
1506
1507static const struct file_operations proc_pid_sched_autogroup_operations = {
1508 .open = sched_autogroup_open,
1509 .read = seq_read,
1510 .write = sched_autogroup_write,
1511 .llseek = seq_lseek,
1512 .release = single_release,
1513};
1514
1515#endif /* CONFIG_SCHED_AUTOGROUP */
1516
1517static ssize_t comm_write(struct file *file, const char __user *buf,
1518 size_t count, loff_t *offset)
1519{
1520 struct inode *inode = file_inode(file);
1521 struct task_struct *p;
1522 char buffer[TASK_COMM_LEN];
1523 const size_t maxlen = sizeof(buffer) - 1;
1524
1525 memset(buffer, 0, sizeof(buffer));
1526 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1527 return -EFAULT;
1528
1529 p = get_proc_task(inode);
1530 if (!p)
1531 return -ESRCH;
1532
1533 if (same_thread_group(current, p))
1534 set_task_comm(p, buffer);
1535 else
1536 count = -EINVAL;
1537
1538 put_task_struct(p);
1539
1540 return count;
1541}
1542
1543static int comm_show(struct seq_file *m, void *v)
1544{
1545 struct inode *inode = m->private;
1546 struct task_struct *p;
1547
1548 p = get_proc_task(inode);
1549 if (!p)
1550 return -ESRCH;
1551
1552 task_lock(p);
1553 seq_printf(m, "%s\n", p->comm);
1554 task_unlock(p);
1555
1556 put_task_struct(p);
1557
1558 return 0;
1559}
1560
1561static int comm_open(struct inode *inode, struct file *filp)
1562{
1563 return single_open(filp, comm_show, inode);
1564}
1565
1566static const struct file_operations proc_pid_set_comm_operations = {
1567 .open = comm_open,
1568 .read = seq_read,
1569 .write = comm_write,
1570 .llseek = seq_lseek,
1571 .release = single_release,
1572};
1573
1574static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1575{
1576 struct task_struct *task;
1577 struct file *exe_file;
1578
1579 task = get_proc_task(d_inode(dentry));
1580 if (!task)
1581 return -ENOENT;
1582 exe_file = get_task_exe_file(task);
1583 put_task_struct(task);
1584 if (exe_file) {
1585 *exe_path = exe_file->f_path;
1586 path_get(&exe_file->f_path);
1587 fput(exe_file);
1588 return 0;
1589 } else
1590 return -ENOENT;
1591}
1592
1593static const char *proc_pid_get_link(struct dentry *dentry,
1594 struct inode *inode,
1595 struct delayed_call *done)
1596{
1597 struct path path;
1598 int error = -EACCES;
1599
1600 if (!dentry)
1601 return ERR_PTR(-ECHILD);
1602
1603 /* Are we allowed to snoop on the tasks file descriptors? */
1604 if (!proc_fd_access_allowed(inode))
1605 goto out;
1606
1607 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1608 if (error)
1609 goto out;
1610
1611 nd_jump_link(&path);
1612 return NULL;
1613out:
1614 return ERR_PTR(error);
1615}
1616
1617static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1618{
1619 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1620 char *pathname;
1621 int len;
1622
1623 if (!tmp)
1624 return -ENOMEM;
1625
1626 pathname = d_path(path, tmp, PAGE_SIZE);
1627 len = PTR_ERR(pathname);
1628 if (IS_ERR(pathname))
1629 goto out;
1630 len = tmp + PAGE_SIZE - 1 - pathname;
1631
1632 if (len > buflen)
1633 len = buflen;
1634 if (copy_to_user(buffer, pathname, len))
1635 len = -EFAULT;
1636 out:
1637 free_page((unsigned long)tmp);
1638 return len;
1639}
1640
1641static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1642{
1643 int error = -EACCES;
1644 struct inode *inode = d_inode(dentry);
1645 struct path path;
1646
1647 /* Are we allowed to snoop on the tasks file descriptors? */
1648 if (!proc_fd_access_allowed(inode))
1649 goto out;
1650
1651 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1652 if (error)
1653 goto out;
1654
1655 error = do_proc_readlink(&path, buffer, buflen);
1656 path_put(&path);
1657out:
1658 return error;
1659}
1660
1661const struct inode_operations proc_pid_link_inode_operations = {
1662 .readlink = proc_pid_readlink,
1663 .get_link = proc_pid_get_link,
1664 .setattr = proc_setattr,
1665};
1666
1667
1668/* building an inode */
1669
1670struct inode *proc_pid_make_inode(struct super_block * sb,
1671 struct task_struct *task, umode_t mode)
1672{
1673 struct inode * inode;
1674 struct proc_inode *ei;
1675 const struct cred *cred;
1676
1677 /* We need a new inode */
1678
1679 inode = new_inode(sb);
1680 if (!inode)
1681 goto out;
1682
1683 /* Common stuff */
1684 ei = PROC_I(inode);
1685 inode->i_mode = mode;
1686 inode->i_ino = get_next_ino();
1687 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1688 inode->i_op = &proc_def_inode_operations;
1689
1690 /*
1691 * grab the reference to task.
1692 */
1693 ei->pid = get_task_pid(task, PIDTYPE_PID);
1694 if (!ei->pid)
1695 goto out_unlock;
1696
1697 if (task_dumpable(task)) {
1698 rcu_read_lock();
1699 cred = __task_cred(task);
1700 inode->i_uid = cred->euid;
1701 inode->i_gid = cred->egid;
1702 rcu_read_unlock();
1703 }
1704 security_task_to_inode(task, inode);
1705
1706out:
1707 return inode;
1708
1709out_unlock:
1710 iput(inode);
1711 return NULL;
1712}
1713
1714int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1715{
1716 struct inode *inode = d_inode(dentry);
1717 struct task_struct *task;
1718 const struct cred *cred;
1719 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1720
1721 generic_fillattr(inode, stat);
1722
1723 rcu_read_lock();
1724 stat->uid = GLOBAL_ROOT_UID;
1725 stat->gid = GLOBAL_ROOT_GID;
1726 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1727 if (task) {
1728 if (!has_pid_permissions(pid, task, 2)) {
1729 rcu_read_unlock();
1730 /*
1731 * This doesn't prevent learning whether PID exists,
1732 * it only makes getattr() consistent with readdir().
1733 */
1734 return -ENOENT;
1735 }
1736 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1737 task_dumpable(task)) {
1738 cred = __task_cred(task);
1739 stat->uid = cred->euid;
1740 stat->gid = cred->egid;
1741 }
1742 }
1743 rcu_read_unlock();
1744 return 0;
1745}
1746
1747/* dentry stuff */
1748
1749/*
1750 * Exceptional case: normally we are not allowed to unhash a busy
1751 * directory. In this case, however, we can do it - no aliasing problems
1752 * due to the way we treat inodes.
1753 *
1754 * Rewrite the inode's ownerships here because the owning task may have
1755 * performed a setuid(), etc.
1756 *
1757 * Before the /proc/pid/status file was created the only way to read
1758 * the effective uid of a /process was to stat /proc/pid. Reading
1759 * /proc/pid/status is slow enough that procps and other packages
1760 * kept stating /proc/pid. To keep the rules in /proc simple I have
1761 * made this apply to all per process world readable and executable
1762 * directories.
1763 */
1764int pid_revalidate(struct dentry *dentry, unsigned int flags)
1765{
1766 struct inode *inode;
1767 struct task_struct *task;
1768 const struct cred *cred;
1769
1770 if (flags & LOOKUP_RCU)
1771 return -ECHILD;
1772
1773 inode = d_inode(dentry);
1774 task = get_proc_task(inode);
1775
1776 if (task) {
1777 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1778 task_dumpable(task)) {
1779 rcu_read_lock();
1780 cred = __task_cred(task);
1781 inode->i_uid = cred->euid;
1782 inode->i_gid = cred->egid;
1783 rcu_read_unlock();
1784 } else {
1785 inode->i_uid = GLOBAL_ROOT_UID;
1786 inode->i_gid = GLOBAL_ROOT_GID;
1787 }
1788 inode->i_mode &= ~(S_ISUID | S_ISGID);
1789 security_task_to_inode(task, inode);
1790 put_task_struct(task);
1791 return 1;
1792 }
1793 return 0;
1794}
1795
1796static inline bool proc_inode_is_dead(struct inode *inode)
1797{
1798 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1799}
1800
1801int pid_delete_dentry(const struct dentry *dentry)
1802{
1803 /* Is the task we represent dead?
1804 * If so, then don't put the dentry on the lru list,
1805 * kill it immediately.
1806 */
1807 return proc_inode_is_dead(d_inode(dentry));
1808}
1809
1810const struct dentry_operations pid_dentry_operations =
1811{
1812 .d_revalidate = pid_revalidate,
1813 .d_delete = pid_delete_dentry,
1814};
1815
1816/* Lookups */
1817
1818/*
1819 * Fill a directory entry.
1820 *
1821 * If possible create the dcache entry and derive our inode number and
1822 * file type from dcache entry.
1823 *
1824 * Since all of the proc inode numbers are dynamically generated, the inode
1825 * numbers do not exist until the inode is cache. This means creating the
1826 * the dcache entry in readdir is necessary to keep the inode numbers
1827 * reported by readdir in sync with the inode numbers reported
1828 * by stat.
1829 */
1830bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1831 const char *name, int len,
1832 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1833{
1834 struct dentry *child, *dir = file->f_path.dentry;
1835 struct qstr qname = QSTR_INIT(name, len);
1836 struct inode *inode;
1837 unsigned type;
1838 ino_t ino;
1839
1840 child = d_hash_and_lookup(dir, &qname);
1841 if (!child) {
1842 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1843 child = d_alloc_parallel(dir, &qname, &wq);
1844 if (IS_ERR(child))
1845 goto end_instantiate;
1846 if (d_in_lookup(child)) {
1847 int err = instantiate(d_inode(dir), child, task, ptr);
1848 d_lookup_done(child);
1849 if (err < 0) {
1850 dput(child);
1851 goto end_instantiate;
1852 }
1853 }
1854 }
1855 inode = d_inode(child);
1856 ino = inode->i_ino;
1857 type = inode->i_mode >> 12;
1858 dput(child);
1859 return dir_emit(ctx, name, len, ino, type);
1860
1861end_instantiate:
1862 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1863}
1864
1865/*
1866 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1867 * which represent vma start and end addresses.
1868 */
1869static int dname_to_vma_addr(struct dentry *dentry,
1870 unsigned long *start, unsigned long *end)
1871{
1872 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1873 return -EINVAL;
1874
1875 return 0;
1876}
1877
1878static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1879{
1880 unsigned long vm_start, vm_end;
1881 bool exact_vma_exists = false;
1882 struct mm_struct *mm = NULL;
1883 struct task_struct *task;
1884 const struct cred *cred;
1885 struct inode *inode;
1886 int status = 0;
1887
1888 if (flags & LOOKUP_RCU)
1889 return -ECHILD;
1890
1891 inode = d_inode(dentry);
1892 task = get_proc_task(inode);
1893 if (!task)
1894 goto out_notask;
1895
1896 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1897 if (IS_ERR_OR_NULL(mm))
1898 goto out;
1899
1900 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1901 down_read(&mm->mmap_sem);
1902 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1903 up_read(&mm->mmap_sem);
1904 }
1905
1906 mmput(mm);
1907
1908 if (exact_vma_exists) {
1909 if (task_dumpable(task)) {
1910 rcu_read_lock();
1911 cred = __task_cred(task);
1912 inode->i_uid = cred->euid;
1913 inode->i_gid = cred->egid;
1914 rcu_read_unlock();
1915 } else {
1916 inode->i_uid = GLOBAL_ROOT_UID;
1917 inode->i_gid = GLOBAL_ROOT_GID;
1918 }
1919 security_task_to_inode(task, inode);
1920 status = 1;
1921 }
1922
1923out:
1924 put_task_struct(task);
1925
1926out_notask:
1927 return status;
1928}
1929
1930static const struct dentry_operations tid_map_files_dentry_operations = {
1931 .d_revalidate = map_files_d_revalidate,
1932 .d_delete = pid_delete_dentry,
1933};
1934
1935static int map_files_get_link(struct dentry *dentry, struct path *path)
1936{
1937 unsigned long vm_start, vm_end;
1938 struct vm_area_struct *vma;
1939 struct task_struct *task;
1940 struct mm_struct *mm;
1941 int rc;
1942
1943 rc = -ENOENT;
1944 task = get_proc_task(d_inode(dentry));
1945 if (!task)
1946 goto out;
1947
1948 mm = get_task_mm(task);
1949 put_task_struct(task);
1950 if (!mm)
1951 goto out;
1952
1953 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1954 if (rc)
1955 goto out_mmput;
1956
1957 rc = -ENOENT;
1958 down_read(&mm->mmap_sem);
1959 vma = find_exact_vma(mm, vm_start, vm_end);
1960 if (vma && vma->vm_file) {
1961 *path = vma->vm_file->f_path;
1962 path_get(path);
1963 rc = 0;
1964 }
1965 up_read(&mm->mmap_sem);
1966
1967out_mmput:
1968 mmput(mm);
1969out:
1970 return rc;
1971}
1972
1973struct map_files_info {
1974 fmode_t mode;
1975 unsigned int len;
1976 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1977};
1978
1979/*
1980 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1981 * symlinks may be used to bypass permissions on ancestor directories in the
1982 * path to the file in question.
1983 */
1984static const char *
1985proc_map_files_get_link(struct dentry *dentry,
1986 struct inode *inode,
1987 struct delayed_call *done)
1988{
1989 if (!capable(CAP_SYS_ADMIN))
1990 return ERR_PTR(-EPERM);
1991
1992 return proc_pid_get_link(dentry, inode, done);
1993}
1994
1995/*
1996 * Identical to proc_pid_link_inode_operations except for get_link()
1997 */
1998static const struct inode_operations proc_map_files_link_inode_operations = {
1999 .readlink = proc_pid_readlink,
2000 .get_link = proc_map_files_get_link,
2001 .setattr = proc_setattr,
2002};
2003
2004static int
2005proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2006 struct task_struct *task, const void *ptr)
2007{
2008 fmode_t mode = (fmode_t)(unsigned long)ptr;
2009 struct proc_inode *ei;
2010 struct inode *inode;
2011
2012 inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2013 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2014 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2015 if (!inode)
2016 return -ENOENT;
2017
2018 ei = PROC_I(inode);
2019 ei->op.proc_get_link = map_files_get_link;
2020
2021 inode->i_op = &proc_map_files_link_inode_operations;
2022 inode->i_size = 64;
2023
2024 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2025 d_add(dentry, inode);
2026
2027 return 0;
2028}
2029
2030static struct dentry *proc_map_files_lookup(struct inode *dir,
2031 struct dentry *dentry, unsigned int flags)
2032{
2033 unsigned long vm_start, vm_end;
2034 struct vm_area_struct *vma;
2035 struct task_struct *task;
2036 int result;
2037 struct mm_struct *mm;
2038
2039 result = -ENOENT;
2040 task = get_proc_task(dir);
2041 if (!task)
2042 goto out;
2043
2044 result = -EACCES;
2045 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2046 goto out_put_task;
2047
2048 result = -ENOENT;
2049 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2050 goto out_put_task;
2051
2052 mm = get_task_mm(task);
2053 if (!mm)
2054 goto out_put_task;
2055
2056 down_read(&mm->mmap_sem);
2057 vma = find_exact_vma(mm, vm_start, vm_end);
2058 if (!vma)
2059 goto out_no_vma;
2060
2061 if (vma->vm_file)
2062 result = proc_map_files_instantiate(dir, dentry, task,
2063 (void *)(unsigned long)vma->vm_file->f_mode);
2064
2065out_no_vma:
2066 up_read(&mm->mmap_sem);
2067 mmput(mm);
2068out_put_task:
2069 put_task_struct(task);
2070out:
2071 return ERR_PTR(result);
2072}
2073
2074static const struct inode_operations proc_map_files_inode_operations = {
2075 .lookup = proc_map_files_lookup,
2076 .permission = proc_fd_permission,
2077 .setattr = proc_setattr,
2078};
2079
2080static int
2081proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2082{
2083 struct vm_area_struct *vma;
2084 struct task_struct *task;
2085 struct mm_struct *mm;
2086 unsigned long nr_files, pos, i;
2087 struct flex_array *fa = NULL;
2088 struct map_files_info info;
2089 struct map_files_info *p;
2090 int ret;
2091
2092 ret = -ENOENT;
2093 task = get_proc_task(file_inode(file));
2094 if (!task)
2095 goto out;
2096
2097 ret = -EACCES;
2098 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2099 goto out_put_task;
2100
2101 ret = 0;
2102 if (!dir_emit_dots(file, ctx))
2103 goto out_put_task;
2104
2105 mm = get_task_mm(task);
2106 if (!mm)
2107 goto out_put_task;
2108 down_read(&mm->mmap_sem);
2109
2110 nr_files = 0;
2111
2112 /*
2113 * We need two passes here:
2114 *
2115 * 1) Collect vmas of mapped files with mmap_sem taken
2116 * 2) Release mmap_sem and instantiate entries
2117 *
2118 * otherwise we get lockdep complained, since filldir()
2119 * routine might require mmap_sem taken in might_fault().
2120 */
2121
2122 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2123 if (vma->vm_file && ++pos > ctx->pos)
2124 nr_files++;
2125 }
2126
2127 if (nr_files) {
2128 fa = flex_array_alloc(sizeof(info), nr_files,
2129 GFP_KERNEL);
2130 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2131 GFP_KERNEL)) {
2132 ret = -ENOMEM;
2133 if (fa)
2134 flex_array_free(fa);
2135 up_read(&mm->mmap_sem);
2136 mmput(mm);
2137 goto out_put_task;
2138 }
2139 for (i = 0, vma = mm->mmap, pos = 2; vma;
2140 vma = vma->vm_next) {
2141 if (!vma->vm_file)
2142 continue;
2143 if (++pos <= ctx->pos)
2144 continue;
2145
2146 info.mode = vma->vm_file->f_mode;
2147 info.len = snprintf(info.name,
2148 sizeof(info.name), "%lx-%lx",
2149 vma->vm_start, vma->vm_end);
2150 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2151 BUG();
2152 }
2153 }
2154 up_read(&mm->mmap_sem);
2155
2156 for (i = 0; i < nr_files; i++) {
2157 p = flex_array_get(fa, i);
2158 if (!proc_fill_cache(file, ctx,
2159 p->name, p->len,
2160 proc_map_files_instantiate,
2161 task,
2162 (void *)(unsigned long)p->mode))
2163 break;
2164 ctx->pos++;
2165 }
2166 if (fa)
2167 flex_array_free(fa);
2168 mmput(mm);
2169
2170out_put_task:
2171 put_task_struct(task);
2172out:
2173 return ret;
2174}
2175
2176static const struct file_operations proc_map_files_operations = {
2177 .read = generic_read_dir,
2178 .iterate_shared = proc_map_files_readdir,
2179 .llseek = generic_file_llseek,
2180};
2181
2182#ifdef CONFIG_CHECKPOINT_RESTORE
2183struct timers_private {
2184 struct pid *pid;
2185 struct task_struct *task;
2186 struct sighand_struct *sighand;
2187 struct pid_namespace *ns;
2188 unsigned long flags;
2189};
2190
2191static void *timers_start(struct seq_file *m, loff_t *pos)
2192{
2193 struct timers_private *tp = m->private;
2194
2195 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2196 if (!tp->task)
2197 return ERR_PTR(-ESRCH);
2198
2199 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2200 if (!tp->sighand)
2201 return ERR_PTR(-ESRCH);
2202
2203 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2204}
2205
2206static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2207{
2208 struct timers_private *tp = m->private;
2209 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2210}
2211
2212static void timers_stop(struct seq_file *m, void *v)
2213{
2214 struct timers_private *tp = m->private;
2215
2216 if (tp->sighand) {
2217 unlock_task_sighand(tp->task, &tp->flags);
2218 tp->sighand = NULL;
2219 }
2220
2221 if (tp->task) {
2222 put_task_struct(tp->task);
2223 tp->task = NULL;
2224 }
2225}
2226
2227static int show_timer(struct seq_file *m, void *v)
2228{
2229 struct k_itimer *timer;
2230 struct timers_private *tp = m->private;
2231 int notify;
2232 static const char * const nstr[] = {
2233 [SIGEV_SIGNAL] = "signal",
2234 [SIGEV_NONE] = "none",
2235 [SIGEV_THREAD] = "thread",
2236 };
2237
2238 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2239 notify = timer->it_sigev_notify;
2240
2241 seq_printf(m, "ID: %d\n", timer->it_id);
2242 seq_printf(m, "signal: %d/%p\n",
2243 timer->sigq->info.si_signo,
2244 timer->sigq->info.si_value.sival_ptr);
2245 seq_printf(m, "notify: %s/%s.%d\n",
2246 nstr[notify & ~SIGEV_THREAD_ID],
2247 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2248 pid_nr_ns(timer->it_pid, tp->ns));
2249 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2250
2251 return 0;
2252}
2253
2254static const struct seq_operations proc_timers_seq_ops = {
2255 .start = timers_start,
2256 .next = timers_next,
2257 .stop = timers_stop,
2258 .show = show_timer,
2259};
2260
2261static int proc_timers_open(struct inode *inode, struct file *file)
2262{
2263 struct timers_private *tp;
2264
2265 tp = __seq_open_private(file, &proc_timers_seq_ops,
2266 sizeof(struct timers_private));
2267 if (!tp)
2268 return -ENOMEM;
2269
2270 tp->pid = proc_pid(inode);
2271 tp->ns = inode->i_sb->s_fs_info;
2272 return 0;
2273}
2274
2275static const struct file_operations proc_timers_operations = {
2276 .open = proc_timers_open,
2277 .read = seq_read,
2278 .llseek = seq_lseek,
2279 .release = seq_release_private,
2280};
2281#endif
2282
2283static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2284 size_t count, loff_t *offset)
2285{
2286 struct inode *inode = file_inode(file);
2287 struct task_struct *p;
2288 u64 slack_ns;
2289 int err;
2290
2291 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2292 if (err < 0)
2293 return err;
2294
2295 p = get_proc_task(inode);
2296 if (!p)
2297 return -ESRCH;
2298
2299 if (p != current) {
2300 if (!capable(CAP_SYS_NICE)) {
2301 count = -EPERM;
2302 goto out;
2303 }
2304
2305 err = security_task_setscheduler(p);
2306 if (err) {
2307 count = err;
2308 goto out;
2309 }
2310 }
2311
2312 task_lock(p);
2313 if (slack_ns == 0)
2314 p->timer_slack_ns = p->default_timer_slack_ns;
2315 else
2316 p->timer_slack_ns = slack_ns;
2317 task_unlock(p);
2318
2319out:
2320 put_task_struct(p);
2321
2322 return count;
2323}
2324
2325static int timerslack_ns_show(struct seq_file *m, void *v)
2326{
2327 struct inode *inode = m->private;
2328 struct task_struct *p;
2329 int err = 0;
2330
2331 p = get_proc_task(inode);
2332 if (!p)
2333 return -ESRCH;
2334
2335 if (p != current) {
2336
2337 if (!capable(CAP_SYS_NICE)) {
2338 err = -EPERM;
2339 goto out;
2340 }
2341 err = security_task_getscheduler(p);
2342 if (err)
2343 goto out;
2344 }
2345
2346 task_lock(p);
2347 seq_printf(m, "%llu\n", p->timer_slack_ns);
2348 task_unlock(p);
2349
2350out:
2351 put_task_struct(p);
2352
2353 return err;
2354}
2355
2356static int timerslack_ns_open(struct inode *inode, struct file *filp)
2357{
2358 return single_open(filp, timerslack_ns_show, inode);
2359}
2360
2361static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2362 .open = timerslack_ns_open,
2363 .read = seq_read,
2364 .write = timerslack_ns_write,
2365 .llseek = seq_lseek,
2366 .release = single_release,
2367};
2368
2369static int proc_pident_instantiate(struct inode *dir,
2370 struct dentry *dentry, struct task_struct *task, const void *ptr)
2371{
2372 const struct pid_entry *p = ptr;
2373 struct inode *inode;
2374 struct proc_inode *ei;
2375
2376 inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2377 if (!inode)
2378 goto out;
2379
2380 ei = PROC_I(inode);
2381 if (S_ISDIR(inode->i_mode))
2382 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2383 if (p->iop)
2384 inode->i_op = p->iop;
2385 if (p->fop)
2386 inode->i_fop = p->fop;
2387 ei->op = p->op;
2388 d_set_d_op(dentry, &pid_dentry_operations);
2389 d_add(dentry, inode);
2390 /* Close the race of the process dying before we return the dentry */
2391 if (pid_revalidate(dentry, 0))
2392 return 0;
2393out:
2394 return -ENOENT;
2395}
2396
2397static struct dentry *proc_pident_lookup(struct inode *dir,
2398 struct dentry *dentry,
2399 const struct pid_entry *ents,
2400 unsigned int nents)
2401{
2402 int error;
2403 struct task_struct *task = get_proc_task(dir);
2404 const struct pid_entry *p, *last;
2405
2406 error = -ENOENT;
2407
2408 if (!task)
2409 goto out_no_task;
2410
2411 /*
2412 * Yes, it does not scale. And it should not. Don't add
2413 * new entries into /proc/<tgid>/ without very good reasons.
2414 */
2415 last = &ents[nents];
2416 for (p = ents; p < last; p++) {
2417 if (p->len != dentry->d_name.len)
2418 continue;
2419 if (!memcmp(dentry->d_name.name, p->name, p->len))
2420 break;
2421 }
2422 if (p >= last)
2423 goto out;
2424
2425 error = proc_pident_instantiate(dir, dentry, task, p);
2426out:
2427 put_task_struct(task);
2428out_no_task:
2429 return ERR_PTR(error);
2430}
2431
2432static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2433 const struct pid_entry *ents, unsigned int nents)
2434{
2435 struct task_struct *task = get_proc_task(file_inode(file));
2436 const struct pid_entry *p;
2437
2438 if (!task)
2439 return -ENOENT;
2440
2441 if (!dir_emit_dots(file, ctx))
2442 goto out;
2443
2444 if (ctx->pos >= nents + 2)
2445 goto out;
2446
2447 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2448 if (!proc_fill_cache(file, ctx, p->name, p->len,
2449 proc_pident_instantiate, task, p))
2450 break;
2451 ctx->pos++;
2452 }
2453out:
2454 put_task_struct(task);
2455 return 0;
2456}
2457
2458#ifdef CONFIG_SECURITY
2459static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2460 size_t count, loff_t *ppos)
2461{
2462 struct inode * inode = file_inode(file);
2463 char *p = NULL;
2464 ssize_t length;
2465 struct task_struct *task = get_proc_task(inode);
2466
2467 if (!task)
2468 return -ESRCH;
2469
2470 length = security_getprocattr(task,
2471 (char*)file->f_path.dentry->d_name.name,
2472 &p);
2473 put_task_struct(task);
2474 if (length > 0)
2475 length = simple_read_from_buffer(buf, count, ppos, p, length);
2476 kfree(p);
2477 return length;
2478}
2479
2480static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2481 size_t count, loff_t *ppos)
2482{
2483 struct inode * inode = file_inode(file);
2484 void *page;
2485 ssize_t length;
2486 struct task_struct *task = get_proc_task(inode);
2487
2488 length = -ESRCH;
2489 if (!task)
2490 goto out_no_task;
2491 if (count > PAGE_SIZE)
2492 count = PAGE_SIZE;
2493
2494 /* No partial writes. */
2495 length = -EINVAL;
2496 if (*ppos != 0)
2497 goto out;
2498
2499 page = memdup_user(buf, count);
2500 if (IS_ERR(page)) {
2501 length = PTR_ERR(page);
2502 goto out;
2503 }
2504
2505 /* Guard against adverse ptrace interaction */
2506 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2507 if (length < 0)
2508 goto out_free;
2509
2510 length = security_setprocattr(task,
2511 (char*)file->f_path.dentry->d_name.name,
2512 page, count);
2513 mutex_unlock(&task->signal->cred_guard_mutex);
2514out_free:
2515 kfree(page);
2516out:
2517 put_task_struct(task);
2518out_no_task:
2519 return length;
2520}
2521
2522static const struct file_operations proc_pid_attr_operations = {
2523 .read = proc_pid_attr_read,
2524 .write = proc_pid_attr_write,
2525 .llseek = generic_file_llseek,
2526};
2527
2528static const struct pid_entry attr_dir_stuff[] = {
2529 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2530 REG("prev", S_IRUGO, proc_pid_attr_operations),
2531 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2532 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2533 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2534 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2535};
2536
2537static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2538{
2539 return proc_pident_readdir(file, ctx,
2540 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2541}
2542
2543static const struct file_operations proc_attr_dir_operations = {
2544 .read = generic_read_dir,
2545 .iterate_shared = proc_attr_dir_readdir,
2546 .llseek = generic_file_llseek,
2547};
2548
2549static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2550 struct dentry *dentry, unsigned int flags)
2551{
2552 return proc_pident_lookup(dir, dentry,
2553 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2554}
2555
2556static const struct inode_operations proc_attr_dir_inode_operations = {
2557 .lookup = proc_attr_dir_lookup,
2558 .getattr = pid_getattr,
2559 .setattr = proc_setattr,
2560};
2561
2562#endif
2563
2564#ifdef CONFIG_ELF_CORE
2565static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2566 size_t count, loff_t *ppos)
2567{
2568 struct task_struct *task = get_proc_task(file_inode(file));
2569 struct mm_struct *mm;
2570 char buffer[PROC_NUMBUF];
2571 size_t len;
2572 int ret;
2573
2574 if (!task)
2575 return -ESRCH;
2576
2577 ret = 0;
2578 mm = get_task_mm(task);
2579 if (mm) {
2580 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2581 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2582 MMF_DUMP_FILTER_SHIFT));
2583 mmput(mm);
2584 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2585 }
2586
2587 put_task_struct(task);
2588
2589 return ret;
2590}
2591
2592static ssize_t proc_coredump_filter_write(struct file *file,
2593 const char __user *buf,
2594 size_t count,
2595 loff_t *ppos)
2596{
2597 struct task_struct *task;
2598 struct mm_struct *mm;
2599 unsigned int val;
2600 int ret;
2601 int i;
2602 unsigned long mask;
2603
2604 ret = kstrtouint_from_user(buf, count, 0, &val);
2605 if (ret < 0)
2606 return ret;
2607
2608 ret = -ESRCH;
2609 task = get_proc_task(file_inode(file));
2610 if (!task)
2611 goto out_no_task;
2612
2613 mm = get_task_mm(task);
2614 if (!mm)
2615 goto out_no_mm;
2616 ret = 0;
2617
2618 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2619 if (val & mask)
2620 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2621 else
2622 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2623 }
2624
2625 mmput(mm);
2626 out_no_mm:
2627 put_task_struct(task);
2628 out_no_task:
2629 if (ret < 0)
2630 return ret;
2631 return count;
2632}
2633
2634static const struct file_operations proc_coredump_filter_operations = {
2635 .read = proc_coredump_filter_read,
2636 .write = proc_coredump_filter_write,
2637 .llseek = generic_file_llseek,
2638};
2639#endif
2640
2641#ifdef CONFIG_TASK_IO_ACCOUNTING
2642static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2643{
2644 struct task_io_accounting acct = task->ioac;
2645 unsigned long flags;
2646 int result;
2647
2648 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2649 if (result)
2650 return result;
2651
2652 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2653 result = -EACCES;
2654 goto out_unlock;
2655 }
2656
2657 if (whole && lock_task_sighand(task, &flags)) {
2658 struct task_struct *t = task;
2659
2660 task_io_accounting_add(&acct, &task->signal->ioac);
2661 while_each_thread(task, t)
2662 task_io_accounting_add(&acct, &t->ioac);
2663
2664 unlock_task_sighand(task, &flags);
2665 }
2666 seq_printf(m,
2667 "rchar: %llu\n"
2668 "wchar: %llu\n"
2669 "syscr: %llu\n"
2670 "syscw: %llu\n"
2671 "read_bytes: %llu\n"
2672 "write_bytes: %llu\n"
2673 "cancelled_write_bytes: %llu\n",
2674 (unsigned long long)acct.rchar,
2675 (unsigned long long)acct.wchar,
2676 (unsigned long long)acct.syscr,
2677 (unsigned long long)acct.syscw,
2678 (unsigned long long)acct.read_bytes,
2679 (unsigned long long)acct.write_bytes,
2680 (unsigned long long)acct.cancelled_write_bytes);
2681 result = 0;
2682
2683out_unlock:
2684 mutex_unlock(&task->signal->cred_guard_mutex);
2685 return result;
2686}
2687
2688static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2689 struct pid *pid, struct task_struct *task)
2690{
2691 return do_io_accounting(task, m, 0);
2692}
2693
2694static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2695 struct pid *pid, struct task_struct *task)
2696{
2697 return do_io_accounting(task, m, 1);
2698}
2699#endif /* CONFIG_TASK_IO_ACCOUNTING */
2700
2701#ifdef CONFIG_USER_NS
2702static int proc_id_map_open(struct inode *inode, struct file *file,
2703 const struct seq_operations *seq_ops)
2704{
2705 struct user_namespace *ns = NULL;
2706 struct task_struct *task;
2707 struct seq_file *seq;
2708 int ret = -EINVAL;
2709
2710 task = get_proc_task(inode);
2711 if (task) {
2712 rcu_read_lock();
2713 ns = get_user_ns(task_cred_xxx(task, user_ns));
2714 rcu_read_unlock();
2715 put_task_struct(task);
2716 }
2717 if (!ns)
2718 goto err;
2719
2720 ret = seq_open(file, seq_ops);
2721 if (ret)
2722 goto err_put_ns;
2723
2724 seq = file->private_data;
2725 seq->private = ns;
2726
2727 return 0;
2728err_put_ns:
2729 put_user_ns(ns);
2730err:
2731 return ret;
2732}
2733
2734static int proc_id_map_release(struct inode *inode, struct file *file)
2735{
2736 struct seq_file *seq = file->private_data;
2737 struct user_namespace *ns = seq->private;
2738 put_user_ns(ns);
2739 return seq_release(inode, file);
2740}
2741
2742static int proc_uid_map_open(struct inode *inode, struct file *file)
2743{
2744 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2745}
2746
2747static int proc_gid_map_open(struct inode *inode, struct file *file)
2748{
2749 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2750}
2751
2752static int proc_projid_map_open(struct inode *inode, struct file *file)
2753{
2754 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2755}
2756
2757static const struct file_operations proc_uid_map_operations = {
2758 .open = proc_uid_map_open,
2759 .write = proc_uid_map_write,
2760 .read = seq_read,
2761 .llseek = seq_lseek,
2762 .release = proc_id_map_release,
2763};
2764
2765static const struct file_operations proc_gid_map_operations = {
2766 .open = proc_gid_map_open,
2767 .write = proc_gid_map_write,
2768 .read = seq_read,
2769 .llseek = seq_lseek,
2770 .release = proc_id_map_release,
2771};
2772
2773static const struct file_operations proc_projid_map_operations = {
2774 .open = proc_projid_map_open,
2775 .write = proc_projid_map_write,
2776 .read = seq_read,
2777 .llseek = seq_lseek,
2778 .release = proc_id_map_release,
2779};
2780
2781static int proc_setgroups_open(struct inode *inode, struct file *file)
2782{
2783 struct user_namespace *ns = NULL;
2784 struct task_struct *task;
2785 int ret;
2786
2787 ret = -ESRCH;
2788 task = get_proc_task(inode);
2789 if (task) {
2790 rcu_read_lock();
2791 ns = get_user_ns(task_cred_xxx(task, user_ns));
2792 rcu_read_unlock();
2793 put_task_struct(task);
2794 }
2795 if (!ns)
2796 goto err;
2797
2798 if (file->f_mode & FMODE_WRITE) {
2799 ret = -EACCES;
2800 if (!ns_capable(ns, CAP_SYS_ADMIN))
2801 goto err_put_ns;
2802 }
2803
2804 ret = single_open(file, &proc_setgroups_show, ns);
2805 if (ret)
2806 goto err_put_ns;
2807
2808 return 0;
2809err_put_ns:
2810 put_user_ns(ns);
2811err:
2812 return ret;
2813}
2814
2815static int proc_setgroups_release(struct inode *inode, struct file *file)
2816{
2817 struct seq_file *seq = file->private_data;
2818 struct user_namespace *ns = seq->private;
2819 int ret = single_release(inode, file);
2820 put_user_ns(ns);
2821 return ret;
2822}
2823
2824static const struct file_operations proc_setgroups_operations = {
2825 .open = proc_setgroups_open,
2826 .write = proc_setgroups_write,
2827 .read = seq_read,
2828 .llseek = seq_lseek,
2829 .release = proc_setgroups_release,
2830};
2831#endif /* CONFIG_USER_NS */
2832
2833static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2834 struct pid *pid, struct task_struct *task)
2835{
2836 int err = lock_trace(task);
2837 if (!err) {
2838 seq_printf(m, "%08x\n", task->personality);
2839 unlock_trace(task);
2840 }
2841 return err;
2842}
2843
2844/*
2845 * Thread groups
2846 */
2847static const struct file_operations proc_task_operations;
2848static const struct inode_operations proc_task_inode_operations;
2849
2850static const struct pid_entry tgid_base_stuff[] = {
2851 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2852 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2853 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2854 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2855 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2856#ifdef CONFIG_NET
2857 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2858#endif
2859 REG("environ", S_IRUSR, proc_environ_operations),
2860 REG("auxv", S_IRUSR, proc_auxv_operations),
2861 ONE("status", S_IRUGO, proc_pid_status),
2862 ONE("personality", S_IRUSR, proc_pid_personality),
2863 ONE("limits", S_IRUGO, proc_pid_limits),
2864#ifdef CONFIG_SCHED_DEBUG
2865 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2866#endif
2867#ifdef CONFIG_SCHED_AUTOGROUP
2868 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2869#endif
2870 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2871#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2872 ONE("syscall", S_IRUSR, proc_pid_syscall),
2873#endif
2874 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2875 ONE("stat", S_IRUGO, proc_tgid_stat),
2876 ONE("statm", S_IRUGO, proc_pid_statm),
2877 REG("maps", S_IRUGO, proc_pid_maps_operations),
2878#ifdef CONFIG_NUMA
2879 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2880#endif
2881 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2882 LNK("cwd", proc_cwd_link),
2883 LNK("root", proc_root_link),
2884 LNK("exe", proc_exe_link),
2885 REG("mounts", S_IRUGO, proc_mounts_operations),
2886 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2887 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2888#ifdef CONFIG_PROC_PAGE_MONITOR
2889 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2890 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2891 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2892#endif
2893#ifdef CONFIG_SECURITY
2894 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2895#endif
2896#ifdef CONFIG_KALLSYMS
2897 ONE("wchan", S_IRUGO, proc_pid_wchan),
2898#endif
2899#ifdef CONFIG_STACKTRACE
2900 ONE("stack", S_IRUSR, proc_pid_stack),
2901#endif
2902#ifdef CONFIG_SCHED_INFO
2903 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2904#endif
2905#ifdef CONFIG_LATENCYTOP
2906 REG("latency", S_IRUGO, proc_lstats_operations),
2907#endif
2908#ifdef CONFIG_PROC_PID_CPUSET
2909 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2910#endif
2911#ifdef CONFIG_CGROUPS
2912 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2913#endif
2914 ONE("oom_score", S_IRUGO, proc_oom_score),
2915 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2916 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2917#ifdef CONFIG_AUDITSYSCALL
2918 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2919 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2920#endif
2921#ifdef CONFIG_FAULT_INJECTION
2922 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2923#endif
2924#ifdef CONFIG_ELF_CORE
2925 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2926#endif
2927#ifdef CONFIG_TASK_IO_ACCOUNTING
2928 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2929#endif
2930#ifdef CONFIG_HARDWALL
2931 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2932#endif
2933#ifdef CONFIG_USER_NS
2934 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2935 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2936 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2937 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2938#endif
2939#ifdef CONFIG_CHECKPOINT_RESTORE
2940 REG("timers", S_IRUGO, proc_timers_operations),
2941#endif
2942 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2943};
2944
2945static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2946{
2947 return proc_pident_readdir(file, ctx,
2948 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2949}
2950
2951static const struct file_operations proc_tgid_base_operations = {
2952 .read = generic_read_dir,
2953 .iterate_shared = proc_tgid_base_readdir,
2954 .llseek = generic_file_llseek,
2955};
2956
2957static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2958{
2959 return proc_pident_lookup(dir, dentry,
2960 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2961}
2962
2963static const struct inode_operations proc_tgid_base_inode_operations = {
2964 .lookup = proc_tgid_base_lookup,
2965 .getattr = pid_getattr,
2966 .setattr = proc_setattr,
2967 .permission = proc_pid_permission,
2968};
2969
2970static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2971{
2972 struct dentry *dentry, *leader, *dir;
2973 char buf[PROC_NUMBUF];
2974 struct qstr name;
2975
2976 name.name = buf;
2977 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2978 /* no ->d_hash() rejects on procfs */
2979 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2980 if (dentry) {
2981 d_invalidate(dentry);
2982 dput(dentry);
2983 }
2984
2985 if (pid == tgid)
2986 return;
2987
2988 name.name = buf;
2989 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2990 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2991 if (!leader)
2992 goto out;
2993
2994 name.name = "task";
2995 name.len = strlen(name.name);
2996 dir = d_hash_and_lookup(leader, &name);
2997 if (!dir)
2998 goto out_put_leader;
2999
3000 name.name = buf;
3001 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3002 dentry = d_hash_and_lookup(dir, &name);
3003 if (dentry) {
3004 d_invalidate(dentry);
3005 dput(dentry);
3006 }
3007
3008 dput(dir);
3009out_put_leader:
3010 dput(leader);
3011out:
3012 return;
3013}
3014
3015/**
3016 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3017 * @task: task that should be flushed.
3018 *
3019 * When flushing dentries from proc, one needs to flush them from global
3020 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3021 * in. This call is supposed to do all of this job.
3022 *
3023 * Looks in the dcache for
3024 * /proc/@pid
3025 * /proc/@tgid/task/@pid
3026 * if either directory is present flushes it and all of it'ts children
3027 * from the dcache.
3028 *
3029 * It is safe and reasonable to cache /proc entries for a task until
3030 * that task exits. After that they just clog up the dcache with
3031 * useless entries, possibly causing useful dcache entries to be
3032 * flushed instead. This routine is proved to flush those useless
3033 * dcache entries at process exit time.
3034 *
3035 * NOTE: This routine is just an optimization so it does not guarantee
3036 * that no dcache entries will exist at process exit time it
3037 * just makes it very unlikely that any will persist.
3038 */
3039
3040void proc_flush_task(struct task_struct *task)
3041{
3042 int i;
3043 struct pid *pid, *tgid;
3044 struct upid *upid;
3045
3046 pid = task_pid(task);
3047 tgid = task_tgid(task);
3048
3049 for (i = 0; i <= pid->level; i++) {
3050 upid = &pid->numbers[i];
3051 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3052 tgid->numbers[i].nr);
3053 }
3054}
3055
3056static int proc_pid_instantiate(struct inode *dir,
3057 struct dentry * dentry,
3058 struct task_struct *task, const void *ptr)
3059{
3060 struct inode *inode;
3061
3062 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3063 if (!inode)
3064 goto out;
3065
3066 inode->i_op = &proc_tgid_base_inode_operations;
3067 inode->i_fop = &proc_tgid_base_operations;
3068 inode->i_flags|=S_IMMUTABLE;
3069
3070 set_nlink(inode, nlink_tgid);
3071
3072 d_set_d_op(dentry, &pid_dentry_operations);
3073
3074 d_add(dentry, inode);
3075 /* Close the race of the process dying before we return the dentry */
3076 if (pid_revalidate(dentry, 0))
3077 return 0;
3078out:
3079 return -ENOENT;
3080}
3081
3082struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3083{
3084 int result = -ENOENT;
3085 struct task_struct *task;
3086 unsigned tgid;
3087 struct pid_namespace *ns;
3088
3089 tgid = name_to_int(&dentry->d_name);
3090 if (tgid == ~0U)
3091 goto out;
3092
3093 ns = dentry->d_sb->s_fs_info;
3094 rcu_read_lock();
3095 task = find_task_by_pid_ns(tgid, ns);
3096 if (task)
3097 get_task_struct(task);
3098 rcu_read_unlock();
3099 if (!task)
3100 goto out;
3101
3102 result = proc_pid_instantiate(dir, dentry, task, NULL);
3103 put_task_struct(task);
3104out:
3105 return ERR_PTR(result);
3106}
3107
3108/*
3109 * Find the first task with tgid >= tgid
3110 *
3111 */
3112struct tgid_iter {
3113 unsigned int tgid;
3114 struct task_struct *task;
3115};
3116static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3117{
3118 struct pid *pid;
3119
3120 if (iter.task)
3121 put_task_struct(iter.task);
3122 rcu_read_lock();
3123retry:
3124 iter.task = NULL;
3125 pid = find_ge_pid(iter.tgid, ns);
3126 if (pid) {
3127 iter.tgid = pid_nr_ns(pid, ns);
3128 iter.task = pid_task(pid, PIDTYPE_PID);
3129 /* What we to know is if the pid we have find is the
3130 * pid of a thread_group_leader. Testing for task
3131 * being a thread_group_leader is the obvious thing
3132 * todo but there is a window when it fails, due to
3133 * the pid transfer logic in de_thread.
3134 *
3135 * So we perform the straight forward test of seeing
3136 * if the pid we have found is the pid of a thread
3137 * group leader, and don't worry if the task we have
3138 * found doesn't happen to be a thread group leader.
3139 * As we don't care in the case of readdir.
3140 */
3141 if (!iter.task || !has_group_leader_pid(iter.task)) {
3142 iter.tgid += 1;
3143 goto retry;
3144 }
3145 get_task_struct(iter.task);
3146 }
3147 rcu_read_unlock();
3148 return iter;
3149}
3150
3151#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3152
3153/* for the /proc/ directory itself, after non-process stuff has been done */
3154int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3155{
3156 struct tgid_iter iter;
3157 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3158 loff_t pos = ctx->pos;
3159
3160 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3161 return 0;
3162
3163 if (pos == TGID_OFFSET - 2) {
3164 struct inode *inode = d_inode(ns->proc_self);
3165 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3166 return 0;
3167 ctx->pos = pos = pos + 1;
3168 }
3169 if (pos == TGID_OFFSET - 1) {
3170 struct inode *inode = d_inode(ns->proc_thread_self);
3171 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3172 return 0;
3173 ctx->pos = pos = pos + 1;
3174 }
3175 iter.tgid = pos - TGID_OFFSET;
3176 iter.task = NULL;
3177 for (iter = next_tgid(ns, iter);
3178 iter.task;
3179 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3180 char name[PROC_NUMBUF];
3181 int len;
3182
3183 cond_resched();
3184 if (!has_pid_permissions(ns, iter.task, 2))
3185 continue;
3186
3187 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3188 ctx->pos = iter.tgid + TGID_OFFSET;
3189 if (!proc_fill_cache(file, ctx, name, len,
3190 proc_pid_instantiate, iter.task, NULL)) {
3191 put_task_struct(iter.task);
3192 return 0;
3193 }
3194 }
3195 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3196 return 0;
3197}
3198
3199/*
3200 * proc_tid_comm_permission is a special permission function exclusively
3201 * used for the node /proc/<pid>/task/<tid>/comm.
3202 * It bypasses generic permission checks in the case where a task of the same
3203 * task group attempts to access the node.
3204 * The rationale behind this is that glibc and bionic access this node for
3205 * cross thread naming (pthread_set/getname_np(!self)). However, if
3206 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3207 * which locks out the cross thread naming implementation.
3208 * This function makes sure that the node is always accessible for members of
3209 * same thread group.
3210 */
3211static int proc_tid_comm_permission(struct inode *inode, int mask)
3212{
3213 bool is_same_tgroup;
3214 struct task_struct *task;
3215
3216 task = get_proc_task(inode);
3217 if (!task)
3218 return -ESRCH;
3219 is_same_tgroup = same_thread_group(current, task);
3220 put_task_struct(task);
3221
3222 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3223 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3224 * read or written by the members of the corresponding
3225 * thread group.
3226 */
3227 return 0;
3228 }
3229
3230 return generic_permission(inode, mask);
3231}
3232
3233static const struct inode_operations proc_tid_comm_inode_operations = {
3234 .permission = proc_tid_comm_permission,
3235};
3236
3237/*
3238 * Tasks
3239 */
3240static const struct pid_entry tid_base_stuff[] = {
3241 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3242 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3243 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3244#ifdef CONFIG_NET
3245 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3246#endif
3247 REG("environ", S_IRUSR, proc_environ_operations),
3248 REG("auxv", S_IRUSR, proc_auxv_operations),
3249 ONE("status", S_IRUGO, proc_pid_status),
3250 ONE("personality", S_IRUSR, proc_pid_personality),
3251 ONE("limits", S_IRUGO, proc_pid_limits),
3252#ifdef CONFIG_SCHED_DEBUG
3253 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3254#endif
3255 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3256 &proc_tid_comm_inode_operations,
3257 &proc_pid_set_comm_operations, {}),
3258#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3259 ONE("syscall", S_IRUSR, proc_pid_syscall),
3260#endif
3261 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3262 ONE("stat", S_IRUGO, proc_tid_stat),
3263 ONE("statm", S_IRUGO, proc_pid_statm),
3264 REG("maps", S_IRUGO, proc_tid_maps_operations),
3265#ifdef CONFIG_PROC_CHILDREN
3266 REG("children", S_IRUGO, proc_tid_children_operations),
3267#endif
3268#ifdef CONFIG_NUMA
3269 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3270#endif
3271 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3272 LNK("cwd", proc_cwd_link),
3273 LNK("root", proc_root_link),
3274 LNK("exe", proc_exe_link),
3275 REG("mounts", S_IRUGO, proc_mounts_operations),
3276 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3277#ifdef CONFIG_PROC_PAGE_MONITOR
3278 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3279 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3280 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3281#endif
3282#ifdef CONFIG_SECURITY
3283 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3284#endif
3285#ifdef CONFIG_KALLSYMS
3286 ONE("wchan", S_IRUGO, proc_pid_wchan),
3287#endif
3288#ifdef CONFIG_STACKTRACE
3289 ONE("stack", S_IRUSR, proc_pid_stack),
3290#endif
3291#ifdef CONFIG_SCHED_INFO
3292 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3293#endif
3294#ifdef CONFIG_LATENCYTOP
3295 REG("latency", S_IRUGO, proc_lstats_operations),
3296#endif
3297#ifdef CONFIG_PROC_PID_CPUSET
3298 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3299#endif
3300#ifdef CONFIG_CGROUPS
3301 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3302#endif
3303 ONE("oom_score", S_IRUGO, proc_oom_score),
3304 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3305 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3306#ifdef CONFIG_AUDITSYSCALL
3307 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3308 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3309#endif
3310#ifdef CONFIG_FAULT_INJECTION
3311 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3312#endif
3313#ifdef CONFIG_TASK_IO_ACCOUNTING
3314 ONE("io", S_IRUSR, proc_tid_io_accounting),
3315#endif
3316#ifdef CONFIG_HARDWALL
3317 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3318#endif
3319#ifdef CONFIG_USER_NS
3320 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3321 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3322 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3323 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3324#endif
3325};
3326
3327static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3328{
3329 return proc_pident_readdir(file, ctx,
3330 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3331}
3332
3333static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3334{
3335 return proc_pident_lookup(dir, dentry,
3336 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3337}
3338
3339static const struct file_operations proc_tid_base_operations = {
3340 .read = generic_read_dir,
3341 .iterate_shared = proc_tid_base_readdir,
3342 .llseek = generic_file_llseek,
3343};
3344
3345static const struct inode_operations proc_tid_base_inode_operations = {
3346 .lookup = proc_tid_base_lookup,
3347 .getattr = pid_getattr,
3348 .setattr = proc_setattr,
3349};
3350
3351static int proc_task_instantiate(struct inode *dir,
3352 struct dentry *dentry, struct task_struct *task, const void *ptr)
3353{
3354 struct inode *inode;
3355 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3356
3357 if (!inode)
3358 goto out;
3359 inode->i_op = &proc_tid_base_inode_operations;
3360 inode->i_fop = &proc_tid_base_operations;
3361 inode->i_flags|=S_IMMUTABLE;
3362
3363 set_nlink(inode, nlink_tid);
3364
3365 d_set_d_op(dentry, &pid_dentry_operations);
3366
3367 d_add(dentry, inode);
3368 /* Close the race of the process dying before we return the dentry */
3369 if (pid_revalidate(dentry, 0))
3370 return 0;
3371out:
3372 return -ENOENT;
3373}
3374
3375static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3376{
3377 int result = -ENOENT;
3378 struct task_struct *task;
3379 struct task_struct *leader = get_proc_task(dir);
3380 unsigned tid;
3381 struct pid_namespace *ns;
3382
3383 if (!leader)
3384 goto out_no_task;
3385
3386 tid = name_to_int(&dentry->d_name);
3387 if (tid == ~0U)
3388 goto out;
3389
3390 ns = dentry->d_sb->s_fs_info;
3391 rcu_read_lock();
3392 task = find_task_by_pid_ns(tid, ns);
3393 if (task)
3394 get_task_struct(task);
3395 rcu_read_unlock();
3396 if (!task)
3397 goto out;
3398 if (!same_thread_group(leader, task))
3399 goto out_drop_task;
3400
3401 result = proc_task_instantiate(dir, dentry, task, NULL);
3402out_drop_task:
3403 put_task_struct(task);
3404out:
3405 put_task_struct(leader);
3406out_no_task:
3407 return ERR_PTR(result);
3408}
3409
3410/*
3411 * Find the first tid of a thread group to return to user space.
3412 *
3413 * Usually this is just the thread group leader, but if the users
3414 * buffer was too small or there was a seek into the middle of the
3415 * directory we have more work todo.
3416 *
3417 * In the case of a short read we start with find_task_by_pid.
3418 *
3419 * In the case of a seek we start with the leader and walk nr
3420 * threads past it.
3421 */
3422static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3423 struct pid_namespace *ns)
3424{
3425 struct task_struct *pos, *task;
3426 unsigned long nr = f_pos;
3427
3428 if (nr != f_pos) /* 32bit overflow? */
3429 return NULL;
3430
3431 rcu_read_lock();
3432 task = pid_task(pid, PIDTYPE_PID);
3433 if (!task)
3434 goto fail;
3435
3436 /* Attempt to start with the tid of a thread */
3437 if (tid && nr) {
3438 pos = find_task_by_pid_ns(tid, ns);
3439 if (pos && same_thread_group(pos, task))
3440 goto found;
3441 }
3442
3443 /* If nr exceeds the number of threads there is nothing todo */
3444 if (nr >= get_nr_threads(task))
3445 goto fail;
3446
3447 /* If we haven't found our starting place yet start
3448 * with the leader and walk nr threads forward.
3449 */
3450 pos = task = task->group_leader;
3451 do {
3452 if (!nr--)
3453 goto found;
3454 } while_each_thread(task, pos);
3455fail:
3456 pos = NULL;
3457 goto out;
3458found:
3459 get_task_struct(pos);
3460out:
3461 rcu_read_unlock();
3462 return pos;
3463}
3464
3465/*
3466 * Find the next thread in the thread list.
3467 * Return NULL if there is an error or no next thread.
3468 *
3469 * The reference to the input task_struct is released.
3470 */
3471static struct task_struct *next_tid(struct task_struct *start)
3472{
3473 struct task_struct *pos = NULL;
3474 rcu_read_lock();
3475 if (pid_alive(start)) {
3476 pos = next_thread(start);
3477 if (thread_group_leader(pos))
3478 pos = NULL;
3479 else
3480 get_task_struct(pos);
3481 }
3482 rcu_read_unlock();
3483 put_task_struct(start);
3484 return pos;
3485}
3486
3487/* for the /proc/TGID/task/ directories */
3488static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3489{
3490 struct inode *inode = file_inode(file);
3491 struct task_struct *task;
3492 struct pid_namespace *ns;
3493 int tid;
3494
3495 if (proc_inode_is_dead(inode))
3496 return -ENOENT;
3497
3498 if (!dir_emit_dots(file, ctx))
3499 return 0;
3500
3501 /* f_version caches the tgid value that the last readdir call couldn't
3502 * return. lseek aka telldir automagically resets f_version to 0.
3503 */
3504 ns = inode->i_sb->s_fs_info;
3505 tid = (int)file->f_version;
3506 file->f_version = 0;
3507 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3508 task;
3509 task = next_tid(task), ctx->pos++) {
3510 char name[PROC_NUMBUF];
3511 int len;
3512 tid = task_pid_nr_ns(task, ns);
3513 len = snprintf(name, sizeof(name), "%d", tid);
3514 if (!proc_fill_cache(file, ctx, name, len,
3515 proc_task_instantiate, task, NULL)) {
3516 /* returning this tgid failed, save it as the first
3517 * pid for the next readir call */
3518 file->f_version = (u64)tid;
3519 put_task_struct(task);
3520 break;
3521 }
3522 }
3523
3524 return 0;
3525}
3526
3527static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3528{
3529 struct inode *inode = d_inode(dentry);
3530 struct task_struct *p = get_proc_task(inode);
3531 generic_fillattr(inode, stat);
3532
3533 if (p) {
3534 stat->nlink += get_nr_threads(p);
3535 put_task_struct(p);
3536 }
3537
3538 return 0;
3539}
3540
3541static const struct inode_operations proc_task_inode_operations = {
3542 .lookup = proc_task_lookup,
3543 .getattr = proc_task_getattr,
3544 .setattr = proc_setattr,
3545 .permission = proc_pid_permission,
3546};
3547
3548static const struct file_operations proc_task_operations = {
3549 .read = generic_read_dir,
3550 .iterate_shared = proc_task_readdir,
3551 .llseek = generic_file_llseek,
3552};
3553
3554void __init set_proc_pid_nlink(void)
3555{
3556 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3557 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3558}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/proc/base.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * proc base directory handling functions
8 *
9 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10 * Instead of using magical inumbers to determine the kind of object
11 * we allocate and fill in-core inodes upon lookup. They don't even
12 * go into icache. We cache the reference to task_struct upon lookup too.
13 * Eventually it should become a filesystem in its own. We don't use the
14 * rest of procfs anymore.
15 *
16 *
17 * Changelog:
18 * 17-Jan-2005
19 * Allan Bezerra
20 * Bruna Moreira <bruna.moreira@indt.org.br>
21 * Edjard Mota <edjard.mota@indt.org.br>
22 * Ilias Biris <ilias.biris@indt.org.br>
23 * Mauricio Lin <mauricio.lin@indt.org.br>
24 *
25 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 *
27 * A new process specific entry (smaps) included in /proc. It shows the
28 * size of rss for each memory area. The maps entry lacks information
29 * about physical memory size (rss) for each mapped file, i.e.,
30 * rss information for executables and library files.
31 * This additional information is useful for any tools that need to know
32 * about physical memory consumption for a process specific library.
33 *
34 * Changelog:
35 * 21-Feb-2005
36 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37 * Pud inclusion in the page table walking.
38 *
39 * ChangeLog:
40 * 10-Mar-2005
41 * 10LE Instituto Nokia de Tecnologia - INdT:
42 * A better way to walks through the page table as suggested by Hugh Dickins.
43 *
44 * Simo Piiroinen <simo.piiroinen@nokia.com>:
45 * Smaps information related to shared, private, clean and dirty pages.
46 *
47 * Paul Mundt <paul.mundt@nokia.com>:
48 * Overall revision about smaps.
49 */
50
51#include <linux/uaccess.h>
52
53#include <linux/errno.h>
54#include <linux/time.h>
55#include <linux/proc_fs.h>
56#include <linux/stat.h>
57#include <linux/task_io_accounting_ops.h>
58#include <linux/init.h>
59#include <linux/capability.h>
60#include <linux/file.h>
61#include <linux/fdtable.h>
62#include <linux/generic-radix-tree.h>
63#include <linux/string.h>
64#include <linux/seq_file.h>
65#include <linux/namei.h>
66#include <linux/mnt_namespace.h>
67#include <linux/mm.h>
68#include <linux/swap.h>
69#include <linux/rcupdate.h>
70#include <linux/kallsyms.h>
71#include <linux/stacktrace.h>
72#include <linux/resource.h>
73#include <linux/module.h>
74#include <linux/mount.h>
75#include <linux/security.h>
76#include <linux/ptrace.h>
77#include <linux/tracehook.h>
78#include <linux/printk.h>
79#include <linux/cache.h>
80#include <linux/cgroup.h>
81#include <linux/cpuset.h>
82#include <linux/audit.h>
83#include <linux/poll.h>
84#include <linux/nsproxy.h>
85#include <linux/oom.h>
86#include <linux/elf.h>
87#include <linux/pid_namespace.h>
88#include <linux/user_namespace.h>
89#include <linux/fs_struct.h>
90#include <linux/slab.h>
91#include <linux/sched/autogroup.h>
92#include <linux/sched/mm.h>
93#include <linux/sched/coredump.h>
94#include <linux/sched/debug.h>
95#include <linux/sched/stat.h>
96#include <linux/posix-timers.h>
97#include <trace/events/oom.h>
98#include "internal.h"
99#include "fd.h"
100
101#include "../../lib/kstrtox.h"
102
103/* NOTE:
104 * Implementing inode permission operations in /proc is almost
105 * certainly an error. Permission checks need to happen during
106 * each system call not at open time. The reason is that most of
107 * what we wish to check for permissions in /proc varies at runtime.
108 *
109 * The classic example of a problem is opening file descriptors
110 * in /proc for a task before it execs a suid executable.
111 */
112
113static u8 nlink_tid __ro_after_init;
114static u8 nlink_tgid __ro_after_init;
115
116struct pid_entry {
117 const char *name;
118 unsigned int len;
119 umode_t mode;
120 const struct inode_operations *iop;
121 const struct file_operations *fop;
122 union proc_op op;
123};
124
125#define NOD(NAME, MODE, IOP, FOP, OP) { \
126 .name = (NAME), \
127 .len = sizeof(NAME) - 1, \
128 .mode = MODE, \
129 .iop = IOP, \
130 .fop = FOP, \
131 .op = OP, \
132}
133
134#define DIR(NAME, MODE, iops, fops) \
135 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
136#define LNK(NAME, get_link) \
137 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
138 &proc_pid_link_inode_operations, NULL, \
139 { .proc_get_link = get_link } )
140#define REG(NAME, MODE, fops) \
141 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
142#define ONE(NAME, MODE, show) \
143 NOD(NAME, (S_IFREG|(MODE)), \
144 NULL, &proc_single_file_operations, \
145 { .proc_show = show } )
146#define ATTR(LSM, NAME, MODE) \
147 NOD(NAME, (S_IFREG|(MODE)), \
148 NULL, &proc_pid_attr_operations, \
149 { .lsm = LSM })
150
151/*
152 * Count the number of hardlinks for the pid_entry table, excluding the .
153 * and .. links.
154 */
155static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
156 unsigned int n)
157{
158 unsigned int i;
159 unsigned int count;
160
161 count = 2;
162 for (i = 0; i < n; ++i) {
163 if (S_ISDIR(entries[i].mode))
164 ++count;
165 }
166
167 return count;
168}
169
170static int get_task_root(struct task_struct *task, struct path *root)
171{
172 int result = -ENOENT;
173
174 task_lock(task);
175 if (task->fs) {
176 get_fs_root(task->fs, root);
177 result = 0;
178 }
179 task_unlock(task);
180 return result;
181}
182
183static int proc_cwd_link(struct dentry *dentry, struct path *path)
184{
185 struct task_struct *task = get_proc_task(d_inode(dentry));
186 int result = -ENOENT;
187
188 if (task) {
189 task_lock(task);
190 if (task->fs) {
191 get_fs_pwd(task->fs, path);
192 result = 0;
193 }
194 task_unlock(task);
195 put_task_struct(task);
196 }
197 return result;
198}
199
200static int proc_root_link(struct dentry *dentry, struct path *path)
201{
202 struct task_struct *task = get_proc_task(d_inode(dentry));
203 int result = -ENOENT;
204
205 if (task) {
206 result = get_task_root(task, path);
207 put_task_struct(task);
208 }
209 return result;
210}
211
212/*
213 * If the user used setproctitle(), we just get the string from
214 * user space at arg_start, and limit it to a maximum of one page.
215 */
216static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
217 size_t count, unsigned long pos,
218 unsigned long arg_start)
219{
220 char *page;
221 int ret, got;
222
223 if (pos >= PAGE_SIZE)
224 return 0;
225
226 page = (char *)__get_free_page(GFP_KERNEL);
227 if (!page)
228 return -ENOMEM;
229
230 ret = 0;
231 got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
232 if (got > 0) {
233 int len = strnlen(page, got);
234
235 /* Include the NUL character if it was found */
236 if (len < got)
237 len++;
238
239 if (len > pos) {
240 len -= pos;
241 if (len > count)
242 len = count;
243 len -= copy_to_user(buf, page+pos, len);
244 if (!len)
245 len = -EFAULT;
246 ret = len;
247 }
248 }
249 free_page((unsigned long)page);
250 return ret;
251}
252
253static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
254 size_t count, loff_t *ppos)
255{
256 unsigned long arg_start, arg_end, env_start, env_end;
257 unsigned long pos, len;
258 char *page, c;
259
260 /* Check if process spawned far enough to have cmdline. */
261 if (!mm->env_end)
262 return 0;
263
264 spin_lock(&mm->arg_lock);
265 arg_start = mm->arg_start;
266 arg_end = mm->arg_end;
267 env_start = mm->env_start;
268 env_end = mm->env_end;
269 spin_unlock(&mm->arg_lock);
270
271 if (arg_start >= arg_end)
272 return 0;
273
274 /*
275 * We allow setproctitle() to overwrite the argument
276 * strings, and overflow past the original end. But
277 * only when it overflows into the environment area.
278 */
279 if (env_start != arg_end || env_end < env_start)
280 env_start = env_end = arg_end;
281 len = env_end - arg_start;
282
283 /* We're not going to care if "*ppos" has high bits set */
284 pos = *ppos;
285 if (pos >= len)
286 return 0;
287 if (count > len - pos)
288 count = len - pos;
289 if (!count)
290 return 0;
291
292 /*
293 * Magical special case: if the argv[] end byte is not
294 * zero, the user has overwritten it with setproctitle(3).
295 *
296 * Possible future enhancement: do this only once when
297 * pos is 0, and set a flag in the 'struct file'.
298 */
299 if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
300 return get_mm_proctitle(mm, buf, count, pos, arg_start);
301
302 /*
303 * For the non-setproctitle() case we limit things strictly
304 * to the [arg_start, arg_end[ range.
305 */
306 pos += arg_start;
307 if (pos < arg_start || pos >= arg_end)
308 return 0;
309 if (count > arg_end - pos)
310 count = arg_end - pos;
311
312 page = (char *)__get_free_page(GFP_KERNEL);
313 if (!page)
314 return -ENOMEM;
315
316 len = 0;
317 while (count) {
318 int got;
319 size_t size = min_t(size_t, PAGE_SIZE, count);
320
321 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
322 if (got <= 0)
323 break;
324 got -= copy_to_user(buf, page, got);
325 if (unlikely(!got)) {
326 if (!len)
327 len = -EFAULT;
328 break;
329 }
330 pos += got;
331 buf += got;
332 len += got;
333 count -= got;
334 }
335
336 free_page((unsigned long)page);
337 return len;
338}
339
340static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
341 size_t count, loff_t *pos)
342{
343 struct mm_struct *mm;
344 ssize_t ret;
345
346 mm = get_task_mm(tsk);
347 if (!mm)
348 return 0;
349
350 ret = get_mm_cmdline(mm, buf, count, pos);
351 mmput(mm);
352 return ret;
353}
354
355static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
356 size_t count, loff_t *pos)
357{
358 struct task_struct *tsk;
359 ssize_t ret;
360
361 BUG_ON(*pos < 0);
362
363 tsk = get_proc_task(file_inode(file));
364 if (!tsk)
365 return -ESRCH;
366 ret = get_task_cmdline(tsk, buf, count, pos);
367 put_task_struct(tsk);
368 if (ret > 0)
369 *pos += ret;
370 return ret;
371}
372
373static const struct file_operations proc_pid_cmdline_ops = {
374 .read = proc_pid_cmdline_read,
375 .llseek = generic_file_llseek,
376};
377
378#ifdef CONFIG_KALLSYMS
379/*
380 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
381 * Returns the resolved symbol. If that fails, simply return the address.
382 */
383static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
384 struct pid *pid, struct task_struct *task)
385{
386 unsigned long wchan;
387 char symname[KSYM_NAME_LEN];
388
389 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
390 goto print0;
391
392 wchan = get_wchan(task);
393 if (wchan && !lookup_symbol_name(wchan, symname)) {
394 seq_puts(m, symname);
395 return 0;
396 }
397
398print0:
399 seq_putc(m, '0');
400 return 0;
401}
402#endif /* CONFIG_KALLSYMS */
403
404static int lock_trace(struct task_struct *task)
405{
406 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
407 if (err)
408 return err;
409 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
410 mutex_unlock(&task->signal->cred_guard_mutex);
411 return -EPERM;
412 }
413 return 0;
414}
415
416static void unlock_trace(struct task_struct *task)
417{
418 mutex_unlock(&task->signal->cred_guard_mutex);
419}
420
421#ifdef CONFIG_STACKTRACE
422
423#define MAX_STACK_TRACE_DEPTH 64
424
425static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
426 struct pid *pid, struct task_struct *task)
427{
428 unsigned long *entries;
429 int err;
430
431 /*
432 * The ability to racily run the kernel stack unwinder on a running task
433 * and then observe the unwinder output is scary; while it is useful for
434 * debugging kernel issues, it can also allow an attacker to leak kernel
435 * stack contents.
436 * Doing this in a manner that is at least safe from races would require
437 * some work to ensure that the remote task can not be scheduled; and
438 * even then, this would still expose the unwinder as local attack
439 * surface.
440 * Therefore, this interface is restricted to root.
441 */
442 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
443 return -EACCES;
444
445 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
446 GFP_KERNEL);
447 if (!entries)
448 return -ENOMEM;
449
450 err = lock_trace(task);
451 if (!err) {
452 unsigned int i, nr_entries;
453
454 nr_entries = stack_trace_save_tsk(task, entries,
455 MAX_STACK_TRACE_DEPTH, 0);
456
457 for (i = 0; i < nr_entries; i++) {
458 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
459 }
460
461 unlock_trace(task);
462 }
463 kfree(entries);
464
465 return err;
466}
467#endif
468
469#ifdef CONFIG_SCHED_INFO
470/*
471 * Provides /proc/PID/schedstat
472 */
473static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
474 struct pid *pid, struct task_struct *task)
475{
476 if (unlikely(!sched_info_on()))
477 seq_puts(m, "0 0 0\n");
478 else
479 seq_printf(m, "%llu %llu %lu\n",
480 (unsigned long long)task->se.sum_exec_runtime,
481 (unsigned long long)task->sched_info.run_delay,
482 task->sched_info.pcount);
483
484 return 0;
485}
486#endif
487
488#ifdef CONFIG_LATENCYTOP
489static int lstats_show_proc(struct seq_file *m, void *v)
490{
491 int i;
492 struct inode *inode = m->private;
493 struct task_struct *task = get_proc_task(inode);
494
495 if (!task)
496 return -ESRCH;
497 seq_puts(m, "Latency Top version : v0.1\n");
498 for (i = 0; i < LT_SAVECOUNT; i++) {
499 struct latency_record *lr = &task->latency_record[i];
500 if (lr->backtrace[0]) {
501 int q;
502 seq_printf(m, "%i %li %li",
503 lr->count, lr->time, lr->max);
504 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
505 unsigned long bt = lr->backtrace[q];
506
507 if (!bt)
508 break;
509 seq_printf(m, " %ps", (void *)bt);
510 }
511 seq_putc(m, '\n');
512 }
513
514 }
515 put_task_struct(task);
516 return 0;
517}
518
519static int lstats_open(struct inode *inode, struct file *file)
520{
521 return single_open(file, lstats_show_proc, inode);
522}
523
524static ssize_t lstats_write(struct file *file, const char __user *buf,
525 size_t count, loff_t *offs)
526{
527 struct task_struct *task = get_proc_task(file_inode(file));
528
529 if (!task)
530 return -ESRCH;
531 clear_tsk_latency_tracing(task);
532 put_task_struct(task);
533
534 return count;
535}
536
537static const struct file_operations proc_lstats_operations = {
538 .open = lstats_open,
539 .read = seq_read,
540 .write = lstats_write,
541 .llseek = seq_lseek,
542 .release = single_release,
543};
544
545#endif
546
547static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
548 struct pid *pid, struct task_struct *task)
549{
550 unsigned long totalpages = totalram_pages() + total_swap_pages;
551 unsigned long points = 0;
552
553 points = oom_badness(task, totalpages) * 1000 / totalpages;
554 seq_printf(m, "%lu\n", points);
555
556 return 0;
557}
558
559struct limit_names {
560 const char *name;
561 const char *unit;
562};
563
564static const struct limit_names lnames[RLIM_NLIMITS] = {
565 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
566 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
567 [RLIMIT_DATA] = {"Max data size", "bytes"},
568 [RLIMIT_STACK] = {"Max stack size", "bytes"},
569 [RLIMIT_CORE] = {"Max core file size", "bytes"},
570 [RLIMIT_RSS] = {"Max resident set", "bytes"},
571 [RLIMIT_NPROC] = {"Max processes", "processes"},
572 [RLIMIT_NOFILE] = {"Max open files", "files"},
573 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
574 [RLIMIT_AS] = {"Max address space", "bytes"},
575 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
576 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
577 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
578 [RLIMIT_NICE] = {"Max nice priority", NULL},
579 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
580 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
581};
582
583/* Display limits for a process */
584static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
585 struct pid *pid, struct task_struct *task)
586{
587 unsigned int i;
588 unsigned long flags;
589
590 struct rlimit rlim[RLIM_NLIMITS];
591
592 if (!lock_task_sighand(task, &flags))
593 return 0;
594 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
595 unlock_task_sighand(task, &flags);
596
597 /*
598 * print the file header
599 */
600 seq_puts(m, "Limit "
601 "Soft Limit "
602 "Hard Limit "
603 "Units \n");
604
605 for (i = 0; i < RLIM_NLIMITS; i++) {
606 if (rlim[i].rlim_cur == RLIM_INFINITY)
607 seq_printf(m, "%-25s %-20s ",
608 lnames[i].name, "unlimited");
609 else
610 seq_printf(m, "%-25s %-20lu ",
611 lnames[i].name, rlim[i].rlim_cur);
612
613 if (rlim[i].rlim_max == RLIM_INFINITY)
614 seq_printf(m, "%-20s ", "unlimited");
615 else
616 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
617
618 if (lnames[i].unit)
619 seq_printf(m, "%-10s\n", lnames[i].unit);
620 else
621 seq_putc(m, '\n');
622 }
623
624 return 0;
625}
626
627#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
628static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
629 struct pid *pid, struct task_struct *task)
630{
631 struct syscall_info info;
632 u64 *args = &info.data.args[0];
633 int res;
634
635 res = lock_trace(task);
636 if (res)
637 return res;
638
639 if (task_current_syscall(task, &info))
640 seq_puts(m, "running\n");
641 else if (info.data.nr < 0)
642 seq_printf(m, "%d 0x%llx 0x%llx\n",
643 info.data.nr, info.sp, info.data.instruction_pointer);
644 else
645 seq_printf(m,
646 "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
647 info.data.nr,
648 args[0], args[1], args[2], args[3], args[4], args[5],
649 info.sp, info.data.instruction_pointer);
650 unlock_trace(task);
651
652 return 0;
653}
654#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
655
656/************************************************************************/
657/* Here the fs part begins */
658/************************************************************************/
659
660/* permission checks */
661static int proc_fd_access_allowed(struct inode *inode)
662{
663 struct task_struct *task;
664 int allowed = 0;
665 /* Allow access to a task's file descriptors if it is us or we
666 * may use ptrace attach to the process and find out that
667 * information.
668 */
669 task = get_proc_task(inode);
670 if (task) {
671 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
672 put_task_struct(task);
673 }
674 return allowed;
675}
676
677int proc_setattr(struct dentry *dentry, struct iattr *attr)
678{
679 int error;
680 struct inode *inode = d_inode(dentry);
681
682 if (attr->ia_valid & ATTR_MODE)
683 return -EPERM;
684
685 error = setattr_prepare(dentry, attr);
686 if (error)
687 return error;
688
689 setattr_copy(inode, attr);
690 mark_inode_dirty(inode);
691 return 0;
692}
693
694/*
695 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
696 * or euid/egid (for hide_pid_min=2)?
697 */
698static bool has_pid_permissions(struct pid_namespace *pid,
699 struct task_struct *task,
700 int hide_pid_min)
701{
702 if (pid->hide_pid < hide_pid_min)
703 return true;
704 if (in_group_p(pid->pid_gid))
705 return true;
706 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
707}
708
709
710static int proc_pid_permission(struct inode *inode, int mask)
711{
712 struct pid_namespace *pid = proc_pid_ns(inode);
713 struct task_struct *task;
714 bool has_perms;
715
716 task = get_proc_task(inode);
717 if (!task)
718 return -ESRCH;
719 has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
720 put_task_struct(task);
721
722 if (!has_perms) {
723 if (pid->hide_pid == HIDEPID_INVISIBLE) {
724 /*
725 * Let's make getdents(), stat(), and open()
726 * consistent with each other. If a process
727 * may not stat() a file, it shouldn't be seen
728 * in procfs at all.
729 */
730 return -ENOENT;
731 }
732
733 return -EPERM;
734 }
735 return generic_permission(inode, mask);
736}
737
738
739
740static const struct inode_operations proc_def_inode_operations = {
741 .setattr = proc_setattr,
742};
743
744static int proc_single_show(struct seq_file *m, void *v)
745{
746 struct inode *inode = m->private;
747 struct pid_namespace *ns = proc_pid_ns(inode);
748 struct pid *pid = proc_pid(inode);
749 struct task_struct *task;
750 int ret;
751
752 task = get_pid_task(pid, PIDTYPE_PID);
753 if (!task)
754 return -ESRCH;
755
756 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
757
758 put_task_struct(task);
759 return ret;
760}
761
762static int proc_single_open(struct inode *inode, struct file *filp)
763{
764 return single_open(filp, proc_single_show, inode);
765}
766
767static const struct file_operations proc_single_file_operations = {
768 .open = proc_single_open,
769 .read = seq_read,
770 .llseek = seq_lseek,
771 .release = single_release,
772};
773
774
775struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
776{
777 struct task_struct *task = get_proc_task(inode);
778 struct mm_struct *mm = ERR_PTR(-ESRCH);
779
780 if (task) {
781 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
782 put_task_struct(task);
783
784 if (!IS_ERR_OR_NULL(mm)) {
785 /* ensure this mm_struct can't be freed */
786 mmgrab(mm);
787 /* but do not pin its memory */
788 mmput(mm);
789 }
790 }
791
792 return mm;
793}
794
795static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
796{
797 struct mm_struct *mm = proc_mem_open(inode, mode);
798
799 if (IS_ERR(mm))
800 return PTR_ERR(mm);
801
802 file->private_data = mm;
803 return 0;
804}
805
806static int mem_open(struct inode *inode, struct file *file)
807{
808 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
809
810 /* OK to pass negative loff_t, we can catch out-of-range */
811 file->f_mode |= FMODE_UNSIGNED_OFFSET;
812
813 return ret;
814}
815
816static ssize_t mem_rw(struct file *file, char __user *buf,
817 size_t count, loff_t *ppos, int write)
818{
819 struct mm_struct *mm = file->private_data;
820 unsigned long addr = *ppos;
821 ssize_t copied;
822 char *page;
823 unsigned int flags;
824
825 if (!mm)
826 return 0;
827
828 page = (char *)__get_free_page(GFP_KERNEL);
829 if (!page)
830 return -ENOMEM;
831
832 copied = 0;
833 if (!mmget_not_zero(mm))
834 goto free;
835
836 flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
837
838 while (count > 0) {
839 int this_len = min_t(int, count, PAGE_SIZE);
840
841 if (write && copy_from_user(page, buf, this_len)) {
842 copied = -EFAULT;
843 break;
844 }
845
846 this_len = access_remote_vm(mm, addr, page, this_len, flags);
847 if (!this_len) {
848 if (!copied)
849 copied = -EIO;
850 break;
851 }
852
853 if (!write && copy_to_user(buf, page, this_len)) {
854 copied = -EFAULT;
855 break;
856 }
857
858 buf += this_len;
859 addr += this_len;
860 copied += this_len;
861 count -= this_len;
862 }
863 *ppos = addr;
864
865 mmput(mm);
866free:
867 free_page((unsigned long) page);
868 return copied;
869}
870
871static ssize_t mem_read(struct file *file, char __user *buf,
872 size_t count, loff_t *ppos)
873{
874 return mem_rw(file, buf, count, ppos, 0);
875}
876
877static ssize_t mem_write(struct file *file, const char __user *buf,
878 size_t count, loff_t *ppos)
879{
880 return mem_rw(file, (char __user*)buf, count, ppos, 1);
881}
882
883loff_t mem_lseek(struct file *file, loff_t offset, int orig)
884{
885 switch (orig) {
886 case 0:
887 file->f_pos = offset;
888 break;
889 case 1:
890 file->f_pos += offset;
891 break;
892 default:
893 return -EINVAL;
894 }
895 force_successful_syscall_return();
896 return file->f_pos;
897}
898
899static int mem_release(struct inode *inode, struct file *file)
900{
901 struct mm_struct *mm = file->private_data;
902 if (mm)
903 mmdrop(mm);
904 return 0;
905}
906
907static const struct file_operations proc_mem_operations = {
908 .llseek = mem_lseek,
909 .read = mem_read,
910 .write = mem_write,
911 .open = mem_open,
912 .release = mem_release,
913};
914
915static int environ_open(struct inode *inode, struct file *file)
916{
917 return __mem_open(inode, file, PTRACE_MODE_READ);
918}
919
920static ssize_t environ_read(struct file *file, char __user *buf,
921 size_t count, loff_t *ppos)
922{
923 char *page;
924 unsigned long src = *ppos;
925 int ret = 0;
926 struct mm_struct *mm = file->private_data;
927 unsigned long env_start, env_end;
928
929 /* Ensure the process spawned far enough to have an environment. */
930 if (!mm || !mm->env_end)
931 return 0;
932
933 page = (char *)__get_free_page(GFP_KERNEL);
934 if (!page)
935 return -ENOMEM;
936
937 ret = 0;
938 if (!mmget_not_zero(mm))
939 goto free;
940
941 spin_lock(&mm->arg_lock);
942 env_start = mm->env_start;
943 env_end = mm->env_end;
944 spin_unlock(&mm->arg_lock);
945
946 while (count > 0) {
947 size_t this_len, max_len;
948 int retval;
949
950 if (src >= (env_end - env_start))
951 break;
952
953 this_len = env_end - (env_start + src);
954
955 max_len = min_t(size_t, PAGE_SIZE, count);
956 this_len = min(max_len, this_len);
957
958 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
959
960 if (retval <= 0) {
961 ret = retval;
962 break;
963 }
964
965 if (copy_to_user(buf, page, retval)) {
966 ret = -EFAULT;
967 break;
968 }
969
970 ret += retval;
971 src += retval;
972 buf += retval;
973 count -= retval;
974 }
975 *ppos = src;
976 mmput(mm);
977
978free:
979 free_page((unsigned long) page);
980 return ret;
981}
982
983static const struct file_operations proc_environ_operations = {
984 .open = environ_open,
985 .read = environ_read,
986 .llseek = generic_file_llseek,
987 .release = mem_release,
988};
989
990static int auxv_open(struct inode *inode, struct file *file)
991{
992 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
993}
994
995static ssize_t auxv_read(struct file *file, char __user *buf,
996 size_t count, loff_t *ppos)
997{
998 struct mm_struct *mm = file->private_data;
999 unsigned int nwords = 0;
1000
1001 if (!mm)
1002 return 0;
1003 do {
1004 nwords += 2;
1005 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1006 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1007 nwords * sizeof(mm->saved_auxv[0]));
1008}
1009
1010static const struct file_operations proc_auxv_operations = {
1011 .open = auxv_open,
1012 .read = auxv_read,
1013 .llseek = generic_file_llseek,
1014 .release = mem_release,
1015};
1016
1017static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1018 loff_t *ppos)
1019{
1020 struct task_struct *task = get_proc_task(file_inode(file));
1021 char buffer[PROC_NUMBUF];
1022 int oom_adj = OOM_ADJUST_MIN;
1023 size_t len;
1024
1025 if (!task)
1026 return -ESRCH;
1027 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1028 oom_adj = OOM_ADJUST_MAX;
1029 else
1030 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1031 OOM_SCORE_ADJ_MAX;
1032 put_task_struct(task);
1033 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1034 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1035}
1036
1037static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1038{
1039 static DEFINE_MUTEX(oom_adj_mutex);
1040 struct mm_struct *mm = NULL;
1041 struct task_struct *task;
1042 int err = 0;
1043
1044 task = get_proc_task(file_inode(file));
1045 if (!task)
1046 return -ESRCH;
1047
1048 mutex_lock(&oom_adj_mutex);
1049 if (legacy) {
1050 if (oom_adj < task->signal->oom_score_adj &&
1051 !capable(CAP_SYS_RESOURCE)) {
1052 err = -EACCES;
1053 goto err_unlock;
1054 }
1055 /*
1056 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1057 * /proc/pid/oom_score_adj instead.
1058 */
1059 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1060 current->comm, task_pid_nr(current), task_pid_nr(task),
1061 task_pid_nr(task));
1062 } else {
1063 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1064 !capable(CAP_SYS_RESOURCE)) {
1065 err = -EACCES;
1066 goto err_unlock;
1067 }
1068 }
1069
1070 /*
1071 * Make sure we will check other processes sharing the mm if this is
1072 * not vfrok which wants its own oom_score_adj.
1073 * pin the mm so it doesn't go away and get reused after task_unlock
1074 */
1075 if (!task->vfork_done) {
1076 struct task_struct *p = find_lock_task_mm(task);
1077
1078 if (p) {
1079 if (atomic_read(&p->mm->mm_users) > 1) {
1080 mm = p->mm;
1081 mmgrab(mm);
1082 }
1083 task_unlock(p);
1084 }
1085 }
1086
1087 task->signal->oom_score_adj = oom_adj;
1088 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1089 task->signal->oom_score_adj_min = (short)oom_adj;
1090 trace_oom_score_adj_update(task);
1091
1092 if (mm) {
1093 struct task_struct *p;
1094
1095 rcu_read_lock();
1096 for_each_process(p) {
1097 if (same_thread_group(task, p))
1098 continue;
1099
1100 /* do not touch kernel threads or the global init */
1101 if (p->flags & PF_KTHREAD || is_global_init(p))
1102 continue;
1103
1104 task_lock(p);
1105 if (!p->vfork_done && process_shares_mm(p, mm)) {
1106 p->signal->oom_score_adj = oom_adj;
1107 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1108 p->signal->oom_score_adj_min = (short)oom_adj;
1109 }
1110 task_unlock(p);
1111 }
1112 rcu_read_unlock();
1113 mmdrop(mm);
1114 }
1115err_unlock:
1116 mutex_unlock(&oom_adj_mutex);
1117 put_task_struct(task);
1118 return err;
1119}
1120
1121/*
1122 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1123 * kernels. The effective policy is defined by oom_score_adj, which has a
1124 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1125 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1126 * Processes that become oom disabled via oom_adj will still be oom disabled
1127 * with this implementation.
1128 *
1129 * oom_adj cannot be removed since existing userspace binaries use it.
1130 */
1131static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1132 size_t count, loff_t *ppos)
1133{
1134 char buffer[PROC_NUMBUF];
1135 int oom_adj;
1136 int err;
1137
1138 memset(buffer, 0, sizeof(buffer));
1139 if (count > sizeof(buffer) - 1)
1140 count = sizeof(buffer) - 1;
1141 if (copy_from_user(buffer, buf, count)) {
1142 err = -EFAULT;
1143 goto out;
1144 }
1145
1146 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1147 if (err)
1148 goto out;
1149 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1150 oom_adj != OOM_DISABLE) {
1151 err = -EINVAL;
1152 goto out;
1153 }
1154
1155 /*
1156 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1157 * value is always attainable.
1158 */
1159 if (oom_adj == OOM_ADJUST_MAX)
1160 oom_adj = OOM_SCORE_ADJ_MAX;
1161 else
1162 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1163
1164 err = __set_oom_adj(file, oom_adj, true);
1165out:
1166 return err < 0 ? err : count;
1167}
1168
1169static const struct file_operations proc_oom_adj_operations = {
1170 .read = oom_adj_read,
1171 .write = oom_adj_write,
1172 .llseek = generic_file_llseek,
1173};
1174
1175static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1176 size_t count, loff_t *ppos)
1177{
1178 struct task_struct *task = get_proc_task(file_inode(file));
1179 char buffer[PROC_NUMBUF];
1180 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1181 size_t len;
1182
1183 if (!task)
1184 return -ESRCH;
1185 oom_score_adj = task->signal->oom_score_adj;
1186 put_task_struct(task);
1187 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1188 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1189}
1190
1191static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1192 size_t count, loff_t *ppos)
1193{
1194 char buffer[PROC_NUMBUF];
1195 int oom_score_adj;
1196 int err;
1197
1198 memset(buffer, 0, sizeof(buffer));
1199 if (count > sizeof(buffer) - 1)
1200 count = sizeof(buffer) - 1;
1201 if (copy_from_user(buffer, buf, count)) {
1202 err = -EFAULT;
1203 goto out;
1204 }
1205
1206 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1207 if (err)
1208 goto out;
1209 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1210 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1211 err = -EINVAL;
1212 goto out;
1213 }
1214
1215 err = __set_oom_adj(file, oom_score_adj, false);
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_AUDIT
1227#define TMPBUFLEN 11
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
1367static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1368 size_t count, loff_t *ppos)
1369{
1370 struct task_struct *task;
1371 int err;
1372 unsigned int n;
1373
1374 err = kstrtouint_from_user(buf, count, 0, &n);
1375 if (err)
1376 return err;
1377
1378 task = get_proc_task(file_inode(file));
1379 if (!task)
1380 return -ESRCH;
1381 task->fail_nth = n;
1382 put_task_struct(task);
1383
1384 return count;
1385}
1386
1387static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1388 size_t count, loff_t *ppos)
1389{
1390 struct task_struct *task;
1391 char numbuf[PROC_NUMBUF];
1392 ssize_t len;
1393
1394 task = get_proc_task(file_inode(file));
1395 if (!task)
1396 return -ESRCH;
1397 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1398 put_task_struct(task);
1399 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1400}
1401
1402static const struct file_operations proc_fail_nth_operations = {
1403 .read = proc_fail_nth_read,
1404 .write = proc_fail_nth_write,
1405};
1406#endif
1407
1408
1409#ifdef CONFIG_SCHED_DEBUG
1410/*
1411 * Print out various scheduling related per-task fields:
1412 */
1413static int sched_show(struct seq_file *m, void *v)
1414{
1415 struct inode *inode = m->private;
1416 struct pid_namespace *ns = proc_pid_ns(inode);
1417 struct task_struct *p;
1418
1419 p = get_proc_task(inode);
1420 if (!p)
1421 return -ESRCH;
1422 proc_sched_show_task(p, ns, m);
1423
1424 put_task_struct(p);
1425
1426 return 0;
1427}
1428
1429static ssize_t
1430sched_write(struct file *file, const char __user *buf,
1431 size_t count, loff_t *offset)
1432{
1433 struct inode *inode = file_inode(file);
1434 struct task_struct *p;
1435
1436 p = get_proc_task(inode);
1437 if (!p)
1438 return -ESRCH;
1439 proc_sched_set_task(p);
1440
1441 put_task_struct(p);
1442
1443 return count;
1444}
1445
1446static int sched_open(struct inode *inode, struct file *filp)
1447{
1448 return single_open(filp, sched_show, inode);
1449}
1450
1451static const struct file_operations proc_pid_sched_operations = {
1452 .open = sched_open,
1453 .read = seq_read,
1454 .write = sched_write,
1455 .llseek = seq_lseek,
1456 .release = single_release,
1457};
1458
1459#endif
1460
1461#ifdef CONFIG_SCHED_AUTOGROUP
1462/*
1463 * Print out autogroup related information:
1464 */
1465static int sched_autogroup_show(struct seq_file *m, void *v)
1466{
1467 struct inode *inode = m->private;
1468 struct task_struct *p;
1469
1470 p = get_proc_task(inode);
1471 if (!p)
1472 return -ESRCH;
1473 proc_sched_autogroup_show_task(p, m);
1474
1475 put_task_struct(p);
1476
1477 return 0;
1478}
1479
1480static ssize_t
1481sched_autogroup_write(struct file *file, const char __user *buf,
1482 size_t count, loff_t *offset)
1483{
1484 struct inode *inode = file_inode(file);
1485 struct task_struct *p;
1486 char buffer[PROC_NUMBUF];
1487 int nice;
1488 int err;
1489
1490 memset(buffer, 0, sizeof(buffer));
1491 if (count > sizeof(buffer) - 1)
1492 count = sizeof(buffer) - 1;
1493 if (copy_from_user(buffer, buf, count))
1494 return -EFAULT;
1495
1496 err = kstrtoint(strstrip(buffer), 0, &nice);
1497 if (err < 0)
1498 return err;
1499
1500 p = get_proc_task(inode);
1501 if (!p)
1502 return -ESRCH;
1503
1504 err = proc_sched_autogroup_set_nice(p, nice);
1505 if (err)
1506 count = err;
1507
1508 put_task_struct(p);
1509
1510 return count;
1511}
1512
1513static int sched_autogroup_open(struct inode *inode, struct file *filp)
1514{
1515 int ret;
1516
1517 ret = single_open(filp, sched_autogroup_show, NULL);
1518 if (!ret) {
1519 struct seq_file *m = filp->private_data;
1520
1521 m->private = inode;
1522 }
1523 return ret;
1524}
1525
1526static const struct file_operations proc_pid_sched_autogroup_operations = {
1527 .open = sched_autogroup_open,
1528 .read = seq_read,
1529 .write = sched_autogroup_write,
1530 .llseek = seq_lseek,
1531 .release = single_release,
1532};
1533
1534#endif /* CONFIG_SCHED_AUTOGROUP */
1535
1536static ssize_t comm_write(struct file *file, const char __user *buf,
1537 size_t count, loff_t *offset)
1538{
1539 struct inode *inode = file_inode(file);
1540 struct task_struct *p;
1541 char buffer[TASK_COMM_LEN];
1542 const size_t maxlen = sizeof(buffer) - 1;
1543
1544 memset(buffer, 0, sizeof(buffer));
1545 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1546 return -EFAULT;
1547
1548 p = get_proc_task(inode);
1549 if (!p)
1550 return -ESRCH;
1551
1552 if (same_thread_group(current, p))
1553 set_task_comm(p, buffer);
1554 else
1555 count = -EINVAL;
1556
1557 put_task_struct(p);
1558
1559 return count;
1560}
1561
1562static int comm_show(struct seq_file *m, void *v)
1563{
1564 struct inode *inode = m->private;
1565 struct task_struct *p;
1566
1567 p = get_proc_task(inode);
1568 if (!p)
1569 return -ESRCH;
1570
1571 proc_task_name(m, p, false);
1572 seq_putc(m, '\n');
1573
1574 put_task_struct(p);
1575
1576 return 0;
1577}
1578
1579static int comm_open(struct inode *inode, struct file *filp)
1580{
1581 return single_open(filp, comm_show, inode);
1582}
1583
1584static const struct file_operations proc_pid_set_comm_operations = {
1585 .open = comm_open,
1586 .read = seq_read,
1587 .write = comm_write,
1588 .llseek = seq_lseek,
1589 .release = single_release,
1590};
1591
1592static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1593{
1594 struct task_struct *task;
1595 struct file *exe_file;
1596
1597 task = get_proc_task(d_inode(dentry));
1598 if (!task)
1599 return -ENOENT;
1600 exe_file = get_task_exe_file(task);
1601 put_task_struct(task);
1602 if (exe_file) {
1603 *exe_path = exe_file->f_path;
1604 path_get(&exe_file->f_path);
1605 fput(exe_file);
1606 return 0;
1607 } else
1608 return -ENOENT;
1609}
1610
1611static const char *proc_pid_get_link(struct dentry *dentry,
1612 struct inode *inode,
1613 struct delayed_call *done)
1614{
1615 struct path path;
1616 int error = -EACCES;
1617
1618 if (!dentry)
1619 return ERR_PTR(-ECHILD);
1620
1621 /* Are we allowed to snoop on the tasks file descriptors? */
1622 if (!proc_fd_access_allowed(inode))
1623 goto out;
1624
1625 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1626 if (error)
1627 goto out;
1628
1629 nd_jump_link(&path);
1630 return NULL;
1631out:
1632 return ERR_PTR(error);
1633}
1634
1635static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1636{
1637 char *tmp = (char *)__get_free_page(GFP_KERNEL);
1638 char *pathname;
1639 int len;
1640
1641 if (!tmp)
1642 return -ENOMEM;
1643
1644 pathname = d_path(path, tmp, PAGE_SIZE);
1645 len = PTR_ERR(pathname);
1646 if (IS_ERR(pathname))
1647 goto out;
1648 len = tmp + PAGE_SIZE - 1 - pathname;
1649
1650 if (len > buflen)
1651 len = buflen;
1652 if (copy_to_user(buffer, pathname, len))
1653 len = -EFAULT;
1654 out:
1655 free_page((unsigned long)tmp);
1656 return len;
1657}
1658
1659static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1660{
1661 int error = -EACCES;
1662 struct inode *inode = d_inode(dentry);
1663 struct path path;
1664
1665 /* Are we allowed to snoop on the tasks file descriptors? */
1666 if (!proc_fd_access_allowed(inode))
1667 goto out;
1668
1669 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1670 if (error)
1671 goto out;
1672
1673 error = do_proc_readlink(&path, buffer, buflen);
1674 path_put(&path);
1675out:
1676 return error;
1677}
1678
1679const struct inode_operations proc_pid_link_inode_operations = {
1680 .readlink = proc_pid_readlink,
1681 .get_link = proc_pid_get_link,
1682 .setattr = proc_setattr,
1683};
1684
1685
1686/* building an inode */
1687
1688void task_dump_owner(struct task_struct *task, umode_t mode,
1689 kuid_t *ruid, kgid_t *rgid)
1690{
1691 /* Depending on the state of dumpable compute who should own a
1692 * proc file for a task.
1693 */
1694 const struct cred *cred;
1695 kuid_t uid;
1696 kgid_t gid;
1697
1698 if (unlikely(task->flags & PF_KTHREAD)) {
1699 *ruid = GLOBAL_ROOT_UID;
1700 *rgid = GLOBAL_ROOT_GID;
1701 return;
1702 }
1703
1704 /* Default to the tasks effective ownership */
1705 rcu_read_lock();
1706 cred = __task_cred(task);
1707 uid = cred->euid;
1708 gid = cred->egid;
1709 rcu_read_unlock();
1710
1711 /*
1712 * Before the /proc/pid/status file was created the only way to read
1713 * the effective uid of a /process was to stat /proc/pid. Reading
1714 * /proc/pid/status is slow enough that procps and other packages
1715 * kept stating /proc/pid. To keep the rules in /proc simple I have
1716 * made this apply to all per process world readable and executable
1717 * directories.
1718 */
1719 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1720 struct mm_struct *mm;
1721 task_lock(task);
1722 mm = task->mm;
1723 /* Make non-dumpable tasks owned by some root */
1724 if (mm) {
1725 if (get_dumpable(mm) != SUID_DUMP_USER) {
1726 struct user_namespace *user_ns = mm->user_ns;
1727
1728 uid = make_kuid(user_ns, 0);
1729 if (!uid_valid(uid))
1730 uid = GLOBAL_ROOT_UID;
1731
1732 gid = make_kgid(user_ns, 0);
1733 if (!gid_valid(gid))
1734 gid = GLOBAL_ROOT_GID;
1735 }
1736 } else {
1737 uid = GLOBAL_ROOT_UID;
1738 gid = GLOBAL_ROOT_GID;
1739 }
1740 task_unlock(task);
1741 }
1742 *ruid = uid;
1743 *rgid = gid;
1744}
1745
1746struct inode *proc_pid_make_inode(struct super_block * sb,
1747 struct task_struct *task, umode_t mode)
1748{
1749 struct inode * inode;
1750 struct proc_inode *ei;
1751
1752 /* We need a new inode */
1753
1754 inode = new_inode(sb);
1755 if (!inode)
1756 goto out;
1757
1758 /* Common stuff */
1759 ei = PROC_I(inode);
1760 inode->i_mode = mode;
1761 inode->i_ino = get_next_ino();
1762 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1763 inode->i_op = &proc_def_inode_operations;
1764
1765 /*
1766 * grab the reference to task.
1767 */
1768 ei->pid = get_task_pid(task, PIDTYPE_PID);
1769 if (!ei->pid)
1770 goto out_unlock;
1771
1772 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1773 security_task_to_inode(task, inode);
1774
1775out:
1776 return inode;
1777
1778out_unlock:
1779 iput(inode);
1780 return NULL;
1781}
1782
1783int pid_getattr(const struct path *path, struct kstat *stat,
1784 u32 request_mask, unsigned int query_flags)
1785{
1786 struct inode *inode = d_inode(path->dentry);
1787 struct pid_namespace *pid = proc_pid_ns(inode);
1788 struct task_struct *task;
1789
1790 generic_fillattr(inode, stat);
1791
1792 stat->uid = GLOBAL_ROOT_UID;
1793 stat->gid = GLOBAL_ROOT_GID;
1794 rcu_read_lock();
1795 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1796 if (task) {
1797 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1798 rcu_read_unlock();
1799 /*
1800 * This doesn't prevent learning whether PID exists,
1801 * it only makes getattr() consistent with readdir().
1802 */
1803 return -ENOENT;
1804 }
1805 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1806 }
1807 rcu_read_unlock();
1808 return 0;
1809}
1810
1811/* dentry stuff */
1812
1813/*
1814 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1815 */
1816void pid_update_inode(struct task_struct *task, struct inode *inode)
1817{
1818 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1819
1820 inode->i_mode &= ~(S_ISUID | S_ISGID);
1821 security_task_to_inode(task, inode);
1822}
1823
1824/*
1825 * Rewrite the inode's ownerships here because the owning task may have
1826 * performed a setuid(), etc.
1827 *
1828 */
1829static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1830{
1831 struct inode *inode;
1832 struct task_struct *task;
1833
1834 if (flags & LOOKUP_RCU)
1835 return -ECHILD;
1836
1837 inode = d_inode(dentry);
1838 task = get_proc_task(inode);
1839
1840 if (task) {
1841 pid_update_inode(task, inode);
1842 put_task_struct(task);
1843 return 1;
1844 }
1845 return 0;
1846}
1847
1848static inline bool proc_inode_is_dead(struct inode *inode)
1849{
1850 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1851}
1852
1853int pid_delete_dentry(const struct dentry *dentry)
1854{
1855 /* Is the task we represent dead?
1856 * If so, then don't put the dentry on the lru list,
1857 * kill it immediately.
1858 */
1859 return proc_inode_is_dead(d_inode(dentry));
1860}
1861
1862const struct dentry_operations pid_dentry_operations =
1863{
1864 .d_revalidate = pid_revalidate,
1865 .d_delete = pid_delete_dentry,
1866};
1867
1868/* Lookups */
1869
1870/*
1871 * Fill a directory entry.
1872 *
1873 * If possible create the dcache entry and derive our inode number and
1874 * file type from dcache entry.
1875 *
1876 * Since all of the proc inode numbers are dynamically generated, the inode
1877 * numbers do not exist until the inode is cache. This means creating the
1878 * the dcache entry in readdir is necessary to keep the inode numbers
1879 * reported by readdir in sync with the inode numbers reported
1880 * by stat.
1881 */
1882bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1883 const char *name, unsigned int len,
1884 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1885{
1886 struct dentry *child, *dir = file->f_path.dentry;
1887 struct qstr qname = QSTR_INIT(name, len);
1888 struct inode *inode;
1889 unsigned type = DT_UNKNOWN;
1890 ino_t ino = 1;
1891
1892 child = d_hash_and_lookup(dir, &qname);
1893 if (!child) {
1894 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1895 child = d_alloc_parallel(dir, &qname, &wq);
1896 if (IS_ERR(child))
1897 goto end_instantiate;
1898 if (d_in_lookup(child)) {
1899 struct dentry *res;
1900 res = instantiate(child, task, ptr);
1901 d_lookup_done(child);
1902 if (unlikely(res)) {
1903 dput(child);
1904 child = res;
1905 if (IS_ERR(child))
1906 goto end_instantiate;
1907 }
1908 }
1909 }
1910 inode = d_inode(child);
1911 ino = inode->i_ino;
1912 type = inode->i_mode >> 12;
1913 dput(child);
1914end_instantiate:
1915 return dir_emit(ctx, name, len, ino, type);
1916}
1917
1918/*
1919 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1920 * which represent vma start and end addresses.
1921 */
1922static int dname_to_vma_addr(struct dentry *dentry,
1923 unsigned long *start, unsigned long *end)
1924{
1925 const char *str = dentry->d_name.name;
1926 unsigned long long sval, eval;
1927 unsigned int len;
1928
1929 if (str[0] == '0' && str[1] != '-')
1930 return -EINVAL;
1931 len = _parse_integer(str, 16, &sval);
1932 if (len & KSTRTOX_OVERFLOW)
1933 return -EINVAL;
1934 if (sval != (unsigned long)sval)
1935 return -EINVAL;
1936 str += len;
1937
1938 if (*str != '-')
1939 return -EINVAL;
1940 str++;
1941
1942 if (str[0] == '0' && str[1])
1943 return -EINVAL;
1944 len = _parse_integer(str, 16, &eval);
1945 if (len & KSTRTOX_OVERFLOW)
1946 return -EINVAL;
1947 if (eval != (unsigned long)eval)
1948 return -EINVAL;
1949 str += len;
1950
1951 if (*str != '\0')
1952 return -EINVAL;
1953
1954 *start = sval;
1955 *end = eval;
1956
1957 return 0;
1958}
1959
1960static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1961{
1962 unsigned long vm_start, vm_end;
1963 bool exact_vma_exists = false;
1964 struct mm_struct *mm = NULL;
1965 struct task_struct *task;
1966 struct inode *inode;
1967 int status = 0;
1968
1969 if (flags & LOOKUP_RCU)
1970 return -ECHILD;
1971
1972 inode = d_inode(dentry);
1973 task = get_proc_task(inode);
1974 if (!task)
1975 goto out_notask;
1976
1977 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1978 if (IS_ERR_OR_NULL(mm))
1979 goto out;
1980
1981 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1982 status = down_read_killable(&mm->mmap_sem);
1983 if (!status) {
1984 exact_vma_exists = !!find_exact_vma(mm, vm_start,
1985 vm_end);
1986 up_read(&mm->mmap_sem);
1987 }
1988 }
1989
1990 mmput(mm);
1991
1992 if (exact_vma_exists) {
1993 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1994
1995 security_task_to_inode(task, inode);
1996 status = 1;
1997 }
1998
1999out:
2000 put_task_struct(task);
2001
2002out_notask:
2003 return status;
2004}
2005
2006static const struct dentry_operations tid_map_files_dentry_operations = {
2007 .d_revalidate = map_files_d_revalidate,
2008 .d_delete = pid_delete_dentry,
2009};
2010
2011static int map_files_get_link(struct dentry *dentry, struct path *path)
2012{
2013 unsigned long vm_start, vm_end;
2014 struct vm_area_struct *vma;
2015 struct task_struct *task;
2016 struct mm_struct *mm;
2017 int rc;
2018
2019 rc = -ENOENT;
2020 task = get_proc_task(d_inode(dentry));
2021 if (!task)
2022 goto out;
2023
2024 mm = get_task_mm(task);
2025 put_task_struct(task);
2026 if (!mm)
2027 goto out;
2028
2029 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2030 if (rc)
2031 goto out_mmput;
2032
2033 rc = down_read_killable(&mm->mmap_sem);
2034 if (rc)
2035 goto out_mmput;
2036
2037 rc = -ENOENT;
2038 vma = find_exact_vma(mm, vm_start, vm_end);
2039 if (vma && vma->vm_file) {
2040 *path = vma->vm_file->f_path;
2041 path_get(path);
2042 rc = 0;
2043 }
2044 up_read(&mm->mmap_sem);
2045
2046out_mmput:
2047 mmput(mm);
2048out:
2049 return rc;
2050}
2051
2052struct map_files_info {
2053 unsigned long start;
2054 unsigned long end;
2055 fmode_t mode;
2056};
2057
2058/*
2059 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2060 * symlinks may be used to bypass permissions on ancestor directories in the
2061 * path to the file in question.
2062 */
2063static const char *
2064proc_map_files_get_link(struct dentry *dentry,
2065 struct inode *inode,
2066 struct delayed_call *done)
2067{
2068 if (!capable(CAP_SYS_ADMIN))
2069 return ERR_PTR(-EPERM);
2070
2071 return proc_pid_get_link(dentry, inode, done);
2072}
2073
2074/*
2075 * Identical to proc_pid_link_inode_operations except for get_link()
2076 */
2077static const struct inode_operations proc_map_files_link_inode_operations = {
2078 .readlink = proc_pid_readlink,
2079 .get_link = proc_map_files_get_link,
2080 .setattr = proc_setattr,
2081};
2082
2083static struct dentry *
2084proc_map_files_instantiate(struct dentry *dentry,
2085 struct task_struct *task, const void *ptr)
2086{
2087 fmode_t mode = (fmode_t)(unsigned long)ptr;
2088 struct proc_inode *ei;
2089 struct inode *inode;
2090
2091 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2092 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2093 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2094 if (!inode)
2095 return ERR_PTR(-ENOENT);
2096
2097 ei = PROC_I(inode);
2098 ei->op.proc_get_link = map_files_get_link;
2099
2100 inode->i_op = &proc_map_files_link_inode_operations;
2101 inode->i_size = 64;
2102
2103 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2104 return d_splice_alias(inode, dentry);
2105}
2106
2107static struct dentry *proc_map_files_lookup(struct inode *dir,
2108 struct dentry *dentry, unsigned int flags)
2109{
2110 unsigned long vm_start, vm_end;
2111 struct vm_area_struct *vma;
2112 struct task_struct *task;
2113 struct dentry *result;
2114 struct mm_struct *mm;
2115
2116 result = ERR_PTR(-ENOENT);
2117 task = get_proc_task(dir);
2118 if (!task)
2119 goto out;
2120
2121 result = ERR_PTR(-EACCES);
2122 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2123 goto out_put_task;
2124
2125 result = ERR_PTR(-ENOENT);
2126 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2127 goto out_put_task;
2128
2129 mm = get_task_mm(task);
2130 if (!mm)
2131 goto out_put_task;
2132
2133 result = ERR_PTR(-EINTR);
2134 if (down_read_killable(&mm->mmap_sem))
2135 goto out_put_mm;
2136
2137 result = ERR_PTR(-ENOENT);
2138 vma = find_exact_vma(mm, vm_start, vm_end);
2139 if (!vma)
2140 goto out_no_vma;
2141
2142 if (vma->vm_file)
2143 result = proc_map_files_instantiate(dentry, task,
2144 (void *)(unsigned long)vma->vm_file->f_mode);
2145
2146out_no_vma:
2147 up_read(&mm->mmap_sem);
2148out_put_mm:
2149 mmput(mm);
2150out_put_task:
2151 put_task_struct(task);
2152out:
2153 return result;
2154}
2155
2156static const struct inode_operations proc_map_files_inode_operations = {
2157 .lookup = proc_map_files_lookup,
2158 .permission = proc_fd_permission,
2159 .setattr = proc_setattr,
2160};
2161
2162static int
2163proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2164{
2165 struct vm_area_struct *vma;
2166 struct task_struct *task;
2167 struct mm_struct *mm;
2168 unsigned long nr_files, pos, i;
2169 GENRADIX(struct map_files_info) fa;
2170 struct map_files_info *p;
2171 int ret;
2172
2173 genradix_init(&fa);
2174
2175 ret = -ENOENT;
2176 task = get_proc_task(file_inode(file));
2177 if (!task)
2178 goto out;
2179
2180 ret = -EACCES;
2181 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2182 goto out_put_task;
2183
2184 ret = 0;
2185 if (!dir_emit_dots(file, ctx))
2186 goto out_put_task;
2187
2188 mm = get_task_mm(task);
2189 if (!mm)
2190 goto out_put_task;
2191
2192 ret = down_read_killable(&mm->mmap_sem);
2193 if (ret) {
2194 mmput(mm);
2195 goto out_put_task;
2196 }
2197
2198 nr_files = 0;
2199
2200 /*
2201 * We need two passes here:
2202 *
2203 * 1) Collect vmas of mapped files with mmap_sem taken
2204 * 2) Release mmap_sem and instantiate entries
2205 *
2206 * otherwise we get lockdep complained, since filldir()
2207 * routine might require mmap_sem taken in might_fault().
2208 */
2209
2210 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2211 if (!vma->vm_file)
2212 continue;
2213 if (++pos <= ctx->pos)
2214 continue;
2215
2216 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2217 if (!p) {
2218 ret = -ENOMEM;
2219 up_read(&mm->mmap_sem);
2220 mmput(mm);
2221 goto out_put_task;
2222 }
2223
2224 p->start = vma->vm_start;
2225 p->end = vma->vm_end;
2226 p->mode = vma->vm_file->f_mode;
2227 }
2228 up_read(&mm->mmap_sem);
2229 mmput(mm);
2230
2231 for (i = 0; i < nr_files; i++) {
2232 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2233 unsigned int len;
2234
2235 p = genradix_ptr(&fa, i);
2236 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2237 if (!proc_fill_cache(file, ctx,
2238 buf, len,
2239 proc_map_files_instantiate,
2240 task,
2241 (void *)(unsigned long)p->mode))
2242 break;
2243 ctx->pos++;
2244 }
2245
2246out_put_task:
2247 put_task_struct(task);
2248out:
2249 genradix_free(&fa);
2250 return ret;
2251}
2252
2253static const struct file_operations proc_map_files_operations = {
2254 .read = generic_read_dir,
2255 .iterate_shared = proc_map_files_readdir,
2256 .llseek = generic_file_llseek,
2257};
2258
2259#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2260struct timers_private {
2261 struct pid *pid;
2262 struct task_struct *task;
2263 struct sighand_struct *sighand;
2264 struct pid_namespace *ns;
2265 unsigned long flags;
2266};
2267
2268static void *timers_start(struct seq_file *m, loff_t *pos)
2269{
2270 struct timers_private *tp = m->private;
2271
2272 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2273 if (!tp->task)
2274 return ERR_PTR(-ESRCH);
2275
2276 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2277 if (!tp->sighand)
2278 return ERR_PTR(-ESRCH);
2279
2280 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2281}
2282
2283static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2284{
2285 struct timers_private *tp = m->private;
2286 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2287}
2288
2289static void timers_stop(struct seq_file *m, void *v)
2290{
2291 struct timers_private *tp = m->private;
2292
2293 if (tp->sighand) {
2294 unlock_task_sighand(tp->task, &tp->flags);
2295 tp->sighand = NULL;
2296 }
2297
2298 if (tp->task) {
2299 put_task_struct(tp->task);
2300 tp->task = NULL;
2301 }
2302}
2303
2304static int show_timer(struct seq_file *m, void *v)
2305{
2306 struct k_itimer *timer;
2307 struct timers_private *tp = m->private;
2308 int notify;
2309 static const char * const nstr[] = {
2310 [SIGEV_SIGNAL] = "signal",
2311 [SIGEV_NONE] = "none",
2312 [SIGEV_THREAD] = "thread",
2313 };
2314
2315 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2316 notify = timer->it_sigev_notify;
2317
2318 seq_printf(m, "ID: %d\n", timer->it_id);
2319 seq_printf(m, "signal: %d/%px\n",
2320 timer->sigq->info.si_signo,
2321 timer->sigq->info.si_value.sival_ptr);
2322 seq_printf(m, "notify: %s/%s.%d\n",
2323 nstr[notify & ~SIGEV_THREAD_ID],
2324 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2325 pid_nr_ns(timer->it_pid, tp->ns));
2326 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2327
2328 return 0;
2329}
2330
2331static const struct seq_operations proc_timers_seq_ops = {
2332 .start = timers_start,
2333 .next = timers_next,
2334 .stop = timers_stop,
2335 .show = show_timer,
2336};
2337
2338static int proc_timers_open(struct inode *inode, struct file *file)
2339{
2340 struct timers_private *tp;
2341
2342 tp = __seq_open_private(file, &proc_timers_seq_ops,
2343 sizeof(struct timers_private));
2344 if (!tp)
2345 return -ENOMEM;
2346
2347 tp->pid = proc_pid(inode);
2348 tp->ns = proc_pid_ns(inode);
2349 return 0;
2350}
2351
2352static const struct file_operations proc_timers_operations = {
2353 .open = proc_timers_open,
2354 .read = seq_read,
2355 .llseek = seq_lseek,
2356 .release = seq_release_private,
2357};
2358#endif
2359
2360static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2361 size_t count, loff_t *offset)
2362{
2363 struct inode *inode = file_inode(file);
2364 struct task_struct *p;
2365 u64 slack_ns;
2366 int err;
2367
2368 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2369 if (err < 0)
2370 return err;
2371
2372 p = get_proc_task(inode);
2373 if (!p)
2374 return -ESRCH;
2375
2376 if (p != current) {
2377 rcu_read_lock();
2378 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2379 rcu_read_unlock();
2380 count = -EPERM;
2381 goto out;
2382 }
2383 rcu_read_unlock();
2384
2385 err = security_task_setscheduler(p);
2386 if (err) {
2387 count = err;
2388 goto out;
2389 }
2390 }
2391
2392 task_lock(p);
2393 if (slack_ns == 0)
2394 p->timer_slack_ns = p->default_timer_slack_ns;
2395 else
2396 p->timer_slack_ns = slack_ns;
2397 task_unlock(p);
2398
2399out:
2400 put_task_struct(p);
2401
2402 return count;
2403}
2404
2405static int timerslack_ns_show(struct seq_file *m, void *v)
2406{
2407 struct inode *inode = m->private;
2408 struct task_struct *p;
2409 int err = 0;
2410
2411 p = get_proc_task(inode);
2412 if (!p)
2413 return -ESRCH;
2414
2415 if (p != current) {
2416 rcu_read_lock();
2417 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2418 rcu_read_unlock();
2419 err = -EPERM;
2420 goto out;
2421 }
2422 rcu_read_unlock();
2423
2424 err = security_task_getscheduler(p);
2425 if (err)
2426 goto out;
2427 }
2428
2429 task_lock(p);
2430 seq_printf(m, "%llu\n", p->timer_slack_ns);
2431 task_unlock(p);
2432
2433out:
2434 put_task_struct(p);
2435
2436 return err;
2437}
2438
2439static int timerslack_ns_open(struct inode *inode, struct file *filp)
2440{
2441 return single_open(filp, timerslack_ns_show, inode);
2442}
2443
2444static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2445 .open = timerslack_ns_open,
2446 .read = seq_read,
2447 .write = timerslack_ns_write,
2448 .llseek = seq_lseek,
2449 .release = single_release,
2450};
2451
2452static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2453 struct task_struct *task, const void *ptr)
2454{
2455 const struct pid_entry *p = ptr;
2456 struct inode *inode;
2457 struct proc_inode *ei;
2458
2459 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2460 if (!inode)
2461 return ERR_PTR(-ENOENT);
2462
2463 ei = PROC_I(inode);
2464 if (S_ISDIR(inode->i_mode))
2465 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2466 if (p->iop)
2467 inode->i_op = p->iop;
2468 if (p->fop)
2469 inode->i_fop = p->fop;
2470 ei->op = p->op;
2471 pid_update_inode(task, inode);
2472 d_set_d_op(dentry, &pid_dentry_operations);
2473 return d_splice_alias(inode, dentry);
2474}
2475
2476static struct dentry *proc_pident_lookup(struct inode *dir,
2477 struct dentry *dentry,
2478 const struct pid_entry *p,
2479 const struct pid_entry *end)
2480{
2481 struct task_struct *task = get_proc_task(dir);
2482 struct dentry *res = ERR_PTR(-ENOENT);
2483
2484 if (!task)
2485 goto out_no_task;
2486
2487 /*
2488 * Yes, it does not scale. And it should not. Don't add
2489 * new entries into /proc/<tgid>/ without very good reasons.
2490 */
2491 for (; p < end; p++) {
2492 if (p->len != dentry->d_name.len)
2493 continue;
2494 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2495 res = proc_pident_instantiate(dentry, task, p);
2496 break;
2497 }
2498 }
2499 put_task_struct(task);
2500out_no_task:
2501 return res;
2502}
2503
2504static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2505 const struct pid_entry *ents, unsigned int nents)
2506{
2507 struct task_struct *task = get_proc_task(file_inode(file));
2508 const struct pid_entry *p;
2509
2510 if (!task)
2511 return -ENOENT;
2512
2513 if (!dir_emit_dots(file, ctx))
2514 goto out;
2515
2516 if (ctx->pos >= nents + 2)
2517 goto out;
2518
2519 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2520 if (!proc_fill_cache(file, ctx, p->name, p->len,
2521 proc_pident_instantiate, task, p))
2522 break;
2523 ctx->pos++;
2524 }
2525out:
2526 put_task_struct(task);
2527 return 0;
2528}
2529
2530#ifdef CONFIG_SECURITY
2531static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2532 size_t count, loff_t *ppos)
2533{
2534 struct inode * inode = file_inode(file);
2535 char *p = NULL;
2536 ssize_t length;
2537 struct task_struct *task = get_proc_task(inode);
2538
2539 if (!task)
2540 return -ESRCH;
2541
2542 length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2543 (char*)file->f_path.dentry->d_name.name,
2544 &p);
2545 put_task_struct(task);
2546 if (length > 0)
2547 length = simple_read_from_buffer(buf, count, ppos, p, length);
2548 kfree(p);
2549 return length;
2550}
2551
2552static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2553 size_t count, loff_t *ppos)
2554{
2555 struct inode * inode = file_inode(file);
2556 struct task_struct *task;
2557 void *page;
2558 int rv;
2559
2560 rcu_read_lock();
2561 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2562 if (!task) {
2563 rcu_read_unlock();
2564 return -ESRCH;
2565 }
2566 /* A task may only write its own attributes. */
2567 if (current != task) {
2568 rcu_read_unlock();
2569 return -EACCES;
2570 }
2571 /* Prevent changes to overridden credentials. */
2572 if (current_cred() != current_real_cred()) {
2573 rcu_read_unlock();
2574 return -EBUSY;
2575 }
2576 rcu_read_unlock();
2577
2578 if (count > PAGE_SIZE)
2579 count = PAGE_SIZE;
2580
2581 /* No partial writes. */
2582 if (*ppos != 0)
2583 return -EINVAL;
2584
2585 page = memdup_user(buf, count);
2586 if (IS_ERR(page)) {
2587 rv = PTR_ERR(page);
2588 goto out;
2589 }
2590
2591 /* Guard against adverse ptrace interaction */
2592 rv = mutex_lock_interruptible(¤t->signal->cred_guard_mutex);
2593 if (rv < 0)
2594 goto out_free;
2595
2596 rv = security_setprocattr(PROC_I(inode)->op.lsm,
2597 file->f_path.dentry->d_name.name, page,
2598 count);
2599 mutex_unlock(¤t->signal->cred_guard_mutex);
2600out_free:
2601 kfree(page);
2602out:
2603 return rv;
2604}
2605
2606static const struct file_operations proc_pid_attr_operations = {
2607 .read = proc_pid_attr_read,
2608 .write = proc_pid_attr_write,
2609 .llseek = generic_file_llseek,
2610};
2611
2612#define LSM_DIR_OPS(LSM) \
2613static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2614 struct dir_context *ctx) \
2615{ \
2616 return proc_pident_readdir(filp, ctx, \
2617 LSM##_attr_dir_stuff, \
2618 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2619} \
2620\
2621static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2622 .read = generic_read_dir, \
2623 .iterate = proc_##LSM##_attr_dir_iterate, \
2624 .llseek = default_llseek, \
2625}; \
2626\
2627static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2628 struct dentry *dentry, unsigned int flags) \
2629{ \
2630 return proc_pident_lookup(dir, dentry, \
2631 LSM##_attr_dir_stuff, \
2632 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2633} \
2634\
2635static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2636 .lookup = proc_##LSM##_attr_dir_lookup, \
2637 .getattr = pid_getattr, \
2638 .setattr = proc_setattr, \
2639}
2640
2641#ifdef CONFIG_SECURITY_SMACK
2642static const struct pid_entry smack_attr_dir_stuff[] = {
2643 ATTR("smack", "current", 0666),
2644};
2645LSM_DIR_OPS(smack);
2646#endif
2647
2648static const struct pid_entry attr_dir_stuff[] = {
2649 ATTR(NULL, "current", 0666),
2650 ATTR(NULL, "prev", 0444),
2651 ATTR(NULL, "exec", 0666),
2652 ATTR(NULL, "fscreate", 0666),
2653 ATTR(NULL, "keycreate", 0666),
2654 ATTR(NULL, "sockcreate", 0666),
2655#ifdef CONFIG_SECURITY_SMACK
2656 DIR("smack", 0555,
2657 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2658#endif
2659};
2660
2661static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2662{
2663 return proc_pident_readdir(file, ctx,
2664 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2665}
2666
2667static const struct file_operations proc_attr_dir_operations = {
2668 .read = generic_read_dir,
2669 .iterate_shared = proc_attr_dir_readdir,
2670 .llseek = generic_file_llseek,
2671};
2672
2673static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2674 struct dentry *dentry, unsigned int flags)
2675{
2676 return proc_pident_lookup(dir, dentry,
2677 attr_dir_stuff,
2678 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2679}
2680
2681static const struct inode_operations proc_attr_dir_inode_operations = {
2682 .lookup = proc_attr_dir_lookup,
2683 .getattr = pid_getattr,
2684 .setattr = proc_setattr,
2685};
2686
2687#endif
2688
2689#ifdef CONFIG_ELF_CORE
2690static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2691 size_t count, loff_t *ppos)
2692{
2693 struct task_struct *task = get_proc_task(file_inode(file));
2694 struct mm_struct *mm;
2695 char buffer[PROC_NUMBUF];
2696 size_t len;
2697 int ret;
2698
2699 if (!task)
2700 return -ESRCH;
2701
2702 ret = 0;
2703 mm = get_task_mm(task);
2704 if (mm) {
2705 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2706 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2707 MMF_DUMP_FILTER_SHIFT));
2708 mmput(mm);
2709 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2710 }
2711
2712 put_task_struct(task);
2713
2714 return ret;
2715}
2716
2717static ssize_t proc_coredump_filter_write(struct file *file,
2718 const char __user *buf,
2719 size_t count,
2720 loff_t *ppos)
2721{
2722 struct task_struct *task;
2723 struct mm_struct *mm;
2724 unsigned int val;
2725 int ret;
2726 int i;
2727 unsigned long mask;
2728
2729 ret = kstrtouint_from_user(buf, count, 0, &val);
2730 if (ret < 0)
2731 return ret;
2732
2733 ret = -ESRCH;
2734 task = get_proc_task(file_inode(file));
2735 if (!task)
2736 goto out_no_task;
2737
2738 mm = get_task_mm(task);
2739 if (!mm)
2740 goto out_no_mm;
2741 ret = 0;
2742
2743 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2744 if (val & mask)
2745 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2746 else
2747 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2748 }
2749
2750 mmput(mm);
2751 out_no_mm:
2752 put_task_struct(task);
2753 out_no_task:
2754 if (ret < 0)
2755 return ret;
2756 return count;
2757}
2758
2759static const struct file_operations proc_coredump_filter_operations = {
2760 .read = proc_coredump_filter_read,
2761 .write = proc_coredump_filter_write,
2762 .llseek = generic_file_llseek,
2763};
2764#endif
2765
2766#ifdef CONFIG_TASK_IO_ACCOUNTING
2767static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2768{
2769 struct task_io_accounting acct = task->ioac;
2770 unsigned long flags;
2771 int result;
2772
2773 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2774 if (result)
2775 return result;
2776
2777 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2778 result = -EACCES;
2779 goto out_unlock;
2780 }
2781
2782 if (whole && lock_task_sighand(task, &flags)) {
2783 struct task_struct *t = task;
2784
2785 task_io_accounting_add(&acct, &task->signal->ioac);
2786 while_each_thread(task, t)
2787 task_io_accounting_add(&acct, &t->ioac);
2788
2789 unlock_task_sighand(task, &flags);
2790 }
2791 seq_printf(m,
2792 "rchar: %llu\n"
2793 "wchar: %llu\n"
2794 "syscr: %llu\n"
2795 "syscw: %llu\n"
2796 "read_bytes: %llu\n"
2797 "write_bytes: %llu\n"
2798 "cancelled_write_bytes: %llu\n",
2799 (unsigned long long)acct.rchar,
2800 (unsigned long long)acct.wchar,
2801 (unsigned long long)acct.syscr,
2802 (unsigned long long)acct.syscw,
2803 (unsigned long long)acct.read_bytes,
2804 (unsigned long long)acct.write_bytes,
2805 (unsigned long long)acct.cancelled_write_bytes);
2806 result = 0;
2807
2808out_unlock:
2809 mutex_unlock(&task->signal->cred_guard_mutex);
2810 return result;
2811}
2812
2813static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2814 struct pid *pid, struct task_struct *task)
2815{
2816 return do_io_accounting(task, m, 0);
2817}
2818
2819static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2820 struct pid *pid, struct task_struct *task)
2821{
2822 return do_io_accounting(task, m, 1);
2823}
2824#endif /* CONFIG_TASK_IO_ACCOUNTING */
2825
2826#ifdef CONFIG_USER_NS
2827static int proc_id_map_open(struct inode *inode, struct file *file,
2828 const struct seq_operations *seq_ops)
2829{
2830 struct user_namespace *ns = NULL;
2831 struct task_struct *task;
2832 struct seq_file *seq;
2833 int ret = -EINVAL;
2834
2835 task = get_proc_task(inode);
2836 if (task) {
2837 rcu_read_lock();
2838 ns = get_user_ns(task_cred_xxx(task, user_ns));
2839 rcu_read_unlock();
2840 put_task_struct(task);
2841 }
2842 if (!ns)
2843 goto err;
2844
2845 ret = seq_open(file, seq_ops);
2846 if (ret)
2847 goto err_put_ns;
2848
2849 seq = file->private_data;
2850 seq->private = ns;
2851
2852 return 0;
2853err_put_ns:
2854 put_user_ns(ns);
2855err:
2856 return ret;
2857}
2858
2859static int proc_id_map_release(struct inode *inode, struct file *file)
2860{
2861 struct seq_file *seq = file->private_data;
2862 struct user_namespace *ns = seq->private;
2863 put_user_ns(ns);
2864 return seq_release(inode, file);
2865}
2866
2867static int proc_uid_map_open(struct inode *inode, struct file *file)
2868{
2869 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2870}
2871
2872static int proc_gid_map_open(struct inode *inode, struct file *file)
2873{
2874 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2875}
2876
2877static int proc_projid_map_open(struct inode *inode, struct file *file)
2878{
2879 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2880}
2881
2882static const struct file_operations proc_uid_map_operations = {
2883 .open = proc_uid_map_open,
2884 .write = proc_uid_map_write,
2885 .read = seq_read,
2886 .llseek = seq_lseek,
2887 .release = proc_id_map_release,
2888};
2889
2890static const struct file_operations proc_gid_map_operations = {
2891 .open = proc_gid_map_open,
2892 .write = proc_gid_map_write,
2893 .read = seq_read,
2894 .llseek = seq_lseek,
2895 .release = proc_id_map_release,
2896};
2897
2898static const struct file_operations proc_projid_map_operations = {
2899 .open = proc_projid_map_open,
2900 .write = proc_projid_map_write,
2901 .read = seq_read,
2902 .llseek = seq_lseek,
2903 .release = proc_id_map_release,
2904};
2905
2906static int proc_setgroups_open(struct inode *inode, struct file *file)
2907{
2908 struct user_namespace *ns = NULL;
2909 struct task_struct *task;
2910 int ret;
2911
2912 ret = -ESRCH;
2913 task = get_proc_task(inode);
2914 if (task) {
2915 rcu_read_lock();
2916 ns = get_user_ns(task_cred_xxx(task, user_ns));
2917 rcu_read_unlock();
2918 put_task_struct(task);
2919 }
2920 if (!ns)
2921 goto err;
2922
2923 if (file->f_mode & FMODE_WRITE) {
2924 ret = -EACCES;
2925 if (!ns_capable(ns, CAP_SYS_ADMIN))
2926 goto err_put_ns;
2927 }
2928
2929 ret = single_open(file, &proc_setgroups_show, ns);
2930 if (ret)
2931 goto err_put_ns;
2932
2933 return 0;
2934err_put_ns:
2935 put_user_ns(ns);
2936err:
2937 return ret;
2938}
2939
2940static int proc_setgroups_release(struct inode *inode, struct file *file)
2941{
2942 struct seq_file *seq = file->private_data;
2943 struct user_namespace *ns = seq->private;
2944 int ret = single_release(inode, file);
2945 put_user_ns(ns);
2946 return ret;
2947}
2948
2949static const struct file_operations proc_setgroups_operations = {
2950 .open = proc_setgroups_open,
2951 .write = proc_setgroups_write,
2952 .read = seq_read,
2953 .llseek = seq_lseek,
2954 .release = proc_setgroups_release,
2955};
2956#endif /* CONFIG_USER_NS */
2957
2958static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2959 struct pid *pid, struct task_struct *task)
2960{
2961 int err = lock_trace(task);
2962 if (!err) {
2963 seq_printf(m, "%08x\n", task->personality);
2964 unlock_trace(task);
2965 }
2966 return err;
2967}
2968
2969#ifdef CONFIG_LIVEPATCH
2970static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2971 struct pid *pid, struct task_struct *task)
2972{
2973 seq_printf(m, "%d\n", task->patch_state);
2974 return 0;
2975}
2976#endif /* CONFIG_LIVEPATCH */
2977
2978#ifdef CONFIG_STACKLEAK_METRICS
2979static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
2980 struct pid *pid, struct task_struct *task)
2981{
2982 unsigned long prev_depth = THREAD_SIZE -
2983 (task->prev_lowest_stack & (THREAD_SIZE - 1));
2984 unsigned long depth = THREAD_SIZE -
2985 (task->lowest_stack & (THREAD_SIZE - 1));
2986
2987 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
2988 prev_depth, depth);
2989 return 0;
2990}
2991#endif /* CONFIG_STACKLEAK_METRICS */
2992
2993/*
2994 * Thread groups
2995 */
2996static const struct file_operations proc_task_operations;
2997static const struct inode_operations proc_task_inode_operations;
2998
2999static const struct pid_entry tgid_base_stuff[] = {
3000 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3001 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3002 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3003 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3004 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3005#ifdef CONFIG_NET
3006 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3007#endif
3008 REG("environ", S_IRUSR, proc_environ_operations),
3009 REG("auxv", S_IRUSR, proc_auxv_operations),
3010 ONE("status", S_IRUGO, proc_pid_status),
3011 ONE("personality", S_IRUSR, proc_pid_personality),
3012 ONE("limits", S_IRUGO, proc_pid_limits),
3013#ifdef CONFIG_SCHED_DEBUG
3014 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3015#endif
3016#ifdef CONFIG_SCHED_AUTOGROUP
3017 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3018#endif
3019 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3020#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3021 ONE("syscall", S_IRUSR, proc_pid_syscall),
3022#endif
3023 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3024 ONE("stat", S_IRUGO, proc_tgid_stat),
3025 ONE("statm", S_IRUGO, proc_pid_statm),
3026 REG("maps", S_IRUGO, proc_pid_maps_operations),
3027#ifdef CONFIG_NUMA
3028 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3029#endif
3030 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3031 LNK("cwd", proc_cwd_link),
3032 LNK("root", proc_root_link),
3033 LNK("exe", proc_exe_link),
3034 REG("mounts", S_IRUGO, proc_mounts_operations),
3035 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3036 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3037#ifdef CONFIG_PROC_PAGE_MONITOR
3038 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3039 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3040 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3041 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3042#endif
3043#ifdef CONFIG_SECURITY
3044 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3045#endif
3046#ifdef CONFIG_KALLSYMS
3047 ONE("wchan", S_IRUGO, proc_pid_wchan),
3048#endif
3049#ifdef CONFIG_STACKTRACE
3050 ONE("stack", S_IRUSR, proc_pid_stack),
3051#endif
3052#ifdef CONFIG_SCHED_INFO
3053 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3054#endif
3055#ifdef CONFIG_LATENCYTOP
3056 REG("latency", S_IRUGO, proc_lstats_operations),
3057#endif
3058#ifdef CONFIG_PROC_PID_CPUSET
3059 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3060#endif
3061#ifdef CONFIG_CGROUPS
3062 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3063#endif
3064 ONE("oom_score", S_IRUGO, proc_oom_score),
3065 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3066 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3067#ifdef CONFIG_AUDIT
3068 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3069 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3070#endif
3071#ifdef CONFIG_FAULT_INJECTION
3072 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3073 REG("fail-nth", 0644, proc_fail_nth_operations),
3074#endif
3075#ifdef CONFIG_ELF_CORE
3076 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3077#endif
3078#ifdef CONFIG_TASK_IO_ACCOUNTING
3079 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3080#endif
3081#ifdef CONFIG_USER_NS
3082 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3083 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3084 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3085 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3086#endif
3087#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3088 REG("timers", S_IRUGO, proc_timers_operations),
3089#endif
3090 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3091#ifdef CONFIG_LIVEPATCH
3092 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3093#endif
3094#ifdef CONFIG_STACKLEAK_METRICS
3095 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3096#endif
3097#ifdef CONFIG_PROC_PID_ARCH_STATUS
3098 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3099#endif
3100};
3101
3102static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3103{
3104 return proc_pident_readdir(file, ctx,
3105 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3106}
3107
3108static const struct file_operations proc_tgid_base_operations = {
3109 .read = generic_read_dir,
3110 .iterate_shared = proc_tgid_base_readdir,
3111 .llseek = generic_file_llseek,
3112};
3113
3114struct pid *tgid_pidfd_to_pid(const struct file *file)
3115{
3116 if (file->f_op != &proc_tgid_base_operations)
3117 return ERR_PTR(-EBADF);
3118
3119 return proc_pid(file_inode(file));
3120}
3121
3122static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3123{
3124 return proc_pident_lookup(dir, dentry,
3125 tgid_base_stuff,
3126 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3127}
3128
3129static const struct inode_operations proc_tgid_base_inode_operations = {
3130 .lookup = proc_tgid_base_lookup,
3131 .getattr = pid_getattr,
3132 .setattr = proc_setattr,
3133 .permission = proc_pid_permission,
3134};
3135
3136static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3137{
3138 struct dentry *dentry, *leader, *dir;
3139 char buf[10 + 1];
3140 struct qstr name;
3141
3142 name.name = buf;
3143 name.len = snprintf(buf, sizeof(buf), "%u", pid);
3144 /* no ->d_hash() rejects on procfs */
3145 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3146 if (dentry) {
3147 d_invalidate(dentry);
3148 dput(dentry);
3149 }
3150
3151 if (pid == tgid)
3152 return;
3153
3154 name.name = buf;
3155 name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3156 leader = d_hash_and_lookup(mnt->mnt_root, &name);
3157 if (!leader)
3158 goto out;
3159
3160 name.name = "task";
3161 name.len = strlen(name.name);
3162 dir = d_hash_and_lookup(leader, &name);
3163 if (!dir)
3164 goto out_put_leader;
3165
3166 name.name = buf;
3167 name.len = snprintf(buf, sizeof(buf), "%u", pid);
3168 dentry = d_hash_and_lookup(dir, &name);
3169 if (dentry) {
3170 d_invalidate(dentry);
3171 dput(dentry);
3172 }
3173
3174 dput(dir);
3175out_put_leader:
3176 dput(leader);
3177out:
3178 return;
3179}
3180
3181/**
3182 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3183 * @task: task that should be flushed.
3184 *
3185 * When flushing dentries from proc, one needs to flush them from global
3186 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3187 * in. This call is supposed to do all of this job.
3188 *
3189 * Looks in the dcache for
3190 * /proc/@pid
3191 * /proc/@tgid/task/@pid
3192 * if either directory is present flushes it and all of it'ts children
3193 * from the dcache.
3194 *
3195 * It is safe and reasonable to cache /proc entries for a task until
3196 * that task exits. After that they just clog up the dcache with
3197 * useless entries, possibly causing useful dcache entries to be
3198 * flushed instead. This routine is proved to flush those useless
3199 * dcache entries at process exit time.
3200 *
3201 * NOTE: This routine is just an optimization so it does not guarantee
3202 * that no dcache entries will exist at process exit time it
3203 * just makes it very unlikely that any will persist.
3204 */
3205
3206void proc_flush_task(struct task_struct *task)
3207{
3208 int i;
3209 struct pid *pid, *tgid;
3210 struct upid *upid;
3211
3212 pid = task_pid(task);
3213 tgid = task_tgid(task);
3214
3215 for (i = 0; i <= pid->level; i++) {
3216 upid = &pid->numbers[i];
3217 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3218 tgid->numbers[i].nr);
3219 }
3220}
3221
3222static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3223 struct task_struct *task, const void *ptr)
3224{
3225 struct inode *inode;
3226
3227 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3228 if (!inode)
3229 return ERR_PTR(-ENOENT);
3230
3231 inode->i_op = &proc_tgid_base_inode_operations;
3232 inode->i_fop = &proc_tgid_base_operations;
3233 inode->i_flags|=S_IMMUTABLE;
3234
3235 set_nlink(inode, nlink_tgid);
3236 pid_update_inode(task, inode);
3237
3238 d_set_d_op(dentry, &pid_dentry_operations);
3239 return d_splice_alias(inode, dentry);
3240}
3241
3242struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3243{
3244 struct task_struct *task;
3245 unsigned tgid;
3246 struct pid_namespace *ns;
3247 struct dentry *result = ERR_PTR(-ENOENT);
3248
3249 tgid = name_to_int(&dentry->d_name);
3250 if (tgid == ~0U)
3251 goto out;
3252
3253 ns = dentry->d_sb->s_fs_info;
3254 rcu_read_lock();
3255 task = find_task_by_pid_ns(tgid, ns);
3256 if (task)
3257 get_task_struct(task);
3258 rcu_read_unlock();
3259 if (!task)
3260 goto out;
3261
3262 result = proc_pid_instantiate(dentry, task, NULL);
3263 put_task_struct(task);
3264out:
3265 return result;
3266}
3267
3268/*
3269 * Find the first task with tgid >= tgid
3270 *
3271 */
3272struct tgid_iter {
3273 unsigned int tgid;
3274 struct task_struct *task;
3275};
3276static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3277{
3278 struct pid *pid;
3279
3280 if (iter.task)
3281 put_task_struct(iter.task);
3282 rcu_read_lock();
3283retry:
3284 iter.task = NULL;
3285 pid = find_ge_pid(iter.tgid, ns);
3286 if (pid) {
3287 iter.tgid = pid_nr_ns(pid, ns);
3288 iter.task = pid_task(pid, PIDTYPE_PID);
3289 /* What we to know is if the pid we have find is the
3290 * pid of a thread_group_leader. Testing for task
3291 * being a thread_group_leader is the obvious thing
3292 * todo but there is a window when it fails, due to
3293 * the pid transfer logic in de_thread.
3294 *
3295 * So we perform the straight forward test of seeing
3296 * if the pid we have found is the pid of a thread
3297 * group leader, and don't worry if the task we have
3298 * found doesn't happen to be a thread group leader.
3299 * As we don't care in the case of readdir.
3300 */
3301 if (!iter.task || !has_group_leader_pid(iter.task)) {
3302 iter.tgid += 1;
3303 goto retry;
3304 }
3305 get_task_struct(iter.task);
3306 }
3307 rcu_read_unlock();
3308 return iter;
3309}
3310
3311#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3312
3313/* for the /proc/ directory itself, after non-process stuff has been done */
3314int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3315{
3316 struct tgid_iter iter;
3317 struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3318 loff_t pos = ctx->pos;
3319
3320 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3321 return 0;
3322
3323 if (pos == TGID_OFFSET - 2) {
3324 struct inode *inode = d_inode(ns->proc_self);
3325 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3326 return 0;
3327 ctx->pos = pos = pos + 1;
3328 }
3329 if (pos == TGID_OFFSET - 1) {
3330 struct inode *inode = d_inode(ns->proc_thread_self);
3331 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3332 return 0;
3333 ctx->pos = pos = pos + 1;
3334 }
3335 iter.tgid = pos - TGID_OFFSET;
3336 iter.task = NULL;
3337 for (iter = next_tgid(ns, iter);
3338 iter.task;
3339 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3340 char name[10 + 1];
3341 unsigned int len;
3342
3343 cond_resched();
3344 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3345 continue;
3346
3347 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3348 ctx->pos = iter.tgid + TGID_OFFSET;
3349 if (!proc_fill_cache(file, ctx, name, len,
3350 proc_pid_instantiate, iter.task, NULL)) {
3351 put_task_struct(iter.task);
3352 return 0;
3353 }
3354 }
3355 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3356 return 0;
3357}
3358
3359/*
3360 * proc_tid_comm_permission is a special permission function exclusively
3361 * used for the node /proc/<pid>/task/<tid>/comm.
3362 * It bypasses generic permission checks in the case where a task of the same
3363 * task group attempts to access the node.
3364 * The rationale behind this is that glibc and bionic access this node for
3365 * cross thread naming (pthread_set/getname_np(!self)). However, if
3366 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3367 * which locks out the cross thread naming implementation.
3368 * This function makes sure that the node is always accessible for members of
3369 * same thread group.
3370 */
3371static int proc_tid_comm_permission(struct inode *inode, int mask)
3372{
3373 bool is_same_tgroup;
3374 struct task_struct *task;
3375
3376 task = get_proc_task(inode);
3377 if (!task)
3378 return -ESRCH;
3379 is_same_tgroup = same_thread_group(current, task);
3380 put_task_struct(task);
3381
3382 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3383 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3384 * read or written by the members of the corresponding
3385 * thread group.
3386 */
3387 return 0;
3388 }
3389
3390 return generic_permission(inode, mask);
3391}
3392
3393static const struct inode_operations proc_tid_comm_inode_operations = {
3394 .permission = proc_tid_comm_permission,
3395};
3396
3397/*
3398 * Tasks
3399 */
3400static const struct pid_entry tid_base_stuff[] = {
3401 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3402 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3403 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3404#ifdef CONFIG_NET
3405 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3406#endif
3407 REG("environ", S_IRUSR, proc_environ_operations),
3408 REG("auxv", S_IRUSR, proc_auxv_operations),
3409 ONE("status", S_IRUGO, proc_pid_status),
3410 ONE("personality", S_IRUSR, proc_pid_personality),
3411 ONE("limits", S_IRUGO, proc_pid_limits),
3412#ifdef CONFIG_SCHED_DEBUG
3413 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3414#endif
3415 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3416 &proc_tid_comm_inode_operations,
3417 &proc_pid_set_comm_operations, {}),
3418#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3419 ONE("syscall", S_IRUSR, proc_pid_syscall),
3420#endif
3421 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3422 ONE("stat", S_IRUGO, proc_tid_stat),
3423 ONE("statm", S_IRUGO, proc_pid_statm),
3424 REG("maps", S_IRUGO, proc_pid_maps_operations),
3425#ifdef CONFIG_PROC_CHILDREN
3426 REG("children", S_IRUGO, proc_tid_children_operations),
3427#endif
3428#ifdef CONFIG_NUMA
3429 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3430#endif
3431 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3432 LNK("cwd", proc_cwd_link),
3433 LNK("root", proc_root_link),
3434 LNK("exe", proc_exe_link),
3435 REG("mounts", S_IRUGO, proc_mounts_operations),
3436 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3437#ifdef CONFIG_PROC_PAGE_MONITOR
3438 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3439 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3440 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3441 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3442#endif
3443#ifdef CONFIG_SECURITY
3444 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3445#endif
3446#ifdef CONFIG_KALLSYMS
3447 ONE("wchan", S_IRUGO, proc_pid_wchan),
3448#endif
3449#ifdef CONFIG_STACKTRACE
3450 ONE("stack", S_IRUSR, proc_pid_stack),
3451#endif
3452#ifdef CONFIG_SCHED_INFO
3453 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3454#endif
3455#ifdef CONFIG_LATENCYTOP
3456 REG("latency", S_IRUGO, proc_lstats_operations),
3457#endif
3458#ifdef CONFIG_PROC_PID_CPUSET
3459 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3460#endif
3461#ifdef CONFIG_CGROUPS
3462 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3463#endif
3464 ONE("oom_score", S_IRUGO, proc_oom_score),
3465 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3466 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3467#ifdef CONFIG_AUDIT
3468 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3469 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3470#endif
3471#ifdef CONFIG_FAULT_INJECTION
3472 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3473 REG("fail-nth", 0644, proc_fail_nth_operations),
3474#endif
3475#ifdef CONFIG_TASK_IO_ACCOUNTING
3476 ONE("io", S_IRUSR, proc_tid_io_accounting),
3477#endif
3478#ifdef CONFIG_USER_NS
3479 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3480 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3481 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3482 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3483#endif
3484#ifdef CONFIG_LIVEPATCH
3485 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3486#endif
3487#ifdef CONFIG_PROC_PID_ARCH_STATUS
3488 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3489#endif
3490};
3491
3492static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3493{
3494 return proc_pident_readdir(file, ctx,
3495 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3496}
3497
3498static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3499{
3500 return proc_pident_lookup(dir, dentry,
3501 tid_base_stuff,
3502 tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3503}
3504
3505static const struct file_operations proc_tid_base_operations = {
3506 .read = generic_read_dir,
3507 .iterate_shared = proc_tid_base_readdir,
3508 .llseek = generic_file_llseek,
3509};
3510
3511static const struct inode_operations proc_tid_base_inode_operations = {
3512 .lookup = proc_tid_base_lookup,
3513 .getattr = pid_getattr,
3514 .setattr = proc_setattr,
3515};
3516
3517static struct dentry *proc_task_instantiate(struct dentry *dentry,
3518 struct task_struct *task, const void *ptr)
3519{
3520 struct inode *inode;
3521 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3522 if (!inode)
3523 return ERR_PTR(-ENOENT);
3524
3525 inode->i_op = &proc_tid_base_inode_operations;
3526 inode->i_fop = &proc_tid_base_operations;
3527 inode->i_flags |= S_IMMUTABLE;
3528
3529 set_nlink(inode, nlink_tid);
3530 pid_update_inode(task, inode);
3531
3532 d_set_d_op(dentry, &pid_dentry_operations);
3533 return d_splice_alias(inode, dentry);
3534}
3535
3536static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3537{
3538 struct task_struct *task;
3539 struct task_struct *leader = get_proc_task(dir);
3540 unsigned tid;
3541 struct pid_namespace *ns;
3542 struct dentry *result = ERR_PTR(-ENOENT);
3543
3544 if (!leader)
3545 goto out_no_task;
3546
3547 tid = name_to_int(&dentry->d_name);
3548 if (tid == ~0U)
3549 goto out;
3550
3551 ns = dentry->d_sb->s_fs_info;
3552 rcu_read_lock();
3553 task = find_task_by_pid_ns(tid, ns);
3554 if (task)
3555 get_task_struct(task);
3556 rcu_read_unlock();
3557 if (!task)
3558 goto out;
3559 if (!same_thread_group(leader, task))
3560 goto out_drop_task;
3561
3562 result = proc_task_instantiate(dentry, task, NULL);
3563out_drop_task:
3564 put_task_struct(task);
3565out:
3566 put_task_struct(leader);
3567out_no_task:
3568 return result;
3569}
3570
3571/*
3572 * Find the first tid of a thread group to return to user space.
3573 *
3574 * Usually this is just the thread group leader, but if the users
3575 * buffer was too small or there was a seek into the middle of the
3576 * directory we have more work todo.
3577 *
3578 * In the case of a short read we start with find_task_by_pid.
3579 *
3580 * In the case of a seek we start with the leader and walk nr
3581 * threads past it.
3582 */
3583static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3584 struct pid_namespace *ns)
3585{
3586 struct task_struct *pos, *task;
3587 unsigned long nr = f_pos;
3588
3589 if (nr != f_pos) /* 32bit overflow? */
3590 return NULL;
3591
3592 rcu_read_lock();
3593 task = pid_task(pid, PIDTYPE_PID);
3594 if (!task)
3595 goto fail;
3596
3597 /* Attempt to start with the tid of a thread */
3598 if (tid && nr) {
3599 pos = find_task_by_pid_ns(tid, ns);
3600 if (pos && same_thread_group(pos, task))
3601 goto found;
3602 }
3603
3604 /* If nr exceeds the number of threads there is nothing todo */
3605 if (nr >= get_nr_threads(task))
3606 goto fail;
3607
3608 /* If we haven't found our starting place yet start
3609 * with the leader and walk nr threads forward.
3610 */
3611 pos = task = task->group_leader;
3612 do {
3613 if (!nr--)
3614 goto found;
3615 } while_each_thread(task, pos);
3616fail:
3617 pos = NULL;
3618 goto out;
3619found:
3620 get_task_struct(pos);
3621out:
3622 rcu_read_unlock();
3623 return pos;
3624}
3625
3626/*
3627 * Find the next thread in the thread list.
3628 * Return NULL if there is an error or no next thread.
3629 *
3630 * The reference to the input task_struct is released.
3631 */
3632static struct task_struct *next_tid(struct task_struct *start)
3633{
3634 struct task_struct *pos = NULL;
3635 rcu_read_lock();
3636 if (pid_alive(start)) {
3637 pos = next_thread(start);
3638 if (thread_group_leader(pos))
3639 pos = NULL;
3640 else
3641 get_task_struct(pos);
3642 }
3643 rcu_read_unlock();
3644 put_task_struct(start);
3645 return pos;
3646}
3647
3648/* for the /proc/TGID/task/ directories */
3649static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3650{
3651 struct inode *inode = file_inode(file);
3652 struct task_struct *task;
3653 struct pid_namespace *ns;
3654 int tid;
3655
3656 if (proc_inode_is_dead(inode))
3657 return -ENOENT;
3658
3659 if (!dir_emit_dots(file, ctx))
3660 return 0;
3661
3662 /* f_version caches the tgid value that the last readdir call couldn't
3663 * return. lseek aka telldir automagically resets f_version to 0.
3664 */
3665 ns = proc_pid_ns(inode);
3666 tid = (int)file->f_version;
3667 file->f_version = 0;
3668 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3669 task;
3670 task = next_tid(task), ctx->pos++) {
3671 char name[10 + 1];
3672 unsigned int len;
3673 tid = task_pid_nr_ns(task, ns);
3674 len = snprintf(name, sizeof(name), "%u", tid);
3675 if (!proc_fill_cache(file, ctx, name, len,
3676 proc_task_instantiate, task, NULL)) {
3677 /* returning this tgid failed, save it as the first
3678 * pid for the next readir call */
3679 file->f_version = (u64)tid;
3680 put_task_struct(task);
3681 break;
3682 }
3683 }
3684
3685 return 0;
3686}
3687
3688static int proc_task_getattr(const struct path *path, struct kstat *stat,
3689 u32 request_mask, unsigned int query_flags)
3690{
3691 struct inode *inode = d_inode(path->dentry);
3692 struct task_struct *p = get_proc_task(inode);
3693 generic_fillattr(inode, stat);
3694
3695 if (p) {
3696 stat->nlink += get_nr_threads(p);
3697 put_task_struct(p);
3698 }
3699
3700 return 0;
3701}
3702
3703static const struct inode_operations proc_task_inode_operations = {
3704 .lookup = proc_task_lookup,
3705 .getattr = proc_task_getattr,
3706 .setattr = proc_setattr,
3707 .permission = proc_pid_permission,
3708};
3709
3710static const struct file_operations proc_task_operations = {
3711 .read = generic_read_dir,
3712 .iterate_shared = proc_task_readdir,
3713 .llseek = generic_file_llseek,
3714};
3715
3716void __init set_proc_pid_nlink(void)
3717{
3718 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3719 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3720}