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