Loading...
1/*
2 * linux/fs/proc/base.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * proc base directory handling functions
7 *
8 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 * Instead of using magical inumbers to determine the kind of object
10 * we allocate and fill in-core inodes upon lookup. They don't even
11 * go into icache. We cache the reference to task_struct upon lookup too.
12 * Eventually it should become a filesystem in its own. We don't use the
13 * rest of procfs anymore.
14 *
15 *
16 * Changelog:
17 * 17-Jan-2005
18 * Allan Bezerra
19 * Bruna Moreira <bruna.moreira@indt.org.br>
20 * Edjard Mota <edjard.mota@indt.org.br>
21 * Ilias Biris <ilias.biris@indt.org.br>
22 * Mauricio Lin <mauricio.lin@indt.org.br>
23 *
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25 *
26 * A new process specific entry (smaps) included in /proc. It shows the
27 * size of rss for each memory area. The maps entry lacks information
28 * about physical memory size (rss) for each mapped file, i.e.,
29 * rss information for executables and library files.
30 * This additional information is useful for any tools that need to know
31 * about physical memory consumption for a process specific library.
32 *
33 * Changelog:
34 * 21-Feb-2005
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
37 *
38 * ChangeLog:
39 * 10-Mar-2005
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
42 *
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
45 *
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
48 */
49
50#include <asm/uaccess.h>
51
52#include <linux/errno.h>
53#include <linux/time.h>
54#include <linux/proc_fs.h>
55#include <linux/stat.h>
56#include <linux/task_io_accounting_ops.h>
57#include <linux/init.h>
58#include <linux/capability.h>
59#include <linux/file.h>
60#include <linux/fdtable.h>
61#include <linux/string.h>
62#include <linux/seq_file.h>
63#include <linux/namei.h>
64#include <linux/mnt_namespace.h>
65#include <linux/mm.h>
66#include <linux/swap.h>
67#include <linux/rcupdate.h>
68#include <linux/kallsyms.h>
69#include <linux/stacktrace.h>
70#include <linux/resource.h>
71#include <linux/module.h>
72#include <linux/mount.h>
73#include <linux/security.h>
74#include <linux/ptrace.h>
75#include <linux/tracehook.h>
76#include <linux/cgroup.h>
77#include <linux/cpuset.h>
78#include <linux/audit.h>
79#include <linux/poll.h>
80#include <linux/nsproxy.h>
81#include <linux/oom.h>
82#include <linux/elf.h>
83#include <linux/pid_namespace.h>
84#include <linux/user_namespace.h>
85#include <linux/fs_struct.h>
86#include <linux/slab.h>
87#include <linux/flex_array.h>
88#ifdef CONFIG_HARDWALL
89#include <asm/hardwall.h>
90#endif
91#include <trace/events/oom.h>
92#include "internal.h"
93
94/* NOTE:
95 * Implementing inode permission operations in /proc is almost
96 * certainly an error. Permission checks need to happen during
97 * each system call not at open time. The reason is that most of
98 * what we wish to check for permissions in /proc varies at runtime.
99 *
100 * The classic example of a problem is opening file descriptors
101 * in /proc for a task before it execs a suid executable.
102 */
103
104struct pid_entry {
105 char *name;
106 int len;
107 umode_t mode;
108 const struct inode_operations *iop;
109 const struct file_operations *fop;
110 union proc_op op;
111};
112
113#define NOD(NAME, MODE, IOP, FOP, OP) { \
114 .name = (NAME), \
115 .len = sizeof(NAME) - 1, \
116 .mode = MODE, \
117 .iop = IOP, \
118 .fop = FOP, \
119 .op = OP, \
120}
121
122#define DIR(NAME, MODE, iops, fops) \
123 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
124#define LNK(NAME, get_link) \
125 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
126 &proc_pid_link_inode_operations, NULL, \
127 { .proc_get_link = get_link } )
128#define REG(NAME, MODE, fops) \
129 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
130#define INF(NAME, MODE, read) \
131 NOD(NAME, (S_IFREG|(MODE)), \
132 NULL, &proc_info_file_operations, \
133 { .proc_read = read } )
134#define ONE(NAME, MODE, show) \
135 NOD(NAME, (S_IFREG|(MODE)), \
136 NULL, &proc_single_file_operations, \
137 { .proc_show = show } )
138
139static int proc_fd_permission(struct inode *inode, int mask);
140
141/*
142 * Count the number of hardlinks for the pid_entry table, excluding the .
143 * and .. links.
144 */
145static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
146 unsigned int n)
147{
148 unsigned int i;
149 unsigned int count;
150
151 count = 0;
152 for (i = 0; i < n; ++i) {
153 if (S_ISDIR(entries[i].mode))
154 ++count;
155 }
156
157 return count;
158}
159
160static int get_task_root(struct task_struct *task, struct path *root)
161{
162 int result = -ENOENT;
163
164 task_lock(task);
165 if (task->fs) {
166 get_fs_root(task->fs, root);
167 result = 0;
168 }
169 task_unlock(task);
170 return result;
171}
172
173static int proc_cwd_link(struct dentry *dentry, struct path *path)
174{
175 struct task_struct *task = get_proc_task(dentry->d_inode);
176 int result = -ENOENT;
177
178 if (task) {
179 task_lock(task);
180 if (task->fs) {
181 get_fs_pwd(task->fs, path);
182 result = 0;
183 }
184 task_unlock(task);
185 put_task_struct(task);
186 }
187 return result;
188}
189
190static int proc_root_link(struct dentry *dentry, struct path *path)
191{
192 struct task_struct *task = get_proc_task(dentry->d_inode);
193 int result = -ENOENT;
194
195 if (task) {
196 result = get_task_root(task, path);
197 put_task_struct(task);
198 }
199 return result;
200}
201
202static int proc_pid_cmdline(struct task_struct *task, char * buffer)
203{
204 int res = 0;
205 unsigned int len;
206 struct mm_struct *mm = get_task_mm(task);
207 if (!mm)
208 goto out;
209 if (!mm->arg_end)
210 goto out_mm; /* Shh! No looking before we're done */
211
212 len = mm->arg_end - mm->arg_start;
213
214 if (len > PAGE_SIZE)
215 len = PAGE_SIZE;
216
217 res = access_process_vm(task, mm->arg_start, buffer, len, 0);
218
219 // If the nul at the end of args has been overwritten, then
220 // assume application is using setproctitle(3).
221 if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
222 len = strnlen(buffer, res);
223 if (len < res) {
224 res = len;
225 } else {
226 len = mm->env_end - mm->env_start;
227 if (len > PAGE_SIZE - res)
228 len = PAGE_SIZE - res;
229 res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
230 res = strnlen(buffer, res);
231 }
232 }
233out_mm:
234 mmput(mm);
235out:
236 return res;
237}
238
239static int proc_pid_auxv(struct task_struct *task, char *buffer)
240{
241 struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
242 int res = PTR_ERR(mm);
243 if (mm && !IS_ERR(mm)) {
244 unsigned int nwords = 0;
245 do {
246 nwords += 2;
247 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
248 res = nwords * sizeof(mm->saved_auxv[0]);
249 if (res > PAGE_SIZE)
250 res = PAGE_SIZE;
251 memcpy(buffer, mm->saved_auxv, res);
252 mmput(mm);
253 }
254 return res;
255}
256
257
258#ifdef CONFIG_KALLSYMS
259/*
260 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
261 * Returns the resolved symbol. If that fails, simply return the address.
262 */
263static int proc_pid_wchan(struct task_struct *task, char *buffer)
264{
265 unsigned long wchan;
266 char symname[KSYM_NAME_LEN];
267
268 wchan = get_wchan(task);
269
270 if (lookup_symbol_name(wchan, symname) < 0)
271 if (!ptrace_may_access(task, PTRACE_MODE_READ))
272 return 0;
273 else
274 return sprintf(buffer, "%lu", wchan);
275 else
276 return sprintf(buffer, "%s", symname);
277}
278#endif /* CONFIG_KALLSYMS */
279
280static int lock_trace(struct task_struct *task)
281{
282 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
283 if (err)
284 return err;
285 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
286 mutex_unlock(&task->signal->cred_guard_mutex);
287 return -EPERM;
288 }
289 return 0;
290}
291
292static void unlock_trace(struct task_struct *task)
293{
294 mutex_unlock(&task->signal->cred_guard_mutex);
295}
296
297#ifdef CONFIG_STACKTRACE
298
299#define MAX_STACK_TRACE_DEPTH 64
300
301static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
302 struct pid *pid, struct task_struct *task)
303{
304 struct stack_trace trace;
305 unsigned long *entries;
306 int err;
307 int i;
308
309 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
310 if (!entries)
311 return -ENOMEM;
312
313 trace.nr_entries = 0;
314 trace.max_entries = MAX_STACK_TRACE_DEPTH;
315 trace.entries = entries;
316 trace.skip = 0;
317
318 err = lock_trace(task);
319 if (!err) {
320 save_stack_trace_tsk(task, &trace);
321
322 for (i = 0; i < trace.nr_entries; i++) {
323 seq_printf(m, "[<%pK>] %pS\n",
324 (void *)entries[i], (void *)entries[i]);
325 }
326 unlock_trace(task);
327 }
328 kfree(entries);
329
330 return err;
331}
332#endif
333
334#ifdef CONFIG_SCHEDSTATS
335/*
336 * Provides /proc/PID/schedstat
337 */
338static int proc_pid_schedstat(struct task_struct *task, char *buffer)
339{
340 return sprintf(buffer, "%llu %llu %lu\n",
341 (unsigned long long)task->se.sum_exec_runtime,
342 (unsigned long long)task->sched_info.run_delay,
343 task->sched_info.pcount);
344}
345#endif
346
347#ifdef CONFIG_LATENCYTOP
348static int lstats_show_proc(struct seq_file *m, void *v)
349{
350 int i;
351 struct inode *inode = m->private;
352 struct task_struct *task = get_proc_task(inode);
353
354 if (!task)
355 return -ESRCH;
356 seq_puts(m, "Latency Top version : v0.1\n");
357 for (i = 0; i < 32; i++) {
358 struct latency_record *lr = &task->latency_record[i];
359 if (lr->backtrace[0]) {
360 int q;
361 seq_printf(m, "%i %li %li",
362 lr->count, lr->time, lr->max);
363 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
364 unsigned long bt = lr->backtrace[q];
365 if (!bt)
366 break;
367 if (bt == ULONG_MAX)
368 break;
369 seq_printf(m, " %ps", (void *)bt);
370 }
371 seq_putc(m, '\n');
372 }
373
374 }
375 put_task_struct(task);
376 return 0;
377}
378
379static int lstats_open(struct inode *inode, struct file *file)
380{
381 return single_open(file, lstats_show_proc, inode);
382}
383
384static ssize_t lstats_write(struct file *file, const char __user *buf,
385 size_t count, loff_t *offs)
386{
387 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
388
389 if (!task)
390 return -ESRCH;
391 clear_all_latency_tracing(task);
392 put_task_struct(task);
393
394 return count;
395}
396
397static const struct file_operations proc_lstats_operations = {
398 .open = lstats_open,
399 .read = seq_read,
400 .write = lstats_write,
401 .llseek = seq_lseek,
402 .release = single_release,
403};
404
405#endif
406
407static int proc_oom_score(struct task_struct *task, char *buffer)
408{
409 unsigned long totalpages = totalram_pages + total_swap_pages;
410 unsigned long points = 0;
411
412 read_lock(&tasklist_lock);
413 if (pid_alive(task))
414 points = oom_badness(task, NULL, NULL, totalpages) *
415 1000 / totalpages;
416 read_unlock(&tasklist_lock);
417 return sprintf(buffer, "%lu\n", points);
418}
419
420struct limit_names {
421 char *name;
422 char *unit;
423};
424
425static const struct limit_names lnames[RLIM_NLIMITS] = {
426 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
427 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
428 [RLIMIT_DATA] = {"Max data size", "bytes"},
429 [RLIMIT_STACK] = {"Max stack size", "bytes"},
430 [RLIMIT_CORE] = {"Max core file size", "bytes"},
431 [RLIMIT_RSS] = {"Max resident set", "bytes"},
432 [RLIMIT_NPROC] = {"Max processes", "processes"},
433 [RLIMIT_NOFILE] = {"Max open files", "files"},
434 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
435 [RLIMIT_AS] = {"Max address space", "bytes"},
436 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
437 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
438 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
439 [RLIMIT_NICE] = {"Max nice priority", NULL},
440 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
441 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
442};
443
444/* Display limits for a process */
445static int proc_pid_limits(struct task_struct *task, char *buffer)
446{
447 unsigned int i;
448 int count = 0;
449 unsigned long flags;
450 char *bufptr = buffer;
451
452 struct rlimit rlim[RLIM_NLIMITS];
453
454 if (!lock_task_sighand(task, &flags))
455 return 0;
456 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
457 unlock_task_sighand(task, &flags);
458
459 /*
460 * print the file header
461 */
462 count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
463 "Limit", "Soft Limit", "Hard Limit", "Units");
464
465 for (i = 0; i < RLIM_NLIMITS; i++) {
466 if (rlim[i].rlim_cur == RLIM_INFINITY)
467 count += sprintf(&bufptr[count], "%-25s %-20s ",
468 lnames[i].name, "unlimited");
469 else
470 count += sprintf(&bufptr[count], "%-25s %-20lu ",
471 lnames[i].name, rlim[i].rlim_cur);
472
473 if (rlim[i].rlim_max == RLIM_INFINITY)
474 count += sprintf(&bufptr[count], "%-20s ", "unlimited");
475 else
476 count += sprintf(&bufptr[count], "%-20lu ",
477 rlim[i].rlim_max);
478
479 if (lnames[i].unit)
480 count += sprintf(&bufptr[count], "%-10s\n",
481 lnames[i].unit);
482 else
483 count += sprintf(&bufptr[count], "\n");
484 }
485
486 return count;
487}
488
489#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
490static int proc_pid_syscall(struct task_struct *task, char *buffer)
491{
492 long nr;
493 unsigned long args[6], sp, pc;
494 int res = lock_trace(task);
495 if (res)
496 return res;
497
498 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
499 res = sprintf(buffer, "running\n");
500 else if (nr < 0)
501 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
502 else
503 res = sprintf(buffer,
504 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
505 nr,
506 args[0], args[1], args[2], args[3], args[4], args[5],
507 sp, pc);
508 unlock_trace(task);
509 return res;
510}
511#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
512
513/************************************************************************/
514/* Here the fs part begins */
515/************************************************************************/
516
517/* permission checks */
518static int proc_fd_access_allowed(struct inode *inode)
519{
520 struct task_struct *task;
521 int allowed = 0;
522 /* Allow access to a task's file descriptors if it is us or we
523 * may use ptrace attach to the process and find out that
524 * information.
525 */
526 task = get_proc_task(inode);
527 if (task) {
528 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
529 put_task_struct(task);
530 }
531 return allowed;
532}
533
534int proc_setattr(struct dentry *dentry, struct iattr *attr)
535{
536 int error;
537 struct inode *inode = dentry->d_inode;
538
539 if (attr->ia_valid & ATTR_MODE)
540 return -EPERM;
541
542 error = inode_change_ok(inode, attr);
543 if (error)
544 return error;
545
546 if ((attr->ia_valid & ATTR_SIZE) &&
547 attr->ia_size != i_size_read(inode)) {
548 error = vmtruncate(inode, attr->ia_size);
549 if (error)
550 return error;
551 }
552
553 setattr_copy(inode, attr);
554 mark_inode_dirty(inode);
555 return 0;
556}
557
558/*
559 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
560 * or euid/egid (for hide_pid_min=2)?
561 */
562static bool has_pid_permissions(struct pid_namespace *pid,
563 struct task_struct *task,
564 int hide_pid_min)
565{
566 if (pid->hide_pid < hide_pid_min)
567 return true;
568 if (in_group_p(pid->pid_gid))
569 return true;
570 return ptrace_may_access(task, PTRACE_MODE_READ);
571}
572
573
574static int proc_pid_permission(struct inode *inode, int mask)
575{
576 struct pid_namespace *pid = inode->i_sb->s_fs_info;
577 struct task_struct *task;
578 bool has_perms;
579
580 task = get_proc_task(inode);
581 if (!task)
582 return -ESRCH;
583 has_perms = has_pid_permissions(pid, task, 1);
584 put_task_struct(task);
585
586 if (!has_perms) {
587 if (pid->hide_pid == 2) {
588 /*
589 * Let's make getdents(), stat(), and open()
590 * consistent with each other. If a process
591 * may not stat() a file, it shouldn't be seen
592 * in procfs at all.
593 */
594 return -ENOENT;
595 }
596
597 return -EPERM;
598 }
599 return generic_permission(inode, mask);
600}
601
602
603
604static const struct inode_operations proc_def_inode_operations = {
605 .setattr = proc_setattr,
606};
607
608#define PROC_BLOCK_SIZE (3*1024) /* 4K page size but our output routines use some slack for overruns */
609
610static ssize_t proc_info_read(struct file * file, char __user * buf,
611 size_t count, loff_t *ppos)
612{
613 struct inode * inode = file->f_path.dentry->d_inode;
614 unsigned long page;
615 ssize_t length;
616 struct task_struct *task = get_proc_task(inode);
617
618 length = -ESRCH;
619 if (!task)
620 goto out_no_task;
621
622 if (count > PROC_BLOCK_SIZE)
623 count = PROC_BLOCK_SIZE;
624
625 length = -ENOMEM;
626 if (!(page = __get_free_page(GFP_TEMPORARY)))
627 goto out;
628
629 length = PROC_I(inode)->op.proc_read(task, (char*)page);
630
631 if (length >= 0)
632 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
633 free_page(page);
634out:
635 put_task_struct(task);
636out_no_task:
637 return length;
638}
639
640static const struct file_operations proc_info_file_operations = {
641 .read = proc_info_read,
642 .llseek = generic_file_llseek,
643};
644
645static int proc_single_show(struct seq_file *m, void *v)
646{
647 struct inode *inode = m->private;
648 struct pid_namespace *ns;
649 struct pid *pid;
650 struct task_struct *task;
651 int ret;
652
653 ns = inode->i_sb->s_fs_info;
654 pid = proc_pid(inode);
655 task = get_pid_task(pid, PIDTYPE_PID);
656 if (!task)
657 return -ESRCH;
658
659 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
660
661 put_task_struct(task);
662 return ret;
663}
664
665static int proc_single_open(struct inode *inode, struct file *filp)
666{
667 return single_open(filp, proc_single_show, inode);
668}
669
670static const struct file_operations proc_single_file_operations = {
671 .open = proc_single_open,
672 .read = seq_read,
673 .llseek = seq_lseek,
674 .release = single_release,
675};
676
677static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
678{
679 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
680 struct mm_struct *mm;
681
682 if (!task)
683 return -ESRCH;
684
685 mm = mm_access(task, mode);
686 put_task_struct(task);
687
688 if (IS_ERR(mm))
689 return PTR_ERR(mm);
690
691 if (mm) {
692 /* ensure this mm_struct can't be freed */
693 atomic_inc(&mm->mm_count);
694 /* but do not pin its memory */
695 mmput(mm);
696 }
697
698 /* OK to pass negative loff_t, we can catch out-of-range */
699 file->f_mode |= FMODE_UNSIGNED_OFFSET;
700 file->private_data = mm;
701
702 return 0;
703}
704
705static int mem_open(struct inode *inode, struct file *file)
706{
707 return __mem_open(inode, file, PTRACE_MODE_ATTACH);
708}
709
710static ssize_t mem_rw(struct file *file, char __user *buf,
711 size_t count, loff_t *ppos, int write)
712{
713 struct mm_struct *mm = file->private_data;
714 unsigned long addr = *ppos;
715 ssize_t copied;
716 char *page;
717
718 if (!mm)
719 return 0;
720
721 page = (char *)__get_free_page(GFP_TEMPORARY);
722 if (!page)
723 return -ENOMEM;
724
725 copied = 0;
726 if (!atomic_inc_not_zero(&mm->mm_users))
727 goto free;
728
729 while (count > 0) {
730 int this_len = min_t(int, count, PAGE_SIZE);
731
732 if (write && copy_from_user(page, buf, this_len)) {
733 copied = -EFAULT;
734 break;
735 }
736
737 this_len = access_remote_vm(mm, addr, page, this_len, write);
738 if (!this_len) {
739 if (!copied)
740 copied = -EIO;
741 break;
742 }
743
744 if (!write && copy_to_user(buf, page, this_len)) {
745 copied = -EFAULT;
746 break;
747 }
748
749 buf += this_len;
750 addr += this_len;
751 copied += this_len;
752 count -= this_len;
753 }
754 *ppos = addr;
755
756 mmput(mm);
757free:
758 free_page((unsigned long) page);
759 return copied;
760}
761
762static ssize_t mem_read(struct file *file, char __user *buf,
763 size_t count, loff_t *ppos)
764{
765 return mem_rw(file, buf, count, ppos, 0);
766}
767
768static ssize_t mem_write(struct file *file, const char __user *buf,
769 size_t count, loff_t *ppos)
770{
771 return mem_rw(file, (char __user*)buf, count, ppos, 1);
772}
773
774loff_t mem_lseek(struct file *file, loff_t offset, int orig)
775{
776 switch (orig) {
777 case 0:
778 file->f_pos = offset;
779 break;
780 case 1:
781 file->f_pos += offset;
782 break;
783 default:
784 return -EINVAL;
785 }
786 force_successful_syscall_return();
787 return file->f_pos;
788}
789
790static int mem_release(struct inode *inode, struct file *file)
791{
792 struct mm_struct *mm = file->private_data;
793 if (mm)
794 mmdrop(mm);
795 return 0;
796}
797
798static const struct file_operations proc_mem_operations = {
799 .llseek = mem_lseek,
800 .read = mem_read,
801 .write = mem_write,
802 .open = mem_open,
803 .release = mem_release,
804};
805
806static int environ_open(struct inode *inode, struct file *file)
807{
808 return __mem_open(inode, file, PTRACE_MODE_READ);
809}
810
811static ssize_t environ_read(struct file *file, char __user *buf,
812 size_t count, loff_t *ppos)
813{
814 char *page;
815 unsigned long src = *ppos;
816 int ret = 0;
817 struct mm_struct *mm = file->private_data;
818
819 if (!mm)
820 return 0;
821
822 page = (char *)__get_free_page(GFP_TEMPORARY);
823 if (!page)
824 return -ENOMEM;
825
826 ret = 0;
827 if (!atomic_inc_not_zero(&mm->mm_users))
828 goto free;
829 while (count > 0) {
830 int this_len, retval, max_len;
831
832 this_len = mm->env_end - (mm->env_start + src);
833
834 if (this_len <= 0)
835 break;
836
837 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
838 this_len = (this_len > max_len) ? max_len : this_len;
839
840 retval = access_remote_vm(mm, (mm->env_start + src),
841 page, this_len, 0);
842
843 if (retval <= 0) {
844 ret = retval;
845 break;
846 }
847
848 if (copy_to_user(buf, page, retval)) {
849 ret = -EFAULT;
850 break;
851 }
852
853 ret += retval;
854 src += retval;
855 buf += retval;
856 count -= retval;
857 }
858 *ppos = src;
859 mmput(mm);
860
861free:
862 free_page((unsigned long) page);
863 return ret;
864}
865
866static const struct file_operations proc_environ_operations = {
867 .open = environ_open,
868 .read = environ_read,
869 .llseek = generic_file_llseek,
870 .release = mem_release,
871};
872
873static ssize_t oom_adjust_read(struct file *file, char __user *buf,
874 size_t count, loff_t *ppos)
875{
876 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
877 char buffer[PROC_NUMBUF];
878 size_t len;
879 int oom_adjust = OOM_DISABLE;
880 unsigned long flags;
881
882 if (!task)
883 return -ESRCH;
884
885 if (lock_task_sighand(task, &flags)) {
886 oom_adjust = task->signal->oom_adj;
887 unlock_task_sighand(task, &flags);
888 }
889
890 put_task_struct(task);
891
892 len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
893
894 return simple_read_from_buffer(buf, count, ppos, buffer, len);
895}
896
897static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
898 size_t count, loff_t *ppos)
899{
900 struct task_struct *task;
901 char buffer[PROC_NUMBUF];
902 int oom_adjust;
903 unsigned long flags;
904 int err;
905
906 memset(buffer, 0, sizeof(buffer));
907 if (count > sizeof(buffer) - 1)
908 count = sizeof(buffer) - 1;
909 if (copy_from_user(buffer, buf, count)) {
910 err = -EFAULT;
911 goto out;
912 }
913
914 err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
915 if (err)
916 goto out;
917 if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
918 oom_adjust != OOM_DISABLE) {
919 err = -EINVAL;
920 goto out;
921 }
922
923 task = get_proc_task(file->f_path.dentry->d_inode);
924 if (!task) {
925 err = -ESRCH;
926 goto out;
927 }
928
929 task_lock(task);
930 if (!task->mm) {
931 err = -EINVAL;
932 goto err_task_lock;
933 }
934
935 if (!lock_task_sighand(task, &flags)) {
936 err = -ESRCH;
937 goto err_task_lock;
938 }
939
940 if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
941 err = -EACCES;
942 goto err_sighand;
943 }
944
945 /*
946 * Warn that /proc/pid/oom_adj is deprecated, see
947 * Documentation/feature-removal-schedule.txt.
948 */
949 printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
950 current->comm, task_pid_nr(current), task_pid_nr(task),
951 task_pid_nr(task));
952 task->signal->oom_adj = oom_adjust;
953 /*
954 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
955 * value is always attainable.
956 */
957 if (task->signal->oom_adj == OOM_ADJUST_MAX)
958 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
959 else
960 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
961 -OOM_DISABLE;
962 trace_oom_score_adj_update(task);
963err_sighand:
964 unlock_task_sighand(task, &flags);
965err_task_lock:
966 task_unlock(task);
967 put_task_struct(task);
968out:
969 return err < 0 ? err : count;
970}
971
972static const struct file_operations proc_oom_adjust_operations = {
973 .read = oom_adjust_read,
974 .write = oom_adjust_write,
975 .llseek = generic_file_llseek,
976};
977
978static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
979 size_t count, loff_t *ppos)
980{
981 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
982 char buffer[PROC_NUMBUF];
983 int oom_score_adj = OOM_SCORE_ADJ_MIN;
984 unsigned long flags;
985 size_t len;
986
987 if (!task)
988 return -ESRCH;
989 if (lock_task_sighand(task, &flags)) {
990 oom_score_adj = task->signal->oom_score_adj;
991 unlock_task_sighand(task, &flags);
992 }
993 put_task_struct(task);
994 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
995 return simple_read_from_buffer(buf, count, ppos, buffer, len);
996}
997
998static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
999 size_t count, loff_t *ppos)
1000{
1001 struct task_struct *task;
1002 char buffer[PROC_NUMBUF];
1003 unsigned long flags;
1004 int oom_score_adj;
1005 int err;
1006
1007 memset(buffer, 0, sizeof(buffer));
1008 if (count > sizeof(buffer) - 1)
1009 count = sizeof(buffer) - 1;
1010 if (copy_from_user(buffer, buf, count)) {
1011 err = -EFAULT;
1012 goto out;
1013 }
1014
1015 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1016 if (err)
1017 goto out;
1018 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1019 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1020 err = -EINVAL;
1021 goto out;
1022 }
1023
1024 task = get_proc_task(file->f_path.dentry->d_inode);
1025 if (!task) {
1026 err = -ESRCH;
1027 goto out;
1028 }
1029
1030 task_lock(task);
1031 if (!task->mm) {
1032 err = -EINVAL;
1033 goto err_task_lock;
1034 }
1035
1036 if (!lock_task_sighand(task, &flags)) {
1037 err = -ESRCH;
1038 goto err_task_lock;
1039 }
1040
1041 if (oom_score_adj < task->signal->oom_score_adj_min &&
1042 !capable(CAP_SYS_RESOURCE)) {
1043 err = -EACCES;
1044 goto err_sighand;
1045 }
1046
1047 task->signal->oom_score_adj = oom_score_adj;
1048 if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1049 task->signal->oom_score_adj_min = oom_score_adj;
1050 trace_oom_score_adj_update(task);
1051 /*
1052 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1053 * always attainable.
1054 */
1055 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1056 task->signal->oom_adj = OOM_DISABLE;
1057 else
1058 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1059 OOM_SCORE_ADJ_MAX;
1060err_sighand:
1061 unlock_task_sighand(task, &flags);
1062err_task_lock:
1063 task_unlock(task);
1064 put_task_struct(task);
1065out:
1066 return err < 0 ? err : count;
1067}
1068
1069static const struct file_operations proc_oom_score_adj_operations = {
1070 .read = oom_score_adj_read,
1071 .write = oom_score_adj_write,
1072 .llseek = default_llseek,
1073};
1074
1075#ifdef CONFIG_AUDITSYSCALL
1076#define TMPBUFLEN 21
1077static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1078 size_t count, loff_t *ppos)
1079{
1080 struct inode * inode = file->f_path.dentry->d_inode;
1081 struct task_struct *task = get_proc_task(inode);
1082 ssize_t length;
1083 char tmpbuf[TMPBUFLEN];
1084
1085 if (!task)
1086 return -ESRCH;
1087 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1088 audit_get_loginuid(task));
1089 put_task_struct(task);
1090 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1091}
1092
1093static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1094 size_t count, loff_t *ppos)
1095{
1096 struct inode * inode = file->f_path.dentry->d_inode;
1097 char *page, *tmp;
1098 ssize_t length;
1099 uid_t loginuid;
1100
1101 rcu_read_lock();
1102 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1103 rcu_read_unlock();
1104 return -EPERM;
1105 }
1106 rcu_read_unlock();
1107
1108 if (count >= PAGE_SIZE)
1109 count = PAGE_SIZE - 1;
1110
1111 if (*ppos != 0) {
1112 /* No partial writes. */
1113 return -EINVAL;
1114 }
1115 page = (char*)__get_free_page(GFP_TEMPORARY);
1116 if (!page)
1117 return -ENOMEM;
1118 length = -EFAULT;
1119 if (copy_from_user(page, buf, count))
1120 goto out_free_page;
1121
1122 page[count] = '\0';
1123 loginuid = simple_strtoul(page, &tmp, 10);
1124 if (tmp == page) {
1125 length = -EINVAL;
1126 goto out_free_page;
1127
1128 }
1129 length = audit_set_loginuid(loginuid);
1130 if (likely(length == 0))
1131 length = count;
1132
1133out_free_page:
1134 free_page((unsigned long) page);
1135 return length;
1136}
1137
1138static const struct file_operations proc_loginuid_operations = {
1139 .read = proc_loginuid_read,
1140 .write = proc_loginuid_write,
1141 .llseek = generic_file_llseek,
1142};
1143
1144static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1145 size_t count, loff_t *ppos)
1146{
1147 struct inode * inode = file->f_path.dentry->d_inode;
1148 struct task_struct *task = get_proc_task(inode);
1149 ssize_t length;
1150 char tmpbuf[TMPBUFLEN];
1151
1152 if (!task)
1153 return -ESRCH;
1154 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1155 audit_get_sessionid(task));
1156 put_task_struct(task);
1157 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1158}
1159
1160static const struct file_operations proc_sessionid_operations = {
1161 .read = proc_sessionid_read,
1162 .llseek = generic_file_llseek,
1163};
1164#endif
1165
1166#ifdef CONFIG_FAULT_INJECTION
1167static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1168 size_t count, loff_t *ppos)
1169{
1170 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1171 char buffer[PROC_NUMBUF];
1172 size_t len;
1173 int make_it_fail;
1174
1175 if (!task)
1176 return -ESRCH;
1177 make_it_fail = task->make_it_fail;
1178 put_task_struct(task);
1179
1180 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1181
1182 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1183}
1184
1185static ssize_t proc_fault_inject_write(struct file * file,
1186 const char __user * buf, size_t count, loff_t *ppos)
1187{
1188 struct task_struct *task;
1189 char buffer[PROC_NUMBUF], *end;
1190 int make_it_fail;
1191
1192 if (!capable(CAP_SYS_RESOURCE))
1193 return -EPERM;
1194 memset(buffer, 0, sizeof(buffer));
1195 if (count > sizeof(buffer) - 1)
1196 count = sizeof(buffer) - 1;
1197 if (copy_from_user(buffer, buf, count))
1198 return -EFAULT;
1199 make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1200 if (*end)
1201 return -EINVAL;
1202 task = get_proc_task(file->f_dentry->d_inode);
1203 if (!task)
1204 return -ESRCH;
1205 task->make_it_fail = make_it_fail;
1206 put_task_struct(task);
1207
1208 return count;
1209}
1210
1211static const struct file_operations proc_fault_inject_operations = {
1212 .read = proc_fault_inject_read,
1213 .write = proc_fault_inject_write,
1214 .llseek = generic_file_llseek,
1215};
1216#endif
1217
1218
1219#ifdef CONFIG_SCHED_DEBUG
1220/*
1221 * Print out various scheduling related per-task fields:
1222 */
1223static int sched_show(struct seq_file *m, void *v)
1224{
1225 struct inode *inode = m->private;
1226 struct task_struct *p;
1227
1228 p = get_proc_task(inode);
1229 if (!p)
1230 return -ESRCH;
1231 proc_sched_show_task(p, m);
1232
1233 put_task_struct(p);
1234
1235 return 0;
1236}
1237
1238static ssize_t
1239sched_write(struct file *file, const char __user *buf,
1240 size_t count, loff_t *offset)
1241{
1242 struct inode *inode = file->f_path.dentry->d_inode;
1243 struct task_struct *p;
1244
1245 p = get_proc_task(inode);
1246 if (!p)
1247 return -ESRCH;
1248 proc_sched_set_task(p);
1249
1250 put_task_struct(p);
1251
1252 return count;
1253}
1254
1255static int sched_open(struct inode *inode, struct file *filp)
1256{
1257 return single_open(filp, sched_show, inode);
1258}
1259
1260static const struct file_operations proc_pid_sched_operations = {
1261 .open = sched_open,
1262 .read = seq_read,
1263 .write = sched_write,
1264 .llseek = seq_lseek,
1265 .release = single_release,
1266};
1267
1268#endif
1269
1270#ifdef CONFIG_SCHED_AUTOGROUP
1271/*
1272 * Print out autogroup related information:
1273 */
1274static int sched_autogroup_show(struct seq_file *m, void *v)
1275{
1276 struct inode *inode = m->private;
1277 struct task_struct *p;
1278
1279 p = get_proc_task(inode);
1280 if (!p)
1281 return -ESRCH;
1282 proc_sched_autogroup_show_task(p, m);
1283
1284 put_task_struct(p);
1285
1286 return 0;
1287}
1288
1289static ssize_t
1290sched_autogroup_write(struct file *file, const char __user *buf,
1291 size_t count, loff_t *offset)
1292{
1293 struct inode *inode = file->f_path.dentry->d_inode;
1294 struct task_struct *p;
1295 char buffer[PROC_NUMBUF];
1296 int nice;
1297 int err;
1298
1299 memset(buffer, 0, sizeof(buffer));
1300 if (count > sizeof(buffer) - 1)
1301 count = sizeof(buffer) - 1;
1302 if (copy_from_user(buffer, buf, count))
1303 return -EFAULT;
1304
1305 err = kstrtoint(strstrip(buffer), 0, &nice);
1306 if (err < 0)
1307 return err;
1308
1309 p = get_proc_task(inode);
1310 if (!p)
1311 return -ESRCH;
1312
1313 err = proc_sched_autogroup_set_nice(p, nice);
1314 if (err)
1315 count = err;
1316
1317 put_task_struct(p);
1318
1319 return count;
1320}
1321
1322static int sched_autogroup_open(struct inode *inode, struct file *filp)
1323{
1324 int ret;
1325
1326 ret = single_open(filp, sched_autogroup_show, NULL);
1327 if (!ret) {
1328 struct seq_file *m = filp->private_data;
1329
1330 m->private = inode;
1331 }
1332 return ret;
1333}
1334
1335static const struct file_operations proc_pid_sched_autogroup_operations = {
1336 .open = sched_autogroup_open,
1337 .read = seq_read,
1338 .write = sched_autogroup_write,
1339 .llseek = seq_lseek,
1340 .release = single_release,
1341};
1342
1343#endif /* CONFIG_SCHED_AUTOGROUP */
1344
1345static ssize_t comm_write(struct file *file, const char __user *buf,
1346 size_t count, loff_t *offset)
1347{
1348 struct inode *inode = file->f_path.dentry->d_inode;
1349 struct task_struct *p;
1350 char buffer[TASK_COMM_LEN];
1351
1352 memset(buffer, 0, sizeof(buffer));
1353 if (count > sizeof(buffer) - 1)
1354 count = sizeof(buffer) - 1;
1355 if (copy_from_user(buffer, buf, count))
1356 return -EFAULT;
1357
1358 p = get_proc_task(inode);
1359 if (!p)
1360 return -ESRCH;
1361
1362 if (same_thread_group(current, p))
1363 set_task_comm(p, buffer);
1364 else
1365 count = -EINVAL;
1366
1367 put_task_struct(p);
1368
1369 return count;
1370}
1371
1372static int comm_show(struct seq_file *m, void *v)
1373{
1374 struct inode *inode = m->private;
1375 struct task_struct *p;
1376
1377 p = get_proc_task(inode);
1378 if (!p)
1379 return -ESRCH;
1380
1381 task_lock(p);
1382 seq_printf(m, "%s\n", p->comm);
1383 task_unlock(p);
1384
1385 put_task_struct(p);
1386
1387 return 0;
1388}
1389
1390static int comm_open(struct inode *inode, struct file *filp)
1391{
1392 return single_open(filp, comm_show, inode);
1393}
1394
1395static const struct file_operations proc_pid_set_comm_operations = {
1396 .open = comm_open,
1397 .read = seq_read,
1398 .write = comm_write,
1399 .llseek = seq_lseek,
1400 .release = single_release,
1401};
1402
1403static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1404{
1405 struct task_struct *task;
1406 struct mm_struct *mm;
1407 struct file *exe_file;
1408
1409 task = get_proc_task(dentry->d_inode);
1410 if (!task)
1411 return -ENOENT;
1412 mm = get_task_mm(task);
1413 put_task_struct(task);
1414 if (!mm)
1415 return -ENOENT;
1416 exe_file = get_mm_exe_file(mm);
1417 mmput(mm);
1418 if (exe_file) {
1419 *exe_path = exe_file->f_path;
1420 path_get(&exe_file->f_path);
1421 fput(exe_file);
1422 return 0;
1423 } else
1424 return -ENOENT;
1425}
1426
1427static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1428{
1429 struct inode *inode = dentry->d_inode;
1430 int error = -EACCES;
1431
1432 /* We don't need a base pointer in the /proc filesystem */
1433 path_put(&nd->path);
1434
1435 /* Are we allowed to snoop on the tasks file descriptors? */
1436 if (!proc_fd_access_allowed(inode))
1437 goto out;
1438
1439 error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1440out:
1441 return ERR_PTR(error);
1442}
1443
1444static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1445{
1446 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1447 char *pathname;
1448 int len;
1449
1450 if (!tmp)
1451 return -ENOMEM;
1452
1453 pathname = d_path(path, tmp, PAGE_SIZE);
1454 len = PTR_ERR(pathname);
1455 if (IS_ERR(pathname))
1456 goto out;
1457 len = tmp + PAGE_SIZE - 1 - pathname;
1458
1459 if (len > buflen)
1460 len = buflen;
1461 if (copy_to_user(buffer, pathname, len))
1462 len = -EFAULT;
1463 out:
1464 free_page((unsigned long)tmp);
1465 return len;
1466}
1467
1468static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1469{
1470 int error = -EACCES;
1471 struct inode *inode = dentry->d_inode;
1472 struct path path;
1473
1474 /* Are we allowed to snoop on the tasks file descriptors? */
1475 if (!proc_fd_access_allowed(inode))
1476 goto out;
1477
1478 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1479 if (error)
1480 goto out;
1481
1482 error = do_proc_readlink(&path, buffer, buflen);
1483 path_put(&path);
1484out:
1485 return error;
1486}
1487
1488static const struct inode_operations proc_pid_link_inode_operations = {
1489 .readlink = proc_pid_readlink,
1490 .follow_link = proc_pid_follow_link,
1491 .setattr = proc_setattr,
1492};
1493
1494
1495/* building an inode */
1496
1497static int task_dumpable(struct task_struct *task)
1498{
1499 int dumpable = 0;
1500 struct mm_struct *mm;
1501
1502 task_lock(task);
1503 mm = task->mm;
1504 if (mm)
1505 dumpable = get_dumpable(mm);
1506 task_unlock(task);
1507 if(dumpable == 1)
1508 return 1;
1509 return 0;
1510}
1511
1512struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1513{
1514 struct inode * inode;
1515 struct proc_inode *ei;
1516 const struct cred *cred;
1517
1518 /* We need a new inode */
1519
1520 inode = new_inode(sb);
1521 if (!inode)
1522 goto out;
1523
1524 /* Common stuff */
1525 ei = PROC_I(inode);
1526 inode->i_ino = get_next_ino();
1527 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1528 inode->i_op = &proc_def_inode_operations;
1529
1530 /*
1531 * grab the reference to task.
1532 */
1533 ei->pid = get_task_pid(task, PIDTYPE_PID);
1534 if (!ei->pid)
1535 goto out_unlock;
1536
1537 if (task_dumpable(task)) {
1538 rcu_read_lock();
1539 cred = __task_cred(task);
1540 inode->i_uid = cred->euid;
1541 inode->i_gid = cred->egid;
1542 rcu_read_unlock();
1543 }
1544 security_task_to_inode(task, inode);
1545
1546out:
1547 return inode;
1548
1549out_unlock:
1550 iput(inode);
1551 return NULL;
1552}
1553
1554int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1555{
1556 struct inode *inode = dentry->d_inode;
1557 struct task_struct *task;
1558 const struct cred *cred;
1559 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1560
1561 generic_fillattr(inode, stat);
1562
1563 rcu_read_lock();
1564 stat->uid = GLOBAL_ROOT_UID;
1565 stat->gid = GLOBAL_ROOT_GID;
1566 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1567 if (task) {
1568 if (!has_pid_permissions(pid, task, 2)) {
1569 rcu_read_unlock();
1570 /*
1571 * This doesn't prevent learning whether PID exists,
1572 * it only makes getattr() consistent with readdir().
1573 */
1574 return -ENOENT;
1575 }
1576 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1577 task_dumpable(task)) {
1578 cred = __task_cred(task);
1579 stat->uid = cred->euid;
1580 stat->gid = cred->egid;
1581 }
1582 }
1583 rcu_read_unlock();
1584 return 0;
1585}
1586
1587/* dentry stuff */
1588
1589/*
1590 * Exceptional case: normally we are not allowed to unhash a busy
1591 * directory. In this case, however, we can do it - no aliasing problems
1592 * due to the way we treat inodes.
1593 *
1594 * Rewrite the inode's ownerships here because the owning task may have
1595 * performed a setuid(), etc.
1596 *
1597 * Before the /proc/pid/status file was created the only way to read
1598 * the effective uid of a /process was to stat /proc/pid. Reading
1599 * /proc/pid/status is slow enough that procps and other packages
1600 * kept stating /proc/pid. To keep the rules in /proc simple I have
1601 * made this apply to all per process world readable and executable
1602 * directories.
1603 */
1604int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1605{
1606 struct inode *inode;
1607 struct task_struct *task;
1608 const struct cred *cred;
1609
1610 if (nd && nd->flags & LOOKUP_RCU)
1611 return -ECHILD;
1612
1613 inode = dentry->d_inode;
1614 task = get_proc_task(inode);
1615
1616 if (task) {
1617 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1618 task_dumpable(task)) {
1619 rcu_read_lock();
1620 cred = __task_cred(task);
1621 inode->i_uid = cred->euid;
1622 inode->i_gid = cred->egid;
1623 rcu_read_unlock();
1624 } else {
1625 inode->i_uid = GLOBAL_ROOT_UID;
1626 inode->i_gid = GLOBAL_ROOT_GID;
1627 }
1628 inode->i_mode &= ~(S_ISUID | S_ISGID);
1629 security_task_to_inode(task, inode);
1630 put_task_struct(task);
1631 return 1;
1632 }
1633 d_drop(dentry);
1634 return 0;
1635}
1636
1637static int pid_delete_dentry(const struct dentry * dentry)
1638{
1639 /* Is the task we represent dead?
1640 * If so, then don't put the dentry on the lru list,
1641 * kill it immediately.
1642 */
1643 return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1644}
1645
1646const struct dentry_operations pid_dentry_operations =
1647{
1648 .d_revalidate = pid_revalidate,
1649 .d_delete = pid_delete_dentry,
1650};
1651
1652/* Lookups */
1653
1654/*
1655 * Fill a directory entry.
1656 *
1657 * If possible create the dcache entry and derive our inode number and
1658 * file type from dcache entry.
1659 *
1660 * Since all of the proc inode numbers are dynamically generated, the inode
1661 * numbers do not exist until the inode is cache. This means creating the
1662 * the dcache entry in readdir is necessary to keep the inode numbers
1663 * reported by readdir in sync with the inode numbers reported
1664 * by stat.
1665 */
1666int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1667 const char *name, int len,
1668 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1669{
1670 struct dentry *child, *dir = filp->f_path.dentry;
1671 struct inode *inode;
1672 struct qstr qname;
1673 ino_t ino = 0;
1674 unsigned type = DT_UNKNOWN;
1675
1676 qname.name = name;
1677 qname.len = len;
1678 qname.hash = full_name_hash(name, len);
1679
1680 child = d_lookup(dir, &qname);
1681 if (!child) {
1682 struct dentry *new;
1683 new = d_alloc(dir, &qname);
1684 if (new) {
1685 child = instantiate(dir->d_inode, new, task, ptr);
1686 if (child)
1687 dput(new);
1688 else
1689 child = new;
1690 }
1691 }
1692 if (!child || IS_ERR(child) || !child->d_inode)
1693 goto end_instantiate;
1694 inode = child->d_inode;
1695 if (inode) {
1696 ino = inode->i_ino;
1697 type = inode->i_mode >> 12;
1698 }
1699 dput(child);
1700end_instantiate:
1701 if (!ino)
1702 ino = find_inode_number(dir, &qname);
1703 if (!ino)
1704 ino = 1;
1705 return filldir(dirent, name, len, filp->f_pos, ino, type);
1706}
1707
1708static unsigned name_to_int(struct dentry *dentry)
1709{
1710 const char *name = dentry->d_name.name;
1711 int len = dentry->d_name.len;
1712 unsigned n = 0;
1713
1714 if (len > 1 && *name == '0')
1715 goto out;
1716 while (len-- > 0) {
1717 unsigned c = *name++ - '0';
1718 if (c > 9)
1719 goto out;
1720 if (n >= (~0U-9)/10)
1721 goto out;
1722 n *= 10;
1723 n += c;
1724 }
1725 return n;
1726out:
1727 return ~0U;
1728}
1729
1730#define PROC_FDINFO_MAX 64
1731
1732static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1733{
1734 struct task_struct *task = get_proc_task(inode);
1735 struct files_struct *files = NULL;
1736 struct file *file;
1737 int fd = proc_fd(inode);
1738
1739 if (task) {
1740 files = get_files_struct(task);
1741 put_task_struct(task);
1742 }
1743 if (files) {
1744 /*
1745 * We are not taking a ref to the file structure, so we must
1746 * hold ->file_lock.
1747 */
1748 spin_lock(&files->file_lock);
1749 file = fcheck_files(files, fd);
1750 if (file) {
1751 unsigned int f_flags;
1752 struct fdtable *fdt;
1753
1754 fdt = files_fdtable(files);
1755 f_flags = file->f_flags & ~O_CLOEXEC;
1756 if (close_on_exec(fd, fdt))
1757 f_flags |= O_CLOEXEC;
1758
1759 if (path) {
1760 *path = file->f_path;
1761 path_get(&file->f_path);
1762 }
1763 if (info)
1764 snprintf(info, PROC_FDINFO_MAX,
1765 "pos:\t%lli\n"
1766 "flags:\t0%o\n",
1767 (long long) file->f_pos,
1768 f_flags);
1769 spin_unlock(&files->file_lock);
1770 put_files_struct(files);
1771 return 0;
1772 }
1773 spin_unlock(&files->file_lock);
1774 put_files_struct(files);
1775 }
1776 return -ENOENT;
1777}
1778
1779static int proc_fd_link(struct dentry *dentry, struct path *path)
1780{
1781 return proc_fd_info(dentry->d_inode, path, NULL);
1782}
1783
1784static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1785{
1786 struct inode *inode;
1787 struct task_struct *task;
1788 int fd;
1789 struct files_struct *files;
1790 const struct cred *cred;
1791
1792 if (nd && nd->flags & LOOKUP_RCU)
1793 return -ECHILD;
1794
1795 inode = dentry->d_inode;
1796 task = get_proc_task(inode);
1797 fd = proc_fd(inode);
1798
1799 if (task) {
1800 files = get_files_struct(task);
1801 if (files) {
1802 struct file *file;
1803 rcu_read_lock();
1804 file = fcheck_files(files, fd);
1805 if (file) {
1806 unsigned f_mode = file->f_mode;
1807
1808 rcu_read_unlock();
1809 put_files_struct(files);
1810
1811 if (task_dumpable(task)) {
1812 rcu_read_lock();
1813 cred = __task_cred(task);
1814 inode->i_uid = cred->euid;
1815 inode->i_gid = cred->egid;
1816 rcu_read_unlock();
1817 } else {
1818 inode->i_uid = GLOBAL_ROOT_UID;
1819 inode->i_gid = GLOBAL_ROOT_GID;
1820 }
1821
1822 if (S_ISLNK(inode->i_mode)) {
1823 unsigned i_mode = S_IFLNK;
1824 if (f_mode & FMODE_READ)
1825 i_mode |= S_IRUSR | S_IXUSR;
1826 if (f_mode & FMODE_WRITE)
1827 i_mode |= S_IWUSR | S_IXUSR;
1828 inode->i_mode = i_mode;
1829 }
1830
1831 security_task_to_inode(task, inode);
1832 put_task_struct(task);
1833 return 1;
1834 }
1835 rcu_read_unlock();
1836 put_files_struct(files);
1837 }
1838 put_task_struct(task);
1839 }
1840 d_drop(dentry);
1841 return 0;
1842}
1843
1844static const struct dentry_operations tid_fd_dentry_operations =
1845{
1846 .d_revalidate = tid_fd_revalidate,
1847 .d_delete = pid_delete_dentry,
1848};
1849
1850static struct dentry *proc_fd_instantiate(struct inode *dir,
1851 struct dentry *dentry, struct task_struct *task, const void *ptr)
1852{
1853 unsigned fd = (unsigned long)ptr;
1854 struct inode *inode;
1855 struct proc_inode *ei;
1856 struct dentry *error = ERR_PTR(-ENOENT);
1857
1858 inode = proc_pid_make_inode(dir->i_sb, task);
1859 if (!inode)
1860 goto out;
1861 ei = PROC_I(inode);
1862 ei->fd = fd;
1863
1864 inode->i_mode = S_IFLNK;
1865 inode->i_op = &proc_pid_link_inode_operations;
1866 inode->i_size = 64;
1867 ei->op.proc_get_link = proc_fd_link;
1868 d_set_d_op(dentry, &tid_fd_dentry_operations);
1869 d_add(dentry, inode);
1870 /* Close the race of the process dying before we return the dentry */
1871 if (tid_fd_revalidate(dentry, NULL))
1872 error = NULL;
1873
1874 out:
1875 return error;
1876}
1877
1878static struct dentry *proc_lookupfd_common(struct inode *dir,
1879 struct dentry *dentry,
1880 instantiate_t instantiate)
1881{
1882 struct task_struct *task = get_proc_task(dir);
1883 unsigned fd = name_to_int(dentry);
1884 struct dentry *result = ERR_PTR(-ENOENT);
1885
1886 if (!task)
1887 goto out_no_task;
1888 if (fd == ~0U)
1889 goto out;
1890
1891 result = instantiate(dir, dentry, task, (void *)(unsigned long)fd);
1892out:
1893 put_task_struct(task);
1894out_no_task:
1895 return result;
1896}
1897
1898static int proc_readfd_common(struct file * filp, void * dirent,
1899 filldir_t filldir, instantiate_t instantiate)
1900{
1901 struct dentry *dentry = filp->f_path.dentry;
1902 struct inode *inode = dentry->d_inode;
1903 struct task_struct *p = get_proc_task(inode);
1904 unsigned int fd, ino;
1905 int retval;
1906 struct files_struct * files;
1907
1908 retval = -ENOENT;
1909 if (!p)
1910 goto out_no_task;
1911 retval = 0;
1912
1913 fd = filp->f_pos;
1914 switch (fd) {
1915 case 0:
1916 if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1917 goto out;
1918 filp->f_pos++;
1919 case 1:
1920 ino = parent_ino(dentry);
1921 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1922 goto out;
1923 filp->f_pos++;
1924 default:
1925 files = get_files_struct(p);
1926 if (!files)
1927 goto out;
1928 rcu_read_lock();
1929 for (fd = filp->f_pos-2;
1930 fd < files_fdtable(files)->max_fds;
1931 fd++, filp->f_pos++) {
1932 char name[PROC_NUMBUF];
1933 int len;
1934 int rv;
1935
1936 if (!fcheck_files(files, fd))
1937 continue;
1938 rcu_read_unlock();
1939
1940 len = snprintf(name, sizeof(name), "%d", fd);
1941 rv = proc_fill_cache(filp, dirent, filldir,
1942 name, len, instantiate, p,
1943 (void *)(unsigned long)fd);
1944 if (rv < 0)
1945 goto out_fd_loop;
1946 rcu_read_lock();
1947 }
1948 rcu_read_unlock();
1949out_fd_loop:
1950 put_files_struct(files);
1951 }
1952out:
1953 put_task_struct(p);
1954out_no_task:
1955 return retval;
1956}
1957
1958static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1959 struct nameidata *nd)
1960{
1961 return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1962}
1963
1964static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1965{
1966 return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1967}
1968
1969static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1970 size_t len, loff_t *ppos)
1971{
1972 char tmp[PROC_FDINFO_MAX];
1973 int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1974 if (!err)
1975 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1976 return err;
1977}
1978
1979static const struct file_operations proc_fdinfo_file_operations = {
1980 .open = nonseekable_open,
1981 .read = proc_fdinfo_read,
1982 .llseek = no_llseek,
1983};
1984
1985static const struct file_operations proc_fd_operations = {
1986 .read = generic_read_dir,
1987 .readdir = proc_readfd,
1988 .llseek = default_llseek,
1989};
1990
1991#ifdef CONFIG_CHECKPOINT_RESTORE
1992
1993/*
1994 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1995 * which represent vma start and end addresses.
1996 */
1997static int dname_to_vma_addr(struct dentry *dentry,
1998 unsigned long *start, unsigned long *end)
1999{
2000 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2001 return -EINVAL;
2002
2003 return 0;
2004}
2005
2006static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2007{
2008 unsigned long vm_start, vm_end;
2009 bool exact_vma_exists = false;
2010 struct mm_struct *mm = NULL;
2011 struct task_struct *task;
2012 const struct cred *cred;
2013 struct inode *inode;
2014 int status = 0;
2015
2016 if (nd && nd->flags & LOOKUP_RCU)
2017 return -ECHILD;
2018
2019 if (!capable(CAP_SYS_ADMIN)) {
2020 status = -EACCES;
2021 goto out_notask;
2022 }
2023
2024 inode = dentry->d_inode;
2025 task = get_proc_task(inode);
2026 if (!task)
2027 goto out_notask;
2028
2029 mm = mm_access(task, PTRACE_MODE_READ);
2030 if (IS_ERR_OR_NULL(mm))
2031 goto out;
2032
2033 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2034 down_read(&mm->mmap_sem);
2035 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2036 up_read(&mm->mmap_sem);
2037 }
2038
2039 mmput(mm);
2040
2041 if (exact_vma_exists) {
2042 if (task_dumpable(task)) {
2043 rcu_read_lock();
2044 cred = __task_cred(task);
2045 inode->i_uid = cred->euid;
2046 inode->i_gid = cred->egid;
2047 rcu_read_unlock();
2048 } else {
2049 inode->i_uid = GLOBAL_ROOT_UID;
2050 inode->i_gid = GLOBAL_ROOT_GID;
2051 }
2052 security_task_to_inode(task, inode);
2053 status = 1;
2054 }
2055
2056out:
2057 put_task_struct(task);
2058
2059out_notask:
2060 if (status <= 0)
2061 d_drop(dentry);
2062
2063 return status;
2064}
2065
2066static const struct dentry_operations tid_map_files_dentry_operations = {
2067 .d_revalidate = map_files_d_revalidate,
2068 .d_delete = pid_delete_dentry,
2069};
2070
2071static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2072{
2073 unsigned long vm_start, vm_end;
2074 struct vm_area_struct *vma;
2075 struct task_struct *task;
2076 struct mm_struct *mm;
2077 int rc;
2078
2079 rc = -ENOENT;
2080 task = get_proc_task(dentry->d_inode);
2081 if (!task)
2082 goto out;
2083
2084 mm = get_task_mm(task);
2085 put_task_struct(task);
2086 if (!mm)
2087 goto out;
2088
2089 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2090 if (rc)
2091 goto out_mmput;
2092
2093 down_read(&mm->mmap_sem);
2094 vma = find_exact_vma(mm, vm_start, vm_end);
2095 if (vma && vma->vm_file) {
2096 *path = vma->vm_file->f_path;
2097 path_get(path);
2098 rc = 0;
2099 }
2100 up_read(&mm->mmap_sem);
2101
2102out_mmput:
2103 mmput(mm);
2104out:
2105 return rc;
2106}
2107
2108struct map_files_info {
2109 struct file *file;
2110 unsigned long len;
2111 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2112};
2113
2114static struct dentry *
2115proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2116 struct task_struct *task, const void *ptr)
2117{
2118 const struct file *file = ptr;
2119 struct proc_inode *ei;
2120 struct inode *inode;
2121
2122 if (!file)
2123 return ERR_PTR(-ENOENT);
2124
2125 inode = proc_pid_make_inode(dir->i_sb, task);
2126 if (!inode)
2127 return ERR_PTR(-ENOENT);
2128
2129 ei = PROC_I(inode);
2130 ei->op.proc_get_link = proc_map_files_get_link;
2131
2132 inode->i_op = &proc_pid_link_inode_operations;
2133 inode->i_size = 64;
2134 inode->i_mode = S_IFLNK;
2135
2136 if (file->f_mode & FMODE_READ)
2137 inode->i_mode |= S_IRUSR;
2138 if (file->f_mode & FMODE_WRITE)
2139 inode->i_mode |= S_IWUSR;
2140
2141 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2142 d_add(dentry, inode);
2143
2144 return NULL;
2145}
2146
2147static struct dentry *proc_map_files_lookup(struct inode *dir,
2148 struct dentry *dentry, struct nameidata *nd)
2149{
2150 unsigned long vm_start, vm_end;
2151 struct vm_area_struct *vma;
2152 struct task_struct *task;
2153 struct dentry *result;
2154 struct mm_struct *mm;
2155
2156 result = ERR_PTR(-EACCES);
2157 if (!capable(CAP_SYS_ADMIN))
2158 goto out;
2159
2160 result = ERR_PTR(-ENOENT);
2161 task = get_proc_task(dir);
2162 if (!task)
2163 goto out;
2164
2165 result = ERR_PTR(-EACCES);
2166 if (!ptrace_may_access(task, PTRACE_MODE_READ))
2167 goto out_put_task;
2168
2169 result = ERR_PTR(-ENOENT);
2170 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2171 goto out_put_task;
2172
2173 mm = get_task_mm(task);
2174 if (!mm)
2175 goto out_put_task;
2176
2177 down_read(&mm->mmap_sem);
2178 vma = find_exact_vma(mm, vm_start, vm_end);
2179 if (!vma)
2180 goto out_no_vma;
2181
2182 result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2183
2184out_no_vma:
2185 up_read(&mm->mmap_sem);
2186 mmput(mm);
2187out_put_task:
2188 put_task_struct(task);
2189out:
2190 return result;
2191}
2192
2193static const struct inode_operations proc_map_files_inode_operations = {
2194 .lookup = proc_map_files_lookup,
2195 .permission = proc_fd_permission,
2196 .setattr = proc_setattr,
2197};
2198
2199static int
2200proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2201{
2202 struct dentry *dentry = filp->f_path.dentry;
2203 struct inode *inode = dentry->d_inode;
2204 struct vm_area_struct *vma;
2205 struct task_struct *task;
2206 struct mm_struct *mm;
2207 ino_t ino;
2208 int ret;
2209
2210 ret = -EACCES;
2211 if (!capable(CAP_SYS_ADMIN))
2212 goto out;
2213
2214 ret = -ENOENT;
2215 task = get_proc_task(inode);
2216 if (!task)
2217 goto out;
2218
2219 ret = -EACCES;
2220 if (!ptrace_may_access(task, PTRACE_MODE_READ))
2221 goto out_put_task;
2222
2223 ret = 0;
2224 switch (filp->f_pos) {
2225 case 0:
2226 ino = inode->i_ino;
2227 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2228 goto out_put_task;
2229 filp->f_pos++;
2230 case 1:
2231 ino = parent_ino(dentry);
2232 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2233 goto out_put_task;
2234 filp->f_pos++;
2235 default:
2236 {
2237 unsigned long nr_files, pos, i;
2238 struct flex_array *fa = NULL;
2239 struct map_files_info info;
2240 struct map_files_info *p;
2241
2242 mm = get_task_mm(task);
2243 if (!mm)
2244 goto out_put_task;
2245 down_read(&mm->mmap_sem);
2246
2247 nr_files = 0;
2248
2249 /*
2250 * We need two passes here:
2251 *
2252 * 1) Collect vmas of mapped files with mmap_sem taken
2253 * 2) Release mmap_sem and instantiate entries
2254 *
2255 * otherwise we get lockdep complained, since filldir()
2256 * routine might require mmap_sem taken in might_fault().
2257 */
2258
2259 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2260 if (vma->vm_file && ++pos > filp->f_pos)
2261 nr_files++;
2262 }
2263
2264 if (nr_files) {
2265 fa = flex_array_alloc(sizeof(info), nr_files,
2266 GFP_KERNEL);
2267 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2268 GFP_KERNEL)) {
2269 ret = -ENOMEM;
2270 if (fa)
2271 flex_array_free(fa);
2272 up_read(&mm->mmap_sem);
2273 mmput(mm);
2274 goto out_put_task;
2275 }
2276 for (i = 0, vma = mm->mmap, pos = 2; vma;
2277 vma = vma->vm_next) {
2278 if (!vma->vm_file)
2279 continue;
2280 if (++pos <= filp->f_pos)
2281 continue;
2282
2283 get_file(vma->vm_file);
2284 info.file = vma->vm_file;
2285 info.len = snprintf(info.name,
2286 sizeof(info.name), "%lx-%lx",
2287 vma->vm_start, vma->vm_end);
2288 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2289 BUG();
2290 }
2291 }
2292 up_read(&mm->mmap_sem);
2293
2294 for (i = 0; i < nr_files; i++) {
2295 p = flex_array_get(fa, i);
2296 ret = proc_fill_cache(filp, dirent, filldir,
2297 p->name, p->len,
2298 proc_map_files_instantiate,
2299 task, p->file);
2300 if (ret)
2301 break;
2302 filp->f_pos++;
2303 fput(p->file);
2304 }
2305 for (; i < nr_files; i++) {
2306 /*
2307 * In case of error don't forget
2308 * to put rest of file refs.
2309 */
2310 p = flex_array_get(fa, i);
2311 fput(p->file);
2312 }
2313 if (fa)
2314 flex_array_free(fa);
2315 mmput(mm);
2316 }
2317 }
2318
2319out_put_task:
2320 put_task_struct(task);
2321out:
2322 return ret;
2323}
2324
2325static const struct file_operations proc_map_files_operations = {
2326 .read = generic_read_dir,
2327 .readdir = proc_map_files_readdir,
2328 .llseek = default_llseek,
2329};
2330
2331#endif /* CONFIG_CHECKPOINT_RESTORE */
2332
2333/*
2334 * /proc/pid/fd needs a special permission handler so that a process can still
2335 * access /proc/self/fd after it has executed a setuid().
2336 */
2337static int proc_fd_permission(struct inode *inode, int mask)
2338{
2339 int rv = generic_permission(inode, mask);
2340 if (rv == 0)
2341 return 0;
2342 if (task_pid(current) == proc_pid(inode))
2343 rv = 0;
2344 return rv;
2345}
2346
2347/*
2348 * proc directories can do almost nothing..
2349 */
2350static const struct inode_operations proc_fd_inode_operations = {
2351 .lookup = proc_lookupfd,
2352 .permission = proc_fd_permission,
2353 .setattr = proc_setattr,
2354};
2355
2356static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2357 struct dentry *dentry, struct task_struct *task, const void *ptr)
2358{
2359 unsigned fd = (unsigned long)ptr;
2360 struct inode *inode;
2361 struct proc_inode *ei;
2362 struct dentry *error = ERR_PTR(-ENOENT);
2363
2364 inode = proc_pid_make_inode(dir->i_sb, task);
2365 if (!inode)
2366 goto out;
2367 ei = PROC_I(inode);
2368 ei->fd = fd;
2369 inode->i_mode = S_IFREG | S_IRUSR;
2370 inode->i_fop = &proc_fdinfo_file_operations;
2371 d_set_d_op(dentry, &tid_fd_dentry_operations);
2372 d_add(dentry, inode);
2373 /* Close the race of the process dying before we return the dentry */
2374 if (tid_fd_revalidate(dentry, NULL))
2375 error = NULL;
2376
2377 out:
2378 return error;
2379}
2380
2381static struct dentry *proc_lookupfdinfo(struct inode *dir,
2382 struct dentry *dentry,
2383 struct nameidata *nd)
2384{
2385 return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2386}
2387
2388static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2389{
2390 return proc_readfd_common(filp, dirent, filldir,
2391 proc_fdinfo_instantiate);
2392}
2393
2394static const struct file_operations proc_fdinfo_operations = {
2395 .read = generic_read_dir,
2396 .readdir = proc_readfdinfo,
2397 .llseek = default_llseek,
2398};
2399
2400/*
2401 * proc directories can do almost nothing..
2402 */
2403static const struct inode_operations proc_fdinfo_inode_operations = {
2404 .lookup = proc_lookupfdinfo,
2405 .setattr = proc_setattr,
2406};
2407
2408
2409static struct dentry *proc_pident_instantiate(struct inode *dir,
2410 struct dentry *dentry, struct task_struct *task, const void *ptr)
2411{
2412 const struct pid_entry *p = ptr;
2413 struct inode *inode;
2414 struct proc_inode *ei;
2415 struct dentry *error = ERR_PTR(-ENOENT);
2416
2417 inode = proc_pid_make_inode(dir->i_sb, task);
2418 if (!inode)
2419 goto out;
2420
2421 ei = PROC_I(inode);
2422 inode->i_mode = p->mode;
2423 if (S_ISDIR(inode->i_mode))
2424 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2425 if (p->iop)
2426 inode->i_op = p->iop;
2427 if (p->fop)
2428 inode->i_fop = p->fop;
2429 ei->op = p->op;
2430 d_set_d_op(dentry, &pid_dentry_operations);
2431 d_add(dentry, inode);
2432 /* Close the race of the process dying before we return the dentry */
2433 if (pid_revalidate(dentry, NULL))
2434 error = NULL;
2435out:
2436 return error;
2437}
2438
2439static struct dentry *proc_pident_lookup(struct inode *dir,
2440 struct dentry *dentry,
2441 const struct pid_entry *ents,
2442 unsigned int nents)
2443{
2444 struct dentry *error;
2445 struct task_struct *task = get_proc_task(dir);
2446 const struct pid_entry *p, *last;
2447
2448 error = ERR_PTR(-ENOENT);
2449
2450 if (!task)
2451 goto out_no_task;
2452
2453 /*
2454 * Yes, it does not scale. And it should not. Don't add
2455 * new entries into /proc/<tgid>/ without very good reasons.
2456 */
2457 last = &ents[nents - 1];
2458 for (p = ents; p <= last; p++) {
2459 if (p->len != dentry->d_name.len)
2460 continue;
2461 if (!memcmp(dentry->d_name.name, p->name, p->len))
2462 break;
2463 }
2464 if (p > last)
2465 goto out;
2466
2467 error = proc_pident_instantiate(dir, dentry, task, p);
2468out:
2469 put_task_struct(task);
2470out_no_task:
2471 return error;
2472}
2473
2474static int proc_pident_fill_cache(struct file *filp, void *dirent,
2475 filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2476{
2477 return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2478 proc_pident_instantiate, task, p);
2479}
2480
2481static int proc_pident_readdir(struct file *filp,
2482 void *dirent, filldir_t filldir,
2483 const struct pid_entry *ents, unsigned int nents)
2484{
2485 int i;
2486 struct dentry *dentry = filp->f_path.dentry;
2487 struct inode *inode = dentry->d_inode;
2488 struct task_struct *task = get_proc_task(inode);
2489 const struct pid_entry *p, *last;
2490 ino_t ino;
2491 int ret;
2492
2493 ret = -ENOENT;
2494 if (!task)
2495 goto out_no_task;
2496
2497 ret = 0;
2498 i = filp->f_pos;
2499 switch (i) {
2500 case 0:
2501 ino = inode->i_ino;
2502 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2503 goto out;
2504 i++;
2505 filp->f_pos++;
2506 /* fall through */
2507 case 1:
2508 ino = parent_ino(dentry);
2509 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2510 goto out;
2511 i++;
2512 filp->f_pos++;
2513 /* fall through */
2514 default:
2515 i -= 2;
2516 if (i >= nents) {
2517 ret = 1;
2518 goto out;
2519 }
2520 p = ents + i;
2521 last = &ents[nents - 1];
2522 while (p <= last) {
2523 if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2524 goto out;
2525 filp->f_pos++;
2526 p++;
2527 }
2528 }
2529
2530 ret = 1;
2531out:
2532 put_task_struct(task);
2533out_no_task:
2534 return ret;
2535}
2536
2537#ifdef CONFIG_SECURITY
2538static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2539 size_t count, loff_t *ppos)
2540{
2541 struct inode * inode = file->f_path.dentry->d_inode;
2542 char *p = NULL;
2543 ssize_t length;
2544 struct task_struct *task = get_proc_task(inode);
2545
2546 if (!task)
2547 return -ESRCH;
2548
2549 length = security_getprocattr(task,
2550 (char*)file->f_path.dentry->d_name.name,
2551 &p);
2552 put_task_struct(task);
2553 if (length > 0)
2554 length = simple_read_from_buffer(buf, count, ppos, p, length);
2555 kfree(p);
2556 return length;
2557}
2558
2559static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2560 size_t count, loff_t *ppos)
2561{
2562 struct inode * inode = file->f_path.dentry->d_inode;
2563 char *page;
2564 ssize_t length;
2565 struct task_struct *task = get_proc_task(inode);
2566
2567 length = -ESRCH;
2568 if (!task)
2569 goto out_no_task;
2570 if (count > PAGE_SIZE)
2571 count = PAGE_SIZE;
2572
2573 /* No partial writes. */
2574 length = -EINVAL;
2575 if (*ppos != 0)
2576 goto out;
2577
2578 length = -ENOMEM;
2579 page = (char*)__get_free_page(GFP_TEMPORARY);
2580 if (!page)
2581 goto out;
2582
2583 length = -EFAULT;
2584 if (copy_from_user(page, buf, count))
2585 goto out_free;
2586
2587 /* Guard against adverse ptrace interaction */
2588 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2589 if (length < 0)
2590 goto out_free;
2591
2592 length = security_setprocattr(task,
2593 (char*)file->f_path.dentry->d_name.name,
2594 (void*)page, count);
2595 mutex_unlock(&task->signal->cred_guard_mutex);
2596out_free:
2597 free_page((unsigned long) page);
2598out:
2599 put_task_struct(task);
2600out_no_task:
2601 return length;
2602}
2603
2604static const struct file_operations proc_pid_attr_operations = {
2605 .read = proc_pid_attr_read,
2606 .write = proc_pid_attr_write,
2607 .llseek = generic_file_llseek,
2608};
2609
2610static const struct pid_entry attr_dir_stuff[] = {
2611 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2612 REG("prev", S_IRUGO, proc_pid_attr_operations),
2613 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2614 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2615 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2616 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2617};
2618
2619static int proc_attr_dir_readdir(struct file * filp,
2620 void * dirent, filldir_t filldir)
2621{
2622 return proc_pident_readdir(filp,dirent,filldir,
2623 attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2624}
2625
2626static const struct file_operations proc_attr_dir_operations = {
2627 .read = generic_read_dir,
2628 .readdir = proc_attr_dir_readdir,
2629 .llseek = default_llseek,
2630};
2631
2632static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2633 struct dentry *dentry, struct nameidata *nd)
2634{
2635 return proc_pident_lookup(dir, dentry,
2636 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2637}
2638
2639static const struct inode_operations proc_attr_dir_inode_operations = {
2640 .lookup = proc_attr_dir_lookup,
2641 .getattr = pid_getattr,
2642 .setattr = proc_setattr,
2643};
2644
2645#endif
2646
2647#ifdef CONFIG_ELF_CORE
2648static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2649 size_t count, loff_t *ppos)
2650{
2651 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2652 struct mm_struct *mm;
2653 char buffer[PROC_NUMBUF];
2654 size_t len;
2655 int ret;
2656
2657 if (!task)
2658 return -ESRCH;
2659
2660 ret = 0;
2661 mm = get_task_mm(task);
2662 if (mm) {
2663 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2664 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2665 MMF_DUMP_FILTER_SHIFT));
2666 mmput(mm);
2667 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2668 }
2669
2670 put_task_struct(task);
2671
2672 return ret;
2673}
2674
2675static ssize_t proc_coredump_filter_write(struct file *file,
2676 const char __user *buf,
2677 size_t count,
2678 loff_t *ppos)
2679{
2680 struct task_struct *task;
2681 struct mm_struct *mm;
2682 char buffer[PROC_NUMBUF], *end;
2683 unsigned int val;
2684 int ret;
2685 int i;
2686 unsigned long mask;
2687
2688 ret = -EFAULT;
2689 memset(buffer, 0, sizeof(buffer));
2690 if (count > sizeof(buffer) - 1)
2691 count = sizeof(buffer) - 1;
2692 if (copy_from_user(buffer, buf, count))
2693 goto out_no_task;
2694
2695 ret = -EINVAL;
2696 val = (unsigned int)simple_strtoul(buffer, &end, 0);
2697 if (*end == '\n')
2698 end++;
2699 if (end - buffer == 0)
2700 goto out_no_task;
2701
2702 ret = -ESRCH;
2703 task = get_proc_task(file->f_dentry->d_inode);
2704 if (!task)
2705 goto out_no_task;
2706
2707 ret = end - buffer;
2708 mm = get_task_mm(task);
2709 if (!mm)
2710 goto out_no_mm;
2711
2712 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2713 if (val & mask)
2714 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2715 else
2716 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2717 }
2718
2719 mmput(mm);
2720 out_no_mm:
2721 put_task_struct(task);
2722 out_no_task:
2723 return ret;
2724}
2725
2726static const struct file_operations proc_coredump_filter_operations = {
2727 .read = proc_coredump_filter_read,
2728 .write = proc_coredump_filter_write,
2729 .llseek = generic_file_llseek,
2730};
2731#endif
2732
2733/*
2734 * /proc/self:
2735 */
2736static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2737 int buflen)
2738{
2739 struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2740 pid_t tgid = task_tgid_nr_ns(current, ns);
2741 char tmp[PROC_NUMBUF];
2742 if (!tgid)
2743 return -ENOENT;
2744 sprintf(tmp, "%d", tgid);
2745 return vfs_readlink(dentry,buffer,buflen,tmp);
2746}
2747
2748static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2749{
2750 struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2751 pid_t tgid = task_tgid_nr_ns(current, ns);
2752 char *name = ERR_PTR(-ENOENT);
2753 if (tgid) {
2754 name = __getname();
2755 if (!name)
2756 name = ERR_PTR(-ENOMEM);
2757 else
2758 sprintf(name, "%d", tgid);
2759 }
2760 nd_set_link(nd, name);
2761 return NULL;
2762}
2763
2764static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2765 void *cookie)
2766{
2767 char *s = nd_get_link(nd);
2768 if (!IS_ERR(s))
2769 __putname(s);
2770}
2771
2772static const struct inode_operations proc_self_inode_operations = {
2773 .readlink = proc_self_readlink,
2774 .follow_link = proc_self_follow_link,
2775 .put_link = proc_self_put_link,
2776};
2777
2778/*
2779 * proc base
2780 *
2781 * These are the directory entries in the root directory of /proc
2782 * that properly belong to the /proc filesystem, as they describe
2783 * describe something that is process related.
2784 */
2785static const struct pid_entry proc_base_stuff[] = {
2786 NOD("self", S_IFLNK|S_IRWXUGO,
2787 &proc_self_inode_operations, NULL, {}),
2788};
2789
2790static struct dentry *proc_base_instantiate(struct inode *dir,
2791 struct dentry *dentry, struct task_struct *task, const void *ptr)
2792{
2793 const struct pid_entry *p = ptr;
2794 struct inode *inode;
2795 struct proc_inode *ei;
2796 struct dentry *error;
2797
2798 /* Allocate the inode */
2799 error = ERR_PTR(-ENOMEM);
2800 inode = new_inode(dir->i_sb);
2801 if (!inode)
2802 goto out;
2803
2804 /* Initialize the inode */
2805 ei = PROC_I(inode);
2806 inode->i_ino = get_next_ino();
2807 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2808
2809 /*
2810 * grab the reference to the task.
2811 */
2812 ei->pid = get_task_pid(task, PIDTYPE_PID);
2813 if (!ei->pid)
2814 goto out_iput;
2815
2816 inode->i_mode = p->mode;
2817 if (S_ISDIR(inode->i_mode))
2818 set_nlink(inode, 2);
2819 if (S_ISLNK(inode->i_mode))
2820 inode->i_size = 64;
2821 if (p->iop)
2822 inode->i_op = p->iop;
2823 if (p->fop)
2824 inode->i_fop = p->fop;
2825 ei->op = p->op;
2826 d_add(dentry, inode);
2827 error = NULL;
2828out:
2829 return error;
2830out_iput:
2831 iput(inode);
2832 goto out;
2833}
2834
2835static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2836{
2837 struct dentry *error;
2838 struct task_struct *task = get_proc_task(dir);
2839 const struct pid_entry *p, *last;
2840
2841 error = ERR_PTR(-ENOENT);
2842
2843 if (!task)
2844 goto out_no_task;
2845
2846 /* Lookup the directory entry */
2847 last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2848 for (p = proc_base_stuff; p <= last; p++) {
2849 if (p->len != dentry->d_name.len)
2850 continue;
2851 if (!memcmp(dentry->d_name.name, p->name, p->len))
2852 break;
2853 }
2854 if (p > last)
2855 goto out;
2856
2857 error = proc_base_instantiate(dir, dentry, task, p);
2858
2859out:
2860 put_task_struct(task);
2861out_no_task:
2862 return error;
2863}
2864
2865static int proc_base_fill_cache(struct file *filp, void *dirent,
2866 filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2867{
2868 return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2869 proc_base_instantiate, task, p);
2870}
2871
2872#ifdef CONFIG_TASK_IO_ACCOUNTING
2873static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2874{
2875 struct task_io_accounting acct = task->ioac;
2876 unsigned long flags;
2877 int result;
2878
2879 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2880 if (result)
2881 return result;
2882
2883 if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2884 result = -EACCES;
2885 goto out_unlock;
2886 }
2887
2888 if (whole && lock_task_sighand(task, &flags)) {
2889 struct task_struct *t = task;
2890
2891 task_io_accounting_add(&acct, &task->signal->ioac);
2892 while_each_thread(task, t)
2893 task_io_accounting_add(&acct, &t->ioac);
2894
2895 unlock_task_sighand(task, &flags);
2896 }
2897 result = sprintf(buffer,
2898 "rchar: %llu\n"
2899 "wchar: %llu\n"
2900 "syscr: %llu\n"
2901 "syscw: %llu\n"
2902 "read_bytes: %llu\n"
2903 "write_bytes: %llu\n"
2904 "cancelled_write_bytes: %llu\n",
2905 (unsigned long long)acct.rchar,
2906 (unsigned long long)acct.wchar,
2907 (unsigned long long)acct.syscr,
2908 (unsigned long long)acct.syscw,
2909 (unsigned long long)acct.read_bytes,
2910 (unsigned long long)acct.write_bytes,
2911 (unsigned long long)acct.cancelled_write_bytes);
2912out_unlock:
2913 mutex_unlock(&task->signal->cred_guard_mutex);
2914 return result;
2915}
2916
2917static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2918{
2919 return do_io_accounting(task, buffer, 0);
2920}
2921
2922static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2923{
2924 return do_io_accounting(task, buffer, 1);
2925}
2926#endif /* CONFIG_TASK_IO_ACCOUNTING */
2927
2928#ifdef CONFIG_USER_NS
2929static int proc_id_map_open(struct inode *inode, struct file *file,
2930 struct seq_operations *seq_ops)
2931{
2932 struct user_namespace *ns = NULL;
2933 struct task_struct *task;
2934 struct seq_file *seq;
2935 int ret = -EINVAL;
2936
2937 task = get_proc_task(inode);
2938 if (task) {
2939 rcu_read_lock();
2940 ns = get_user_ns(task_cred_xxx(task, user_ns));
2941 rcu_read_unlock();
2942 put_task_struct(task);
2943 }
2944 if (!ns)
2945 goto err;
2946
2947 ret = seq_open(file, seq_ops);
2948 if (ret)
2949 goto err_put_ns;
2950
2951 seq = file->private_data;
2952 seq->private = ns;
2953
2954 return 0;
2955err_put_ns:
2956 put_user_ns(ns);
2957err:
2958 return ret;
2959}
2960
2961static int proc_id_map_release(struct inode *inode, struct file *file)
2962{
2963 struct seq_file *seq = file->private_data;
2964 struct user_namespace *ns = seq->private;
2965 put_user_ns(ns);
2966 return seq_release(inode, file);
2967}
2968
2969static int proc_uid_map_open(struct inode *inode, struct file *file)
2970{
2971 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2972}
2973
2974static int proc_gid_map_open(struct inode *inode, struct file *file)
2975{
2976 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2977}
2978
2979static const struct file_operations proc_uid_map_operations = {
2980 .open = proc_uid_map_open,
2981 .write = proc_uid_map_write,
2982 .read = seq_read,
2983 .llseek = seq_lseek,
2984 .release = proc_id_map_release,
2985};
2986
2987static const struct file_operations proc_gid_map_operations = {
2988 .open = proc_gid_map_open,
2989 .write = proc_gid_map_write,
2990 .read = seq_read,
2991 .llseek = seq_lseek,
2992 .release = proc_id_map_release,
2993};
2994#endif /* CONFIG_USER_NS */
2995
2996static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2997 struct pid *pid, struct task_struct *task)
2998{
2999 int err = lock_trace(task);
3000 if (!err) {
3001 seq_printf(m, "%08x\n", task->personality);
3002 unlock_trace(task);
3003 }
3004 return err;
3005}
3006
3007/*
3008 * Thread groups
3009 */
3010static const struct file_operations proc_task_operations;
3011static const struct inode_operations proc_task_inode_operations;
3012
3013static const struct pid_entry tgid_base_stuff[] = {
3014 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3015 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3016#ifdef CONFIG_CHECKPOINT_RESTORE
3017 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3018#endif
3019 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3020 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3021#ifdef CONFIG_NET
3022 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3023#endif
3024 REG("environ", S_IRUSR, proc_environ_operations),
3025 INF("auxv", S_IRUSR, proc_pid_auxv),
3026 ONE("status", S_IRUGO, proc_pid_status),
3027 ONE("personality", S_IRUGO, proc_pid_personality),
3028 INF("limits", S_IRUGO, proc_pid_limits),
3029#ifdef CONFIG_SCHED_DEBUG
3030 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3031#endif
3032#ifdef CONFIG_SCHED_AUTOGROUP
3033 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3034#endif
3035 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3036#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3037 INF("syscall", S_IRUGO, proc_pid_syscall),
3038#endif
3039 INF("cmdline", S_IRUGO, proc_pid_cmdline),
3040 ONE("stat", S_IRUGO, proc_tgid_stat),
3041 ONE("statm", S_IRUGO, proc_pid_statm),
3042 REG("maps", S_IRUGO, proc_pid_maps_operations),
3043#ifdef CONFIG_NUMA
3044 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3045#endif
3046 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3047 LNK("cwd", proc_cwd_link),
3048 LNK("root", proc_root_link),
3049 LNK("exe", proc_exe_link),
3050 REG("mounts", S_IRUGO, proc_mounts_operations),
3051 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3052 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3053#ifdef CONFIG_PROC_PAGE_MONITOR
3054 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3055 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3056 REG("pagemap", S_IRUGO, proc_pagemap_operations),
3057#endif
3058#ifdef CONFIG_SECURITY
3059 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3060#endif
3061#ifdef CONFIG_KALLSYMS
3062 INF("wchan", S_IRUGO, proc_pid_wchan),
3063#endif
3064#ifdef CONFIG_STACKTRACE
3065 ONE("stack", S_IRUGO, proc_pid_stack),
3066#endif
3067#ifdef CONFIG_SCHEDSTATS
3068 INF("schedstat", S_IRUGO, proc_pid_schedstat),
3069#endif
3070#ifdef CONFIG_LATENCYTOP
3071 REG("latency", S_IRUGO, proc_lstats_operations),
3072#endif
3073#ifdef CONFIG_PROC_PID_CPUSET
3074 REG("cpuset", S_IRUGO, proc_cpuset_operations),
3075#endif
3076#ifdef CONFIG_CGROUPS
3077 REG("cgroup", S_IRUGO, proc_cgroup_operations),
3078#endif
3079 INF("oom_score", S_IRUGO, proc_oom_score),
3080 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3081 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3082#ifdef CONFIG_AUDITSYSCALL
3083 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3084 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3085#endif
3086#ifdef CONFIG_FAULT_INJECTION
3087 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3088#endif
3089#ifdef CONFIG_ELF_CORE
3090 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3091#endif
3092#ifdef CONFIG_TASK_IO_ACCOUNTING
3093 INF("io", S_IRUSR, proc_tgid_io_accounting),
3094#endif
3095#ifdef CONFIG_HARDWALL
3096 INF("hardwall", S_IRUGO, proc_pid_hardwall),
3097#endif
3098#ifdef CONFIG_USER_NS
3099 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3100 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3101#endif
3102};
3103
3104static int proc_tgid_base_readdir(struct file * filp,
3105 void * dirent, filldir_t filldir)
3106{
3107 return proc_pident_readdir(filp,dirent,filldir,
3108 tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3109}
3110
3111static const struct file_operations proc_tgid_base_operations = {
3112 .read = generic_read_dir,
3113 .readdir = proc_tgid_base_readdir,
3114 .llseek = default_llseek,
3115};
3116
3117static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3118 return proc_pident_lookup(dir, dentry,
3119 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3120}
3121
3122static const struct inode_operations proc_tgid_base_inode_operations = {
3123 .lookup = proc_tgid_base_lookup,
3124 .getattr = pid_getattr,
3125 .setattr = proc_setattr,
3126 .permission = proc_pid_permission,
3127};
3128
3129static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3130{
3131 struct dentry *dentry, *leader, *dir;
3132 char buf[PROC_NUMBUF];
3133 struct qstr name;
3134
3135 name.name = buf;
3136 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3137 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3138 if (dentry) {
3139 shrink_dcache_parent(dentry);
3140 d_drop(dentry);
3141 dput(dentry);
3142 }
3143
3144 name.name = buf;
3145 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3146 leader = d_hash_and_lookup(mnt->mnt_root, &name);
3147 if (!leader)
3148 goto out;
3149
3150 name.name = "task";
3151 name.len = strlen(name.name);
3152 dir = d_hash_and_lookup(leader, &name);
3153 if (!dir)
3154 goto out_put_leader;
3155
3156 name.name = buf;
3157 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3158 dentry = d_hash_and_lookup(dir, &name);
3159 if (dentry) {
3160 shrink_dcache_parent(dentry);
3161 d_drop(dentry);
3162 dput(dentry);
3163 }
3164
3165 dput(dir);
3166out_put_leader:
3167 dput(leader);
3168out:
3169 return;
3170}
3171
3172/**
3173 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3174 * @task: task that should be flushed.
3175 *
3176 * When flushing dentries from proc, one needs to flush them from global
3177 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3178 * in. This call is supposed to do all of this job.
3179 *
3180 * Looks in the dcache for
3181 * /proc/@pid
3182 * /proc/@tgid/task/@pid
3183 * if either directory is present flushes it and all of it'ts children
3184 * from the dcache.
3185 *
3186 * It is safe and reasonable to cache /proc entries for a task until
3187 * that task exits. After that they just clog up the dcache with
3188 * useless entries, possibly causing useful dcache entries to be
3189 * flushed instead. This routine is proved to flush those useless
3190 * dcache entries at process exit time.
3191 *
3192 * NOTE: This routine is just an optimization so it does not guarantee
3193 * that no dcache entries will exist at process exit time it
3194 * just makes it very unlikely that any will persist.
3195 */
3196
3197void proc_flush_task(struct task_struct *task)
3198{
3199 int i;
3200 struct pid *pid, *tgid;
3201 struct upid *upid;
3202
3203 pid = task_pid(task);
3204 tgid = task_tgid(task);
3205
3206 for (i = 0; i <= pid->level; i++) {
3207 upid = &pid->numbers[i];
3208 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3209 tgid->numbers[i].nr);
3210 }
3211
3212 upid = &pid->numbers[pid->level];
3213 if (upid->nr == 1)
3214 pid_ns_release_proc(upid->ns);
3215}
3216
3217static struct dentry *proc_pid_instantiate(struct inode *dir,
3218 struct dentry * dentry,
3219 struct task_struct *task, const void *ptr)
3220{
3221 struct dentry *error = ERR_PTR(-ENOENT);
3222 struct inode *inode;
3223
3224 inode = proc_pid_make_inode(dir->i_sb, task);
3225 if (!inode)
3226 goto out;
3227
3228 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3229 inode->i_op = &proc_tgid_base_inode_operations;
3230 inode->i_fop = &proc_tgid_base_operations;
3231 inode->i_flags|=S_IMMUTABLE;
3232
3233 set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3234 ARRAY_SIZE(tgid_base_stuff)));
3235
3236 d_set_d_op(dentry, &pid_dentry_operations);
3237
3238 d_add(dentry, inode);
3239 /* Close the race of the process dying before we return the dentry */
3240 if (pid_revalidate(dentry, NULL))
3241 error = NULL;
3242out:
3243 return error;
3244}
3245
3246struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3247{
3248 struct dentry *result;
3249 struct task_struct *task;
3250 unsigned tgid;
3251 struct pid_namespace *ns;
3252
3253 result = proc_base_lookup(dir, dentry);
3254 if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3255 goto out;
3256
3257 tgid = name_to_int(dentry);
3258 if (tgid == ~0U)
3259 goto out;
3260
3261 ns = dentry->d_sb->s_fs_info;
3262 rcu_read_lock();
3263 task = find_task_by_pid_ns(tgid, ns);
3264 if (task)
3265 get_task_struct(task);
3266 rcu_read_unlock();
3267 if (!task)
3268 goto out;
3269
3270 result = proc_pid_instantiate(dir, dentry, task, NULL);
3271 put_task_struct(task);
3272out:
3273 return result;
3274}
3275
3276/*
3277 * Find the first task with tgid >= tgid
3278 *
3279 */
3280struct tgid_iter {
3281 unsigned int tgid;
3282 struct task_struct *task;
3283};
3284static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3285{
3286 struct pid *pid;
3287
3288 if (iter.task)
3289 put_task_struct(iter.task);
3290 rcu_read_lock();
3291retry:
3292 iter.task = NULL;
3293 pid = find_ge_pid(iter.tgid, ns);
3294 if (pid) {
3295 iter.tgid = pid_nr_ns(pid, ns);
3296 iter.task = pid_task(pid, PIDTYPE_PID);
3297 /* What we to know is if the pid we have find is the
3298 * pid of a thread_group_leader. Testing for task
3299 * being a thread_group_leader is the obvious thing
3300 * todo but there is a window when it fails, due to
3301 * the pid transfer logic in de_thread.
3302 *
3303 * So we perform the straight forward test of seeing
3304 * if the pid we have found is the pid of a thread
3305 * group leader, and don't worry if the task we have
3306 * found doesn't happen to be a thread group leader.
3307 * As we don't care in the case of readdir.
3308 */
3309 if (!iter.task || !has_group_leader_pid(iter.task)) {
3310 iter.tgid += 1;
3311 goto retry;
3312 }
3313 get_task_struct(iter.task);
3314 }
3315 rcu_read_unlock();
3316 return iter;
3317}
3318
3319#define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3320
3321static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3322 struct tgid_iter iter)
3323{
3324 char name[PROC_NUMBUF];
3325 int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3326 return proc_fill_cache(filp, dirent, filldir, name, len,
3327 proc_pid_instantiate, iter.task, NULL);
3328}
3329
3330static int fake_filldir(void *buf, const char *name, int namelen,
3331 loff_t offset, u64 ino, unsigned d_type)
3332{
3333 return 0;
3334}
3335
3336/* for the /proc/ directory itself, after non-process stuff has been done */
3337int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3338{
3339 unsigned int nr;
3340 struct task_struct *reaper;
3341 struct tgid_iter iter;
3342 struct pid_namespace *ns;
3343 filldir_t __filldir;
3344
3345 if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3346 goto out_no_task;
3347 nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3348
3349 reaper = get_proc_task(filp->f_path.dentry->d_inode);
3350 if (!reaper)
3351 goto out_no_task;
3352
3353 for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3354 const struct pid_entry *p = &proc_base_stuff[nr];
3355 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3356 goto out;
3357 }
3358
3359 ns = filp->f_dentry->d_sb->s_fs_info;
3360 iter.task = NULL;
3361 iter.tgid = filp->f_pos - TGID_OFFSET;
3362 for (iter = next_tgid(ns, iter);
3363 iter.task;
3364 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3365 if (has_pid_permissions(ns, iter.task, 2))
3366 __filldir = filldir;
3367 else
3368 __filldir = fake_filldir;
3369
3370 filp->f_pos = iter.tgid + TGID_OFFSET;
3371 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3372 put_task_struct(iter.task);
3373 goto out;
3374 }
3375 }
3376 filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3377out:
3378 put_task_struct(reaper);
3379out_no_task:
3380 return 0;
3381}
3382
3383/*
3384 * Tasks
3385 */
3386static const struct pid_entry tid_base_stuff[] = {
3387 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3388 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3389 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3390 REG("environ", S_IRUSR, proc_environ_operations),
3391 INF("auxv", S_IRUSR, proc_pid_auxv),
3392 ONE("status", S_IRUGO, proc_pid_status),
3393 ONE("personality", S_IRUGO, proc_pid_personality),
3394 INF("limits", S_IRUGO, proc_pid_limits),
3395#ifdef CONFIG_SCHED_DEBUG
3396 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3397#endif
3398 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3399#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3400 INF("syscall", S_IRUGO, proc_pid_syscall),
3401#endif
3402 INF("cmdline", S_IRUGO, proc_pid_cmdline),
3403 ONE("stat", S_IRUGO, proc_tid_stat),
3404 ONE("statm", S_IRUGO, proc_pid_statm),
3405 REG("maps", S_IRUGO, proc_tid_maps_operations),
3406#ifdef CONFIG_CHECKPOINT_RESTORE
3407 REG("children", S_IRUGO, proc_tid_children_operations),
3408#endif
3409#ifdef CONFIG_NUMA
3410 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3411#endif
3412 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3413 LNK("cwd", proc_cwd_link),
3414 LNK("root", proc_root_link),
3415 LNK("exe", proc_exe_link),
3416 REG("mounts", S_IRUGO, proc_mounts_operations),
3417 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3418#ifdef CONFIG_PROC_PAGE_MONITOR
3419 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3420 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3421 REG("pagemap", S_IRUGO, proc_pagemap_operations),
3422#endif
3423#ifdef CONFIG_SECURITY
3424 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3425#endif
3426#ifdef CONFIG_KALLSYMS
3427 INF("wchan", S_IRUGO, proc_pid_wchan),
3428#endif
3429#ifdef CONFIG_STACKTRACE
3430 ONE("stack", S_IRUGO, proc_pid_stack),
3431#endif
3432#ifdef CONFIG_SCHEDSTATS
3433 INF("schedstat", S_IRUGO, proc_pid_schedstat),
3434#endif
3435#ifdef CONFIG_LATENCYTOP
3436 REG("latency", S_IRUGO, proc_lstats_operations),
3437#endif
3438#ifdef CONFIG_PROC_PID_CPUSET
3439 REG("cpuset", S_IRUGO, proc_cpuset_operations),
3440#endif
3441#ifdef CONFIG_CGROUPS
3442 REG("cgroup", S_IRUGO, proc_cgroup_operations),
3443#endif
3444 INF("oom_score", S_IRUGO, proc_oom_score),
3445 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3446 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3447#ifdef CONFIG_AUDITSYSCALL
3448 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3449 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3450#endif
3451#ifdef CONFIG_FAULT_INJECTION
3452 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3453#endif
3454#ifdef CONFIG_TASK_IO_ACCOUNTING
3455 INF("io", S_IRUSR, proc_tid_io_accounting),
3456#endif
3457#ifdef CONFIG_HARDWALL
3458 INF("hardwall", S_IRUGO, proc_pid_hardwall),
3459#endif
3460#ifdef CONFIG_USER_NS
3461 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3462 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3463#endif
3464};
3465
3466static int proc_tid_base_readdir(struct file * filp,
3467 void * dirent, filldir_t filldir)
3468{
3469 return proc_pident_readdir(filp,dirent,filldir,
3470 tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3471}
3472
3473static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3474 return proc_pident_lookup(dir, dentry,
3475 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3476}
3477
3478static const struct file_operations proc_tid_base_operations = {
3479 .read = generic_read_dir,
3480 .readdir = proc_tid_base_readdir,
3481 .llseek = default_llseek,
3482};
3483
3484static const struct inode_operations proc_tid_base_inode_operations = {
3485 .lookup = proc_tid_base_lookup,
3486 .getattr = pid_getattr,
3487 .setattr = proc_setattr,
3488};
3489
3490static struct dentry *proc_task_instantiate(struct inode *dir,
3491 struct dentry *dentry, struct task_struct *task, const void *ptr)
3492{
3493 struct dentry *error = ERR_PTR(-ENOENT);
3494 struct inode *inode;
3495 inode = proc_pid_make_inode(dir->i_sb, task);
3496
3497 if (!inode)
3498 goto out;
3499 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3500 inode->i_op = &proc_tid_base_inode_operations;
3501 inode->i_fop = &proc_tid_base_operations;
3502 inode->i_flags|=S_IMMUTABLE;
3503
3504 set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3505 ARRAY_SIZE(tid_base_stuff)));
3506
3507 d_set_d_op(dentry, &pid_dentry_operations);
3508
3509 d_add(dentry, inode);
3510 /* Close the race of the process dying before we return the dentry */
3511 if (pid_revalidate(dentry, NULL))
3512 error = NULL;
3513out:
3514 return error;
3515}
3516
3517static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3518{
3519 struct dentry *result = ERR_PTR(-ENOENT);
3520 struct task_struct *task;
3521 struct task_struct *leader = get_proc_task(dir);
3522 unsigned tid;
3523 struct pid_namespace *ns;
3524
3525 if (!leader)
3526 goto out_no_task;
3527
3528 tid = name_to_int(dentry);
3529 if (tid == ~0U)
3530 goto out;
3531
3532 ns = dentry->d_sb->s_fs_info;
3533 rcu_read_lock();
3534 task = find_task_by_pid_ns(tid, ns);
3535 if (task)
3536 get_task_struct(task);
3537 rcu_read_unlock();
3538 if (!task)
3539 goto out;
3540 if (!same_thread_group(leader, task))
3541 goto out_drop_task;
3542
3543 result = proc_task_instantiate(dir, dentry, task, NULL);
3544out_drop_task:
3545 put_task_struct(task);
3546out:
3547 put_task_struct(leader);
3548out_no_task:
3549 return result;
3550}
3551
3552/*
3553 * Find the first tid of a thread group to return to user space.
3554 *
3555 * Usually this is just the thread group leader, but if the users
3556 * buffer was too small or there was a seek into the middle of the
3557 * directory we have more work todo.
3558 *
3559 * In the case of a short read we start with find_task_by_pid.
3560 *
3561 * In the case of a seek we start with the leader and walk nr
3562 * threads past it.
3563 */
3564static struct task_struct *first_tid(struct task_struct *leader,
3565 int tid, int nr, struct pid_namespace *ns)
3566{
3567 struct task_struct *pos;
3568
3569 rcu_read_lock();
3570 /* Attempt to start with the pid of a thread */
3571 if (tid && (nr > 0)) {
3572 pos = find_task_by_pid_ns(tid, ns);
3573 if (pos && (pos->group_leader == leader))
3574 goto found;
3575 }
3576
3577 /* If nr exceeds the number of threads there is nothing todo */
3578 pos = NULL;
3579 if (nr && nr >= get_nr_threads(leader))
3580 goto out;
3581
3582 /* If we haven't found our starting place yet start
3583 * with the leader and walk nr threads forward.
3584 */
3585 for (pos = leader; nr > 0; --nr) {
3586 pos = next_thread(pos);
3587 if (pos == leader) {
3588 pos = NULL;
3589 goto out;
3590 }
3591 }
3592found:
3593 get_task_struct(pos);
3594out:
3595 rcu_read_unlock();
3596 return pos;
3597}
3598
3599/*
3600 * Find the next thread in the thread list.
3601 * Return NULL if there is an error or no next thread.
3602 *
3603 * The reference to the input task_struct is released.
3604 */
3605static struct task_struct *next_tid(struct task_struct *start)
3606{
3607 struct task_struct *pos = NULL;
3608 rcu_read_lock();
3609 if (pid_alive(start)) {
3610 pos = next_thread(start);
3611 if (thread_group_leader(pos))
3612 pos = NULL;
3613 else
3614 get_task_struct(pos);
3615 }
3616 rcu_read_unlock();
3617 put_task_struct(start);
3618 return pos;
3619}
3620
3621static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3622 struct task_struct *task, int tid)
3623{
3624 char name[PROC_NUMBUF];
3625 int len = snprintf(name, sizeof(name), "%d", tid);
3626 return proc_fill_cache(filp, dirent, filldir, name, len,
3627 proc_task_instantiate, task, NULL);
3628}
3629
3630/* for the /proc/TGID/task/ directories */
3631static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3632{
3633 struct dentry *dentry = filp->f_path.dentry;
3634 struct inode *inode = dentry->d_inode;
3635 struct task_struct *leader = NULL;
3636 struct task_struct *task;
3637 int retval = -ENOENT;
3638 ino_t ino;
3639 int tid;
3640 struct pid_namespace *ns;
3641
3642 task = get_proc_task(inode);
3643 if (!task)
3644 goto out_no_task;
3645 rcu_read_lock();
3646 if (pid_alive(task)) {
3647 leader = task->group_leader;
3648 get_task_struct(leader);
3649 }
3650 rcu_read_unlock();
3651 put_task_struct(task);
3652 if (!leader)
3653 goto out_no_task;
3654 retval = 0;
3655
3656 switch ((unsigned long)filp->f_pos) {
3657 case 0:
3658 ino = inode->i_ino;
3659 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3660 goto out;
3661 filp->f_pos++;
3662 /* fall through */
3663 case 1:
3664 ino = parent_ino(dentry);
3665 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3666 goto out;
3667 filp->f_pos++;
3668 /* fall through */
3669 }
3670
3671 /* f_version caches the tgid value that the last readdir call couldn't
3672 * return. lseek aka telldir automagically resets f_version to 0.
3673 */
3674 ns = filp->f_dentry->d_sb->s_fs_info;
3675 tid = (int)filp->f_version;
3676 filp->f_version = 0;
3677 for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3678 task;
3679 task = next_tid(task), filp->f_pos++) {
3680 tid = task_pid_nr_ns(task, ns);
3681 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3682 /* returning this tgid failed, save it as the first
3683 * pid for the next readir call */
3684 filp->f_version = (u64)tid;
3685 put_task_struct(task);
3686 break;
3687 }
3688 }
3689out:
3690 put_task_struct(leader);
3691out_no_task:
3692 return retval;
3693}
3694
3695static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3696{
3697 struct inode *inode = dentry->d_inode;
3698 struct task_struct *p = get_proc_task(inode);
3699 generic_fillattr(inode, stat);
3700
3701 if (p) {
3702 stat->nlink += get_nr_threads(p);
3703 put_task_struct(p);
3704 }
3705
3706 return 0;
3707}
3708
3709static const struct inode_operations proc_task_inode_operations = {
3710 .lookup = proc_task_lookup,
3711 .getattr = proc_task_getattr,
3712 .setattr = proc_setattr,
3713 .permission = proc_pid_permission,
3714};
3715
3716static const struct file_operations proc_task_operations = {
3717 .read = generic_read_dir,
3718 .readdir = proc_task_readdir,
3719 .llseek = default_llseek,
3720};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/proc/base.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * proc base directory handling functions
8 *
9 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10 * Instead of using magical inumbers to determine the kind of object
11 * we allocate and fill in-core inodes upon lookup. They don't even
12 * go into icache. We cache the reference to task_struct upon lookup too.
13 * Eventually it should become a filesystem in its own. We don't use the
14 * rest of procfs anymore.
15 *
16 *
17 * Changelog:
18 * 17-Jan-2005
19 * Allan Bezerra
20 * Bruna Moreira <bruna.moreira@indt.org.br>
21 * Edjard Mota <edjard.mota@indt.org.br>
22 * Ilias Biris <ilias.biris@indt.org.br>
23 * Mauricio Lin <mauricio.lin@indt.org.br>
24 *
25 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 *
27 * A new process specific entry (smaps) included in /proc. It shows the
28 * size of rss for each memory area. The maps entry lacks information
29 * about physical memory size (rss) for each mapped file, i.e.,
30 * rss information for executables and library files.
31 * This additional information is useful for any tools that need to know
32 * about physical memory consumption for a process specific library.
33 *
34 * Changelog:
35 * 21-Feb-2005
36 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37 * Pud inclusion in the page table walking.
38 *
39 * ChangeLog:
40 * 10-Mar-2005
41 * 10LE Instituto Nokia de Tecnologia - INdT:
42 * A better way to walks through the page table as suggested by Hugh Dickins.
43 *
44 * Simo Piiroinen <simo.piiroinen@nokia.com>:
45 * Smaps information related to shared, private, clean and dirty pages.
46 *
47 * Paul Mundt <paul.mundt@nokia.com>:
48 * Overall revision about smaps.
49 */
50
51#include <linux/uaccess.h>
52
53#include <linux/errno.h>
54#include <linux/time.h>
55#include <linux/proc_fs.h>
56#include <linux/stat.h>
57#include <linux/task_io_accounting_ops.h>
58#include <linux/init.h>
59#include <linux/capability.h>
60#include <linux/file.h>
61#include <linux/generic-radix-tree.h>
62#include <linux/string.h>
63#include <linux/seq_file.h>
64#include <linux/namei.h>
65#include <linux/mnt_namespace.h>
66#include <linux/mm.h>
67#include <linux/swap.h>
68#include <linux/rcupdate.h>
69#include <linux/kallsyms.h>
70#include <linux/stacktrace.h>
71#include <linux/resource.h>
72#include <linux/module.h>
73#include <linux/mount.h>
74#include <linux/security.h>
75#include <linux/ptrace.h>
76#include <linux/printk.h>
77#include <linux/cache.h>
78#include <linux/cgroup.h>
79#include <linux/cpuset.h>
80#include <linux/audit.h>
81#include <linux/poll.h>
82#include <linux/nsproxy.h>
83#include <linux/oom.h>
84#include <linux/elf.h>
85#include <linux/pid_namespace.h>
86#include <linux/user_namespace.h>
87#include <linux/fs_parser.h>
88#include <linux/fs_struct.h>
89#include <linux/slab.h>
90#include <linux/sched/autogroup.h>
91#include <linux/sched/mm.h>
92#include <linux/sched/coredump.h>
93#include <linux/sched/debug.h>
94#include <linux/sched/stat.h>
95#include <linux/posix-timers.h>
96#include <linux/time_namespace.h>
97#include <linux/resctrl.h>
98#include <linux/cn_proc.h>
99#include <linux/ksm.h>
100#include <uapi/linux/lsm.h>
101#include <trace/events/oom.h>
102#include "internal.h"
103#include "fd.h"
104
105#include "../../lib/kstrtox.h"
106
107/* NOTE:
108 * Implementing inode permission operations in /proc is almost
109 * certainly an error. Permission checks need to happen during
110 * each system call not at open time. The reason is that most of
111 * what we wish to check for permissions in /proc varies at runtime.
112 *
113 * The classic example of a problem is opening file descriptors
114 * in /proc for a task before it execs a suid executable.
115 */
116
117static u8 nlink_tid __ro_after_init;
118static u8 nlink_tgid __ro_after_init;
119
120enum proc_mem_force {
121 PROC_MEM_FORCE_ALWAYS,
122 PROC_MEM_FORCE_PTRACE,
123 PROC_MEM_FORCE_NEVER
124};
125
126static enum proc_mem_force proc_mem_force_override __ro_after_init =
127 IS_ENABLED(CONFIG_PROC_MEM_NO_FORCE) ? PROC_MEM_FORCE_NEVER :
128 IS_ENABLED(CONFIG_PROC_MEM_FORCE_PTRACE) ? PROC_MEM_FORCE_PTRACE :
129 PROC_MEM_FORCE_ALWAYS;
130
131static const struct constant_table proc_mem_force_table[] __initconst = {
132 { "always", PROC_MEM_FORCE_ALWAYS },
133 { "ptrace", PROC_MEM_FORCE_PTRACE },
134 { "never", PROC_MEM_FORCE_NEVER },
135 { }
136};
137
138static int __init early_proc_mem_force_override(char *buf)
139{
140 if (!buf)
141 return -EINVAL;
142
143 /*
144 * lookup_constant() defaults to proc_mem_force_override to preseve
145 * the initial Kconfig choice in case an invalid param gets passed.
146 */
147 proc_mem_force_override = lookup_constant(proc_mem_force_table,
148 buf, proc_mem_force_override);
149
150 return 0;
151}
152early_param("proc_mem.force_override", early_proc_mem_force_override);
153
154struct pid_entry {
155 const char *name;
156 unsigned int len;
157 umode_t mode;
158 const struct inode_operations *iop;
159 const struct file_operations *fop;
160 union proc_op op;
161};
162
163#define NOD(NAME, MODE, IOP, FOP, OP) { \
164 .name = (NAME), \
165 .len = sizeof(NAME) - 1, \
166 .mode = MODE, \
167 .iop = IOP, \
168 .fop = FOP, \
169 .op = OP, \
170}
171
172#define DIR(NAME, MODE, iops, fops) \
173 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
174#define LNK(NAME, get_link) \
175 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
176 &proc_pid_link_inode_operations, NULL, \
177 { .proc_get_link = get_link } )
178#define REG(NAME, MODE, fops) \
179 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
180#define ONE(NAME, MODE, show) \
181 NOD(NAME, (S_IFREG|(MODE)), \
182 NULL, &proc_single_file_operations, \
183 { .proc_show = show } )
184#define ATTR(LSMID, NAME, MODE) \
185 NOD(NAME, (S_IFREG|(MODE)), \
186 NULL, &proc_pid_attr_operations, \
187 { .lsmid = LSMID })
188
189/*
190 * Count the number of hardlinks for the pid_entry table, excluding the .
191 * and .. links.
192 */
193static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
194 unsigned int n)
195{
196 unsigned int i;
197 unsigned int count;
198
199 count = 2;
200 for (i = 0; i < n; ++i) {
201 if (S_ISDIR(entries[i].mode))
202 ++count;
203 }
204
205 return count;
206}
207
208static int get_task_root(struct task_struct *task, struct path *root)
209{
210 int result = -ENOENT;
211
212 task_lock(task);
213 if (task->fs) {
214 get_fs_root(task->fs, root);
215 result = 0;
216 }
217 task_unlock(task);
218 return result;
219}
220
221static int proc_cwd_link(struct dentry *dentry, struct path *path)
222{
223 struct task_struct *task = get_proc_task(d_inode(dentry));
224 int result = -ENOENT;
225
226 if (task) {
227 task_lock(task);
228 if (task->fs) {
229 get_fs_pwd(task->fs, path);
230 result = 0;
231 }
232 task_unlock(task);
233 put_task_struct(task);
234 }
235 return result;
236}
237
238static int proc_root_link(struct dentry *dentry, struct path *path)
239{
240 struct task_struct *task = get_proc_task(d_inode(dentry));
241 int result = -ENOENT;
242
243 if (task) {
244 result = get_task_root(task, path);
245 put_task_struct(task);
246 }
247 return result;
248}
249
250/*
251 * If the user used setproctitle(), we just get the string from
252 * user space at arg_start, and limit it to a maximum of one page.
253 */
254static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
255 size_t count, unsigned long pos,
256 unsigned long arg_start)
257{
258 char *page;
259 int ret, got;
260
261 if (pos >= PAGE_SIZE)
262 return 0;
263
264 page = (char *)__get_free_page(GFP_KERNEL);
265 if (!page)
266 return -ENOMEM;
267
268 ret = 0;
269 got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
270 if (got > 0) {
271 int len = strnlen(page, got);
272
273 /* Include the NUL character if it was found */
274 if (len < got)
275 len++;
276
277 if (len > pos) {
278 len -= pos;
279 if (len > count)
280 len = count;
281 len -= copy_to_user(buf, page+pos, len);
282 if (!len)
283 len = -EFAULT;
284 ret = len;
285 }
286 }
287 free_page((unsigned long)page);
288 return ret;
289}
290
291static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
292 size_t count, loff_t *ppos)
293{
294 unsigned long arg_start, arg_end, env_start, env_end;
295 unsigned long pos, len;
296 char *page, c;
297
298 /* Check if process spawned far enough to have cmdline. */
299 if (!mm->env_end)
300 return 0;
301
302 spin_lock(&mm->arg_lock);
303 arg_start = mm->arg_start;
304 arg_end = mm->arg_end;
305 env_start = mm->env_start;
306 env_end = mm->env_end;
307 spin_unlock(&mm->arg_lock);
308
309 if (arg_start >= arg_end)
310 return 0;
311
312 /*
313 * We allow setproctitle() to overwrite the argument
314 * strings, and overflow past the original end. But
315 * only when it overflows into the environment area.
316 */
317 if (env_start != arg_end || env_end < env_start)
318 env_start = env_end = arg_end;
319 len = env_end - arg_start;
320
321 /* We're not going to care if "*ppos" has high bits set */
322 pos = *ppos;
323 if (pos >= len)
324 return 0;
325 if (count > len - pos)
326 count = len - pos;
327 if (!count)
328 return 0;
329
330 /*
331 * Magical special case: if the argv[] end byte is not
332 * zero, the user has overwritten it with setproctitle(3).
333 *
334 * Possible future enhancement: do this only once when
335 * pos is 0, and set a flag in the 'struct file'.
336 */
337 if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
338 return get_mm_proctitle(mm, buf, count, pos, arg_start);
339
340 /*
341 * For the non-setproctitle() case we limit things strictly
342 * to the [arg_start, arg_end[ range.
343 */
344 pos += arg_start;
345 if (pos < arg_start || pos >= arg_end)
346 return 0;
347 if (count > arg_end - pos)
348 count = arg_end - pos;
349
350 page = (char *)__get_free_page(GFP_KERNEL);
351 if (!page)
352 return -ENOMEM;
353
354 len = 0;
355 while (count) {
356 int got;
357 size_t size = min_t(size_t, PAGE_SIZE, count);
358
359 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
360 if (got <= 0)
361 break;
362 got -= copy_to_user(buf, page, got);
363 if (unlikely(!got)) {
364 if (!len)
365 len = -EFAULT;
366 break;
367 }
368 pos += got;
369 buf += got;
370 len += got;
371 count -= got;
372 }
373
374 free_page((unsigned long)page);
375 return len;
376}
377
378static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
379 size_t count, loff_t *pos)
380{
381 struct mm_struct *mm;
382 ssize_t ret;
383
384 mm = get_task_mm(tsk);
385 if (!mm)
386 return 0;
387
388 ret = get_mm_cmdline(mm, buf, count, pos);
389 mmput(mm);
390 return ret;
391}
392
393static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
394 size_t count, loff_t *pos)
395{
396 struct task_struct *tsk;
397 ssize_t ret;
398
399 BUG_ON(*pos < 0);
400
401 tsk = get_proc_task(file_inode(file));
402 if (!tsk)
403 return -ESRCH;
404 ret = get_task_cmdline(tsk, buf, count, pos);
405 put_task_struct(tsk);
406 if (ret > 0)
407 *pos += ret;
408 return ret;
409}
410
411static const struct file_operations proc_pid_cmdline_ops = {
412 .read = proc_pid_cmdline_read,
413 .llseek = generic_file_llseek,
414};
415
416#ifdef CONFIG_KALLSYMS
417/*
418 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
419 * Returns the resolved symbol. If that fails, simply return the address.
420 */
421static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
422 struct pid *pid, struct task_struct *task)
423{
424 unsigned long wchan;
425 char symname[KSYM_NAME_LEN];
426
427 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
428 goto print0;
429
430 wchan = get_wchan(task);
431 if (wchan && !lookup_symbol_name(wchan, symname)) {
432 seq_puts(m, symname);
433 return 0;
434 }
435
436print0:
437 seq_putc(m, '0');
438 return 0;
439}
440#endif /* CONFIG_KALLSYMS */
441
442static int lock_trace(struct task_struct *task)
443{
444 int err = down_read_killable(&task->signal->exec_update_lock);
445 if (err)
446 return err;
447 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
448 up_read(&task->signal->exec_update_lock);
449 return -EPERM;
450 }
451 return 0;
452}
453
454static void unlock_trace(struct task_struct *task)
455{
456 up_read(&task->signal->exec_update_lock);
457}
458
459#ifdef CONFIG_STACKTRACE
460
461#define MAX_STACK_TRACE_DEPTH 64
462
463static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
464 struct pid *pid, struct task_struct *task)
465{
466 unsigned long *entries;
467 int err;
468
469 /*
470 * The ability to racily run the kernel stack unwinder on a running task
471 * and then observe the unwinder output is scary; while it is useful for
472 * debugging kernel issues, it can also allow an attacker to leak kernel
473 * stack contents.
474 * Doing this in a manner that is at least safe from races would require
475 * some work to ensure that the remote task can not be scheduled; and
476 * even then, this would still expose the unwinder as local attack
477 * surface.
478 * Therefore, this interface is restricted to root.
479 */
480 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
481 return -EACCES;
482
483 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
484 GFP_KERNEL);
485 if (!entries)
486 return -ENOMEM;
487
488 err = lock_trace(task);
489 if (!err) {
490 unsigned int i, nr_entries;
491
492 nr_entries = stack_trace_save_tsk(task, entries,
493 MAX_STACK_TRACE_DEPTH, 0);
494
495 for (i = 0; i < nr_entries; i++) {
496 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
497 }
498
499 unlock_trace(task);
500 }
501 kfree(entries);
502
503 return err;
504}
505#endif
506
507#ifdef CONFIG_SCHED_INFO
508/*
509 * Provides /proc/PID/schedstat
510 */
511static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
512 struct pid *pid, struct task_struct *task)
513{
514 if (unlikely(!sched_info_on()))
515 seq_puts(m, "0 0 0\n");
516 else
517 seq_printf(m, "%llu %llu %lu\n",
518 (unsigned long long)task->se.sum_exec_runtime,
519 (unsigned long long)task->sched_info.run_delay,
520 task->sched_info.pcount);
521
522 return 0;
523}
524#endif
525
526#ifdef CONFIG_LATENCYTOP
527static int lstats_show_proc(struct seq_file *m, void *v)
528{
529 int i;
530 struct inode *inode = m->private;
531 struct task_struct *task = get_proc_task(inode);
532
533 if (!task)
534 return -ESRCH;
535 seq_puts(m, "Latency Top version : v0.1\n");
536 for (i = 0; i < LT_SAVECOUNT; i++) {
537 struct latency_record *lr = &task->latency_record[i];
538 if (lr->backtrace[0]) {
539 int q;
540 seq_printf(m, "%i %li %li",
541 lr->count, lr->time, lr->max);
542 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
543 unsigned long bt = lr->backtrace[q];
544
545 if (!bt)
546 break;
547 seq_printf(m, " %ps", (void *)bt);
548 }
549 seq_putc(m, '\n');
550 }
551
552 }
553 put_task_struct(task);
554 return 0;
555}
556
557static int lstats_open(struct inode *inode, struct file *file)
558{
559 return single_open(file, lstats_show_proc, inode);
560}
561
562static ssize_t lstats_write(struct file *file, const char __user *buf,
563 size_t count, loff_t *offs)
564{
565 struct task_struct *task = get_proc_task(file_inode(file));
566
567 if (!task)
568 return -ESRCH;
569 clear_tsk_latency_tracing(task);
570 put_task_struct(task);
571
572 return count;
573}
574
575static const struct file_operations proc_lstats_operations = {
576 .open = lstats_open,
577 .read = seq_read,
578 .write = lstats_write,
579 .llseek = seq_lseek,
580 .release = single_release,
581};
582
583#endif
584
585static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
586 struct pid *pid, struct task_struct *task)
587{
588 unsigned long totalpages = totalram_pages() + total_swap_pages;
589 unsigned long points = 0;
590 long badness;
591
592 badness = oom_badness(task, totalpages);
593 /*
594 * Special case OOM_SCORE_ADJ_MIN for all others scale the
595 * badness value into [0, 2000] range which we have been
596 * exporting for a long time so userspace might depend on it.
597 */
598 if (badness != LONG_MIN)
599 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
600
601 seq_printf(m, "%lu\n", points);
602
603 return 0;
604}
605
606struct limit_names {
607 const char *name;
608 const char *unit;
609};
610
611static const struct limit_names lnames[RLIM_NLIMITS] = {
612 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
613 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
614 [RLIMIT_DATA] = {"Max data size", "bytes"},
615 [RLIMIT_STACK] = {"Max stack size", "bytes"},
616 [RLIMIT_CORE] = {"Max core file size", "bytes"},
617 [RLIMIT_RSS] = {"Max resident set", "bytes"},
618 [RLIMIT_NPROC] = {"Max processes", "processes"},
619 [RLIMIT_NOFILE] = {"Max open files", "files"},
620 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
621 [RLIMIT_AS] = {"Max address space", "bytes"},
622 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
623 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
624 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
625 [RLIMIT_NICE] = {"Max nice priority", NULL},
626 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
627 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
628};
629
630/* Display limits for a process */
631static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
632 struct pid *pid, struct task_struct *task)
633{
634 unsigned int i;
635 unsigned long flags;
636
637 struct rlimit rlim[RLIM_NLIMITS];
638
639 if (!lock_task_sighand(task, &flags))
640 return 0;
641 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
642 unlock_task_sighand(task, &flags);
643
644 /*
645 * print the file header
646 */
647 seq_puts(m, "Limit "
648 "Soft Limit "
649 "Hard Limit "
650 "Units \n");
651
652 for (i = 0; i < RLIM_NLIMITS; i++) {
653 if (rlim[i].rlim_cur == RLIM_INFINITY)
654 seq_printf(m, "%-25s %-20s ",
655 lnames[i].name, "unlimited");
656 else
657 seq_printf(m, "%-25s %-20lu ",
658 lnames[i].name, rlim[i].rlim_cur);
659
660 if (rlim[i].rlim_max == RLIM_INFINITY)
661 seq_printf(m, "%-20s ", "unlimited");
662 else
663 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
664
665 if (lnames[i].unit)
666 seq_printf(m, "%-10s\n", lnames[i].unit);
667 else
668 seq_putc(m, '\n');
669 }
670
671 return 0;
672}
673
674#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
675static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
676 struct pid *pid, struct task_struct *task)
677{
678 struct syscall_info info;
679 u64 *args = &info.data.args[0];
680 int res;
681
682 res = lock_trace(task);
683 if (res)
684 return res;
685
686 if (task_current_syscall(task, &info))
687 seq_puts(m, "running\n");
688 else if (info.data.nr < 0)
689 seq_printf(m, "%d 0x%llx 0x%llx\n",
690 info.data.nr, info.sp, info.data.instruction_pointer);
691 else
692 seq_printf(m,
693 "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
694 info.data.nr,
695 args[0], args[1], args[2], args[3], args[4], args[5],
696 info.sp, info.data.instruction_pointer);
697 unlock_trace(task);
698
699 return 0;
700}
701#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
702
703/************************************************************************/
704/* Here the fs part begins */
705/************************************************************************/
706
707/* permission checks */
708static bool proc_fd_access_allowed(struct inode *inode)
709{
710 struct task_struct *task;
711 bool allowed = false;
712 /* Allow access to a task's file descriptors if it is us or we
713 * may use ptrace attach to the process and find out that
714 * information.
715 */
716 task = get_proc_task(inode);
717 if (task) {
718 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
719 put_task_struct(task);
720 }
721 return allowed;
722}
723
724int proc_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
725 struct iattr *attr)
726{
727 int error;
728 struct inode *inode = d_inode(dentry);
729
730 if (attr->ia_valid & ATTR_MODE)
731 return -EPERM;
732
733 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
734 if (error)
735 return error;
736
737 setattr_copy(&nop_mnt_idmap, inode, attr);
738 return 0;
739}
740
741/*
742 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
743 * or euid/egid (for hide_pid_min=2)?
744 */
745static bool has_pid_permissions(struct proc_fs_info *fs_info,
746 struct task_struct *task,
747 enum proc_hidepid hide_pid_min)
748{
749 /*
750 * If 'hidpid' mount option is set force a ptrace check,
751 * we indicate that we are using a filesystem syscall
752 * by passing PTRACE_MODE_READ_FSCREDS
753 */
754 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
755 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
756
757 if (fs_info->hide_pid < hide_pid_min)
758 return true;
759 if (in_group_p(fs_info->pid_gid))
760 return true;
761 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
762}
763
764
765static int proc_pid_permission(struct mnt_idmap *idmap,
766 struct inode *inode, int mask)
767{
768 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
769 struct task_struct *task;
770 bool has_perms;
771
772 task = get_proc_task(inode);
773 if (!task)
774 return -ESRCH;
775 has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
776 put_task_struct(task);
777
778 if (!has_perms) {
779 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
780 /*
781 * Let's make getdents(), stat(), and open()
782 * consistent with each other. If a process
783 * may not stat() a file, it shouldn't be seen
784 * in procfs at all.
785 */
786 return -ENOENT;
787 }
788
789 return -EPERM;
790 }
791 return generic_permission(&nop_mnt_idmap, inode, mask);
792}
793
794
795
796static const struct inode_operations proc_def_inode_operations = {
797 .setattr = proc_setattr,
798};
799
800static int proc_single_show(struct seq_file *m, void *v)
801{
802 struct inode *inode = m->private;
803 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
804 struct pid *pid = proc_pid(inode);
805 struct task_struct *task;
806 int ret;
807
808 task = get_pid_task(pid, PIDTYPE_PID);
809 if (!task)
810 return -ESRCH;
811
812 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
813
814 put_task_struct(task);
815 return ret;
816}
817
818static int proc_single_open(struct inode *inode, struct file *filp)
819{
820 return single_open(filp, proc_single_show, inode);
821}
822
823static const struct file_operations proc_single_file_operations = {
824 .open = proc_single_open,
825 .read = seq_read,
826 .llseek = seq_lseek,
827 .release = single_release,
828};
829
830
831struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
832{
833 struct task_struct *task = get_proc_task(inode);
834 struct mm_struct *mm;
835
836 if (!task)
837 return ERR_PTR(-ESRCH);
838
839 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
840 put_task_struct(task);
841
842 if (IS_ERR(mm))
843 return mm == ERR_PTR(-ESRCH) ? NULL : mm;
844
845 /* ensure this mm_struct can't be freed */
846 mmgrab(mm);
847 /* but do not pin its memory */
848 mmput(mm);
849
850 return mm;
851}
852
853static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
854{
855 struct mm_struct *mm = proc_mem_open(inode, mode);
856
857 if (IS_ERR(mm))
858 return PTR_ERR(mm);
859
860 file->private_data = mm;
861 return 0;
862}
863
864static int mem_open(struct inode *inode, struct file *file)
865{
866 if (WARN_ON_ONCE(!(file->f_op->fop_flags & FOP_UNSIGNED_OFFSET)))
867 return -EINVAL;
868 return __mem_open(inode, file, PTRACE_MODE_ATTACH);
869}
870
871static bool proc_mem_foll_force(struct file *file, struct mm_struct *mm)
872{
873 struct task_struct *task;
874 bool ptrace_active = false;
875
876 switch (proc_mem_force_override) {
877 case PROC_MEM_FORCE_NEVER:
878 return false;
879 case PROC_MEM_FORCE_PTRACE:
880 task = get_proc_task(file_inode(file));
881 if (task) {
882 ptrace_active = READ_ONCE(task->ptrace) &&
883 READ_ONCE(task->mm) == mm &&
884 READ_ONCE(task->parent) == current;
885 put_task_struct(task);
886 }
887 return ptrace_active;
888 default:
889 return true;
890 }
891}
892
893static ssize_t mem_rw(struct file *file, char __user *buf,
894 size_t count, loff_t *ppos, int write)
895{
896 struct mm_struct *mm = file->private_data;
897 unsigned long addr = *ppos;
898 ssize_t copied;
899 char *page;
900 unsigned int flags;
901
902 if (!mm)
903 return 0;
904
905 page = (char *)__get_free_page(GFP_KERNEL);
906 if (!page)
907 return -ENOMEM;
908
909 copied = 0;
910 if (!mmget_not_zero(mm))
911 goto free;
912
913 flags = write ? FOLL_WRITE : 0;
914 if (proc_mem_foll_force(file, mm))
915 flags |= FOLL_FORCE;
916
917 while (count > 0) {
918 size_t this_len = min_t(size_t, count, PAGE_SIZE);
919
920 if (write && copy_from_user(page, buf, this_len)) {
921 copied = -EFAULT;
922 break;
923 }
924
925 this_len = access_remote_vm(mm, addr, page, this_len, flags);
926 if (!this_len) {
927 if (!copied)
928 copied = -EIO;
929 break;
930 }
931
932 if (!write && copy_to_user(buf, page, this_len)) {
933 copied = -EFAULT;
934 break;
935 }
936
937 buf += this_len;
938 addr += this_len;
939 copied += this_len;
940 count -= this_len;
941 }
942 *ppos = addr;
943
944 mmput(mm);
945free:
946 free_page((unsigned long) page);
947 return copied;
948}
949
950static ssize_t mem_read(struct file *file, char __user *buf,
951 size_t count, loff_t *ppos)
952{
953 return mem_rw(file, buf, count, ppos, 0);
954}
955
956static ssize_t mem_write(struct file *file, const char __user *buf,
957 size_t count, loff_t *ppos)
958{
959 return mem_rw(file, (char __user*)buf, count, ppos, 1);
960}
961
962loff_t mem_lseek(struct file *file, loff_t offset, int orig)
963{
964 switch (orig) {
965 case 0:
966 file->f_pos = offset;
967 break;
968 case 1:
969 file->f_pos += offset;
970 break;
971 default:
972 return -EINVAL;
973 }
974 force_successful_syscall_return();
975 return file->f_pos;
976}
977
978static int mem_release(struct inode *inode, struct file *file)
979{
980 struct mm_struct *mm = file->private_data;
981 if (mm)
982 mmdrop(mm);
983 return 0;
984}
985
986static const struct file_operations proc_mem_operations = {
987 .llseek = mem_lseek,
988 .read = mem_read,
989 .write = mem_write,
990 .open = mem_open,
991 .release = mem_release,
992 .fop_flags = FOP_UNSIGNED_OFFSET,
993};
994
995static int environ_open(struct inode *inode, struct file *file)
996{
997 return __mem_open(inode, file, PTRACE_MODE_READ);
998}
999
1000static ssize_t environ_read(struct file *file, char __user *buf,
1001 size_t count, loff_t *ppos)
1002{
1003 char *page;
1004 unsigned long src = *ppos;
1005 int ret = 0;
1006 struct mm_struct *mm = file->private_data;
1007 unsigned long env_start, env_end;
1008
1009 /* Ensure the process spawned far enough to have an environment. */
1010 if (!mm || !mm->env_end)
1011 return 0;
1012
1013 page = (char *)__get_free_page(GFP_KERNEL);
1014 if (!page)
1015 return -ENOMEM;
1016
1017 ret = 0;
1018 if (!mmget_not_zero(mm))
1019 goto free;
1020
1021 spin_lock(&mm->arg_lock);
1022 env_start = mm->env_start;
1023 env_end = mm->env_end;
1024 spin_unlock(&mm->arg_lock);
1025
1026 while (count > 0) {
1027 size_t this_len, max_len;
1028 int retval;
1029
1030 if (src >= (env_end - env_start))
1031 break;
1032
1033 this_len = env_end - (env_start + src);
1034
1035 max_len = min_t(size_t, PAGE_SIZE, count);
1036 this_len = min(max_len, this_len);
1037
1038 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
1039
1040 if (retval <= 0) {
1041 ret = retval;
1042 break;
1043 }
1044
1045 if (copy_to_user(buf, page, retval)) {
1046 ret = -EFAULT;
1047 break;
1048 }
1049
1050 ret += retval;
1051 src += retval;
1052 buf += retval;
1053 count -= retval;
1054 }
1055 *ppos = src;
1056 mmput(mm);
1057
1058free:
1059 free_page((unsigned long) page);
1060 return ret;
1061}
1062
1063static const struct file_operations proc_environ_operations = {
1064 .open = environ_open,
1065 .read = environ_read,
1066 .llseek = generic_file_llseek,
1067 .release = mem_release,
1068};
1069
1070static int auxv_open(struct inode *inode, struct file *file)
1071{
1072 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1073}
1074
1075static ssize_t auxv_read(struct file *file, char __user *buf,
1076 size_t count, loff_t *ppos)
1077{
1078 struct mm_struct *mm = file->private_data;
1079 unsigned int nwords = 0;
1080
1081 if (!mm)
1082 return 0;
1083 do {
1084 nwords += 2;
1085 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1086 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1087 nwords * sizeof(mm->saved_auxv[0]));
1088}
1089
1090static const struct file_operations proc_auxv_operations = {
1091 .open = auxv_open,
1092 .read = auxv_read,
1093 .llseek = generic_file_llseek,
1094 .release = mem_release,
1095};
1096
1097static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1098 loff_t *ppos)
1099{
1100 struct task_struct *task = get_proc_task(file_inode(file));
1101 char buffer[PROC_NUMBUF];
1102 int oom_adj = OOM_ADJUST_MIN;
1103 size_t len;
1104
1105 if (!task)
1106 return -ESRCH;
1107 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1108 oom_adj = OOM_ADJUST_MAX;
1109 else
1110 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1111 OOM_SCORE_ADJ_MAX;
1112 put_task_struct(task);
1113 if (oom_adj > OOM_ADJUST_MAX)
1114 oom_adj = OOM_ADJUST_MAX;
1115 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1116 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1117}
1118
1119static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1120{
1121 struct mm_struct *mm = NULL;
1122 struct task_struct *task;
1123 int err = 0;
1124
1125 task = get_proc_task(file_inode(file));
1126 if (!task)
1127 return -ESRCH;
1128
1129 mutex_lock(&oom_adj_mutex);
1130 if (legacy) {
1131 if (oom_adj < task->signal->oom_score_adj &&
1132 !capable(CAP_SYS_RESOURCE)) {
1133 err = -EACCES;
1134 goto err_unlock;
1135 }
1136 /*
1137 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1138 * /proc/pid/oom_score_adj instead.
1139 */
1140 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1141 current->comm, task_pid_nr(current), task_pid_nr(task),
1142 task_pid_nr(task));
1143 } else {
1144 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1145 !capable(CAP_SYS_RESOURCE)) {
1146 err = -EACCES;
1147 goto err_unlock;
1148 }
1149 }
1150
1151 /*
1152 * Make sure we will check other processes sharing the mm if this is
1153 * not vfrok which wants its own oom_score_adj.
1154 * pin the mm so it doesn't go away and get reused after task_unlock
1155 */
1156 if (!task->vfork_done) {
1157 struct task_struct *p = find_lock_task_mm(task);
1158
1159 if (p) {
1160 if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1161 mm = p->mm;
1162 mmgrab(mm);
1163 }
1164 task_unlock(p);
1165 }
1166 }
1167
1168 task->signal->oom_score_adj = oom_adj;
1169 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1170 task->signal->oom_score_adj_min = (short)oom_adj;
1171 trace_oom_score_adj_update(task);
1172
1173 if (mm) {
1174 struct task_struct *p;
1175
1176 rcu_read_lock();
1177 for_each_process(p) {
1178 if (same_thread_group(task, p))
1179 continue;
1180
1181 /* do not touch kernel threads or the global init */
1182 if (p->flags & PF_KTHREAD || is_global_init(p))
1183 continue;
1184
1185 task_lock(p);
1186 if (!p->vfork_done && process_shares_mm(p, mm)) {
1187 p->signal->oom_score_adj = oom_adj;
1188 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1189 p->signal->oom_score_adj_min = (short)oom_adj;
1190 }
1191 task_unlock(p);
1192 }
1193 rcu_read_unlock();
1194 mmdrop(mm);
1195 }
1196err_unlock:
1197 mutex_unlock(&oom_adj_mutex);
1198 put_task_struct(task);
1199 return err;
1200}
1201
1202/*
1203 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1204 * kernels. The effective policy is defined by oom_score_adj, which has a
1205 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1206 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1207 * Processes that become oom disabled via oom_adj will still be oom disabled
1208 * with this implementation.
1209 *
1210 * oom_adj cannot be removed since existing userspace binaries use it.
1211 */
1212static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1213 size_t count, loff_t *ppos)
1214{
1215 char buffer[PROC_NUMBUF] = {};
1216 int oom_adj;
1217 int err;
1218
1219 if (count > sizeof(buffer) - 1)
1220 count = sizeof(buffer) - 1;
1221 if (copy_from_user(buffer, buf, count)) {
1222 err = -EFAULT;
1223 goto out;
1224 }
1225
1226 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1227 if (err)
1228 goto out;
1229 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1230 oom_adj != OOM_DISABLE) {
1231 err = -EINVAL;
1232 goto out;
1233 }
1234
1235 /*
1236 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1237 * value is always attainable.
1238 */
1239 if (oom_adj == OOM_ADJUST_MAX)
1240 oom_adj = OOM_SCORE_ADJ_MAX;
1241 else
1242 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1243
1244 err = __set_oom_adj(file, oom_adj, true);
1245out:
1246 return err < 0 ? err : count;
1247}
1248
1249static const struct file_operations proc_oom_adj_operations = {
1250 .read = oom_adj_read,
1251 .write = oom_adj_write,
1252 .llseek = generic_file_llseek,
1253};
1254
1255static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1256 size_t count, loff_t *ppos)
1257{
1258 struct task_struct *task = get_proc_task(file_inode(file));
1259 char buffer[PROC_NUMBUF];
1260 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1261 size_t len;
1262
1263 if (!task)
1264 return -ESRCH;
1265 oom_score_adj = task->signal->oom_score_adj;
1266 put_task_struct(task);
1267 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1268 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1269}
1270
1271static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1272 size_t count, loff_t *ppos)
1273{
1274 char buffer[PROC_NUMBUF] = {};
1275 int oom_score_adj;
1276 int err;
1277
1278 if (count > sizeof(buffer) - 1)
1279 count = sizeof(buffer) - 1;
1280 if (copy_from_user(buffer, buf, count)) {
1281 err = -EFAULT;
1282 goto out;
1283 }
1284
1285 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1286 if (err)
1287 goto out;
1288 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1289 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1290 err = -EINVAL;
1291 goto out;
1292 }
1293
1294 err = __set_oom_adj(file, oom_score_adj, false);
1295out:
1296 return err < 0 ? err : count;
1297}
1298
1299static const struct file_operations proc_oom_score_adj_operations = {
1300 .read = oom_score_adj_read,
1301 .write = oom_score_adj_write,
1302 .llseek = default_llseek,
1303};
1304
1305#ifdef CONFIG_AUDIT
1306#define TMPBUFLEN 11
1307static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1308 size_t count, loff_t *ppos)
1309{
1310 struct inode * inode = file_inode(file);
1311 struct task_struct *task = get_proc_task(inode);
1312 ssize_t length;
1313 char tmpbuf[TMPBUFLEN];
1314
1315 if (!task)
1316 return -ESRCH;
1317 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1318 from_kuid(file->f_cred->user_ns,
1319 audit_get_loginuid(task)));
1320 put_task_struct(task);
1321 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1322}
1323
1324static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1325 size_t count, loff_t *ppos)
1326{
1327 struct inode * inode = file_inode(file);
1328 uid_t loginuid;
1329 kuid_t kloginuid;
1330 int rv;
1331
1332 /* Don't let kthreads write their own loginuid */
1333 if (current->flags & PF_KTHREAD)
1334 return -EPERM;
1335
1336 rcu_read_lock();
1337 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1338 rcu_read_unlock();
1339 return -EPERM;
1340 }
1341 rcu_read_unlock();
1342
1343 if (*ppos != 0) {
1344 /* No partial writes. */
1345 return -EINVAL;
1346 }
1347
1348 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1349 if (rv < 0)
1350 return rv;
1351
1352 /* is userspace tring to explicitly UNSET the loginuid? */
1353 if (loginuid == AUDIT_UID_UNSET) {
1354 kloginuid = INVALID_UID;
1355 } else {
1356 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1357 if (!uid_valid(kloginuid))
1358 return -EINVAL;
1359 }
1360
1361 rv = audit_set_loginuid(kloginuid);
1362 if (rv < 0)
1363 return rv;
1364 return count;
1365}
1366
1367static const struct file_operations proc_loginuid_operations = {
1368 .read = proc_loginuid_read,
1369 .write = proc_loginuid_write,
1370 .llseek = generic_file_llseek,
1371};
1372
1373static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1374 size_t count, loff_t *ppos)
1375{
1376 struct inode * inode = file_inode(file);
1377 struct task_struct *task = get_proc_task(inode);
1378 ssize_t length;
1379 char tmpbuf[TMPBUFLEN];
1380
1381 if (!task)
1382 return -ESRCH;
1383 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1384 audit_get_sessionid(task));
1385 put_task_struct(task);
1386 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1387}
1388
1389static const struct file_operations proc_sessionid_operations = {
1390 .read = proc_sessionid_read,
1391 .llseek = generic_file_llseek,
1392};
1393#endif
1394
1395#ifdef CONFIG_FAULT_INJECTION
1396static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1397 size_t count, loff_t *ppos)
1398{
1399 struct task_struct *task = get_proc_task(file_inode(file));
1400 char buffer[PROC_NUMBUF];
1401 size_t len;
1402 int make_it_fail;
1403
1404 if (!task)
1405 return -ESRCH;
1406 make_it_fail = task->make_it_fail;
1407 put_task_struct(task);
1408
1409 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1410
1411 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1412}
1413
1414static ssize_t proc_fault_inject_write(struct file * file,
1415 const char __user * buf, size_t count, loff_t *ppos)
1416{
1417 struct task_struct *task;
1418 char buffer[PROC_NUMBUF] = {};
1419 int make_it_fail;
1420 int rv;
1421
1422 if (!capable(CAP_SYS_RESOURCE))
1423 return -EPERM;
1424
1425 if (count > sizeof(buffer) - 1)
1426 count = sizeof(buffer) - 1;
1427 if (copy_from_user(buffer, buf, count))
1428 return -EFAULT;
1429 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1430 if (rv < 0)
1431 return rv;
1432 if (make_it_fail < 0 || make_it_fail > 1)
1433 return -EINVAL;
1434
1435 task = get_proc_task(file_inode(file));
1436 if (!task)
1437 return -ESRCH;
1438 task->make_it_fail = make_it_fail;
1439 put_task_struct(task);
1440
1441 return count;
1442}
1443
1444static const struct file_operations proc_fault_inject_operations = {
1445 .read = proc_fault_inject_read,
1446 .write = proc_fault_inject_write,
1447 .llseek = generic_file_llseek,
1448};
1449
1450static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1451 size_t count, loff_t *ppos)
1452{
1453 struct task_struct *task;
1454 int err;
1455 unsigned int n;
1456
1457 err = kstrtouint_from_user(buf, count, 0, &n);
1458 if (err)
1459 return err;
1460
1461 task = get_proc_task(file_inode(file));
1462 if (!task)
1463 return -ESRCH;
1464 task->fail_nth = n;
1465 put_task_struct(task);
1466
1467 return count;
1468}
1469
1470static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1471 size_t count, loff_t *ppos)
1472{
1473 struct task_struct *task;
1474 char numbuf[PROC_NUMBUF];
1475 ssize_t len;
1476
1477 task = get_proc_task(file_inode(file));
1478 if (!task)
1479 return -ESRCH;
1480 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1481 put_task_struct(task);
1482 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1483}
1484
1485static const struct file_operations proc_fail_nth_operations = {
1486 .read = proc_fail_nth_read,
1487 .write = proc_fail_nth_write,
1488};
1489#endif
1490
1491
1492#ifdef CONFIG_SCHED_DEBUG
1493/*
1494 * Print out various scheduling related per-task fields:
1495 */
1496static int sched_show(struct seq_file *m, void *v)
1497{
1498 struct inode *inode = m->private;
1499 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1500 struct task_struct *p;
1501
1502 p = get_proc_task(inode);
1503 if (!p)
1504 return -ESRCH;
1505 proc_sched_show_task(p, ns, m);
1506
1507 put_task_struct(p);
1508
1509 return 0;
1510}
1511
1512static ssize_t
1513sched_write(struct file *file, const char __user *buf,
1514 size_t count, loff_t *offset)
1515{
1516 struct inode *inode = file_inode(file);
1517 struct task_struct *p;
1518
1519 p = get_proc_task(inode);
1520 if (!p)
1521 return -ESRCH;
1522 proc_sched_set_task(p);
1523
1524 put_task_struct(p);
1525
1526 return count;
1527}
1528
1529static int sched_open(struct inode *inode, struct file *filp)
1530{
1531 return single_open(filp, sched_show, inode);
1532}
1533
1534static const struct file_operations proc_pid_sched_operations = {
1535 .open = sched_open,
1536 .read = seq_read,
1537 .write = sched_write,
1538 .llseek = seq_lseek,
1539 .release = single_release,
1540};
1541
1542#endif
1543
1544#ifdef CONFIG_SCHED_AUTOGROUP
1545/*
1546 * Print out autogroup related information:
1547 */
1548static int sched_autogroup_show(struct seq_file *m, void *v)
1549{
1550 struct inode *inode = m->private;
1551 struct task_struct *p;
1552
1553 p = get_proc_task(inode);
1554 if (!p)
1555 return -ESRCH;
1556 proc_sched_autogroup_show_task(p, m);
1557
1558 put_task_struct(p);
1559
1560 return 0;
1561}
1562
1563static ssize_t
1564sched_autogroup_write(struct file *file, const char __user *buf,
1565 size_t count, loff_t *offset)
1566{
1567 struct inode *inode = file_inode(file);
1568 struct task_struct *p;
1569 char buffer[PROC_NUMBUF] = {};
1570 int nice;
1571 int err;
1572
1573 if (count > sizeof(buffer) - 1)
1574 count = sizeof(buffer) - 1;
1575 if (copy_from_user(buffer, buf, count))
1576 return -EFAULT;
1577
1578 err = kstrtoint(strstrip(buffer), 0, &nice);
1579 if (err < 0)
1580 return err;
1581
1582 p = get_proc_task(inode);
1583 if (!p)
1584 return -ESRCH;
1585
1586 err = proc_sched_autogroup_set_nice(p, nice);
1587 if (err)
1588 count = err;
1589
1590 put_task_struct(p);
1591
1592 return count;
1593}
1594
1595static int sched_autogroup_open(struct inode *inode, struct file *filp)
1596{
1597 int ret;
1598
1599 ret = single_open(filp, sched_autogroup_show, NULL);
1600 if (!ret) {
1601 struct seq_file *m = filp->private_data;
1602
1603 m->private = inode;
1604 }
1605 return ret;
1606}
1607
1608static const struct file_operations proc_pid_sched_autogroup_operations = {
1609 .open = sched_autogroup_open,
1610 .read = seq_read,
1611 .write = sched_autogroup_write,
1612 .llseek = seq_lseek,
1613 .release = single_release,
1614};
1615
1616#endif /* CONFIG_SCHED_AUTOGROUP */
1617
1618#ifdef CONFIG_TIME_NS
1619static int timens_offsets_show(struct seq_file *m, void *v)
1620{
1621 struct task_struct *p;
1622
1623 p = get_proc_task(file_inode(m->file));
1624 if (!p)
1625 return -ESRCH;
1626 proc_timens_show_offsets(p, m);
1627
1628 put_task_struct(p);
1629
1630 return 0;
1631}
1632
1633static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1634 size_t count, loff_t *ppos)
1635{
1636 struct inode *inode = file_inode(file);
1637 struct proc_timens_offset offsets[2];
1638 char *kbuf = NULL, *pos, *next_line;
1639 struct task_struct *p;
1640 int ret, noffsets;
1641
1642 /* Only allow < page size writes at the beginning of the file */
1643 if ((*ppos != 0) || (count >= PAGE_SIZE))
1644 return -EINVAL;
1645
1646 /* Slurp in the user data */
1647 kbuf = memdup_user_nul(buf, count);
1648 if (IS_ERR(kbuf))
1649 return PTR_ERR(kbuf);
1650
1651 /* Parse the user data */
1652 ret = -EINVAL;
1653 noffsets = 0;
1654 for (pos = kbuf; pos; pos = next_line) {
1655 struct proc_timens_offset *off = &offsets[noffsets];
1656 char clock[10];
1657 int err;
1658
1659 /* Find the end of line and ensure we don't look past it */
1660 next_line = strchr(pos, '\n');
1661 if (next_line) {
1662 *next_line = '\0';
1663 next_line++;
1664 if (*next_line == '\0')
1665 next_line = NULL;
1666 }
1667
1668 err = sscanf(pos, "%9s %lld %lu", clock,
1669 &off->val.tv_sec, &off->val.tv_nsec);
1670 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1671 goto out;
1672
1673 clock[sizeof(clock) - 1] = 0;
1674 if (strcmp(clock, "monotonic") == 0 ||
1675 strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1676 off->clockid = CLOCK_MONOTONIC;
1677 else if (strcmp(clock, "boottime") == 0 ||
1678 strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1679 off->clockid = CLOCK_BOOTTIME;
1680 else
1681 goto out;
1682
1683 noffsets++;
1684 if (noffsets == ARRAY_SIZE(offsets)) {
1685 if (next_line)
1686 count = next_line - kbuf;
1687 break;
1688 }
1689 }
1690
1691 ret = -ESRCH;
1692 p = get_proc_task(inode);
1693 if (!p)
1694 goto out;
1695 ret = proc_timens_set_offset(file, p, offsets, noffsets);
1696 put_task_struct(p);
1697 if (ret)
1698 goto out;
1699
1700 ret = count;
1701out:
1702 kfree(kbuf);
1703 return ret;
1704}
1705
1706static int timens_offsets_open(struct inode *inode, struct file *filp)
1707{
1708 return single_open(filp, timens_offsets_show, inode);
1709}
1710
1711static const struct file_operations proc_timens_offsets_operations = {
1712 .open = timens_offsets_open,
1713 .read = seq_read,
1714 .write = timens_offsets_write,
1715 .llseek = seq_lseek,
1716 .release = single_release,
1717};
1718#endif /* CONFIG_TIME_NS */
1719
1720static ssize_t comm_write(struct file *file, const char __user *buf,
1721 size_t count, loff_t *offset)
1722{
1723 struct inode *inode = file_inode(file);
1724 struct task_struct *p;
1725 char buffer[TASK_COMM_LEN] = {};
1726 const size_t maxlen = sizeof(buffer) - 1;
1727
1728 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1729 return -EFAULT;
1730
1731 p = get_proc_task(inode);
1732 if (!p)
1733 return -ESRCH;
1734
1735 if (same_thread_group(current, p)) {
1736 set_task_comm(p, buffer);
1737 proc_comm_connector(p);
1738 }
1739 else
1740 count = -EINVAL;
1741
1742 put_task_struct(p);
1743
1744 return count;
1745}
1746
1747static int comm_show(struct seq_file *m, void *v)
1748{
1749 struct inode *inode = m->private;
1750 struct task_struct *p;
1751
1752 p = get_proc_task(inode);
1753 if (!p)
1754 return -ESRCH;
1755
1756 proc_task_name(m, p, false);
1757 seq_putc(m, '\n');
1758
1759 put_task_struct(p);
1760
1761 return 0;
1762}
1763
1764static int comm_open(struct inode *inode, struct file *filp)
1765{
1766 return single_open(filp, comm_show, inode);
1767}
1768
1769static const struct file_operations proc_pid_set_comm_operations = {
1770 .open = comm_open,
1771 .read = seq_read,
1772 .write = comm_write,
1773 .llseek = seq_lseek,
1774 .release = single_release,
1775};
1776
1777static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1778{
1779 struct task_struct *task;
1780 struct file *exe_file;
1781
1782 task = get_proc_task(d_inode(dentry));
1783 if (!task)
1784 return -ENOENT;
1785 exe_file = get_task_exe_file(task);
1786 put_task_struct(task);
1787 if (exe_file) {
1788 *exe_path = exe_file->f_path;
1789 path_get(&exe_file->f_path);
1790 fput(exe_file);
1791 return 0;
1792 } else
1793 return -ENOENT;
1794}
1795
1796static const char *proc_pid_get_link(struct dentry *dentry,
1797 struct inode *inode,
1798 struct delayed_call *done)
1799{
1800 struct path path;
1801 int error = -EACCES;
1802
1803 if (!dentry)
1804 return ERR_PTR(-ECHILD);
1805
1806 /* Are we allowed to snoop on the tasks file descriptors? */
1807 if (!proc_fd_access_allowed(inode))
1808 goto out;
1809
1810 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1811 if (error)
1812 goto out;
1813
1814 error = nd_jump_link(&path);
1815out:
1816 return ERR_PTR(error);
1817}
1818
1819static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1820{
1821 char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1822 char *pathname;
1823 int len;
1824
1825 if (!tmp)
1826 return -ENOMEM;
1827
1828 pathname = d_path(path, tmp, PATH_MAX);
1829 len = PTR_ERR(pathname);
1830 if (IS_ERR(pathname))
1831 goto out;
1832 len = tmp + PATH_MAX - 1 - pathname;
1833
1834 if (len > buflen)
1835 len = buflen;
1836 if (copy_to_user(buffer, pathname, len))
1837 len = -EFAULT;
1838 out:
1839 kfree(tmp);
1840 return len;
1841}
1842
1843static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1844{
1845 int error = -EACCES;
1846 struct inode *inode = d_inode(dentry);
1847 struct path path;
1848
1849 /* Are we allowed to snoop on the tasks file descriptors? */
1850 if (!proc_fd_access_allowed(inode))
1851 goto out;
1852
1853 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1854 if (error)
1855 goto out;
1856
1857 error = do_proc_readlink(&path, buffer, buflen);
1858 path_put(&path);
1859out:
1860 return error;
1861}
1862
1863const struct inode_operations proc_pid_link_inode_operations = {
1864 .readlink = proc_pid_readlink,
1865 .get_link = proc_pid_get_link,
1866 .setattr = proc_setattr,
1867};
1868
1869
1870/* building an inode */
1871
1872void task_dump_owner(struct task_struct *task, umode_t mode,
1873 kuid_t *ruid, kgid_t *rgid)
1874{
1875 /* Depending on the state of dumpable compute who should own a
1876 * proc file for a task.
1877 */
1878 const struct cred *cred;
1879 kuid_t uid;
1880 kgid_t gid;
1881
1882 if (unlikely(task->flags & PF_KTHREAD)) {
1883 *ruid = GLOBAL_ROOT_UID;
1884 *rgid = GLOBAL_ROOT_GID;
1885 return;
1886 }
1887
1888 /* Default to the tasks effective ownership */
1889 rcu_read_lock();
1890 cred = __task_cred(task);
1891 uid = cred->euid;
1892 gid = cred->egid;
1893 rcu_read_unlock();
1894
1895 /*
1896 * Before the /proc/pid/status file was created the only way to read
1897 * the effective uid of a /process was to stat /proc/pid. Reading
1898 * /proc/pid/status is slow enough that procps and other packages
1899 * kept stating /proc/pid. To keep the rules in /proc simple I have
1900 * made this apply to all per process world readable and executable
1901 * directories.
1902 */
1903 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1904 struct mm_struct *mm;
1905 task_lock(task);
1906 mm = task->mm;
1907 /* Make non-dumpable tasks owned by some root */
1908 if (mm) {
1909 if (get_dumpable(mm) != SUID_DUMP_USER) {
1910 struct user_namespace *user_ns = mm->user_ns;
1911
1912 uid = make_kuid(user_ns, 0);
1913 if (!uid_valid(uid))
1914 uid = GLOBAL_ROOT_UID;
1915
1916 gid = make_kgid(user_ns, 0);
1917 if (!gid_valid(gid))
1918 gid = GLOBAL_ROOT_GID;
1919 }
1920 } else {
1921 uid = GLOBAL_ROOT_UID;
1922 gid = GLOBAL_ROOT_GID;
1923 }
1924 task_unlock(task);
1925 }
1926 *ruid = uid;
1927 *rgid = gid;
1928}
1929
1930void proc_pid_evict_inode(struct proc_inode *ei)
1931{
1932 struct pid *pid = ei->pid;
1933
1934 if (S_ISDIR(ei->vfs_inode.i_mode)) {
1935 spin_lock(&pid->lock);
1936 hlist_del_init_rcu(&ei->sibling_inodes);
1937 spin_unlock(&pid->lock);
1938 }
1939}
1940
1941struct inode *proc_pid_make_inode(struct super_block *sb,
1942 struct task_struct *task, umode_t mode)
1943{
1944 struct inode * inode;
1945 struct proc_inode *ei;
1946 struct pid *pid;
1947
1948 /* We need a new inode */
1949
1950 inode = new_inode(sb);
1951 if (!inode)
1952 goto out;
1953
1954 /* Common stuff */
1955 ei = PROC_I(inode);
1956 inode->i_mode = mode;
1957 inode->i_ino = get_next_ino();
1958 simple_inode_init_ts(inode);
1959 inode->i_op = &proc_def_inode_operations;
1960
1961 /*
1962 * grab the reference to task.
1963 */
1964 pid = get_task_pid(task, PIDTYPE_PID);
1965 if (!pid)
1966 goto out_unlock;
1967
1968 /* Let the pid remember us for quick removal */
1969 ei->pid = pid;
1970
1971 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1972 security_task_to_inode(task, inode);
1973
1974out:
1975 return inode;
1976
1977out_unlock:
1978 iput(inode);
1979 return NULL;
1980}
1981
1982/*
1983 * Generating an inode and adding it into @pid->inodes, so that task will
1984 * invalidate inode's dentry before being released.
1985 *
1986 * This helper is used for creating dir-type entries under '/proc' and
1987 * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1988 * can be released by invalidating '/proc/<tgid>' dentry.
1989 * In theory, dentries under '/proc/<tgid>/task' can also be released by
1990 * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1991 * thread exiting situation: Any one of threads should invalidate its
1992 * '/proc/<tgid>/task/<pid>' dentry before released.
1993 */
1994static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1995 struct task_struct *task, umode_t mode)
1996{
1997 struct inode *inode;
1998 struct proc_inode *ei;
1999 struct pid *pid;
2000
2001 inode = proc_pid_make_inode(sb, task, mode);
2002 if (!inode)
2003 return NULL;
2004
2005 /* Let proc_flush_pid find this directory inode */
2006 ei = PROC_I(inode);
2007 pid = ei->pid;
2008 spin_lock(&pid->lock);
2009 hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
2010 spin_unlock(&pid->lock);
2011
2012 return inode;
2013}
2014
2015int pid_getattr(struct mnt_idmap *idmap, const struct path *path,
2016 struct kstat *stat, u32 request_mask, unsigned int query_flags)
2017{
2018 struct inode *inode = d_inode(path->dentry);
2019 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
2020 struct task_struct *task;
2021
2022 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
2023
2024 stat->uid = GLOBAL_ROOT_UID;
2025 stat->gid = GLOBAL_ROOT_GID;
2026 rcu_read_lock();
2027 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2028 if (task) {
2029 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
2030 rcu_read_unlock();
2031 /*
2032 * This doesn't prevent learning whether PID exists,
2033 * it only makes getattr() consistent with readdir().
2034 */
2035 return -ENOENT;
2036 }
2037 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
2038 }
2039 rcu_read_unlock();
2040 return 0;
2041}
2042
2043/* dentry stuff */
2044
2045/*
2046 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
2047 */
2048void pid_update_inode(struct task_struct *task, struct inode *inode)
2049{
2050 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
2051
2052 inode->i_mode &= ~(S_ISUID | S_ISGID);
2053 security_task_to_inode(task, inode);
2054}
2055
2056/*
2057 * Rewrite the inode's ownerships here because the owning task may have
2058 * performed a setuid(), etc.
2059 *
2060 */
2061static int pid_revalidate(struct dentry *dentry, unsigned int flags)
2062{
2063 struct inode *inode;
2064 struct task_struct *task;
2065 int ret = 0;
2066
2067 rcu_read_lock();
2068 inode = d_inode_rcu(dentry);
2069 if (!inode)
2070 goto out;
2071 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2072
2073 if (task) {
2074 pid_update_inode(task, inode);
2075 ret = 1;
2076 }
2077out:
2078 rcu_read_unlock();
2079 return ret;
2080}
2081
2082static inline bool proc_inode_is_dead(struct inode *inode)
2083{
2084 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2085}
2086
2087int pid_delete_dentry(const struct dentry *dentry)
2088{
2089 /* Is the task we represent dead?
2090 * If so, then don't put the dentry on the lru list,
2091 * kill it immediately.
2092 */
2093 return proc_inode_is_dead(d_inode(dentry));
2094}
2095
2096const struct dentry_operations pid_dentry_operations =
2097{
2098 .d_revalidate = pid_revalidate,
2099 .d_delete = pid_delete_dentry,
2100};
2101
2102/* Lookups */
2103
2104/*
2105 * Fill a directory entry.
2106 *
2107 * If possible create the dcache entry and derive our inode number and
2108 * file type from dcache entry.
2109 *
2110 * Since all of the proc inode numbers are dynamically generated, the inode
2111 * numbers do not exist until the inode is cache. This means creating
2112 * the dcache entry in readdir is necessary to keep the inode numbers
2113 * reported by readdir in sync with the inode numbers reported
2114 * by stat.
2115 */
2116bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2117 const char *name, unsigned int len,
2118 instantiate_t instantiate, struct task_struct *task, const void *ptr)
2119{
2120 struct dentry *child, *dir = file->f_path.dentry;
2121 struct qstr qname = QSTR_INIT(name, len);
2122 struct inode *inode;
2123 unsigned type = DT_UNKNOWN;
2124 ino_t ino = 1;
2125
2126 child = d_hash_and_lookup(dir, &qname);
2127 if (!child) {
2128 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2129 child = d_alloc_parallel(dir, &qname, &wq);
2130 if (IS_ERR(child))
2131 goto end_instantiate;
2132 if (d_in_lookup(child)) {
2133 struct dentry *res;
2134 res = instantiate(child, task, ptr);
2135 d_lookup_done(child);
2136 if (unlikely(res)) {
2137 dput(child);
2138 child = res;
2139 if (IS_ERR(child))
2140 goto end_instantiate;
2141 }
2142 }
2143 }
2144 inode = d_inode(child);
2145 ino = inode->i_ino;
2146 type = inode->i_mode >> 12;
2147 dput(child);
2148end_instantiate:
2149 return dir_emit(ctx, name, len, ino, type);
2150}
2151
2152/*
2153 * dname_to_vma_addr - maps a dentry name into two unsigned longs
2154 * which represent vma start and end addresses.
2155 */
2156static int dname_to_vma_addr(struct dentry *dentry,
2157 unsigned long *start, unsigned long *end)
2158{
2159 const char *str = dentry->d_name.name;
2160 unsigned long long sval, eval;
2161 unsigned int len;
2162
2163 if (str[0] == '0' && str[1] != '-')
2164 return -EINVAL;
2165 len = _parse_integer(str, 16, &sval);
2166 if (len & KSTRTOX_OVERFLOW)
2167 return -EINVAL;
2168 if (sval != (unsigned long)sval)
2169 return -EINVAL;
2170 str += len;
2171
2172 if (*str != '-')
2173 return -EINVAL;
2174 str++;
2175
2176 if (str[0] == '0' && str[1])
2177 return -EINVAL;
2178 len = _parse_integer(str, 16, &eval);
2179 if (len & KSTRTOX_OVERFLOW)
2180 return -EINVAL;
2181 if (eval != (unsigned long)eval)
2182 return -EINVAL;
2183 str += len;
2184
2185 if (*str != '\0')
2186 return -EINVAL;
2187
2188 *start = sval;
2189 *end = eval;
2190
2191 return 0;
2192}
2193
2194static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2195{
2196 unsigned long vm_start, vm_end;
2197 bool exact_vma_exists = false;
2198 struct mm_struct *mm = NULL;
2199 struct task_struct *task;
2200 struct inode *inode;
2201 int status = 0;
2202
2203 if (flags & LOOKUP_RCU)
2204 return -ECHILD;
2205
2206 inode = d_inode(dentry);
2207 task = get_proc_task(inode);
2208 if (!task)
2209 goto out_notask;
2210
2211 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2212 if (IS_ERR(mm))
2213 goto out;
2214
2215 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2216 status = mmap_read_lock_killable(mm);
2217 if (!status) {
2218 exact_vma_exists = !!find_exact_vma(mm, vm_start,
2219 vm_end);
2220 mmap_read_unlock(mm);
2221 }
2222 }
2223
2224 mmput(mm);
2225
2226 if (exact_vma_exists) {
2227 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2228
2229 security_task_to_inode(task, inode);
2230 status = 1;
2231 }
2232
2233out:
2234 put_task_struct(task);
2235
2236out_notask:
2237 return status;
2238}
2239
2240static const struct dentry_operations tid_map_files_dentry_operations = {
2241 .d_revalidate = map_files_d_revalidate,
2242 .d_delete = pid_delete_dentry,
2243};
2244
2245static int map_files_get_link(struct dentry *dentry, struct path *path)
2246{
2247 unsigned long vm_start, vm_end;
2248 struct vm_area_struct *vma;
2249 struct task_struct *task;
2250 struct mm_struct *mm;
2251 int rc;
2252
2253 rc = -ENOENT;
2254 task = get_proc_task(d_inode(dentry));
2255 if (!task)
2256 goto out;
2257
2258 mm = get_task_mm(task);
2259 put_task_struct(task);
2260 if (!mm)
2261 goto out;
2262
2263 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2264 if (rc)
2265 goto out_mmput;
2266
2267 rc = mmap_read_lock_killable(mm);
2268 if (rc)
2269 goto out_mmput;
2270
2271 rc = -ENOENT;
2272 vma = find_exact_vma(mm, vm_start, vm_end);
2273 if (vma && vma->vm_file) {
2274 *path = *file_user_path(vma->vm_file);
2275 path_get(path);
2276 rc = 0;
2277 }
2278 mmap_read_unlock(mm);
2279
2280out_mmput:
2281 mmput(mm);
2282out:
2283 return rc;
2284}
2285
2286struct map_files_info {
2287 unsigned long start;
2288 unsigned long end;
2289 fmode_t mode;
2290};
2291
2292/*
2293 * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2294 * to concerns about how the symlinks may be used to bypass permissions on
2295 * ancestor directories in the path to the file in question.
2296 */
2297static const char *
2298proc_map_files_get_link(struct dentry *dentry,
2299 struct inode *inode,
2300 struct delayed_call *done)
2301{
2302 if (!checkpoint_restore_ns_capable(&init_user_ns))
2303 return ERR_PTR(-EPERM);
2304
2305 return proc_pid_get_link(dentry, inode, done);
2306}
2307
2308/*
2309 * Identical to proc_pid_link_inode_operations except for get_link()
2310 */
2311static const struct inode_operations proc_map_files_link_inode_operations = {
2312 .readlink = proc_pid_readlink,
2313 .get_link = proc_map_files_get_link,
2314 .setattr = proc_setattr,
2315};
2316
2317static struct dentry *
2318proc_map_files_instantiate(struct dentry *dentry,
2319 struct task_struct *task, const void *ptr)
2320{
2321 fmode_t mode = (fmode_t)(unsigned long)ptr;
2322 struct proc_inode *ei;
2323 struct inode *inode;
2324
2325 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2326 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2327 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2328 if (!inode)
2329 return ERR_PTR(-ENOENT);
2330
2331 ei = PROC_I(inode);
2332 ei->op.proc_get_link = map_files_get_link;
2333
2334 inode->i_op = &proc_map_files_link_inode_operations;
2335 inode->i_size = 64;
2336
2337 return proc_splice_unmountable(inode, dentry,
2338 &tid_map_files_dentry_operations);
2339}
2340
2341static struct dentry *proc_map_files_lookup(struct inode *dir,
2342 struct dentry *dentry, unsigned int flags)
2343{
2344 unsigned long vm_start, vm_end;
2345 struct vm_area_struct *vma;
2346 struct task_struct *task;
2347 struct dentry *result;
2348 struct mm_struct *mm;
2349
2350 result = ERR_PTR(-ENOENT);
2351 task = get_proc_task(dir);
2352 if (!task)
2353 goto out;
2354
2355 result = ERR_PTR(-EACCES);
2356 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2357 goto out_put_task;
2358
2359 result = ERR_PTR(-ENOENT);
2360 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2361 goto out_put_task;
2362
2363 mm = get_task_mm(task);
2364 if (!mm)
2365 goto out_put_task;
2366
2367 result = ERR_PTR(-EINTR);
2368 if (mmap_read_lock_killable(mm))
2369 goto out_put_mm;
2370
2371 result = ERR_PTR(-ENOENT);
2372 vma = find_exact_vma(mm, vm_start, vm_end);
2373 if (!vma)
2374 goto out_no_vma;
2375
2376 if (vma->vm_file)
2377 result = proc_map_files_instantiate(dentry, task,
2378 (void *)(unsigned long)vma->vm_file->f_mode);
2379
2380out_no_vma:
2381 mmap_read_unlock(mm);
2382out_put_mm:
2383 mmput(mm);
2384out_put_task:
2385 put_task_struct(task);
2386out:
2387 return result;
2388}
2389
2390static const struct inode_operations proc_map_files_inode_operations = {
2391 .lookup = proc_map_files_lookup,
2392 .permission = proc_fd_permission,
2393 .setattr = proc_setattr,
2394};
2395
2396static int
2397proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2398{
2399 struct vm_area_struct *vma;
2400 struct task_struct *task;
2401 struct mm_struct *mm;
2402 unsigned long nr_files, pos, i;
2403 GENRADIX(struct map_files_info) fa;
2404 struct map_files_info *p;
2405 int ret;
2406 struct vma_iterator vmi;
2407
2408 genradix_init(&fa);
2409
2410 ret = -ENOENT;
2411 task = get_proc_task(file_inode(file));
2412 if (!task)
2413 goto out;
2414
2415 ret = -EACCES;
2416 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2417 goto out_put_task;
2418
2419 ret = 0;
2420 if (!dir_emit_dots(file, ctx))
2421 goto out_put_task;
2422
2423 mm = get_task_mm(task);
2424 if (!mm)
2425 goto out_put_task;
2426
2427 ret = mmap_read_lock_killable(mm);
2428 if (ret) {
2429 mmput(mm);
2430 goto out_put_task;
2431 }
2432
2433 nr_files = 0;
2434
2435 /*
2436 * We need two passes here:
2437 *
2438 * 1) Collect vmas of mapped files with mmap_lock taken
2439 * 2) Release mmap_lock and instantiate entries
2440 *
2441 * otherwise we get lockdep complained, since filldir()
2442 * routine might require mmap_lock taken in might_fault().
2443 */
2444
2445 pos = 2;
2446 vma_iter_init(&vmi, mm, 0);
2447 for_each_vma(vmi, vma) {
2448 if (!vma->vm_file)
2449 continue;
2450 if (++pos <= ctx->pos)
2451 continue;
2452
2453 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2454 if (!p) {
2455 ret = -ENOMEM;
2456 mmap_read_unlock(mm);
2457 mmput(mm);
2458 goto out_put_task;
2459 }
2460
2461 p->start = vma->vm_start;
2462 p->end = vma->vm_end;
2463 p->mode = vma->vm_file->f_mode;
2464 }
2465 mmap_read_unlock(mm);
2466 mmput(mm);
2467
2468 for (i = 0; i < nr_files; i++) {
2469 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2470 unsigned int len;
2471
2472 p = genradix_ptr(&fa, i);
2473 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2474 if (!proc_fill_cache(file, ctx,
2475 buf, len,
2476 proc_map_files_instantiate,
2477 task,
2478 (void *)(unsigned long)p->mode))
2479 break;
2480 ctx->pos++;
2481 }
2482
2483out_put_task:
2484 put_task_struct(task);
2485out:
2486 genradix_free(&fa);
2487 return ret;
2488}
2489
2490static const struct file_operations proc_map_files_operations = {
2491 .read = generic_read_dir,
2492 .iterate_shared = proc_map_files_readdir,
2493 .llseek = generic_file_llseek,
2494};
2495
2496#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2497struct timers_private {
2498 struct pid *pid;
2499 struct task_struct *task;
2500 struct sighand_struct *sighand;
2501 struct pid_namespace *ns;
2502 unsigned long flags;
2503};
2504
2505static void *timers_start(struct seq_file *m, loff_t *pos)
2506{
2507 struct timers_private *tp = m->private;
2508
2509 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2510 if (!tp->task)
2511 return ERR_PTR(-ESRCH);
2512
2513 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2514 if (!tp->sighand)
2515 return ERR_PTR(-ESRCH);
2516
2517 return seq_hlist_start(&tp->task->signal->posix_timers, *pos);
2518}
2519
2520static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2521{
2522 struct timers_private *tp = m->private;
2523 return seq_hlist_next(v, &tp->task->signal->posix_timers, pos);
2524}
2525
2526static void timers_stop(struct seq_file *m, void *v)
2527{
2528 struct timers_private *tp = m->private;
2529
2530 if (tp->sighand) {
2531 unlock_task_sighand(tp->task, &tp->flags);
2532 tp->sighand = NULL;
2533 }
2534
2535 if (tp->task) {
2536 put_task_struct(tp->task);
2537 tp->task = NULL;
2538 }
2539}
2540
2541static int show_timer(struct seq_file *m, void *v)
2542{
2543 struct k_itimer *timer;
2544 struct timers_private *tp = m->private;
2545 int notify;
2546 static const char * const nstr[] = {
2547 [SIGEV_SIGNAL] = "signal",
2548 [SIGEV_NONE] = "none",
2549 [SIGEV_THREAD] = "thread",
2550 };
2551
2552 timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
2553 notify = timer->it_sigev_notify;
2554
2555 seq_printf(m, "ID: %d\n", timer->it_id);
2556 seq_printf(m, "signal: %d/%px\n",
2557 timer->sigq.info.si_signo,
2558 timer->sigq.info.si_value.sival_ptr);
2559 seq_printf(m, "notify: %s/%s.%d\n",
2560 nstr[notify & ~SIGEV_THREAD_ID],
2561 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2562 pid_nr_ns(timer->it_pid, tp->ns));
2563 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2564
2565 return 0;
2566}
2567
2568static const struct seq_operations proc_timers_seq_ops = {
2569 .start = timers_start,
2570 .next = timers_next,
2571 .stop = timers_stop,
2572 .show = show_timer,
2573};
2574
2575static int proc_timers_open(struct inode *inode, struct file *file)
2576{
2577 struct timers_private *tp;
2578
2579 tp = __seq_open_private(file, &proc_timers_seq_ops,
2580 sizeof(struct timers_private));
2581 if (!tp)
2582 return -ENOMEM;
2583
2584 tp->pid = proc_pid(inode);
2585 tp->ns = proc_pid_ns(inode->i_sb);
2586 return 0;
2587}
2588
2589static const struct file_operations proc_timers_operations = {
2590 .open = proc_timers_open,
2591 .read = seq_read,
2592 .llseek = seq_lseek,
2593 .release = seq_release_private,
2594};
2595#endif
2596
2597static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2598 size_t count, loff_t *offset)
2599{
2600 struct inode *inode = file_inode(file);
2601 struct task_struct *p;
2602 u64 slack_ns;
2603 int err;
2604
2605 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2606 if (err < 0)
2607 return err;
2608
2609 p = get_proc_task(inode);
2610 if (!p)
2611 return -ESRCH;
2612
2613 if (p != current) {
2614 rcu_read_lock();
2615 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2616 rcu_read_unlock();
2617 count = -EPERM;
2618 goto out;
2619 }
2620 rcu_read_unlock();
2621
2622 err = security_task_setscheduler(p);
2623 if (err) {
2624 count = err;
2625 goto out;
2626 }
2627 }
2628
2629 task_lock(p);
2630 if (rt_or_dl_task_policy(p))
2631 slack_ns = 0;
2632 else if (slack_ns == 0)
2633 slack_ns = p->default_timer_slack_ns;
2634 p->timer_slack_ns = slack_ns;
2635 task_unlock(p);
2636
2637out:
2638 put_task_struct(p);
2639
2640 return count;
2641}
2642
2643static int timerslack_ns_show(struct seq_file *m, void *v)
2644{
2645 struct inode *inode = m->private;
2646 struct task_struct *p;
2647 int err = 0;
2648
2649 p = get_proc_task(inode);
2650 if (!p)
2651 return -ESRCH;
2652
2653 if (p != current) {
2654 rcu_read_lock();
2655 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2656 rcu_read_unlock();
2657 err = -EPERM;
2658 goto out;
2659 }
2660 rcu_read_unlock();
2661
2662 err = security_task_getscheduler(p);
2663 if (err)
2664 goto out;
2665 }
2666
2667 task_lock(p);
2668 seq_printf(m, "%llu\n", p->timer_slack_ns);
2669 task_unlock(p);
2670
2671out:
2672 put_task_struct(p);
2673
2674 return err;
2675}
2676
2677static int timerslack_ns_open(struct inode *inode, struct file *filp)
2678{
2679 return single_open(filp, timerslack_ns_show, inode);
2680}
2681
2682static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2683 .open = timerslack_ns_open,
2684 .read = seq_read,
2685 .write = timerslack_ns_write,
2686 .llseek = seq_lseek,
2687 .release = single_release,
2688};
2689
2690static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2691 struct task_struct *task, const void *ptr)
2692{
2693 const struct pid_entry *p = ptr;
2694 struct inode *inode;
2695 struct proc_inode *ei;
2696
2697 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2698 if (!inode)
2699 return ERR_PTR(-ENOENT);
2700
2701 ei = PROC_I(inode);
2702 if (S_ISDIR(inode->i_mode))
2703 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2704 if (p->iop)
2705 inode->i_op = p->iop;
2706 if (p->fop)
2707 inode->i_fop = p->fop;
2708 ei->op = p->op;
2709 pid_update_inode(task, inode);
2710 d_set_d_op(dentry, &pid_dentry_operations);
2711 return d_splice_alias(inode, dentry);
2712}
2713
2714static struct dentry *proc_pident_lookup(struct inode *dir,
2715 struct dentry *dentry,
2716 const struct pid_entry *p,
2717 const struct pid_entry *end)
2718{
2719 struct task_struct *task = get_proc_task(dir);
2720 struct dentry *res = ERR_PTR(-ENOENT);
2721
2722 if (!task)
2723 goto out_no_task;
2724
2725 /*
2726 * Yes, it does not scale. And it should not. Don't add
2727 * new entries into /proc/<tgid>/ without very good reasons.
2728 */
2729 for (; p < end; p++) {
2730 if (p->len != dentry->d_name.len)
2731 continue;
2732 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2733 res = proc_pident_instantiate(dentry, task, p);
2734 break;
2735 }
2736 }
2737 put_task_struct(task);
2738out_no_task:
2739 return res;
2740}
2741
2742static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2743 const struct pid_entry *ents, unsigned int nents)
2744{
2745 struct task_struct *task = get_proc_task(file_inode(file));
2746 const struct pid_entry *p;
2747
2748 if (!task)
2749 return -ENOENT;
2750
2751 if (!dir_emit_dots(file, ctx))
2752 goto out;
2753
2754 if (ctx->pos >= nents + 2)
2755 goto out;
2756
2757 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2758 if (!proc_fill_cache(file, ctx, p->name, p->len,
2759 proc_pident_instantiate, task, p))
2760 break;
2761 ctx->pos++;
2762 }
2763out:
2764 put_task_struct(task);
2765 return 0;
2766}
2767
2768#ifdef CONFIG_SECURITY
2769static int proc_pid_attr_open(struct inode *inode, struct file *file)
2770{
2771 file->private_data = NULL;
2772 __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2773 return 0;
2774}
2775
2776static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2777 size_t count, loff_t *ppos)
2778{
2779 struct inode * inode = file_inode(file);
2780 char *p = NULL;
2781 ssize_t length;
2782 struct task_struct *task = get_proc_task(inode);
2783
2784 if (!task)
2785 return -ESRCH;
2786
2787 length = security_getprocattr(task, PROC_I(inode)->op.lsmid,
2788 file->f_path.dentry->d_name.name,
2789 &p);
2790 put_task_struct(task);
2791 if (length > 0)
2792 length = simple_read_from_buffer(buf, count, ppos, p, length);
2793 kfree(p);
2794 return length;
2795}
2796
2797static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2798 size_t count, loff_t *ppos)
2799{
2800 struct inode * inode = file_inode(file);
2801 struct task_struct *task;
2802 void *page;
2803 int rv;
2804
2805 /* A task may only write when it was the opener. */
2806 if (file->private_data != current->mm)
2807 return -EPERM;
2808
2809 rcu_read_lock();
2810 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2811 if (!task) {
2812 rcu_read_unlock();
2813 return -ESRCH;
2814 }
2815 /* A task may only write its own attributes. */
2816 if (current != task) {
2817 rcu_read_unlock();
2818 return -EACCES;
2819 }
2820 /* Prevent changes to overridden credentials. */
2821 if (current_cred() != current_real_cred()) {
2822 rcu_read_unlock();
2823 return -EBUSY;
2824 }
2825 rcu_read_unlock();
2826
2827 if (count > PAGE_SIZE)
2828 count = PAGE_SIZE;
2829
2830 /* No partial writes. */
2831 if (*ppos != 0)
2832 return -EINVAL;
2833
2834 page = memdup_user(buf, count);
2835 if (IS_ERR(page)) {
2836 rv = PTR_ERR(page);
2837 goto out;
2838 }
2839
2840 /* Guard against adverse ptrace interaction */
2841 rv = mutex_lock_interruptible(¤t->signal->cred_guard_mutex);
2842 if (rv < 0)
2843 goto out_free;
2844
2845 rv = security_setprocattr(PROC_I(inode)->op.lsmid,
2846 file->f_path.dentry->d_name.name, page,
2847 count);
2848 mutex_unlock(¤t->signal->cred_guard_mutex);
2849out_free:
2850 kfree(page);
2851out:
2852 return rv;
2853}
2854
2855static const struct file_operations proc_pid_attr_operations = {
2856 .open = proc_pid_attr_open,
2857 .read = proc_pid_attr_read,
2858 .write = proc_pid_attr_write,
2859 .llseek = generic_file_llseek,
2860 .release = mem_release,
2861};
2862
2863#define LSM_DIR_OPS(LSM) \
2864static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2865 struct dir_context *ctx) \
2866{ \
2867 return proc_pident_readdir(filp, ctx, \
2868 LSM##_attr_dir_stuff, \
2869 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2870} \
2871\
2872static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2873 .read = generic_read_dir, \
2874 .iterate_shared = proc_##LSM##_attr_dir_iterate, \
2875 .llseek = default_llseek, \
2876}; \
2877\
2878static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2879 struct dentry *dentry, unsigned int flags) \
2880{ \
2881 return proc_pident_lookup(dir, dentry, \
2882 LSM##_attr_dir_stuff, \
2883 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2884} \
2885\
2886static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2887 .lookup = proc_##LSM##_attr_dir_lookup, \
2888 .getattr = pid_getattr, \
2889 .setattr = proc_setattr, \
2890}
2891
2892#ifdef CONFIG_SECURITY_SMACK
2893static const struct pid_entry smack_attr_dir_stuff[] = {
2894 ATTR(LSM_ID_SMACK, "current", 0666),
2895};
2896LSM_DIR_OPS(smack);
2897#endif
2898
2899#ifdef CONFIG_SECURITY_APPARMOR
2900static const struct pid_entry apparmor_attr_dir_stuff[] = {
2901 ATTR(LSM_ID_APPARMOR, "current", 0666),
2902 ATTR(LSM_ID_APPARMOR, "prev", 0444),
2903 ATTR(LSM_ID_APPARMOR, "exec", 0666),
2904};
2905LSM_DIR_OPS(apparmor);
2906#endif
2907
2908static const struct pid_entry attr_dir_stuff[] = {
2909 ATTR(LSM_ID_UNDEF, "current", 0666),
2910 ATTR(LSM_ID_UNDEF, "prev", 0444),
2911 ATTR(LSM_ID_UNDEF, "exec", 0666),
2912 ATTR(LSM_ID_UNDEF, "fscreate", 0666),
2913 ATTR(LSM_ID_UNDEF, "keycreate", 0666),
2914 ATTR(LSM_ID_UNDEF, "sockcreate", 0666),
2915#ifdef CONFIG_SECURITY_SMACK
2916 DIR("smack", 0555,
2917 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2918#endif
2919#ifdef CONFIG_SECURITY_APPARMOR
2920 DIR("apparmor", 0555,
2921 proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2922#endif
2923};
2924
2925static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2926{
2927 return proc_pident_readdir(file, ctx,
2928 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2929}
2930
2931static const struct file_operations proc_attr_dir_operations = {
2932 .read = generic_read_dir,
2933 .iterate_shared = proc_attr_dir_readdir,
2934 .llseek = generic_file_llseek,
2935};
2936
2937static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2938 struct dentry *dentry, unsigned int flags)
2939{
2940 return proc_pident_lookup(dir, dentry,
2941 attr_dir_stuff,
2942 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2943}
2944
2945static const struct inode_operations proc_attr_dir_inode_operations = {
2946 .lookup = proc_attr_dir_lookup,
2947 .getattr = pid_getattr,
2948 .setattr = proc_setattr,
2949};
2950
2951#endif
2952
2953#ifdef CONFIG_ELF_CORE
2954static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2955 size_t count, loff_t *ppos)
2956{
2957 struct task_struct *task = get_proc_task(file_inode(file));
2958 struct mm_struct *mm;
2959 char buffer[PROC_NUMBUF];
2960 size_t len;
2961 int ret;
2962
2963 if (!task)
2964 return -ESRCH;
2965
2966 ret = 0;
2967 mm = get_task_mm(task);
2968 if (mm) {
2969 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2970 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2971 MMF_DUMP_FILTER_SHIFT));
2972 mmput(mm);
2973 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2974 }
2975
2976 put_task_struct(task);
2977
2978 return ret;
2979}
2980
2981static ssize_t proc_coredump_filter_write(struct file *file,
2982 const char __user *buf,
2983 size_t count,
2984 loff_t *ppos)
2985{
2986 struct task_struct *task;
2987 struct mm_struct *mm;
2988 unsigned int val;
2989 int ret;
2990 int i;
2991 unsigned long mask;
2992
2993 ret = kstrtouint_from_user(buf, count, 0, &val);
2994 if (ret < 0)
2995 return ret;
2996
2997 ret = -ESRCH;
2998 task = get_proc_task(file_inode(file));
2999 if (!task)
3000 goto out_no_task;
3001
3002 mm = get_task_mm(task);
3003 if (!mm)
3004 goto out_no_mm;
3005 ret = 0;
3006
3007 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
3008 if (val & mask)
3009 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3010 else
3011 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3012 }
3013
3014 mmput(mm);
3015 out_no_mm:
3016 put_task_struct(task);
3017 out_no_task:
3018 if (ret < 0)
3019 return ret;
3020 return count;
3021}
3022
3023static const struct file_operations proc_coredump_filter_operations = {
3024 .read = proc_coredump_filter_read,
3025 .write = proc_coredump_filter_write,
3026 .llseek = generic_file_llseek,
3027};
3028#endif
3029
3030#ifdef CONFIG_TASK_IO_ACCOUNTING
3031static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
3032{
3033 struct task_io_accounting acct;
3034 int result;
3035
3036 result = down_read_killable(&task->signal->exec_update_lock);
3037 if (result)
3038 return result;
3039
3040 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
3041 result = -EACCES;
3042 goto out_unlock;
3043 }
3044
3045 if (whole) {
3046 struct signal_struct *sig = task->signal;
3047 struct task_struct *t;
3048 unsigned int seq = 1;
3049 unsigned long flags;
3050
3051 rcu_read_lock();
3052 do {
3053 seq++; /* 2 on the 1st/lockless path, otherwise odd */
3054 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
3055
3056 acct = sig->ioac;
3057 __for_each_thread(sig, t)
3058 task_io_accounting_add(&acct, &t->ioac);
3059
3060 } while (need_seqretry(&sig->stats_lock, seq));
3061 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
3062 rcu_read_unlock();
3063 } else {
3064 acct = task->ioac;
3065 }
3066
3067 seq_printf(m,
3068 "rchar: %llu\n"
3069 "wchar: %llu\n"
3070 "syscr: %llu\n"
3071 "syscw: %llu\n"
3072 "read_bytes: %llu\n"
3073 "write_bytes: %llu\n"
3074 "cancelled_write_bytes: %llu\n",
3075 (unsigned long long)acct.rchar,
3076 (unsigned long long)acct.wchar,
3077 (unsigned long long)acct.syscr,
3078 (unsigned long long)acct.syscw,
3079 (unsigned long long)acct.read_bytes,
3080 (unsigned long long)acct.write_bytes,
3081 (unsigned long long)acct.cancelled_write_bytes);
3082 result = 0;
3083
3084out_unlock:
3085 up_read(&task->signal->exec_update_lock);
3086 return result;
3087}
3088
3089static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3090 struct pid *pid, struct task_struct *task)
3091{
3092 return do_io_accounting(task, m, 0);
3093}
3094
3095static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3096 struct pid *pid, struct task_struct *task)
3097{
3098 return do_io_accounting(task, m, 1);
3099}
3100#endif /* CONFIG_TASK_IO_ACCOUNTING */
3101
3102#ifdef CONFIG_USER_NS
3103static int proc_id_map_open(struct inode *inode, struct file *file,
3104 const struct seq_operations *seq_ops)
3105{
3106 struct user_namespace *ns = NULL;
3107 struct task_struct *task;
3108 struct seq_file *seq;
3109 int ret = -EINVAL;
3110
3111 task = get_proc_task(inode);
3112 if (task) {
3113 rcu_read_lock();
3114 ns = get_user_ns(task_cred_xxx(task, user_ns));
3115 rcu_read_unlock();
3116 put_task_struct(task);
3117 }
3118 if (!ns)
3119 goto err;
3120
3121 ret = seq_open(file, seq_ops);
3122 if (ret)
3123 goto err_put_ns;
3124
3125 seq = file->private_data;
3126 seq->private = ns;
3127
3128 return 0;
3129err_put_ns:
3130 put_user_ns(ns);
3131err:
3132 return ret;
3133}
3134
3135static int proc_id_map_release(struct inode *inode, struct file *file)
3136{
3137 struct seq_file *seq = file->private_data;
3138 struct user_namespace *ns = seq->private;
3139 put_user_ns(ns);
3140 return seq_release(inode, file);
3141}
3142
3143static int proc_uid_map_open(struct inode *inode, struct file *file)
3144{
3145 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3146}
3147
3148static int proc_gid_map_open(struct inode *inode, struct file *file)
3149{
3150 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3151}
3152
3153static int proc_projid_map_open(struct inode *inode, struct file *file)
3154{
3155 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3156}
3157
3158static const struct file_operations proc_uid_map_operations = {
3159 .open = proc_uid_map_open,
3160 .write = proc_uid_map_write,
3161 .read = seq_read,
3162 .llseek = seq_lseek,
3163 .release = proc_id_map_release,
3164};
3165
3166static const struct file_operations proc_gid_map_operations = {
3167 .open = proc_gid_map_open,
3168 .write = proc_gid_map_write,
3169 .read = seq_read,
3170 .llseek = seq_lseek,
3171 .release = proc_id_map_release,
3172};
3173
3174static const struct file_operations proc_projid_map_operations = {
3175 .open = proc_projid_map_open,
3176 .write = proc_projid_map_write,
3177 .read = seq_read,
3178 .llseek = seq_lseek,
3179 .release = proc_id_map_release,
3180};
3181
3182static int proc_setgroups_open(struct inode *inode, struct file *file)
3183{
3184 struct user_namespace *ns = NULL;
3185 struct task_struct *task;
3186 int ret;
3187
3188 ret = -ESRCH;
3189 task = get_proc_task(inode);
3190 if (task) {
3191 rcu_read_lock();
3192 ns = get_user_ns(task_cred_xxx(task, user_ns));
3193 rcu_read_unlock();
3194 put_task_struct(task);
3195 }
3196 if (!ns)
3197 goto err;
3198
3199 if (file->f_mode & FMODE_WRITE) {
3200 ret = -EACCES;
3201 if (!ns_capable(ns, CAP_SYS_ADMIN))
3202 goto err_put_ns;
3203 }
3204
3205 ret = single_open(file, &proc_setgroups_show, ns);
3206 if (ret)
3207 goto err_put_ns;
3208
3209 return 0;
3210err_put_ns:
3211 put_user_ns(ns);
3212err:
3213 return ret;
3214}
3215
3216static int proc_setgroups_release(struct inode *inode, struct file *file)
3217{
3218 struct seq_file *seq = file->private_data;
3219 struct user_namespace *ns = seq->private;
3220 int ret = single_release(inode, file);
3221 put_user_ns(ns);
3222 return ret;
3223}
3224
3225static const struct file_operations proc_setgroups_operations = {
3226 .open = proc_setgroups_open,
3227 .write = proc_setgroups_write,
3228 .read = seq_read,
3229 .llseek = seq_lseek,
3230 .release = proc_setgroups_release,
3231};
3232#endif /* CONFIG_USER_NS */
3233
3234static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3235 struct pid *pid, struct task_struct *task)
3236{
3237 int err = lock_trace(task);
3238 if (!err) {
3239 seq_printf(m, "%08x\n", task->personality);
3240 unlock_trace(task);
3241 }
3242 return err;
3243}
3244
3245#ifdef CONFIG_LIVEPATCH
3246static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3247 struct pid *pid, struct task_struct *task)
3248{
3249 seq_printf(m, "%d\n", task->patch_state);
3250 return 0;
3251}
3252#endif /* CONFIG_LIVEPATCH */
3253
3254#ifdef CONFIG_KSM
3255static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3256 struct pid *pid, struct task_struct *task)
3257{
3258 struct mm_struct *mm;
3259
3260 mm = get_task_mm(task);
3261 if (mm) {
3262 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3263 mmput(mm);
3264 }
3265
3266 return 0;
3267}
3268static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3269 struct pid *pid, struct task_struct *task)
3270{
3271 struct mm_struct *mm;
3272
3273 mm = get_task_mm(task);
3274 if (mm) {
3275 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3276 seq_printf(m, "ksm_zero_pages %ld\n", mm_ksm_zero_pages(mm));
3277 seq_printf(m, "ksm_merging_pages %lu\n", mm->ksm_merging_pages);
3278 seq_printf(m, "ksm_process_profit %ld\n", ksm_process_profit(mm));
3279 mmput(mm);
3280 }
3281
3282 return 0;
3283}
3284#endif /* CONFIG_KSM */
3285
3286#ifdef CONFIG_STACKLEAK_METRICS
3287static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3288 struct pid *pid, struct task_struct *task)
3289{
3290 unsigned long prev_depth = THREAD_SIZE -
3291 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3292 unsigned long depth = THREAD_SIZE -
3293 (task->lowest_stack & (THREAD_SIZE - 1));
3294
3295 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3296 prev_depth, depth);
3297 return 0;
3298}
3299#endif /* CONFIG_STACKLEAK_METRICS */
3300
3301/*
3302 * Thread groups
3303 */
3304static const struct file_operations proc_task_operations;
3305static const struct inode_operations proc_task_inode_operations;
3306
3307static const struct pid_entry tgid_base_stuff[] = {
3308 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3309 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3310 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3311 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3312 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3313#ifdef CONFIG_NET
3314 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3315#endif
3316 REG("environ", S_IRUSR, proc_environ_operations),
3317 REG("auxv", S_IRUSR, proc_auxv_operations),
3318 ONE("status", S_IRUGO, proc_pid_status),
3319 ONE("personality", S_IRUSR, proc_pid_personality),
3320 ONE("limits", S_IRUGO, proc_pid_limits),
3321#ifdef CONFIG_SCHED_DEBUG
3322 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3323#endif
3324#ifdef CONFIG_SCHED_AUTOGROUP
3325 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3326#endif
3327#ifdef CONFIG_TIME_NS
3328 REG("timens_offsets", S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3329#endif
3330 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3331#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3332 ONE("syscall", S_IRUSR, proc_pid_syscall),
3333#endif
3334 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3335 ONE("stat", S_IRUGO, proc_tgid_stat),
3336 ONE("statm", S_IRUGO, proc_pid_statm),
3337 REG("maps", S_IRUGO, proc_pid_maps_operations),
3338#ifdef CONFIG_NUMA
3339 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3340#endif
3341 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3342 LNK("cwd", proc_cwd_link),
3343 LNK("root", proc_root_link),
3344 LNK("exe", proc_exe_link),
3345 REG("mounts", S_IRUGO, proc_mounts_operations),
3346 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3347 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3348#ifdef CONFIG_PROC_PAGE_MONITOR
3349 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3350 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3351 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3352 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3353#endif
3354#ifdef CONFIG_SECURITY
3355 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3356#endif
3357#ifdef CONFIG_KALLSYMS
3358 ONE("wchan", S_IRUGO, proc_pid_wchan),
3359#endif
3360#ifdef CONFIG_STACKTRACE
3361 ONE("stack", S_IRUSR, proc_pid_stack),
3362#endif
3363#ifdef CONFIG_SCHED_INFO
3364 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3365#endif
3366#ifdef CONFIG_LATENCYTOP
3367 REG("latency", S_IRUGO, proc_lstats_operations),
3368#endif
3369#ifdef CONFIG_PROC_PID_CPUSET
3370 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3371#endif
3372#ifdef CONFIG_CGROUPS
3373 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3374#endif
3375#ifdef CONFIG_PROC_CPU_RESCTRL
3376 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3377#endif
3378 ONE("oom_score", S_IRUGO, proc_oom_score),
3379 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3380 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3381#ifdef CONFIG_AUDIT
3382 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3383 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3384#endif
3385#ifdef CONFIG_FAULT_INJECTION
3386 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3387 REG("fail-nth", 0644, proc_fail_nth_operations),
3388#endif
3389#ifdef CONFIG_ELF_CORE
3390 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3391#endif
3392#ifdef CONFIG_TASK_IO_ACCOUNTING
3393 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3394#endif
3395#ifdef CONFIG_USER_NS
3396 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3397 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3398 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3399 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3400#endif
3401#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3402 REG("timers", S_IRUGO, proc_timers_operations),
3403#endif
3404 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3405#ifdef CONFIG_LIVEPATCH
3406 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3407#endif
3408#ifdef CONFIG_STACKLEAK_METRICS
3409 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3410#endif
3411#ifdef CONFIG_PROC_PID_ARCH_STATUS
3412 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3413#endif
3414#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3415 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3416#endif
3417#ifdef CONFIG_KSM
3418 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3419 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3420#endif
3421};
3422
3423static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3424{
3425 return proc_pident_readdir(file, ctx,
3426 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3427}
3428
3429static const struct file_operations proc_tgid_base_operations = {
3430 .read = generic_read_dir,
3431 .iterate_shared = proc_tgid_base_readdir,
3432 .llseek = generic_file_llseek,
3433};
3434
3435struct pid *tgid_pidfd_to_pid(const struct file *file)
3436{
3437 if (file->f_op != &proc_tgid_base_operations)
3438 return ERR_PTR(-EBADF);
3439
3440 return proc_pid(file_inode(file));
3441}
3442
3443static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3444{
3445 return proc_pident_lookup(dir, dentry,
3446 tgid_base_stuff,
3447 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3448}
3449
3450static const struct inode_operations proc_tgid_base_inode_operations = {
3451 .lookup = proc_tgid_base_lookup,
3452 .getattr = pid_getattr,
3453 .setattr = proc_setattr,
3454 .permission = proc_pid_permission,
3455};
3456
3457/**
3458 * proc_flush_pid - Remove dcache entries for @pid from the /proc dcache.
3459 * @pid: pid that should be flushed.
3460 *
3461 * This function walks a list of inodes (that belong to any proc
3462 * filesystem) that are attached to the pid and flushes them from
3463 * the dentry cache.
3464 *
3465 * It is safe and reasonable to cache /proc entries for a task until
3466 * that task exits. After that they just clog up the dcache with
3467 * useless entries, possibly causing useful dcache entries to be
3468 * flushed instead. This routine is provided to flush those useless
3469 * dcache entries when a process is reaped.
3470 *
3471 * NOTE: This routine is just an optimization so it does not guarantee
3472 * that no dcache entries will exist after a process is reaped
3473 * it just makes it very unlikely that any will persist.
3474 */
3475
3476void proc_flush_pid(struct pid *pid)
3477{
3478 proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3479}
3480
3481static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3482 struct task_struct *task, const void *ptr)
3483{
3484 struct inode *inode;
3485
3486 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3487 S_IFDIR | S_IRUGO | S_IXUGO);
3488 if (!inode)
3489 return ERR_PTR(-ENOENT);
3490
3491 inode->i_op = &proc_tgid_base_inode_operations;
3492 inode->i_fop = &proc_tgid_base_operations;
3493 inode->i_flags|=S_IMMUTABLE;
3494
3495 set_nlink(inode, nlink_tgid);
3496 pid_update_inode(task, inode);
3497
3498 d_set_d_op(dentry, &pid_dentry_operations);
3499 return d_splice_alias(inode, dentry);
3500}
3501
3502struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3503{
3504 struct task_struct *task;
3505 unsigned tgid;
3506 struct proc_fs_info *fs_info;
3507 struct pid_namespace *ns;
3508 struct dentry *result = ERR_PTR(-ENOENT);
3509
3510 tgid = name_to_int(&dentry->d_name);
3511 if (tgid == ~0U)
3512 goto out;
3513
3514 fs_info = proc_sb_info(dentry->d_sb);
3515 ns = fs_info->pid_ns;
3516 rcu_read_lock();
3517 task = find_task_by_pid_ns(tgid, ns);
3518 if (task)
3519 get_task_struct(task);
3520 rcu_read_unlock();
3521 if (!task)
3522 goto out;
3523
3524 /* Limit procfs to only ptraceable tasks */
3525 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3526 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3527 goto out_put_task;
3528 }
3529
3530 result = proc_pid_instantiate(dentry, task, NULL);
3531out_put_task:
3532 put_task_struct(task);
3533out:
3534 return result;
3535}
3536
3537/*
3538 * Find the first task with tgid >= tgid
3539 *
3540 */
3541struct tgid_iter {
3542 unsigned int tgid;
3543 struct task_struct *task;
3544};
3545static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3546{
3547 struct pid *pid;
3548
3549 if (iter.task)
3550 put_task_struct(iter.task);
3551 rcu_read_lock();
3552retry:
3553 iter.task = NULL;
3554 pid = find_ge_pid(iter.tgid, ns);
3555 if (pid) {
3556 iter.tgid = pid_nr_ns(pid, ns);
3557 iter.task = pid_task(pid, PIDTYPE_TGID);
3558 if (!iter.task) {
3559 iter.tgid += 1;
3560 goto retry;
3561 }
3562 get_task_struct(iter.task);
3563 }
3564 rcu_read_unlock();
3565 return iter;
3566}
3567
3568#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3569
3570/* for the /proc/ directory itself, after non-process stuff has been done */
3571int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3572{
3573 struct tgid_iter iter;
3574 struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3575 struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3576 loff_t pos = ctx->pos;
3577
3578 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3579 return 0;
3580
3581 if (pos == TGID_OFFSET - 2) {
3582 struct inode *inode = d_inode(fs_info->proc_self);
3583 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3584 return 0;
3585 ctx->pos = pos = pos + 1;
3586 }
3587 if (pos == TGID_OFFSET - 1) {
3588 struct inode *inode = d_inode(fs_info->proc_thread_self);
3589 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3590 return 0;
3591 ctx->pos = pos = pos + 1;
3592 }
3593 iter.tgid = pos - TGID_OFFSET;
3594 iter.task = NULL;
3595 for (iter = next_tgid(ns, iter);
3596 iter.task;
3597 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3598 char name[10 + 1];
3599 unsigned int len;
3600
3601 cond_resched();
3602 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3603 continue;
3604
3605 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3606 ctx->pos = iter.tgid + TGID_OFFSET;
3607 if (!proc_fill_cache(file, ctx, name, len,
3608 proc_pid_instantiate, iter.task, NULL)) {
3609 put_task_struct(iter.task);
3610 return 0;
3611 }
3612 }
3613 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3614 return 0;
3615}
3616
3617/*
3618 * proc_tid_comm_permission is a special permission function exclusively
3619 * used for the node /proc/<pid>/task/<tid>/comm.
3620 * It bypasses generic permission checks in the case where a task of the same
3621 * task group attempts to access the node.
3622 * The rationale behind this is that glibc and bionic access this node for
3623 * cross thread naming (pthread_set/getname_np(!self)). However, if
3624 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3625 * which locks out the cross thread naming implementation.
3626 * This function makes sure that the node is always accessible for members of
3627 * same thread group.
3628 */
3629static int proc_tid_comm_permission(struct mnt_idmap *idmap,
3630 struct inode *inode, int mask)
3631{
3632 bool is_same_tgroup;
3633 struct task_struct *task;
3634
3635 task = get_proc_task(inode);
3636 if (!task)
3637 return -ESRCH;
3638 is_same_tgroup = same_thread_group(current, task);
3639 put_task_struct(task);
3640
3641 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3642 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3643 * read or written by the members of the corresponding
3644 * thread group.
3645 */
3646 return 0;
3647 }
3648
3649 return generic_permission(&nop_mnt_idmap, inode, mask);
3650}
3651
3652static const struct inode_operations proc_tid_comm_inode_operations = {
3653 .setattr = proc_setattr,
3654 .permission = proc_tid_comm_permission,
3655};
3656
3657/*
3658 * Tasks
3659 */
3660static const struct pid_entry tid_base_stuff[] = {
3661 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3662 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3663 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3664#ifdef CONFIG_NET
3665 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3666#endif
3667 REG("environ", S_IRUSR, proc_environ_operations),
3668 REG("auxv", S_IRUSR, proc_auxv_operations),
3669 ONE("status", S_IRUGO, proc_pid_status),
3670 ONE("personality", S_IRUSR, proc_pid_personality),
3671 ONE("limits", S_IRUGO, proc_pid_limits),
3672#ifdef CONFIG_SCHED_DEBUG
3673 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3674#endif
3675 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3676 &proc_tid_comm_inode_operations,
3677 &proc_pid_set_comm_operations, {}),
3678#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3679 ONE("syscall", S_IRUSR, proc_pid_syscall),
3680#endif
3681 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3682 ONE("stat", S_IRUGO, proc_tid_stat),
3683 ONE("statm", S_IRUGO, proc_pid_statm),
3684 REG("maps", S_IRUGO, proc_pid_maps_operations),
3685#ifdef CONFIG_PROC_CHILDREN
3686 REG("children", S_IRUGO, proc_tid_children_operations),
3687#endif
3688#ifdef CONFIG_NUMA
3689 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3690#endif
3691 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3692 LNK("cwd", proc_cwd_link),
3693 LNK("root", proc_root_link),
3694 LNK("exe", proc_exe_link),
3695 REG("mounts", S_IRUGO, proc_mounts_operations),
3696 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3697#ifdef CONFIG_PROC_PAGE_MONITOR
3698 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3699 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3700 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3701 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3702#endif
3703#ifdef CONFIG_SECURITY
3704 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3705#endif
3706#ifdef CONFIG_KALLSYMS
3707 ONE("wchan", S_IRUGO, proc_pid_wchan),
3708#endif
3709#ifdef CONFIG_STACKTRACE
3710 ONE("stack", S_IRUSR, proc_pid_stack),
3711#endif
3712#ifdef CONFIG_SCHED_INFO
3713 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3714#endif
3715#ifdef CONFIG_LATENCYTOP
3716 REG("latency", S_IRUGO, proc_lstats_operations),
3717#endif
3718#ifdef CONFIG_PROC_PID_CPUSET
3719 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3720#endif
3721#ifdef CONFIG_CGROUPS
3722 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3723#endif
3724#ifdef CONFIG_PROC_CPU_RESCTRL
3725 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3726#endif
3727 ONE("oom_score", S_IRUGO, proc_oom_score),
3728 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3729 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3730#ifdef CONFIG_AUDIT
3731 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3732 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3733#endif
3734#ifdef CONFIG_FAULT_INJECTION
3735 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3736 REG("fail-nth", 0644, proc_fail_nth_operations),
3737#endif
3738#ifdef CONFIG_TASK_IO_ACCOUNTING
3739 ONE("io", S_IRUSR, proc_tid_io_accounting),
3740#endif
3741#ifdef CONFIG_USER_NS
3742 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3743 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3744 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3745 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3746#endif
3747#ifdef CONFIG_LIVEPATCH
3748 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3749#endif
3750#ifdef CONFIG_PROC_PID_ARCH_STATUS
3751 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3752#endif
3753#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3754 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3755#endif
3756#ifdef CONFIG_KSM
3757 ONE("ksm_merging_pages", S_IRUSR, proc_pid_ksm_merging_pages),
3758 ONE("ksm_stat", S_IRUSR, proc_pid_ksm_stat),
3759#endif
3760};
3761
3762static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3763{
3764 return proc_pident_readdir(file, ctx,
3765 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3766}
3767
3768static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3769{
3770 return proc_pident_lookup(dir, dentry,
3771 tid_base_stuff,
3772 tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3773}
3774
3775static const struct file_operations proc_tid_base_operations = {
3776 .read = generic_read_dir,
3777 .iterate_shared = proc_tid_base_readdir,
3778 .llseek = generic_file_llseek,
3779};
3780
3781static const struct inode_operations proc_tid_base_inode_operations = {
3782 .lookup = proc_tid_base_lookup,
3783 .getattr = pid_getattr,
3784 .setattr = proc_setattr,
3785};
3786
3787static struct dentry *proc_task_instantiate(struct dentry *dentry,
3788 struct task_struct *task, const void *ptr)
3789{
3790 struct inode *inode;
3791 inode = proc_pid_make_base_inode(dentry->d_sb, task,
3792 S_IFDIR | S_IRUGO | S_IXUGO);
3793 if (!inode)
3794 return ERR_PTR(-ENOENT);
3795
3796 inode->i_op = &proc_tid_base_inode_operations;
3797 inode->i_fop = &proc_tid_base_operations;
3798 inode->i_flags |= S_IMMUTABLE;
3799
3800 set_nlink(inode, nlink_tid);
3801 pid_update_inode(task, inode);
3802
3803 d_set_d_op(dentry, &pid_dentry_operations);
3804 return d_splice_alias(inode, dentry);
3805}
3806
3807static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3808{
3809 struct task_struct *task;
3810 struct task_struct *leader = get_proc_task(dir);
3811 unsigned tid;
3812 struct proc_fs_info *fs_info;
3813 struct pid_namespace *ns;
3814 struct dentry *result = ERR_PTR(-ENOENT);
3815
3816 if (!leader)
3817 goto out_no_task;
3818
3819 tid = name_to_int(&dentry->d_name);
3820 if (tid == ~0U)
3821 goto out;
3822
3823 fs_info = proc_sb_info(dentry->d_sb);
3824 ns = fs_info->pid_ns;
3825 rcu_read_lock();
3826 task = find_task_by_pid_ns(tid, ns);
3827 if (task)
3828 get_task_struct(task);
3829 rcu_read_unlock();
3830 if (!task)
3831 goto out;
3832 if (!same_thread_group(leader, task))
3833 goto out_drop_task;
3834
3835 result = proc_task_instantiate(dentry, task, NULL);
3836out_drop_task:
3837 put_task_struct(task);
3838out:
3839 put_task_struct(leader);
3840out_no_task:
3841 return result;
3842}
3843
3844/*
3845 * Find the first tid of a thread group to return to user space.
3846 *
3847 * Usually this is just the thread group leader, but if the users
3848 * buffer was too small or there was a seek into the middle of the
3849 * directory we have more work todo.
3850 *
3851 * In the case of a short read we start with find_task_by_pid.
3852 *
3853 * In the case of a seek we start with the leader and walk nr
3854 * threads past it.
3855 */
3856static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3857 struct pid_namespace *ns)
3858{
3859 struct task_struct *pos, *task;
3860 unsigned long nr = f_pos;
3861
3862 if (nr != f_pos) /* 32bit overflow? */
3863 return NULL;
3864
3865 rcu_read_lock();
3866 task = pid_task(pid, PIDTYPE_PID);
3867 if (!task)
3868 goto fail;
3869
3870 /* Attempt to start with the tid of a thread */
3871 if (tid && nr) {
3872 pos = find_task_by_pid_ns(tid, ns);
3873 if (pos && same_thread_group(pos, task))
3874 goto found;
3875 }
3876
3877 /* If nr exceeds the number of threads there is nothing todo */
3878 if (nr >= get_nr_threads(task))
3879 goto fail;
3880
3881 /* If we haven't found our starting place yet start
3882 * with the leader and walk nr threads forward.
3883 */
3884 for_each_thread(task, pos) {
3885 if (!nr--)
3886 goto found;
3887 }
3888fail:
3889 pos = NULL;
3890 goto out;
3891found:
3892 get_task_struct(pos);
3893out:
3894 rcu_read_unlock();
3895 return pos;
3896}
3897
3898/*
3899 * Find the next thread in the thread list.
3900 * Return NULL if there is an error or no next thread.
3901 *
3902 * The reference to the input task_struct is released.
3903 */
3904static struct task_struct *next_tid(struct task_struct *start)
3905{
3906 struct task_struct *pos = NULL;
3907 rcu_read_lock();
3908 if (pid_alive(start)) {
3909 pos = __next_thread(start);
3910 if (pos)
3911 get_task_struct(pos);
3912 }
3913 rcu_read_unlock();
3914 put_task_struct(start);
3915 return pos;
3916}
3917
3918/* for the /proc/TGID/task/ directories */
3919static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3920{
3921 struct inode *inode = file_inode(file);
3922 struct task_struct *task;
3923 struct pid_namespace *ns;
3924 int tid;
3925
3926 if (proc_inode_is_dead(inode))
3927 return -ENOENT;
3928
3929 if (!dir_emit_dots(file, ctx))
3930 return 0;
3931
3932 /* We cache the tgid value that the last readdir call couldn't
3933 * return and lseek resets it to 0.
3934 */
3935 ns = proc_pid_ns(inode->i_sb);
3936 tid = (int)(intptr_t)file->private_data;
3937 file->private_data = NULL;
3938 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3939 task;
3940 task = next_tid(task), ctx->pos++) {
3941 char name[10 + 1];
3942 unsigned int len;
3943
3944 tid = task_pid_nr_ns(task, ns);
3945 if (!tid)
3946 continue; /* The task has just exited. */
3947 len = snprintf(name, sizeof(name), "%u", tid);
3948 if (!proc_fill_cache(file, ctx, name, len,
3949 proc_task_instantiate, task, NULL)) {
3950 /* returning this tgid failed, save it as the first
3951 * pid for the next readir call */
3952 file->private_data = (void *)(intptr_t)tid;
3953 put_task_struct(task);
3954 break;
3955 }
3956 }
3957
3958 return 0;
3959}
3960
3961static int proc_task_getattr(struct mnt_idmap *idmap,
3962 const struct path *path, struct kstat *stat,
3963 u32 request_mask, unsigned int query_flags)
3964{
3965 struct inode *inode = d_inode(path->dentry);
3966 struct task_struct *p = get_proc_task(inode);
3967 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
3968
3969 if (p) {
3970 stat->nlink += get_nr_threads(p);
3971 put_task_struct(p);
3972 }
3973
3974 return 0;
3975}
3976
3977/*
3978 * proc_task_readdir() set @file->private_data to a positive integer
3979 * value, so casting that to u64 is safe. generic_llseek_cookie() will
3980 * set @cookie to 0, so casting to an int is safe. The WARN_ON_ONCE() is
3981 * here to catch any unexpected change in behavior either in
3982 * proc_task_readdir() or generic_llseek_cookie().
3983 */
3984static loff_t proc_dir_llseek(struct file *file, loff_t offset, int whence)
3985{
3986 u64 cookie = (u64)(intptr_t)file->private_data;
3987 loff_t off;
3988
3989 off = generic_llseek_cookie(file, offset, whence, &cookie);
3990 WARN_ON_ONCE(cookie > INT_MAX);
3991 file->private_data = (void *)(intptr_t)cookie; /* serialized by f_pos_lock */
3992 return off;
3993}
3994
3995static const struct inode_operations proc_task_inode_operations = {
3996 .lookup = proc_task_lookup,
3997 .getattr = proc_task_getattr,
3998 .setattr = proc_setattr,
3999 .permission = proc_pid_permission,
4000};
4001
4002static const struct file_operations proc_task_operations = {
4003 .read = generic_read_dir,
4004 .iterate_shared = proc_task_readdir,
4005 .llseek = proc_dir_llseek,
4006};
4007
4008void __init set_proc_pid_nlink(void)
4009{
4010 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
4011 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
4012}