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