projects / deliverable / linux.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
proc, oom_adj: extract oom_score_adj setting into a helper
[deliverable/linux.git] / fs / proc / base.c
... / ...
CommitLineData
1/*
2 * linux/fs/proc/base.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * proc base directory handling functions
7 *
8 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 * Instead of using magical inumbers to determine the kind of object
10 * we allocate and fill in-core inodes upon lookup. They don't even
11 * go into icache. We cache the reference to task_struct upon lookup too.
12 * Eventually it should become a filesystem in its own. We don't use the
13 * rest of procfs anymore.
14 *
15 *
16 * Changelog:
17 * 17-Jan-2005
18 * Allan Bezerra
19 * Bruna Moreira <bruna.moreira@indt.org.br>
20 * Edjard Mota <edjard.mota@indt.org.br>
21 * Ilias Biris <ilias.biris@indt.org.br>
22 * Mauricio Lin <mauricio.lin@indt.org.br>
23 *
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25 *
26 * A new process specific entry (smaps) included in /proc. It shows the
27 * size of rss for each memory area. The maps entry lacks information
28 * about physical memory size (rss) for each mapped file, i.e.,
29 * rss information for executables and library files.
30 * This additional information is useful for any tools that need to know
31 * about physical memory consumption for a process specific library.
32 *
33 * Changelog:
34 * 21-Feb-2005
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
37 *
38 * ChangeLog:
39 * 10-Mar-2005
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
42 *
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
45 *
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
48 */
49
50#include <asm/uaccess.h>
51
52#include <linux/errno.h>
53#include <linux/time.h>
54#include <linux/proc_fs.h>
55#include <linux/stat.h>
56#include <linux/task_io_accounting_ops.h>
57#include <linux/init.h>
58#include <linux/capability.h>
59#include <linux/file.h>
60#include <linux/fdtable.h>
61#include <linux/string.h>
62#include <linux/seq_file.h>
63#include <linux/namei.h>
64#include <linux/mnt_namespace.h>
65#include <linux/mm.h>
66#include <linux/swap.h>
67#include <linux/rcupdate.h>
68#include <linux/kallsyms.h>
69#include <linux/stacktrace.h>
70#include <linux/resource.h>
71#include <linux/module.h>
72#include <linux/mount.h>
73#include <linux/security.h>
74#include <linux/ptrace.h>
75#include <linux/tracehook.h>
76#include <linux/printk.h>
77#include <linux/cgroup.h>
78#include <linux/cpuset.h>
79#include <linux/audit.h>
80#include <linux/poll.h>
81#include <linux/nsproxy.h>
82#include <linux/oom.h>
83#include <linux/elf.h>
84#include <linux/pid_namespace.h>
85#include <linux/user_namespace.h>
86#include <linux/fs_struct.h>
87#include <linux/slab.h>
88#include <linux/flex_array.h>
89#include <linux/posix-timers.h>
90#ifdef CONFIG_HARDWALL
91#include <asm/hardwall.h>
92#endif
93#include <trace/events/oom.h>
94#include "internal.h"
95#include "fd.h"
96
97/* NOTE:
98 * Implementing inode permission operations in /proc is almost
99 * certainly an error. Permission checks need to happen during
100 * each system call not at open time. The reason is that most of
101 * what we wish to check for permissions in /proc varies at runtime.
102 *
103 * The classic example of a problem is opening file descriptors
104 * in /proc for a task before it execs a suid executable.
105 */
106
107struct pid_entry {
108 const char *name;
109 int len;
110 umode_t mode;
111 const struct inode_operations *iop;
112 const struct file_operations *fop;
113 union proc_op op;
114};
115
116#define NOD(NAME, MODE, IOP, FOP, OP) { \
117 .name = (NAME), \
118 .len = sizeof(NAME) - 1, \
119 .mode = MODE, \
120 .iop = IOP, \
121 .fop = FOP, \
122 .op = OP, \
123}
124
125#define DIR(NAME, MODE, iops, fops) \
126 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
127#define LNK(NAME, get_link) \
128 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
129 &proc_pid_link_inode_operations, NULL, \
130 { .proc_get_link = get_link } )
131#define REG(NAME, MODE, fops) \
132 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
133#define ONE(NAME, MODE, show) \
134 NOD(NAME, (S_IFREG|(MODE)), \
135 NULL, &proc_single_file_operations, \
136 { .proc_show = show } )
137
138/*
139 * Count the number of hardlinks for the pid_entry table, excluding the .
140 * and .. links.
141 */
142static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
143 unsigned int n)
144{
145 unsigned int i;
146 unsigned int count;
147
148 count = 0;
149 for (i = 0; i < n; ++i) {
150 if (S_ISDIR(entries[i].mode))
151 ++count;
152 }
153
154 return count;
155}
156
157static int get_task_root(struct task_struct *task, struct path *root)
158{
159 int result = -ENOENT;
160
161 task_lock(task);
162 if (task->fs) {
163 get_fs_root(task->fs, root);
164 result = 0;
165 }
166 task_unlock(task);
167 return result;
168}
169
170static int proc_cwd_link(struct dentry *dentry, struct path *path)
171{
172 struct task_struct *task = get_proc_task(d_inode(dentry));
173 int result = -ENOENT;
174
175 if (task) {
176 task_lock(task);
177 if (task->fs) {
178 get_fs_pwd(task->fs, path);
179 result = 0;
180 }
181 task_unlock(task);
182 put_task_struct(task);
183 }
184 return result;
185}
186
187static int proc_root_link(struct dentry *dentry, struct path *path)
188{
189 struct task_struct *task = get_proc_task(d_inode(dentry));
190 int result = -ENOENT;
191
192 if (task) {
193 result = get_task_root(task, path);
194 put_task_struct(task);
195 }
196 return result;
197}
198
199static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
200 size_t _count, loff_t *pos)
201{
202 struct task_struct *tsk;
203 struct mm_struct *mm;
204 char *page;
205 unsigned long count = _count;
206 unsigned long arg_start, arg_end, env_start, env_end;
207 unsigned long len1, len2, len;
208 unsigned long p;
209 char c;
210 ssize_t rv;
211
212 BUG_ON(*pos < 0);
213
214 tsk = get_proc_task(file_inode(file));
215 if (!tsk)
216 return -ESRCH;
217 mm = get_task_mm(tsk);
218 put_task_struct(tsk);
219 if (!mm)
220 return 0;
221 /* Check if process spawned far enough to have cmdline. */
222 if (!mm->env_end) {
223 rv = 0;
224 goto out_mmput;
225 }
226
227 page = (char *)__get_free_page(GFP_TEMPORARY);
228 if (!page) {
229 rv = -ENOMEM;
230 goto out_mmput;
231 }
232
233 down_read(&mm->mmap_sem);
234 arg_start = mm->arg_start;
235 arg_end = mm->arg_end;
236 env_start = mm->env_start;
237 env_end = mm->env_end;
238 up_read(&mm->mmap_sem);
239
240 BUG_ON(arg_start > arg_end);
241 BUG_ON(env_start > env_end);
242
243 len1 = arg_end - arg_start;
244 len2 = env_end - env_start;
245
246 /* Empty ARGV. */
247 if (len1 == 0) {
248 rv = 0;
249 goto out_free_page;
250 }
251 /*
252 * Inherently racy -- command line shares address space
253 * with code and data.
254 */
255 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
256 if (rv <= 0)
257 goto out_free_page;
258
259 rv = 0;
260
261 if (c == '\0') {
262 /* Command line (set of strings) occupies whole ARGV. */
263 if (len1 <= *pos)
264 goto out_free_page;
265
266 p = arg_start + *pos;
267 len = len1 - *pos;
268 while (count > 0 && len > 0) {
269 unsigned int _count;
270 int nr_read;
271
272 _count = min3(count, len, PAGE_SIZE);
273 nr_read = access_remote_vm(mm, p, page, _count, 0);
274 if (nr_read < 0)
275 rv = nr_read;
276 if (nr_read <= 0)
277 goto out_free_page;
278
279 if (copy_to_user(buf, page, nr_read)) {
280 rv = -EFAULT;
281 goto out_free_page;
282 }
283
284 p += nr_read;
285 len -= nr_read;
286 buf += nr_read;
287 count -= nr_read;
288 rv += nr_read;
289 }
290 } else {
291 /*
292 * Command line (1 string) occupies ARGV and maybe
293 * extends into ENVP.
294 */
295 if (len1 + len2 <= *pos)
296 goto skip_argv_envp;
297 if (len1 <= *pos)
298 goto skip_argv;
299
300 p = arg_start + *pos;
301 len = len1 - *pos;
302 while (count > 0 && len > 0) {
303 unsigned int _count, l;
304 int nr_read;
305 bool final;
306
307 _count = min3(count, len, PAGE_SIZE);
308 nr_read = access_remote_vm(mm, p, page, _count, 0);
309 if (nr_read < 0)
310 rv = nr_read;
311 if (nr_read <= 0)
312 goto out_free_page;
313
314 /*
315 * Command line can be shorter than whole ARGV
316 * even if last "marker" byte says it is not.
317 */
318 final = false;
319 l = strnlen(page, nr_read);
320 if (l < nr_read) {
321 nr_read = l;
322 final = true;
323 }
324
325 if (copy_to_user(buf, page, nr_read)) {
326 rv = -EFAULT;
327 goto out_free_page;
328 }
329
330 p += nr_read;
331 len -= nr_read;
332 buf += nr_read;
333 count -= nr_read;
334 rv += nr_read;
335
336 if (final)
337 goto out_free_page;
338 }
339skip_argv:
340 /*
341 * Command line (1 string) occupies ARGV and
342 * extends into ENVP.
343 */
344 if (len1 <= *pos) {
345 p = env_start + *pos - len1;
346 len = len1 + len2 - *pos;
347 } else {
348 p = env_start;
349 len = len2;
350 }
351 while (count > 0 && len > 0) {
352 unsigned int _count, l;
353 int nr_read;
354 bool final;
355
356 _count = min3(count, len, PAGE_SIZE);
357 nr_read = access_remote_vm(mm, p, page, _count, 0);
358 if (nr_read < 0)
359 rv = nr_read;
360 if (nr_read <= 0)
361 goto out_free_page;
362
363 /* Find EOS. */
364 final = false;
365 l = strnlen(page, nr_read);
366 if (l < nr_read) {
367 nr_read = l;
368 final = true;
369 }
370
371 if (copy_to_user(buf, page, nr_read)) {
372 rv = -EFAULT;
373 goto out_free_page;
374 }
375
376 p += nr_read;
377 len -= nr_read;
378 buf += nr_read;
379 count -= nr_read;
380 rv += nr_read;
381
382 if (final)
383 goto out_free_page;
384 }
385skip_argv_envp:
386 ;
387 }
388
389out_free_page:
390 free_page((unsigned long)page);
391out_mmput:
392 mmput(mm);
393 if (rv > 0)
394 *pos += rv;
395 return rv;
396}
397
398static const struct file_operations proc_pid_cmdline_ops = {
399 .read = proc_pid_cmdline_read,
400 .llseek = generic_file_llseek,
401};
402
403static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
404 struct pid *pid, struct task_struct *task)
405{
406 struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
407 if (mm && !IS_ERR(mm)) {
408 unsigned int nwords = 0;
409 do {
410 nwords += 2;
411 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
412 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
413 mmput(mm);
414 return 0;
415 } else
416 return PTR_ERR(mm);
417}
418
419
420#ifdef CONFIG_KALLSYMS
421/*
422 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
423 * Returns the resolved symbol. If that fails, simply return the address.
424 */
425static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
426 struct pid *pid, struct task_struct *task)
427{
428 unsigned long wchan;
429 char symname[KSYM_NAME_LEN];
430
431 wchan = get_wchan(task);
432
433 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
434 && !lookup_symbol_name(wchan, symname))
435 seq_printf(m, "%s", symname);
436 else
437 seq_putc(m, '0');
438
439 return 0;
440}
441#endif /* CONFIG_KALLSYMS */
442
443static int lock_trace(struct task_struct *task)
444{
445 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
446 if (err)
447 return err;
448 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
449 mutex_unlock(&task->signal->cred_guard_mutex);
450 return -EPERM;
451 }
452 return 0;
453}
454
455static void unlock_trace(struct task_struct *task)
456{
457 mutex_unlock(&task->signal->cred_guard_mutex);
458}
459
460#ifdef CONFIG_STACKTRACE
461
462#define MAX_STACK_TRACE_DEPTH 64
463
464static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
465 struct pid *pid, struct task_struct *task)
466{
467 struct stack_trace trace;
468 unsigned long *entries;
469 int err;
470 int i;
471
472 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
473 if (!entries)
474 return -ENOMEM;
475
476 trace.nr_entries = 0;
477 trace.max_entries = MAX_STACK_TRACE_DEPTH;
478 trace.entries = entries;
479 trace.skip = 0;
480
481 err = lock_trace(task);
482 if (!err) {
483 save_stack_trace_tsk(task, &trace);
484
485 for (i = 0; i < trace.nr_entries; i++) {
486 seq_printf(m, "[<%pK>] %pS\n",
487 (void *)entries[i], (void *)entries[i]);
488 }
489 unlock_trace(task);
490 }
491 kfree(entries);
492
493 return err;
494}
495#endif
496
497#ifdef CONFIG_SCHED_INFO
498/*
499 * Provides /proc/PID/schedstat
500 */
501static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
502 struct pid *pid, struct task_struct *task)
503{
504 if (unlikely(!sched_info_on()))
505 seq_printf(m, "0 0 0\n");
506 else
507 seq_printf(m, "%llu %llu %lu\n",
508 (unsigned long long)task->se.sum_exec_runtime,
509 (unsigned long long)task->sched_info.run_delay,
510 task->sched_info.pcount);
511
512 return 0;
513}
514#endif
515
516#ifdef CONFIG_LATENCYTOP
517static int lstats_show_proc(struct seq_file *m, void *v)
518{
519 int i;
520 struct inode *inode = m->private;
521 struct task_struct *task = get_proc_task(inode);
522
523 if (!task)
524 return -ESRCH;
525 seq_puts(m, "Latency Top version : v0.1\n");
526 for (i = 0; i < 32; i++) {
527 struct latency_record *lr = &task->latency_record[i];
528 if (lr->backtrace[0]) {
529 int q;
530 seq_printf(m, "%i %li %li",
531 lr->count, lr->time, lr->max);
532 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
533 unsigned long bt = lr->backtrace[q];
534 if (!bt)
535 break;
536 if (bt == ULONG_MAX)
537 break;
538 seq_printf(m, " %ps", (void *)bt);
539 }
540 seq_putc(m, '\n');
541 }
542
543 }
544 put_task_struct(task);
545 return 0;
546}
547
548static int lstats_open(struct inode *inode, struct file *file)
549{
550 return single_open(file, lstats_show_proc, inode);
551}
552
553static ssize_t lstats_write(struct file *file, const char __user *buf,
554 size_t count, loff_t *offs)
555{
556 struct task_struct *task = get_proc_task(file_inode(file));
557
558 if (!task)
559 return -ESRCH;
560 clear_all_latency_tracing(task);
561 put_task_struct(task);
562
563 return count;
564}
565
566static const struct file_operations proc_lstats_operations = {
567 .open = lstats_open,
568 .read = seq_read,
569 .write = lstats_write,
570 .llseek = seq_lseek,
571 .release = single_release,
572};
573
574#endif
575
576static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
577 struct pid *pid, struct task_struct *task)
578{
579 unsigned long totalpages = totalram_pages + total_swap_pages;
580 unsigned long points = 0;
581
582 read_lock(&tasklist_lock);
583 if (pid_alive(task))
584 points = oom_badness(task, NULL, NULL, totalpages) *
585 1000 / totalpages;
586 read_unlock(&tasklist_lock);
587 seq_printf(m, "%lu\n", points);
588
589 return 0;
590}
591
592struct limit_names {
593 const char *name;
594 const char *unit;
595};
596
597static const struct limit_names lnames[RLIM_NLIMITS] = {
598 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
599 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
600 [RLIMIT_DATA] = {"Max data size", "bytes"},
601 [RLIMIT_STACK] = {"Max stack size", "bytes"},
602 [RLIMIT_CORE] = {"Max core file size", "bytes"},
603 [RLIMIT_RSS] = {"Max resident set", "bytes"},
604 [RLIMIT_NPROC] = {"Max processes", "processes"},
605 [RLIMIT_NOFILE] = {"Max open files", "files"},
606 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
607 [RLIMIT_AS] = {"Max address space", "bytes"},
608 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
609 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
610 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
611 [RLIMIT_NICE] = {"Max nice priority", NULL},
612 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
613 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
614};
615
616/* Display limits for a process */
617static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
618 struct pid *pid, struct task_struct *task)
619{
620 unsigned int i;
621 unsigned long flags;
622
623 struct rlimit rlim[RLIM_NLIMITS];
624
625 if (!lock_task_sighand(task, &flags))
626 return 0;
627 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
628 unlock_task_sighand(task, &flags);
629
630 /*
631 * print the file header
632 */
633 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
634 "Limit", "Soft Limit", "Hard Limit", "Units");
635
636 for (i = 0; i < RLIM_NLIMITS; i++) {
637 if (rlim[i].rlim_cur == RLIM_INFINITY)
638 seq_printf(m, "%-25s %-20s ",
639 lnames[i].name, "unlimited");
640 else
641 seq_printf(m, "%-25s %-20lu ",
642 lnames[i].name, rlim[i].rlim_cur);
643
644 if (rlim[i].rlim_max == RLIM_INFINITY)
645 seq_printf(m, "%-20s ", "unlimited");
646 else
647 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
648
649 if (lnames[i].unit)
650 seq_printf(m, "%-10s\n", lnames[i].unit);
651 else
652 seq_putc(m, '\n');
653 }
654
655 return 0;
656}
657
658#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
659static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
660 struct pid *pid, struct task_struct *task)
661{
662 long nr;
663 unsigned long args[6], sp, pc;
664 int res;
665
666 res = lock_trace(task);
667 if (res)
668 return res;
669
670 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
671 seq_puts(m, "running\n");
672 else if (nr < 0)
673 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
674 else
675 seq_printf(m,
676 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
677 nr,
678 args[0], args[1], args[2], args[3], args[4], args[5],
679 sp, pc);
680 unlock_trace(task);
681
682 return 0;
683}
684#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
685
686/************************************************************************/
687/* Here the fs part begins */
688/************************************************************************/
689
690/* permission checks */
691static int proc_fd_access_allowed(struct inode *inode)
692{
693 struct task_struct *task;
694 int allowed = 0;
695 /* Allow access to a task's file descriptors if it is us or we
696 * may use ptrace attach to the process and find out that
697 * information.
698 */
699 task = get_proc_task(inode);
700 if (task) {
701 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
702 put_task_struct(task);
703 }
704 return allowed;
705}
706
707int proc_setattr(struct dentry *dentry, struct iattr *attr)
708{
709 int error;
710 struct inode *inode = d_inode(dentry);
711
712 if (attr->ia_valid & ATTR_MODE)
713 return -EPERM;
714
715 error = inode_change_ok(inode, attr);
716 if (error)
717 return error;
718
719 setattr_copy(inode, attr);
720 mark_inode_dirty(inode);
721 return 0;
722}
723
724/*
725 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
726 * or euid/egid (for hide_pid_min=2)?
727 */
728static bool has_pid_permissions(struct pid_namespace *pid,
729 struct task_struct *task,
730 int hide_pid_min)
731{
732 if (pid->hide_pid < hide_pid_min)
733 return true;
734 if (in_group_p(pid->pid_gid))
735 return true;
736 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
737}
738
739
740static int proc_pid_permission(struct inode *inode, int mask)
741{
742 struct pid_namespace *pid = inode->i_sb->s_fs_info;
743 struct task_struct *task;
744 bool has_perms;
745
746 task = get_proc_task(inode);
747 if (!task)
748 return -ESRCH;
749 has_perms = has_pid_permissions(pid, task, 1);
750 put_task_struct(task);
751
752 if (!has_perms) {
753 if (pid->hide_pid == 2) {
754 /*
755 * Let's make getdents(), stat(), and open()
756 * consistent with each other. If a process
757 * may not stat() a file, it shouldn't be seen
758 * in procfs at all.
759 */
760 return -ENOENT;
761 }
762
763 return -EPERM;
764 }
765 return generic_permission(inode, mask);
766}
767
768
769
770static const struct inode_operations proc_def_inode_operations = {
771 .setattr = proc_setattr,
772};
773
774static int proc_single_show(struct seq_file *m, void *v)
775{
776 struct inode *inode = m->private;
777 struct pid_namespace *ns;
778 struct pid *pid;
779 struct task_struct *task;
780 int ret;
781
782 ns = inode->i_sb->s_fs_info;
783 pid = proc_pid(inode);
784 task = get_pid_task(pid, PIDTYPE_PID);
785 if (!task)
786 return -ESRCH;
787
788 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
789
790 put_task_struct(task);
791 return ret;
792}
793
794static int proc_single_open(struct inode *inode, struct file *filp)
795{
796 return single_open(filp, proc_single_show, inode);
797}
798
799static const struct file_operations proc_single_file_operations = {
800 .open = proc_single_open,
801 .read = seq_read,
802 .llseek = seq_lseek,
803 .release = single_release,
804};
805
806
807struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
808{
809 struct task_struct *task = get_proc_task(inode);
810 struct mm_struct *mm = ERR_PTR(-ESRCH);
811
812 if (task) {
813 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
814 put_task_struct(task);
815
816 if (!IS_ERR_OR_NULL(mm)) {
817 /* ensure this mm_struct can't be freed */
818 atomic_inc(&mm->mm_count);
819 /* but do not pin its memory */
820 mmput(mm);
821 }
822 }
823
824 return mm;
825}
826
827static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
828{
829 struct mm_struct *mm = proc_mem_open(inode, mode);
830
831 if (IS_ERR(mm))
832 return PTR_ERR(mm);
833
834 file->private_data = mm;
835 return 0;
836}
837
838static int mem_open(struct inode *inode, struct file *file)
839{
840 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
841
842 /* OK to pass negative loff_t, we can catch out-of-range */
843 file->f_mode |= FMODE_UNSIGNED_OFFSET;
844
845 return ret;
846}
847
848static ssize_t mem_rw(struct file *file, char __user *buf,
849 size_t count, loff_t *ppos, int write)
850{
851 struct mm_struct *mm = file->private_data;
852 unsigned long addr = *ppos;
853 ssize_t copied;
854 char *page;
855
856 if (!mm)
857 return 0;
858
859 page = (char *)__get_free_page(GFP_TEMPORARY);
860 if (!page)
861 return -ENOMEM;
862
863 copied = 0;
864 if (!atomic_inc_not_zero(&mm->mm_users))
865 goto free;
866
867 while (count > 0) {
868 int this_len = min_t(int, count, PAGE_SIZE);
869
870 if (write && copy_from_user(page, buf, this_len)) {
871 copied = -EFAULT;
872 break;
873 }
874
875 this_len = access_remote_vm(mm, addr, page, this_len, write);
876 if (!this_len) {
877 if (!copied)
878 copied = -EIO;
879 break;
880 }
881
882 if (!write && copy_to_user(buf, page, this_len)) {
883 copied = -EFAULT;
884 break;
885 }
886
887 buf += this_len;
888 addr += this_len;
889 copied += this_len;
890 count -= this_len;
891 }
892 *ppos = addr;
893
894 mmput(mm);
895free:
896 free_page((unsigned long) page);
897 return copied;
898}
899
900static ssize_t mem_read(struct file *file, char __user *buf,
901 size_t count, loff_t *ppos)
902{
903 return mem_rw(file, buf, count, ppos, 0);
904}
905
906static ssize_t mem_write(struct file *file, const char __user *buf,
907 size_t count, loff_t *ppos)
908{
909 return mem_rw(file, (char __user*)buf, count, ppos, 1);
910}
911
912loff_t mem_lseek(struct file *file, loff_t offset, int orig)
913{
914 switch (orig) {
915 case 0:
916 file->f_pos = offset;
917 break;
918 case 1:
919 file->f_pos += offset;
920 break;
921 default:
922 return -EINVAL;
923 }
924 force_successful_syscall_return();
925 return file->f_pos;
926}
927
928static int mem_release(struct inode *inode, struct file *file)
929{
930 struct mm_struct *mm = file->private_data;
931 if (mm)
932 mmdrop(mm);
933 return 0;
934}
935
936static const struct file_operations proc_mem_operations = {
937 .llseek = mem_lseek,
938 .read = mem_read,
939 .write = mem_write,
940 .open = mem_open,
941 .release = mem_release,
942};
943
944static int environ_open(struct inode *inode, struct file *file)
945{
946 return __mem_open(inode, file, PTRACE_MODE_READ);
947}
948
949static ssize_t environ_read(struct file *file, char __user *buf,
950 size_t count, loff_t *ppos)
951{
952 char *page;
953 unsigned long src = *ppos;
954 int ret = 0;
955 struct mm_struct *mm = file->private_data;
956 unsigned long env_start, env_end;
957
958 /* Ensure the process spawned far enough to have an environment. */
959 if (!mm || !mm->env_end)
960 return 0;
961
962 page = (char *)__get_free_page(GFP_TEMPORARY);
963 if (!page)
964 return -ENOMEM;
965
966 ret = 0;
967 if (!atomic_inc_not_zero(&mm->mm_users))
968 goto free;
969
970 down_read(&mm->mmap_sem);
971 env_start = mm->env_start;
972 env_end = mm->env_end;
973 up_read(&mm->mmap_sem);
974
975 while (count > 0) {
976 size_t this_len, max_len;
977 int retval;
978
979 if (src >= (env_end - env_start))
980 break;
981
982 this_len = env_end - (env_start + src);
983
984 max_len = min_t(size_t, PAGE_SIZE, count);
985 this_len = min(max_len, this_len);
986
987 retval = access_remote_vm(mm, (env_start + src),
988 page, this_len, 0);
989
990 if (retval <= 0) {
991 ret = retval;
992 break;
993 }
994
995 if (copy_to_user(buf, page, retval)) {
996 ret = -EFAULT;
997 break;
998 }
999
1000 ret += retval;
1001 src += retval;
1002 buf += retval;
1003 count -= retval;
1004 }
1005 *ppos = src;
1006 mmput(mm);
1007
1008free:
1009 free_page((unsigned long) page);
1010 return ret;
1011}
1012
1013static const struct file_operations proc_environ_operations = {
1014 .open = environ_open,
1015 .read = environ_read,
1016 .llseek = generic_file_llseek,
1017 .release = mem_release,
1018};
1019
1020static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1021 loff_t *ppos)
1022{
1023 struct task_struct *task = get_proc_task(file_inode(file));
1024 char buffer[PROC_NUMBUF];
1025 int oom_adj = OOM_ADJUST_MIN;
1026 size_t len;
1027
1028 if (!task)
1029 return -ESRCH;
1030 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1031 oom_adj = OOM_ADJUST_MAX;
1032 else
1033 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1034 OOM_SCORE_ADJ_MAX;
1035 put_task_struct(task);
1036 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1037 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1038}
1039
1040static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1041{
1042 static DEFINE_MUTEX(oom_adj_mutex);
1043 struct task_struct *task;
1044 int err = 0;
1045
1046 task = get_proc_task(file_inode(file));
1047 if (!task)
1048 return -ESRCH;
1049
1050 mutex_lock(&oom_adj_mutex);
1051 if (legacy) {
1052 if (oom_adj < task->signal->oom_score_adj &&
1053 !capable(CAP_SYS_RESOURCE)) {
1054 err = -EACCES;
1055 goto err_unlock;
1056 }
1057 /*
1058 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1059 * /proc/pid/oom_score_adj instead.
1060 */
1061 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1062 current->comm, task_pid_nr(current), task_pid_nr(task),
1063 task_pid_nr(task));
1064 } else {
1065 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1066 !capable(CAP_SYS_RESOURCE)) {
1067 err = -EACCES;
1068 goto err_unlock;
1069 }
1070 }
1071
1072 task->signal->oom_score_adj = oom_adj;
1073 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1074 task->signal->oom_score_adj_min = (short)oom_adj;
1075 trace_oom_score_adj_update(task);
1076err_unlock:
1077 mutex_unlock(&oom_adj_mutex);
1078 put_task_struct(task);
1079 return err;
1080}
1081
1082/*
1083 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1084 * kernels. The effective policy is defined by oom_score_adj, which has a
1085 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1086 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1087 * Processes that become oom disabled via oom_adj will still be oom disabled
1088 * with this implementation.
1089 *
1090 * oom_adj cannot be removed since existing userspace binaries use it.
1091 */
1092static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1093 size_t count, loff_t *ppos)
1094{
1095 char buffer[PROC_NUMBUF];
1096 int oom_adj;
1097 int err;
1098
1099 memset(buffer, 0, sizeof(buffer));
1100 if (count > sizeof(buffer) - 1)
1101 count = sizeof(buffer) - 1;
1102 if (copy_from_user(buffer, buf, count)) {
1103 err = -EFAULT;
1104 goto out;
1105 }
1106
1107 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1108 if (err)
1109 goto out;
1110 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1111 oom_adj != OOM_DISABLE) {
1112 err = -EINVAL;
1113 goto out;
1114 }
1115
1116 /*
1117 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1118 * value is always attainable.
1119 */
1120 if (oom_adj == OOM_ADJUST_MAX)
1121 oom_adj = OOM_SCORE_ADJ_MAX;
1122 else
1123 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1124
1125 err = __set_oom_adj(file, oom_adj, true);
1126out:
1127 return err < 0 ? err : count;
1128}
1129
1130static const struct file_operations proc_oom_adj_operations = {
1131 .read = oom_adj_read,
1132 .write = oom_adj_write,
1133 .llseek = generic_file_llseek,
1134};
1135
1136static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1137 size_t count, loff_t *ppos)
1138{
1139 struct task_struct *task = get_proc_task(file_inode(file));
1140 char buffer[PROC_NUMBUF];
1141 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1142 size_t len;
1143
1144 if (!task)
1145 return -ESRCH;
1146 oom_score_adj = task->signal->oom_score_adj;
1147 put_task_struct(task);
1148 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1149 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1150}
1151
1152static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1153 size_t count, loff_t *ppos)
1154{
1155 char buffer[PROC_NUMBUF];
1156 int oom_score_adj;
1157 int err;
1158
1159 memset(buffer, 0, sizeof(buffer));
1160 if (count > sizeof(buffer) - 1)
1161 count = sizeof(buffer) - 1;
1162 if (copy_from_user(buffer, buf, count)) {
1163 err = -EFAULT;
1164 goto out;
1165 }
1166
1167 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1168 if (err)
1169 goto out;
1170 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1171 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1172 err = -EINVAL;
1173 goto out;
1174 }
1175
1176 err = __set_oom_adj(file, oom_score_adj, false);
1177out:
1178 return err < 0 ? err : count;
1179}
1180
1181static const struct file_operations proc_oom_score_adj_operations = {
1182 .read = oom_score_adj_read,
1183 .write = oom_score_adj_write,
1184 .llseek = default_llseek,
1185};
1186
1187#ifdef CONFIG_AUDITSYSCALL
1188#define TMPBUFLEN 21
1189static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1190 size_t count, loff_t *ppos)
1191{
1192 struct inode * inode = file_inode(file);
1193 struct task_struct *task = get_proc_task(inode);
1194 ssize_t length;
1195 char tmpbuf[TMPBUFLEN];
1196
1197 if (!task)
1198 return -ESRCH;
1199 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1200 from_kuid(file->f_cred->user_ns,
1201 audit_get_loginuid(task)));
1202 put_task_struct(task);
1203 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1204}
1205
1206static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1207 size_t count, loff_t *ppos)
1208{
1209 struct inode * inode = file_inode(file);
1210 uid_t loginuid;
1211 kuid_t kloginuid;
1212 int rv;
1213
1214 rcu_read_lock();
1215 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1216 rcu_read_unlock();
1217 return -EPERM;
1218 }
1219 rcu_read_unlock();
1220
1221 if (*ppos != 0) {
1222 /* No partial writes. */
1223 return -EINVAL;
1224 }
1225
1226 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1227 if (rv < 0)
1228 return rv;
1229
1230 /* is userspace tring to explicitly UNSET the loginuid? */
1231 if (loginuid == AUDIT_UID_UNSET) {
1232 kloginuid = INVALID_UID;
1233 } else {
1234 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1235 if (!uid_valid(kloginuid))
1236 return -EINVAL;
1237 }
1238
1239 rv = audit_set_loginuid(kloginuid);
1240 if (rv < 0)
1241 return rv;
1242 return count;
1243}
1244
1245static const struct file_operations proc_loginuid_operations = {
1246 .read = proc_loginuid_read,
1247 .write = proc_loginuid_write,
1248 .llseek = generic_file_llseek,
1249};
1250
1251static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1252 size_t count, loff_t *ppos)
1253{
1254 struct inode * inode = file_inode(file);
1255 struct task_struct *task = get_proc_task(inode);
1256 ssize_t length;
1257 char tmpbuf[TMPBUFLEN];
1258
1259 if (!task)
1260 return -ESRCH;
1261 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1262 audit_get_sessionid(task));
1263 put_task_struct(task);
1264 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1265}
1266
1267static const struct file_operations proc_sessionid_operations = {
1268 .read = proc_sessionid_read,
1269 .llseek = generic_file_llseek,
1270};
1271#endif
1272
1273#ifdef CONFIG_FAULT_INJECTION
1274static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1275 size_t count, loff_t *ppos)
1276{
1277 struct task_struct *task = get_proc_task(file_inode(file));
1278 char buffer[PROC_NUMBUF];
1279 size_t len;
1280 int make_it_fail;
1281
1282 if (!task)
1283 return -ESRCH;
1284 make_it_fail = task->make_it_fail;
1285 put_task_struct(task);
1286
1287 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1288
1289 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1290}
1291
1292static ssize_t proc_fault_inject_write(struct file * file,
1293 const char __user * buf, size_t count, loff_t *ppos)
1294{
1295 struct task_struct *task;
1296 char buffer[PROC_NUMBUF];
1297 int make_it_fail;
1298 int rv;
1299
1300 if (!capable(CAP_SYS_RESOURCE))
1301 return -EPERM;
1302 memset(buffer, 0, sizeof(buffer));
1303 if (count > sizeof(buffer) - 1)
1304 count = sizeof(buffer) - 1;
1305 if (copy_from_user(buffer, buf, count))
1306 return -EFAULT;
1307 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1308 if (rv < 0)
1309 return rv;
1310 if (make_it_fail < 0 || make_it_fail > 1)
1311 return -EINVAL;
1312
1313 task = get_proc_task(file_inode(file));
1314 if (!task)
1315 return -ESRCH;
1316 task->make_it_fail = make_it_fail;
1317 put_task_struct(task);
1318
1319 return count;
1320}
1321
1322static const struct file_operations proc_fault_inject_operations = {
1323 .read = proc_fault_inject_read,
1324 .write = proc_fault_inject_write,
1325 .llseek = generic_file_llseek,
1326};
1327#endif
1328
1329
1330#ifdef CONFIG_SCHED_DEBUG
1331/*
1332 * Print out various scheduling related per-task fields:
1333 */
1334static int sched_show(struct seq_file *m, void *v)
1335{
1336 struct inode *inode = m->private;
1337 struct task_struct *p;
1338
1339 p = get_proc_task(inode);
1340 if (!p)
1341 return -ESRCH;
1342 proc_sched_show_task(p, m);
1343
1344 put_task_struct(p);
1345
1346 return 0;
1347}
1348
1349static ssize_t
1350sched_write(struct file *file, const char __user *buf,
1351 size_t count, loff_t *offset)
1352{
1353 struct inode *inode = file_inode(file);
1354 struct task_struct *p;
1355
1356 p = get_proc_task(inode);
1357 if (!p)
1358 return -ESRCH;
1359 proc_sched_set_task(p);
1360
1361 put_task_struct(p);
1362
1363 return count;
1364}
1365
1366static int sched_open(struct inode *inode, struct file *filp)
1367{
1368 return single_open(filp, sched_show, inode);
1369}
1370
1371static const struct file_operations proc_pid_sched_operations = {
1372 .open = sched_open,
1373 .read = seq_read,
1374 .write = sched_write,
1375 .llseek = seq_lseek,
1376 .release = single_release,
1377};
1378
1379#endif
1380
1381#ifdef CONFIG_SCHED_AUTOGROUP
1382/*
1383 * Print out autogroup related information:
1384 */
1385static int sched_autogroup_show(struct seq_file *m, void *v)
1386{
1387 struct inode *inode = m->private;
1388 struct task_struct *p;
1389
1390 p = get_proc_task(inode);
1391 if (!p)
1392 return -ESRCH;
1393 proc_sched_autogroup_show_task(p, m);
1394
1395 put_task_struct(p);
1396
1397 return 0;
1398}
1399
1400static ssize_t
1401sched_autogroup_write(struct file *file, const char __user *buf,
1402 size_t count, loff_t *offset)
1403{
1404 struct inode *inode = file_inode(file);
1405 struct task_struct *p;
1406 char buffer[PROC_NUMBUF];
1407 int nice;
1408 int err;
1409
1410 memset(buffer, 0, sizeof(buffer));
1411 if (count > sizeof(buffer) - 1)
1412 count = sizeof(buffer) - 1;
1413 if (copy_from_user(buffer, buf, count))
1414 return -EFAULT;
1415
1416 err = kstrtoint(strstrip(buffer), 0, &nice);
1417 if (err < 0)
1418 return err;
1419
1420 p = get_proc_task(inode);
1421 if (!p)
1422 return -ESRCH;
1423
1424 err = proc_sched_autogroup_set_nice(p, nice);
1425 if (err)
1426 count = err;
1427
1428 put_task_struct(p);
1429
1430 return count;
1431}
1432
1433static int sched_autogroup_open(struct inode *inode, struct file *filp)
1434{
1435 int ret;
1436
1437 ret = single_open(filp, sched_autogroup_show, NULL);
1438 if (!ret) {
1439 struct seq_file *m = filp->private_data;
1440
1441 m->private = inode;
1442 }
1443 return ret;
1444}
1445
1446static const struct file_operations proc_pid_sched_autogroup_operations = {
1447 .open = sched_autogroup_open,
1448 .read = seq_read,
1449 .write = sched_autogroup_write,
1450 .llseek = seq_lseek,
1451 .release = single_release,
1452};
1453
1454#endif /* CONFIG_SCHED_AUTOGROUP */
1455
1456static ssize_t comm_write(struct file *file, const char __user *buf,
1457 size_t count, loff_t *offset)
1458{
1459 struct inode *inode = file_inode(file);
1460 struct task_struct *p;
1461 char buffer[TASK_COMM_LEN];
1462 const size_t maxlen = sizeof(buffer) - 1;
1463
1464 memset(buffer, 0, sizeof(buffer));
1465 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1466 return -EFAULT;
1467
1468 p = get_proc_task(inode);
1469 if (!p)
1470 return -ESRCH;
1471
1472 if (same_thread_group(current, p))
1473 set_task_comm(p, buffer);
1474 else
1475 count = -EINVAL;
1476
1477 put_task_struct(p);
1478
1479 return count;
1480}
1481
1482static int comm_show(struct seq_file *m, void *v)
1483{
1484 struct inode *inode = m->private;
1485 struct task_struct *p;
1486
1487 p = get_proc_task(inode);
1488 if (!p)
1489 return -ESRCH;
1490
1491 task_lock(p);
1492 seq_printf(m, "%s\n", p->comm);
1493 task_unlock(p);
1494
1495 put_task_struct(p);
1496
1497 return 0;
1498}
1499
1500static int comm_open(struct inode *inode, struct file *filp)
1501{
1502 return single_open(filp, comm_show, inode);
1503}
1504
1505static const struct file_operations proc_pid_set_comm_operations = {
1506 .open = comm_open,
1507 .read = seq_read,
1508 .write = comm_write,
1509 .llseek = seq_lseek,
1510 .release = single_release,
1511};
1512
1513static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1514{
1515 struct task_struct *task;
1516 struct mm_struct *mm;
1517 struct file *exe_file;
1518
1519 task = get_proc_task(d_inode(dentry));
1520 if (!task)
1521 return -ENOENT;
1522 mm = get_task_mm(task);
1523 put_task_struct(task);
1524 if (!mm)
1525 return -ENOENT;
1526 exe_file = get_mm_exe_file(mm);
1527 mmput(mm);
1528 if (exe_file) {
1529 *exe_path = exe_file->f_path;
1530 path_get(&exe_file->f_path);
1531 fput(exe_file);
1532 return 0;
1533 } else
1534 return -ENOENT;
1535}
1536
1537static const char *proc_pid_get_link(struct dentry *dentry,
1538 struct inode *inode,
1539 struct delayed_call *done)
1540{
1541 struct path path;
1542 int error = -EACCES;
1543
1544 if (!dentry)
1545 return ERR_PTR(-ECHILD);
1546
1547 /* Are we allowed to snoop on the tasks file descriptors? */
1548 if (!proc_fd_access_allowed(inode))
1549 goto out;
1550
1551 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1552 if (error)
1553 goto out;
1554
1555 nd_jump_link(&path);
1556 return NULL;
1557out:
1558 return ERR_PTR(error);
1559}
1560
1561static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1562{
1563 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1564 char *pathname;
1565 int len;
1566
1567 if (!tmp)
1568 return -ENOMEM;
1569
1570 pathname = d_path(path, tmp, PAGE_SIZE);
1571 len = PTR_ERR(pathname);
1572 if (IS_ERR(pathname))
1573 goto out;
1574 len = tmp + PAGE_SIZE - 1 - pathname;
1575
1576 if (len > buflen)
1577 len = buflen;
1578 if (copy_to_user(buffer, pathname, len))
1579 len = -EFAULT;
1580 out:
1581 free_page((unsigned long)tmp);
1582 return len;
1583}
1584
1585static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1586{
1587 int error = -EACCES;
1588 struct inode *inode = d_inode(dentry);
1589 struct path path;
1590
1591 /* Are we allowed to snoop on the tasks file descriptors? */
1592 if (!proc_fd_access_allowed(inode))
1593 goto out;
1594
1595 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1596 if (error)
1597 goto out;
1598
1599 error = do_proc_readlink(&path, buffer, buflen);
1600 path_put(&path);
1601out:
1602 return error;
1603}
1604
1605const struct inode_operations proc_pid_link_inode_operations = {
1606 .readlink = proc_pid_readlink,
1607 .get_link = proc_pid_get_link,
1608 .setattr = proc_setattr,
1609};
1610
1611
1612/* building an inode */
1613
1614struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1615{
1616 struct inode * inode;
1617 struct proc_inode *ei;
1618 const struct cred *cred;
1619
1620 /* We need a new inode */
1621
1622 inode = new_inode(sb);
1623 if (!inode)
1624 goto out;
1625
1626 /* Common stuff */
1627 ei = PROC_I(inode);
1628 inode->i_ino = get_next_ino();
1629 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1630 inode->i_op = &proc_def_inode_operations;
1631
1632 /*
1633 * grab the reference to task.
1634 */
1635 ei->pid = get_task_pid(task, PIDTYPE_PID);
1636 if (!ei->pid)
1637 goto out_unlock;
1638
1639 if (task_dumpable(task)) {
1640 rcu_read_lock();
1641 cred = __task_cred(task);
1642 inode->i_uid = cred->euid;
1643 inode->i_gid = cred->egid;
1644 rcu_read_unlock();
1645 }
1646 security_task_to_inode(task, inode);
1647
1648out:
1649 return inode;
1650
1651out_unlock:
1652 iput(inode);
1653 return NULL;
1654}
1655
1656int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1657{
1658 struct inode *inode = d_inode(dentry);
1659 struct task_struct *task;
1660 const struct cred *cred;
1661 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1662
1663 generic_fillattr(inode, stat);
1664
1665 rcu_read_lock();
1666 stat->uid = GLOBAL_ROOT_UID;
1667 stat->gid = GLOBAL_ROOT_GID;
1668 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1669 if (task) {
1670 if (!has_pid_permissions(pid, task, 2)) {
1671 rcu_read_unlock();
1672 /*
1673 * This doesn't prevent learning whether PID exists,
1674 * it only makes getattr() consistent with readdir().
1675 */
1676 return -ENOENT;
1677 }
1678 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1679 task_dumpable(task)) {
1680 cred = __task_cred(task);
1681 stat->uid = cred->euid;
1682 stat->gid = cred->egid;
1683 }
1684 }
1685 rcu_read_unlock();
1686 return 0;
1687}
1688
1689/* dentry stuff */
1690
1691/*
1692 * Exceptional case: normally we are not allowed to unhash a busy
1693 * directory. In this case, however, we can do it - no aliasing problems
1694 * due to the way we treat inodes.
1695 *
1696 * Rewrite the inode's ownerships here because the owning task may have
1697 * performed a setuid(), etc.
1698 *
1699 * Before the /proc/pid/status file was created the only way to read
1700 * the effective uid of a /process was to stat /proc/pid. Reading
1701 * /proc/pid/status is slow enough that procps and other packages
1702 * kept stating /proc/pid. To keep the rules in /proc simple I have
1703 * made this apply to all per process world readable and executable
1704 * directories.
1705 */
1706int pid_revalidate(struct dentry *dentry, unsigned int flags)
1707{
1708 struct inode *inode;
1709 struct task_struct *task;
1710 const struct cred *cred;
1711
1712 if (flags & LOOKUP_RCU)
1713 return -ECHILD;
1714
1715 inode = d_inode(dentry);
1716 task = get_proc_task(inode);
1717
1718 if (task) {
1719 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1720 task_dumpable(task)) {
1721 rcu_read_lock();
1722 cred = __task_cred(task);
1723 inode->i_uid = cred->euid;
1724 inode->i_gid = cred->egid;
1725 rcu_read_unlock();
1726 } else {
1727 inode->i_uid = GLOBAL_ROOT_UID;
1728 inode->i_gid = GLOBAL_ROOT_GID;
1729 }
1730 inode->i_mode &= ~(S_ISUID | S_ISGID);
1731 security_task_to_inode(task, inode);
1732 put_task_struct(task);
1733 return 1;
1734 }
1735 return 0;
1736}
1737
1738static inline bool proc_inode_is_dead(struct inode *inode)
1739{
1740 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1741}
1742
1743int pid_delete_dentry(const struct dentry *dentry)
1744{
1745 /* Is the task we represent dead?
1746 * If so, then don't put the dentry on the lru list,
1747 * kill it immediately.
1748 */
1749 return proc_inode_is_dead(d_inode(dentry));
1750}
1751
1752const struct dentry_operations pid_dentry_operations =
1753{
1754 .d_revalidate = pid_revalidate,
1755 .d_delete = pid_delete_dentry,
1756};
1757
1758/* Lookups */
1759
1760/*
1761 * Fill a directory entry.
1762 *
1763 * If possible create the dcache entry and derive our inode number and
1764 * file type from dcache entry.
1765 *
1766 * Since all of the proc inode numbers are dynamically generated, the inode
1767 * numbers do not exist until the inode is cache. This means creating the
1768 * the dcache entry in readdir is necessary to keep the inode numbers
1769 * reported by readdir in sync with the inode numbers reported
1770 * by stat.
1771 */
1772bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1773 const char *name, int len,
1774 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1775{
1776 struct dentry *child, *dir = file->f_path.dentry;
1777 struct qstr qname = QSTR_INIT(name, len);
1778 struct inode *inode;
1779 unsigned type;
1780 ino_t ino;
1781
1782 child = d_hash_and_lookup(dir, &qname);
1783 if (!child) {
1784 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1785 child = d_alloc_parallel(dir, &qname, &wq);
1786 if (IS_ERR(child))
1787 goto end_instantiate;
1788 if (d_in_lookup(child)) {
1789 int err = instantiate(d_inode(dir), child, task, ptr);
1790 d_lookup_done(child);
1791 if (err < 0) {
1792 dput(child);
1793 goto end_instantiate;
1794 }
1795 }
1796 }
1797 inode = d_inode(child);
1798 ino = inode->i_ino;
1799 type = inode->i_mode >> 12;
1800 dput(child);
1801 return dir_emit(ctx, name, len, ino, type);
1802
1803end_instantiate:
1804 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1805}
1806
1807/*
1808 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1809 * which represent vma start and end addresses.
1810 */
1811static int dname_to_vma_addr(struct dentry *dentry,
1812 unsigned long *start, unsigned long *end)
1813{
1814 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1815 return -EINVAL;
1816
1817 return 0;
1818}
1819
1820static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1821{
1822 unsigned long vm_start, vm_end;
1823 bool exact_vma_exists = false;
1824 struct mm_struct *mm = NULL;
1825 struct task_struct *task;
1826 const struct cred *cred;
1827 struct inode *inode;
1828 int status = 0;
1829
1830 if (flags & LOOKUP_RCU)
1831 return -ECHILD;
1832
1833 inode = d_inode(dentry);
1834 task = get_proc_task(inode);
1835 if (!task)
1836 goto out_notask;
1837
1838 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1839 if (IS_ERR_OR_NULL(mm))
1840 goto out;
1841
1842 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1843 down_read(&mm->mmap_sem);
1844 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1845 up_read(&mm->mmap_sem);
1846 }
1847
1848 mmput(mm);
1849
1850 if (exact_vma_exists) {
1851 if (task_dumpable(task)) {
1852 rcu_read_lock();
1853 cred = __task_cred(task);
1854 inode->i_uid = cred->euid;
1855 inode->i_gid = cred->egid;
1856 rcu_read_unlock();
1857 } else {
1858 inode->i_uid = GLOBAL_ROOT_UID;
1859 inode->i_gid = GLOBAL_ROOT_GID;
1860 }
1861 security_task_to_inode(task, inode);
1862 status = 1;
1863 }
1864
1865out:
1866 put_task_struct(task);
1867
1868out_notask:
1869 return status;
1870}
1871
1872static const struct dentry_operations tid_map_files_dentry_operations = {
1873 .d_revalidate = map_files_d_revalidate,
1874 .d_delete = pid_delete_dentry,
1875};
1876
1877static int map_files_get_link(struct dentry *dentry, struct path *path)
1878{
1879 unsigned long vm_start, vm_end;
1880 struct vm_area_struct *vma;
1881 struct task_struct *task;
1882 struct mm_struct *mm;
1883 int rc;
1884
1885 rc = -ENOENT;
1886 task = get_proc_task(d_inode(dentry));
1887 if (!task)
1888 goto out;
1889
1890 mm = get_task_mm(task);
1891 put_task_struct(task);
1892 if (!mm)
1893 goto out;
1894
1895 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1896 if (rc)
1897 goto out_mmput;
1898
1899 rc = -ENOENT;
1900 down_read(&mm->mmap_sem);
1901 vma = find_exact_vma(mm, vm_start, vm_end);
1902 if (vma && vma->vm_file) {
1903 *path = vma->vm_file->f_path;
1904 path_get(path);
1905 rc = 0;
1906 }
1907 up_read(&mm->mmap_sem);
1908
1909out_mmput:
1910 mmput(mm);
1911out:
1912 return rc;
1913}
1914
1915struct map_files_info {
1916 fmode_t mode;
1917 unsigned long len;
1918 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1919};
1920
1921/*
1922 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1923 * symlinks may be used to bypass permissions on ancestor directories in the
1924 * path to the file in question.
1925 */
1926static const char *
1927proc_map_files_get_link(struct dentry *dentry,
1928 struct inode *inode,
1929 struct delayed_call *done)
1930{
1931 if (!capable(CAP_SYS_ADMIN))
1932 return ERR_PTR(-EPERM);
1933
1934 return proc_pid_get_link(dentry, inode, done);
1935}
1936
1937/*
1938 * Identical to proc_pid_link_inode_operations except for get_link()
1939 */
1940static const struct inode_operations proc_map_files_link_inode_operations = {
1941 .readlink = proc_pid_readlink,
1942 .get_link = proc_map_files_get_link,
1943 .setattr = proc_setattr,
1944};
1945
1946static int
1947proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1948 struct task_struct *task, const void *ptr)
1949{
1950 fmode_t mode = (fmode_t)(unsigned long)ptr;
1951 struct proc_inode *ei;
1952 struct inode *inode;
1953
1954 inode = proc_pid_make_inode(dir->i_sb, task);
1955 if (!inode)
1956 return -ENOENT;
1957
1958 ei = PROC_I(inode);
1959 ei->op.proc_get_link = map_files_get_link;
1960
1961 inode->i_op = &proc_map_files_link_inode_operations;
1962 inode->i_size = 64;
1963 inode->i_mode = S_IFLNK;
1964
1965 if (mode & FMODE_READ)
1966 inode->i_mode |= S_IRUSR;
1967 if (mode & FMODE_WRITE)
1968 inode->i_mode |= S_IWUSR;
1969
1970 d_set_d_op(dentry, &tid_map_files_dentry_operations);
1971 d_add(dentry, inode);
1972
1973 return 0;
1974}
1975
1976static struct dentry *proc_map_files_lookup(struct inode *dir,
1977 struct dentry *dentry, unsigned int flags)
1978{
1979 unsigned long vm_start, vm_end;
1980 struct vm_area_struct *vma;
1981 struct task_struct *task;
1982 int result;
1983 struct mm_struct *mm;
1984
1985 result = -ENOENT;
1986 task = get_proc_task(dir);
1987 if (!task)
1988 goto out;
1989
1990 result = -EACCES;
1991 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
1992 goto out_put_task;
1993
1994 result = -ENOENT;
1995 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1996 goto out_put_task;
1997
1998 mm = get_task_mm(task);
1999 if (!mm)
2000 goto out_put_task;
2001
2002 down_read(&mm->mmap_sem);
2003 vma = find_exact_vma(mm, vm_start, vm_end);
2004 if (!vma)
2005 goto out_no_vma;
2006
2007 if (vma->vm_file)
2008 result = proc_map_files_instantiate(dir, dentry, task,
2009 (void *)(unsigned long)vma->vm_file->f_mode);
2010
2011out_no_vma:
2012 up_read(&mm->mmap_sem);
2013 mmput(mm);
2014out_put_task:
2015 put_task_struct(task);
2016out:
2017 return ERR_PTR(result);
2018}
2019
2020static const struct inode_operations proc_map_files_inode_operations = {
2021 .lookup = proc_map_files_lookup,
2022 .permission = proc_fd_permission,
2023 .setattr = proc_setattr,
2024};
2025
2026static int
2027proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2028{
2029 struct vm_area_struct *vma;
2030 struct task_struct *task;
2031 struct mm_struct *mm;
2032 unsigned long nr_files, pos, i;
2033 struct flex_array *fa = NULL;
2034 struct map_files_info info;
2035 struct map_files_info *p;
2036 int ret;
2037
2038 ret = -ENOENT;
2039 task = get_proc_task(file_inode(file));
2040 if (!task)
2041 goto out;
2042
2043 ret = -EACCES;
2044 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2045 goto out_put_task;
2046
2047 ret = 0;
2048 if (!dir_emit_dots(file, ctx))
2049 goto out_put_task;
2050
2051 mm = get_task_mm(task);
2052 if (!mm)
2053 goto out_put_task;
2054 down_read(&mm->mmap_sem);
2055
2056 nr_files = 0;
2057
2058 /*
2059 * We need two passes here:
2060 *
2061 * 1) Collect vmas of mapped files with mmap_sem taken
2062 * 2) Release mmap_sem and instantiate entries
2063 *
2064 * otherwise we get lockdep complained, since filldir()
2065 * routine might require mmap_sem taken in might_fault().
2066 */
2067
2068 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2069 if (vma->vm_file && ++pos > ctx->pos)
2070 nr_files++;
2071 }
2072
2073 if (nr_files) {
2074 fa = flex_array_alloc(sizeof(info), nr_files,
2075 GFP_KERNEL);
2076 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2077 GFP_KERNEL)) {
2078 ret = -ENOMEM;
2079 if (fa)
2080 flex_array_free(fa);
2081 up_read(&mm->mmap_sem);
2082 mmput(mm);
2083 goto out_put_task;
2084 }
2085 for (i = 0, vma = mm->mmap, pos = 2; vma;
2086 vma = vma->vm_next) {
2087 if (!vma->vm_file)
2088 continue;
2089 if (++pos <= ctx->pos)
2090 continue;
2091
2092 info.mode = vma->vm_file->f_mode;
2093 info.len = snprintf(info.name,
2094 sizeof(info.name), "%lx-%lx",
2095 vma->vm_start, vma->vm_end);
2096 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2097 BUG();
2098 }
2099 }
2100 up_read(&mm->mmap_sem);
2101
2102 for (i = 0; i < nr_files; i++) {
2103 p = flex_array_get(fa, i);
2104 if (!proc_fill_cache(file, ctx,
2105 p->name, p->len,
2106 proc_map_files_instantiate,
2107 task,
2108 (void *)(unsigned long)p->mode))
2109 break;
2110 ctx->pos++;
2111 }
2112 if (fa)
2113 flex_array_free(fa);
2114 mmput(mm);
2115
2116out_put_task:
2117 put_task_struct(task);
2118out:
2119 return ret;
2120}
2121
2122static const struct file_operations proc_map_files_operations = {
2123 .read = generic_read_dir,
2124 .iterate_shared = proc_map_files_readdir,
2125 .llseek = generic_file_llseek,
2126};
2127
2128#ifdef CONFIG_CHECKPOINT_RESTORE
2129struct timers_private {
2130 struct pid *pid;
2131 struct task_struct *task;
2132 struct sighand_struct *sighand;
2133 struct pid_namespace *ns;
2134 unsigned long flags;
2135};
2136
2137static void *timers_start(struct seq_file *m, loff_t *pos)
2138{
2139 struct timers_private *tp = m->private;
2140
2141 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2142 if (!tp->task)
2143 return ERR_PTR(-ESRCH);
2144
2145 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2146 if (!tp->sighand)
2147 return ERR_PTR(-ESRCH);
2148
2149 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2150}
2151
2152static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2153{
2154 struct timers_private *tp = m->private;
2155 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2156}
2157
2158static void timers_stop(struct seq_file *m, void *v)
2159{
2160 struct timers_private *tp = m->private;
2161
2162 if (tp->sighand) {
2163 unlock_task_sighand(tp->task, &tp->flags);
2164 tp->sighand = NULL;
2165 }
2166
2167 if (tp->task) {
2168 put_task_struct(tp->task);
2169 tp->task = NULL;
2170 }
2171}
2172
2173static int show_timer(struct seq_file *m, void *v)
2174{
2175 struct k_itimer *timer;
2176 struct timers_private *tp = m->private;
2177 int notify;
2178 static const char * const nstr[] = {
2179 [SIGEV_SIGNAL] = "signal",
2180 [SIGEV_NONE] = "none",
2181 [SIGEV_THREAD] = "thread",
2182 };
2183
2184 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2185 notify = timer->it_sigev_notify;
2186
2187 seq_printf(m, "ID: %d\n", timer->it_id);
2188 seq_printf(m, "signal: %d/%p\n",
2189 timer->sigq->info.si_signo,
2190 timer->sigq->info.si_value.sival_ptr);
2191 seq_printf(m, "notify: %s/%s.%d\n",
2192 nstr[notify & ~SIGEV_THREAD_ID],
2193 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2194 pid_nr_ns(timer->it_pid, tp->ns));
2195 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2196
2197 return 0;
2198}
2199
2200static const struct seq_operations proc_timers_seq_ops = {
2201 .start = timers_start,
2202 .next = timers_next,
2203 .stop = timers_stop,
2204 .show = show_timer,
2205};
2206
2207static int proc_timers_open(struct inode *inode, struct file *file)
2208{
2209 struct timers_private *tp;
2210
2211 tp = __seq_open_private(file, &proc_timers_seq_ops,
2212 sizeof(struct timers_private));
2213 if (!tp)
2214 return -ENOMEM;
2215
2216 tp->pid = proc_pid(inode);
2217 tp->ns = inode->i_sb->s_fs_info;
2218 return 0;
2219}
2220
2221static const struct file_operations proc_timers_operations = {
2222 .open = proc_timers_open,
2223 .read = seq_read,
2224 .llseek = seq_lseek,
2225 .release = seq_release_private,
2226};
2227#endif
2228
2229static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2230 size_t count, loff_t *offset)
2231{
2232 struct inode *inode = file_inode(file);
2233 struct task_struct *p;
2234 u64 slack_ns;
2235 int err;
2236
2237 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2238 if (err < 0)
2239 return err;
2240
2241 p = get_proc_task(inode);
2242 if (!p)
2243 return -ESRCH;
2244
2245 if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2246 task_lock(p);
2247 if (slack_ns == 0)
2248 p->timer_slack_ns = p->default_timer_slack_ns;
2249 else
2250 p->timer_slack_ns = slack_ns;
2251 task_unlock(p);
2252 } else
2253 count = -EPERM;
2254
2255 put_task_struct(p);
2256
2257 return count;
2258}
2259
2260static int timerslack_ns_show(struct seq_file *m, void *v)
2261{
2262 struct inode *inode = m->private;
2263 struct task_struct *p;
2264 int err = 0;
2265
2266 p = get_proc_task(inode);
2267 if (!p)
2268 return -ESRCH;
2269
2270 if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2271 task_lock(p);
2272 seq_printf(m, "%llu\n", p->timer_slack_ns);
2273 task_unlock(p);
2274 } else
2275 err = -EPERM;
2276
2277 put_task_struct(p);
2278
2279 return err;
2280}
2281
2282static int timerslack_ns_open(struct inode *inode, struct file *filp)
2283{
2284 return single_open(filp, timerslack_ns_show, inode);
2285}
2286
2287static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2288 .open = timerslack_ns_open,
2289 .read = seq_read,
2290 .write = timerslack_ns_write,
2291 .llseek = seq_lseek,
2292 .release = single_release,
2293};
2294
2295static int proc_pident_instantiate(struct inode *dir,
2296 struct dentry *dentry, struct task_struct *task, const void *ptr)
2297{
2298 const struct pid_entry *p = ptr;
2299 struct inode *inode;
2300 struct proc_inode *ei;
2301
2302 inode = proc_pid_make_inode(dir->i_sb, task);
2303 if (!inode)
2304 goto out;
2305
2306 ei = PROC_I(inode);
2307 inode->i_mode = p->mode;
2308 if (S_ISDIR(inode->i_mode))
2309 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2310 if (p->iop)
2311 inode->i_op = p->iop;
2312 if (p->fop)
2313 inode->i_fop = p->fop;
2314 ei->op = p->op;
2315 d_set_d_op(dentry, &pid_dentry_operations);
2316 d_add(dentry, inode);
2317 /* Close the race of the process dying before we return the dentry */
2318 if (pid_revalidate(dentry, 0))
2319 return 0;
2320out:
2321 return -ENOENT;
2322}
2323
2324static struct dentry *proc_pident_lookup(struct inode *dir,
2325 struct dentry *dentry,
2326 const struct pid_entry *ents,
2327 unsigned int nents)
2328{
2329 int error;
2330 struct task_struct *task = get_proc_task(dir);
2331 const struct pid_entry *p, *last;
2332
2333 error = -ENOENT;
2334
2335 if (!task)
2336 goto out_no_task;
2337
2338 /*
2339 * Yes, it does not scale. And it should not. Don't add
2340 * new entries into /proc/<tgid>/ without very good reasons.
2341 */
2342 last = &ents[nents - 1];
2343 for (p = ents; p <= last; p++) {
2344 if (p->len != dentry->d_name.len)
2345 continue;
2346 if (!memcmp(dentry->d_name.name, p->name, p->len))
2347 break;
2348 }
2349 if (p > last)
2350 goto out;
2351
2352 error = proc_pident_instantiate(dir, dentry, task, p);
2353out:
2354 put_task_struct(task);
2355out_no_task:
2356 return ERR_PTR(error);
2357}
2358
2359static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2360 const struct pid_entry *ents, unsigned int nents)
2361{
2362 struct task_struct *task = get_proc_task(file_inode(file));
2363 const struct pid_entry *p;
2364
2365 if (!task)
2366 return -ENOENT;
2367
2368 if (!dir_emit_dots(file, ctx))
2369 goto out;
2370
2371 if (ctx->pos >= nents + 2)
2372 goto out;
2373
2374 for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2375 if (!proc_fill_cache(file, ctx, p->name, p->len,
2376 proc_pident_instantiate, task, p))
2377 break;
2378 ctx->pos++;
2379 }
2380out:
2381 put_task_struct(task);
2382 return 0;
2383}
2384
2385#ifdef CONFIG_SECURITY
2386static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2387 size_t count, loff_t *ppos)
2388{
2389 struct inode * inode = file_inode(file);
2390 char *p = NULL;
2391 ssize_t length;
2392 struct task_struct *task = get_proc_task(inode);
2393
2394 if (!task)
2395 return -ESRCH;
2396
2397 length = security_getprocattr(task,
2398 (char*)file->f_path.dentry->d_name.name,
2399 &p);
2400 put_task_struct(task);
2401 if (length > 0)
2402 length = simple_read_from_buffer(buf, count, ppos, p, length);
2403 kfree(p);
2404 return length;
2405}
2406
2407static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2408 size_t count, loff_t *ppos)
2409{
2410 struct inode * inode = file_inode(file);
2411 void *page;
2412 ssize_t length;
2413 struct task_struct *task = get_proc_task(inode);
2414
2415 length = -ESRCH;
2416 if (!task)
2417 goto out_no_task;
2418 if (count > PAGE_SIZE)
2419 count = PAGE_SIZE;
2420
2421 /* No partial writes. */
2422 length = -EINVAL;
2423 if (*ppos != 0)
2424 goto out;
2425
2426 page = memdup_user(buf, count);
2427 if (IS_ERR(page)) {
2428 length = PTR_ERR(page);
2429 goto out;
2430 }
2431
2432 /* Guard against adverse ptrace interaction */
2433 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2434 if (length < 0)
2435 goto out_free;
2436
2437 length = security_setprocattr(task,
2438 (char*)file->f_path.dentry->d_name.name,
2439 page, count);
2440 mutex_unlock(&task->signal->cred_guard_mutex);
2441out_free:
2442 kfree(page);
2443out:
2444 put_task_struct(task);
2445out_no_task:
2446 return length;
2447}
2448
2449static const struct file_operations proc_pid_attr_operations = {
2450 .read = proc_pid_attr_read,
2451 .write = proc_pid_attr_write,
2452 .llseek = generic_file_llseek,
2453};
2454
2455static const struct pid_entry attr_dir_stuff[] = {
2456 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2457 REG("prev", S_IRUGO, proc_pid_attr_operations),
2458 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2459 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2460 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2461 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2462};
2463
2464static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2465{
2466 return proc_pident_readdir(file, ctx,
2467 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2468}
2469
2470static const struct file_operations proc_attr_dir_operations = {
2471 .read = generic_read_dir,
2472 .iterate_shared = proc_attr_dir_readdir,
2473 .llseek = generic_file_llseek,
2474};
2475
2476static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2477 struct dentry *dentry, unsigned int flags)
2478{
2479 return proc_pident_lookup(dir, dentry,
2480 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2481}
2482
2483static const struct inode_operations proc_attr_dir_inode_operations = {
2484 .lookup = proc_attr_dir_lookup,
2485 .getattr = pid_getattr,
2486 .setattr = proc_setattr,
2487};
2488
2489#endif
2490
2491#ifdef CONFIG_ELF_CORE
2492static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2493 size_t count, loff_t *ppos)
2494{
2495 struct task_struct *task = get_proc_task(file_inode(file));
2496 struct mm_struct *mm;
2497 char buffer[PROC_NUMBUF];
2498 size_t len;
2499 int ret;
2500
2501 if (!task)
2502 return -ESRCH;
2503
2504 ret = 0;
2505 mm = get_task_mm(task);
2506 if (mm) {
2507 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2508 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2509 MMF_DUMP_FILTER_SHIFT));
2510 mmput(mm);
2511 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2512 }
2513
2514 put_task_struct(task);
2515
2516 return ret;
2517}
2518
2519static ssize_t proc_coredump_filter_write(struct file *file,
2520 const char __user *buf,
2521 size_t count,
2522 loff_t *ppos)
2523{
2524 struct task_struct *task;
2525 struct mm_struct *mm;
2526 unsigned int val;
2527 int ret;
2528 int i;
2529 unsigned long mask;
2530
2531 ret = kstrtouint_from_user(buf, count, 0, &val);
2532 if (ret < 0)
2533 return ret;
2534
2535 ret = -ESRCH;
2536 task = get_proc_task(file_inode(file));
2537 if (!task)
2538 goto out_no_task;
2539
2540 mm = get_task_mm(task);
2541 if (!mm)
2542 goto out_no_mm;
2543 ret = 0;
2544
2545 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2546 if (val & mask)
2547 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2548 else
2549 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2550 }
2551
2552 mmput(mm);
2553 out_no_mm:
2554 put_task_struct(task);
2555 out_no_task:
2556 if (ret < 0)
2557 return ret;
2558 return count;
2559}
2560
2561static const struct file_operations proc_coredump_filter_operations = {
2562 .read = proc_coredump_filter_read,
2563 .write = proc_coredump_filter_write,
2564 .llseek = generic_file_llseek,
2565};
2566#endif
2567
2568#ifdef CONFIG_TASK_IO_ACCOUNTING
2569static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2570{
2571 struct task_io_accounting acct = task->ioac;
2572 unsigned long flags;
2573 int result;
2574
2575 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2576 if (result)
2577 return result;
2578
2579 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2580 result = -EACCES;
2581 goto out_unlock;
2582 }
2583
2584 if (whole && lock_task_sighand(task, &flags)) {
2585 struct task_struct *t = task;
2586
2587 task_io_accounting_add(&acct, &task->signal->ioac);
2588 while_each_thread(task, t)
2589 task_io_accounting_add(&acct, &t->ioac);
2590
2591 unlock_task_sighand(task, &flags);
2592 }
2593 seq_printf(m,
2594 "rchar: %llu\n"
2595 "wchar: %llu\n"
2596 "syscr: %llu\n"
2597 "syscw: %llu\n"
2598 "read_bytes: %llu\n"
2599 "write_bytes: %llu\n"
2600 "cancelled_write_bytes: %llu\n",
2601 (unsigned long long)acct.rchar,
2602 (unsigned long long)acct.wchar,
2603 (unsigned long long)acct.syscr,
2604 (unsigned long long)acct.syscw,
2605 (unsigned long long)acct.read_bytes,
2606 (unsigned long long)acct.write_bytes,
2607 (unsigned long long)acct.cancelled_write_bytes);
2608 result = 0;
2609
2610out_unlock:
2611 mutex_unlock(&task->signal->cred_guard_mutex);
2612 return result;
2613}
2614
2615static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2616 struct pid *pid, struct task_struct *task)
2617{
2618 return do_io_accounting(task, m, 0);
2619}
2620
2621static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2622 struct pid *pid, struct task_struct *task)
2623{
2624 return do_io_accounting(task, m, 1);
2625}
2626#endif /* CONFIG_TASK_IO_ACCOUNTING */
2627
2628#ifdef CONFIG_USER_NS
2629static int proc_id_map_open(struct inode *inode, struct file *file,
2630 const struct seq_operations *seq_ops)
2631{
2632 struct user_namespace *ns = NULL;
2633 struct task_struct *task;
2634 struct seq_file *seq;
2635 int ret = -EINVAL;
2636
2637 task = get_proc_task(inode);
2638 if (task) {
2639 rcu_read_lock();
2640 ns = get_user_ns(task_cred_xxx(task, user_ns));
2641 rcu_read_unlock();
2642 put_task_struct(task);
2643 }
2644 if (!ns)
2645 goto err;
2646
2647 ret = seq_open(file, seq_ops);
2648 if (ret)
2649 goto err_put_ns;
2650
2651 seq = file->private_data;
2652 seq->private = ns;
2653
2654 return 0;
2655err_put_ns:
2656 put_user_ns(ns);
2657err:
2658 return ret;
2659}
2660
2661static int proc_id_map_release(struct inode *inode, struct file *file)
2662{
2663 struct seq_file *seq = file->private_data;
2664 struct user_namespace *ns = seq->private;
2665 put_user_ns(ns);
2666 return seq_release(inode, file);
2667}
2668
2669static int proc_uid_map_open(struct inode *inode, struct file *file)
2670{
2671 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2672}
2673
2674static int proc_gid_map_open(struct inode *inode, struct file *file)
2675{
2676 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2677}
2678
2679static int proc_projid_map_open(struct inode *inode, struct file *file)
2680{
2681 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2682}
2683
2684static const struct file_operations proc_uid_map_operations = {
2685 .open = proc_uid_map_open,
2686 .write = proc_uid_map_write,
2687 .read = seq_read,
2688 .llseek = seq_lseek,
2689 .release = proc_id_map_release,
2690};
2691
2692static const struct file_operations proc_gid_map_operations = {
2693 .open = proc_gid_map_open,
2694 .write = proc_gid_map_write,
2695 .read = seq_read,
2696 .llseek = seq_lseek,
2697 .release = proc_id_map_release,
2698};
2699
2700static const struct file_operations proc_projid_map_operations = {
2701 .open = proc_projid_map_open,
2702 .write = proc_projid_map_write,
2703 .read = seq_read,
2704 .llseek = seq_lseek,
2705 .release = proc_id_map_release,
2706};
2707
2708static int proc_setgroups_open(struct inode *inode, struct file *file)
2709{
2710 struct user_namespace *ns = NULL;
2711 struct task_struct *task;
2712 int ret;
2713
2714 ret = -ESRCH;
2715 task = get_proc_task(inode);
2716 if (task) {
2717 rcu_read_lock();
2718 ns = get_user_ns(task_cred_xxx(task, user_ns));
2719 rcu_read_unlock();
2720 put_task_struct(task);
2721 }
2722 if (!ns)
2723 goto err;
2724
2725 if (file->f_mode & FMODE_WRITE) {
2726 ret = -EACCES;
2727 if (!ns_capable(ns, CAP_SYS_ADMIN))
2728 goto err_put_ns;
2729 }
2730
2731 ret = single_open(file, &proc_setgroups_show, ns);
2732 if (ret)
2733 goto err_put_ns;
2734
2735 return 0;
2736err_put_ns:
2737 put_user_ns(ns);
2738err:
2739 return ret;
2740}
2741
2742static int proc_setgroups_release(struct inode *inode, struct file *file)
2743{
2744 struct seq_file *seq = file->private_data;
2745 struct user_namespace *ns = seq->private;
2746 int ret = single_release(inode, file);
2747 put_user_ns(ns);
2748 return ret;
2749}
2750
2751static const struct file_operations proc_setgroups_operations = {
2752 .open = proc_setgroups_open,
2753 .write = proc_setgroups_write,
2754 .read = seq_read,
2755 .llseek = seq_lseek,
2756 .release = proc_setgroups_release,
2757};
2758#endif /* CONFIG_USER_NS */
2759
2760static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2761 struct pid *pid, struct task_struct *task)
2762{
2763 int err = lock_trace(task);
2764 if (!err) {
2765 seq_printf(m, "%08x\n", task->personality);
2766 unlock_trace(task);
2767 }
2768 return err;
2769}
2770
2771/*
2772 * Thread groups
2773 */
2774static const struct file_operations proc_task_operations;
2775static const struct inode_operations proc_task_inode_operations;
2776
2777static const struct pid_entry tgid_base_stuff[] = {
2778 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2779 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2780 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2781 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2782 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2783#ifdef CONFIG_NET
2784 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2785#endif
2786 REG("environ", S_IRUSR, proc_environ_operations),
2787 ONE("auxv", S_IRUSR, proc_pid_auxv),
2788 ONE("status", S_IRUGO, proc_pid_status),
2789 ONE("personality", S_IRUSR, proc_pid_personality),
2790 ONE("limits", S_IRUGO, proc_pid_limits),
2791#ifdef CONFIG_SCHED_DEBUG
2792 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2793#endif
2794#ifdef CONFIG_SCHED_AUTOGROUP
2795 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2796#endif
2797 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2798#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2799 ONE("syscall", S_IRUSR, proc_pid_syscall),
2800#endif
2801 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2802 ONE("stat", S_IRUGO, proc_tgid_stat),
2803 ONE("statm", S_IRUGO, proc_pid_statm),
2804 REG("maps", S_IRUGO, proc_pid_maps_operations),
2805#ifdef CONFIG_NUMA
2806 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2807#endif
2808 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2809 LNK("cwd", proc_cwd_link),
2810 LNK("root", proc_root_link),
2811 LNK("exe", proc_exe_link),
2812 REG("mounts", S_IRUGO, proc_mounts_operations),
2813 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2814 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2815#ifdef CONFIG_PROC_PAGE_MONITOR
2816 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2817 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2818 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2819#endif
2820#ifdef CONFIG_SECURITY
2821 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2822#endif
2823#ifdef CONFIG_KALLSYMS
2824 ONE("wchan", S_IRUGO, proc_pid_wchan),
2825#endif
2826#ifdef CONFIG_STACKTRACE
2827 ONE("stack", S_IRUSR, proc_pid_stack),
2828#endif
2829#ifdef CONFIG_SCHED_INFO
2830 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2831#endif
2832#ifdef CONFIG_LATENCYTOP
2833 REG("latency", S_IRUGO, proc_lstats_operations),
2834#endif
2835#ifdef CONFIG_PROC_PID_CPUSET
2836 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2837#endif
2838#ifdef CONFIG_CGROUPS
2839 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2840#endif
2841 ONE("oom_score", S_IRUGO, proc_oom_score),
2842 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2843 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2844#ifdef CONFIG_AUDITSYSCALL
2845 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2846 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2847#endif
2848#ifdef CONFIG_FAULT_INJECTION
2849 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2850#endif
2851#ifdef CONFIG_ELF_CORE
2852 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2853#endif
2854#ifdef CONFIG_TASK_IO_ACCOUNTING
2855 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2856#endif
2857#ifdef CONFIG_HARDWALL
2858 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2859#endif
2860#ifdef CONFIG_USER_NS
2861 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2862 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2863 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2864 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2865#endif
2866#ifdef CONFIG_CHECKPOINT_RESTORE
2867 REG("timers", S_IRUGO, proc_timers_operations),
2868#endif
2869 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2870};
2871
2872static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2873{
2874 return proc_pident_readdir(file, ctx,
2875 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2876}
2877
2878static const struct file_operations proc_tgid_base_operations = {
2879 .read = generic_read_dir,
2880 .iterate_shared = proc_tgid_base_readdir,
2881 .llseek = generic_file_llseek,
2882};
2883
2884static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2885{
2886 return proc_pident_lookup(dir, dentry,
2887 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2888}
2889
2890static const struct inode_operations proc_tgid_base_inode_operations = {
2891 .lookup = proc_tgid_base_lookup,
2892 .getattr = pid_getattr,
2893 .setattr = proc_setattr,
2894 .permission = proc_pid_permission,
2895};
2896
2897static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2898{
2899 struct dentry *dentry, *leader, *dir;
2900 char buf[PROC_NUMBUF];
2901 struct qstr name;
2902
2903 name.name = buf;
2904 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2905 /* no ->d_hash() rejects on procfs */
2906 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2907 if (dentry) {
2908 d_invalidate(dentry);
2909 dput(dentry);
2910 }
2911
2912 if (pid == tgid)
2913 return;
2914
2915 name.name = buf;
2916 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2917 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2918 if (!leader)
2919 goto out;
2920
2921 name.name = "task";
2922 name.len = strlen(name.name);
2923 dir = d_hash_and_lookup(leader, &name);
2924 if (!dir)
2925 goto out_put_leader;
2926
2927 name.name = buf;
2928 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2929 dentry = d_hash_and_lookup(dir, &name);
2930 if (dentry) {
2931 d_invalidate(dentry);
2932 dput(dentry);
2933 }
2934
2935 dput(dir);
2936out_put_leader:
2937 dput(leader);
2938out:
2939 return;
2940}
2941
2942/**
2943 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
2944 * @task: task that should be flushed.
2945 *
2946 * When flushing dentries from proc, one needs to flush them from global
2947 * proc (proc_mnt) and from all the namespaces' procs this task was seen
2948 * in. This call is supposed to do all of this job.
2949 *
2950 * Looks in the dcache for
2951 * /proc/@pid
2952 * /proc/@tgid/task/@pid
2953 * if either directory is present flushes it and all of it'ts children
2954 * from the dcache.
2955 *
2956 * It is safe and reasonable to cache /proc entries for a task until
2957 * that task exits. After that they just clog up the dcache with
2958 * useless entries, possibly causing useful dcache entries to be
2959 * flushed instead. This routine is proved to flush those useless
2960 * dcache entries at process exit time.
2961 *
2962 * NOTE: This routine is just an optimization so it does not guarantee
2963 * that no dcache entries will exist at process exit time it
2964 * just makes it very unlikely that any will persist.
2965 */
2966
2967void proc_flush_task(struct task_struct *task)
2968{
2969 int i;
2970 struct pid *pid, *tgid;
2971 struct upid *upid;
2972
2973 pid = task_pid(task);
2974 tgid = task_tgid(task);
2975
2976 for (i = 0; i <= pid->level; i++) {
2977 upid = &pid->numbers[i];
2978 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2979 tgid->numbers[i].nr);
2980 }
2981}
2982
2983static int proc_pid_instantiate(struct inode *dir,
2984 struct dentry * dentry,
2985 struct task_struct *task, const void *ptr)
2986{
2987 struct inode *inode;
2988
2989 inode = proc_pid_make_inode(dir->i_sb, task);
2990 if (!inode)
2991 goto out;
2992
2993 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2994 inode->i_op = &proc_tgid_base_inode_operations;
2995 inode->i_fop = &proc_tgid_base_operations;
2996 inode->i_flags|=S_IMMUTABLE;
2997
2998 set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2999 ARRAY_SIZE(tgid_base_stuff)));
3000
3001 d_set_d_op(dentry, &pid_dentry_operations);
3002
3003 d_add(dentry, inode);
3004 /* Close the race of the process dying before we return the dentry */
3005 if (pid_revalidate(dentry, 0))
3006 return 0;
3007out:
3008 return -ENOENT;
3009}
3010
3011struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3012{
3013 int result = -ENOENT;
3014 struct task_struct *task;
3015 unsigned tgid;
3016 struct pid_namespace *ns;
3017
3018 tgid = name_to_int(&dentry->d_name);
3019 if (tgid == ~0U)
3020 goto out;
3021
3022 ns = dentry->d_sb->s_fs_info;
3023 rcu_read_lock();
3024 task = find_task_by_pid_ns(tgid, ns);
3025 if (task)
3026 get_task_struct(task);
3027 rcu_read_unlock();
3028 if (!task)
3029 goto out;
3030
3031 result = proc_pid_instantiate(dir, dentry, task, NULL);
3032 put_task_struct(task);
3033out:
3034 return ERR_PTR(result);
3035}
3036
3037/*
3038 * Find the first task with tgid >= tgid
3039 *
3040 */
3041struct tgid_iter {
3042 unsigned int tgid;
3043 struct task_struct *task;
3044};
3045static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3046{
3047 struct pid *pid;
3048
3049 if (iter.task)
3050 put_task_struct(iter.task);
3051 rcu_read_lock();
3052retry:
3053 iter.task = NULL;
3054 pid = find_ge_pid(iter.tgid, ns);
3055 if (pid) {
3056 iter.tgid = pid_nr_ns(pid, ns);
3057 iter.task = pid_task(pid, PIDTYPE_PID);
3058 /* What we to know is if the pid we have find is the
3059 * pid of a thread_group_leader. Testing for task
3060 * being a thread_group_leader is the obvious thing
3061 * todo but there is a window when it fails, due to
3062 * the pid transfer logic in de_thread.
3063 *
3064 * So we perform the straight forward test of seeing
3065 * if the pid we have found is the pid of a thread
3066 * group leader, and don't worry if the task we have
3067 * found doesn't happen to be a thread group leader.
3068 * As we don't care in the case of readdir.
3069 */
3070 if (!iter.task || !has_group_leader_pid(iter.task)) {
3071 iter.tgid += 1;
3072 goto retry;
3073 }
3074 get_task_struct(iter.task);
3075 }
3076 rcu_read_unlock();
3077 return iter;
3078}
3079
3080#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3081
3082/* for the /proc/ directory itself, after non-process stuff has been done */
3083int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3084{
3085 struct tgid_iter iter;
3086 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3087 loff_t pos = ctx->pos;
3088
3089 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3090 return 0;
3091
3092 if (pos == TGID_OFFSET - 2) {
3093 struct inode *inode = d_inode(ns->proc_self);
3094 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3095 return 0;
3096 ctx->pos = pos = pos + 1;
3097 }
3098 if (pos == TGID_OFFSET - 1) {
3099 struct inode *inode = d_inode(ns->proc_thread_self);
3100 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3101 return 0;
3102 ctx->pos = pos = pos + 1;
3103 }
3104 iter.tgid = pos - TGID_OFFSET;
3105 iter.task = NULL;
3106 for (iter = next_tgid(ns, iter);
3107 iter.task;
3108 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3109 char name[PROC_NUMBUF];
3110 int len;
3111 if (!has_pid_permissions(ns, iter.task, 2))
3112 continue;
3113
3114 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3115 ctx->pos = iter.tgid + TGID_OFFSET;
3116 if (!proc_fill_cache(file, ctx, name, len,
3117 proc_pid_instantiate, iter.task, NULL)) {
3118 put_task_struct(iter.task);
3119 return 0;
3120 }
3121 }
3122 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3123 return 0;
3124}
3125
3126/*
3127 * proc_tid_comm_permission is a special permission function exclusively
3128 * used for the node /proc/<pid>/task/<tid>/comm.
3129 * It bypasses generic permission checks in the case where a task of the same
3130 * task group attempts to access the node.
3131 * The rationale behind this is that glibc and bionic access this node for
3132 * cross thread naming (pthread_set/getname_np(!self)). However, if
3133 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3134 * which locks out the cross thread naming implementation.
3135 * This function makes sure that the node is always accessible for members of
3136 * same thread group.
3137 */
3138static int proc_tid_comm_permission(struct inode *inode, int mask)
3139{
3140 bool is_same_tgroup;
3141 struct task_struct *task;
3142
3143 task = get_proc_task(inode);
3144 if (!task)
3145 return -ESRCH;
3146 is_same_tgroup = same_thread_group(current, task);
3147 put_task_struct(task);
3148
3149 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3150 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3151 * read or written by the members of the corresponding
3152 * thread group.
3153 */
3154 return 0;
3155 }
3156
3157 return generic_permission(inode, mask);
3158}
3159
3160static const struct inode_operations proc_tid_comm_inode_operations = {
3161 .permission = proc_tid_comm_permission,
3162};
3163
3164/*
3165 * Tasks
3166 */
3167static const struct pid_entry tid_base_stuff[] = {
3168 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3169 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3170 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3171#ifdef CONFIG_NET
3172 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3173#endif
3174 REG("environ", S_IRUSR, proc_environ_operations),
3175 ONE("auxv", S_IRUSR, proc_pid_auxv),
3176 ONE("status", S_IRUGO, proc_pid_status),
3177 ONE("personality", S_IRUSR, proc_pid_personality),
3178 ONE("limits", S_IRUGO, proc_pid_limits),
3179#ifdef CONFIG_SCHED_DEBUG
3180 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3181#endif
3182 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3183 &proc_tid_comm_inode_operations,
3184 &proc_pid_set_comm_operations, {}),
3185#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3186 ONE("syscall", S_IRUSR, proc_pid_syscall),
3187#endif
3188 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3189 ONE("stat", S_IRUGO, proc_tid_stat),
3190 ONE("statm", S_IRUGO, proc_pid_statm),
3191 REG("maps", S_IRUGO, proc_tid_maps_operations),
3192#ifdef CONFIG_PROC_CHILDREN
3193 REG("children", S_IRUGO, proc_tid_children_operations),
3194#endif
3195#ifdef CONFIG_NUMA
3196 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3197#endif
3198 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3199 LNK("cwd", proc_cwd_link),
3200 LNK("root", proc_root_link),
3201 LNK("exe", proc_exe_link),
3202 REG("mounts", S_IRUGO, proc_mounts_operations),
3203 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3204#ifdef CONFIG_PROC_PAGE_MONITOR
3205 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3206 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3207 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3208#endif
3209#ifdef CONFIG_SECURITY
3210 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3211#endif
3212#ifdef CONFIG_KALLSYMS
3213 ONE("wchan", S_IRUGO, proc_pid_wchan),
3214#endif
3215#ifdef CONFIG_STACKTRACE
3216 ONE("stack", S_IRUSR, proc_pid_stack),
3217#endif
3218#ifdef CONFIG_SCHED_INFO
3219 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3220#endif
3221#ifdef CONFIG_LATENCYTOP
3222 REG("latency", S_IRUGO, proc_lstats_operations),
3223#endif
3224#ifdef CONFIG_PROC_PID_CPUSET
3225 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3226#endif
3227#ifdef CONFIG_CGROUPS
3228 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3229#endif
3230 ONE("oom_score", S_IRUGO, proc_oom_score),
3231 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3232 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3233#ifdef CONFIG_AUDITSYSCALL
3234 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3235 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3236#endif
3237#ifdef CONFIG_FAULT_INJECTION
3238 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3239#endif
3240#ifdef CONFIG_TASK_IO_ACCOUNTING
3241 ONE("io", S_IRUSR, proc_tid_io_accounting),
3242#endif
3243#ifdef CONFIG_HARDWALL
3244 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3245#endif
3246#ifdef CONFIG_USER_NS
3247 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3248 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3249 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3250 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3251#endif
3252};
3253
3254static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3255{
3256 return proc_pident_readdir(file, ctx,
3257 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3258}
3259
3260static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3261{
3262 return proc_pident_lookup(dir, dentry,
3263 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3264}
3265
3266static const struct file_operations proc_tid_base_operations = {
3267 .read = generic_read_dir,
3268 .iterate_shared = proc_tid_base_readdir,
3269 .llseek = generic_file_llseek,
3270};
3271
3272static const struct inode_operations proc_tid_base_inode_operations = {
3273 .lookup = proc_tid_base_lookup,
3274 .getattr = pid_getattr,
3275 .setattr = proc_setattr,
3276};
3277
3278static int proc_task_instantiate(struct inode *dir,
3279 struct dentry *dentry, struct task_struct *task, const void *ptr)
3280{
3281 struct inode *inode;
3282 inode = proc_pid_make_inode(dir->i_sb, task);
3283
3284 if (!inode)
3285 goto out;
3286 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3287 inode->i_op = &proc_tid_base_inode_operations;
3288 inode->i_fop = &proc_tid_base_operations;
3289 inode->i_flags|=S_IMMUTABLE;
3290
3291 set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3292 ARRAY_SIZE(tid_base_stuff)));
3293
3294 d_set_d_op(dentry, &pid_dentry_operations);
3295
3296 d_add(dentry, inode);
3297 /* Close the race of the process dying before we return the dentry */
3298 if (pid_revalidate(dentry, 0))
3299 return 0;
3300out:
3301 return -ENOENT;
3302}
3303
3304static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3305{
3306 int result = -ENOENT;
3307 struct task_struct *task;
3308 struct task_struct *leader = get_proc_task(dir);
3309 unsigned tid;
3310 struct pid_namespace *ns;
3311
3312 if (!leader)
3313 goto out_no_task;
3314
3315 tid = name_to_int(&dentry->d_name);
3316 if (tid == ~0U)
3317 goto out;
3318
3319 ns = dentry->d_sb->s_fs_info;
3320 rcu_read_lock();
3321 task = find_task_by_pid_ns(tid, ns);
3322 if (task)
3323 get_task_struct(task);
3324 rcu_read_unlock();
3325 if (!task)
3326 goto out;
3327 if (!same_thread_group(leader, task))
3328 goto out_drop_task;
3329
3330 result = proc_task_instantiate(dir, dentry, task, NULL);
3331out_drop_task:
3332 put_task_struct(task);
3333out:
3334 put_task_struct(leader);
3335out_no_task:
3336 return ERR_PTR(result);
3337}
3338
3339/*
3340 * Find the first tid of a thread group to return to user space.
3341 *
3342 * Usually this is just the thread group leader, but if the users
3343 * buffer was too small or there was a seek into the middle of the
3344 * directory we have more work todo.
3345 *
3346 * In the case of a short read we start with find_task_by_pid.
3347 *
3348 * In the case of a seek we start with the leader and walk nr
3349 * threads past it.
3350 */
3351static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3352 struct pid_namespace *ns)
3353{
3354 struct task_struct *pos, *task;
3355 unsigned long nr = f_pos;
3356
3357 if (nr != f_pos) /* 32bit overflow? */
3358 return NULL;
3359
3360 rcu_read_lock();
3361 task = pid_task(pid, PIDTYPE_PID);
3362 if (!task)
3363 goto fail;
3364
3365 /* Attempt to start with the tid of a thread */
3366 if (tid && nr) {
3367 pos = find_task_by_pid_ns(tid, ns);
3368 if (pos && same_thread_group(pos, task))
3369 goto found;
3370 }
3371
3372 /* If nr exceeds the number of threads there is nothing todo */
3373 if (nr >= get_nr_threads(task))
3374 goto fail;
3375
3376 /* If we haven't found our starting place yet start
3377 * with the leader and walk nr threads forward.
3378 */
3379 pos = task = task->group_leader;
3380 do {
3381 if (!nr--)
3382 goto found;
3383 } while_each_thread(task, pos);
3384fail:
3385 pos = NULL;
3386 goto out;
3387found:
3388 get_task_struct(pos);
3389out:
3390 rcu_read_unlock();
3391 return pos;
3392}
3393
3394/*
3395 * Find the next thread in the thread list.
3396 * Return NULL if there is an error or no next thread.
3397 *
3398 * The reference to the input task_struct is released.
3399 */
3400static struct task_struct *next_tid(struct task_struct *start)
3401{
3402 struct task_struct *pos = NULL;
3403 rcu_read_lock();
3404 if (pid_alive(start)) {
3405 pos = next_thread(start);
3406 if (thread_group_leader(pos))
3407 pos = NULL;
3408 else
3409 get_task_struct(pos);
3410 }
3411 rcu_read_unlock();
3412 put_task_struct(start);
3413 return pos;
3414}
3415
3416/* for the /proc/TGID/task/ directories */
3417static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3418{
3419 struct inode *inode = file_inode(file);
3420 struct task_struct *task;
3421 struct pid_namespace *ns;
3422 int tid;
3423
3424 if (proc_inode_is_dead(inode))
3425 return -ENOENT;
3426
3427 if (!dir_emit_dots(file, ctx))
3428 return 0;
3429
3430 /* f_version caches the tgid value that the last readdir call couldn't
3431 * return. lseek aka telldir automagically resets f_version to 0.
3432 */
3433 ns = inode->i_sb->s_fs_info;
3434 tid = (int)file->f_version;
3435 file->f_version = 0;
3436 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3437 task;
3438 task = next_tid(task), ctx->pos++) {
3439 char name[PROC_NUMBUF];
3440 int len;
3441 tid = task_pid_nr_ns(task, ns);
3442 len = snprintf(name, sizeof(name), "%d", tid);
3443 if (!proc_fill_cache(file, ctx, name, len,
3444 proc_task_instantiate, task, NULL)) {
3445 /* returning this tgid failed, save it as the first
3446 * pid for the next readir call */
3447 file->f_version = (u64)tid;
3448 put_task_struct(task);
3449 break;
3450 }
3451 }
3452
3453 return 0;
3454}
3455
3456static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3457{
3458 struct inode *inode = d_inode(dentry);
3459 struct task_struct *p = get_proc_task(inode);
3460 generic_fillattr(inode, stat);
3461
3462 if (p) {
3463 stat->nlink += get_nr_threads(p);
3464 put_task_struct(p);
3465 }
3466
3467 return 0;
3468}
3469
3470static const struct inode_operations proc_task_inode_operations = {
3471 .lookup = proc_task_lookup,
3472 .getattr = proc_task_getattr,
3473 .setattr = proc_setattr,
3474 .permission = proc_pid_permission,
3475};
3476
3477static const struct file_operations proc_task_operations = {
3478 .read = generic_read_dir,
3479 .iterate_shared = proc_task_readdir,
3480 .llseek = generic_file_llseek,
3481};
Linux kernel - Deliverables
This page took 0.034586 seconds and 5 git commands to generate.