?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
fs/proc: calculate /proc/* and /proc/*/task/* nlink at init time
[linux-2.6/btrfs-unstable.git] / fs / proc / base.c
blob9b99df4893a49489dd8975e3510cdb7045648e24
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
107 static u8 nlink_tid;
108 static u8 nlink_tgid;
109
110 struct pid_entry {
111 const char *name;
112 unsigned int len;
113 umode_t mode;
114 const struct inode_operations *iop;
115 const struct file_operations *fop;
116 union proc_op op;
117 };
118
119 #define NOD(NAME, MODE, IOP, FOP, OP) { \
120 .name = (NAME), \
121 .len = sizeof(NAME) - 1, \
122 .mode = MODE, \
123 .iop = IOP, \
124 .fop = FOP, \
125 .op = OP, \
126 }
127
128 #define DIR(NAME, MODE, iops, fops) \
129 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
130 #define LNK(NAME, get_link) \
131 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
132 &proc_pid_link_inode_operations, NULL, \
133 { .proc_get_link = get_link } )
134 #define REG(NAME, MODE, fops) \
135 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
136 #define ONE(NAME, MODE, show) \
137 NOD(NAME, (S_IFREG|(MODE)), \
138 NULL, &proc_single_file_operations, \
139 { .proc_show = show } )
140
141 /*
142 * Count the number of hardlinks for the pid_entry table, excluding the .
143 * and .. links.
144 */
145 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
146 unsigned int n)
147 {
148 unsigned int i;
149 unsigned int count;
150
151 count = 2;
152 for (i = 0; i < n; ++i) {
153 if (S_ISDIR(entries[i].mode))
154 ++count;
155 }
156
157 return count;
158 }
159
160 static int get_task_root(struct task_struct *task, struct path *root)
161 {
162 int result = -ENOENT;
163
164 task_lock(task);
165 if (task->fs) {
166 get_fs_root(task->fs, root);
167 result = 0;
168 }
169 task_unlock(task);
170 return result;
171 }
172
173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
174 {
175 struct task_struct *task = get_proc_task(d_inode(dentry));
176 int result = -ENOENT;
177
178 if (task) {
179 task_lock(task);
180 if (task->fs) {
181 get_fs_pwd(task->fs, path);
182 result = 0;
183 }
184 task_unlock(task);
185 put_task_struct(task);
186 }
187 return result;
188 }
189
190 static int proc_root_link(struct dentry *dentry, struct path *path)
191 {
192 struct task_struct *task = get_proc_task(d_inode(dentry));
193 int result = -ENOENT;
194
195 if (task) {
196 result = get_task_root(task, path);
197 put_task_struct(task);
198 }
199 return result;
200 }
201
202 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
203 size_t _count, loff_t *pos)
204 {
205 struct task_struct *tsk;
206 struct mm_struct *mm;
207 char *page;
208 unsigned long count = _count;
209 unsigned long arg_start, arg_end, env_start, env_end;
210 unsigned long len1, len2, len;
211 unsigned long p;
212 char c;
213 ssize_t rv;
214
215 BUG_ON(*pos < 0);
216
217 tsk = get_proc_task(file_inode(file));
218 if (!tsk)
219 return -ESRCH;
220 mm = get_task_mm(tsk);
221 put_task_struct(tsk);
222 if (!mm)
223 return 0;
224 /* Check if process spawned far enough to have cmdline. */
225 if (!mm->env_end) {
226 rv = 0;
227 goto out_mmput;
228 }
229
230 page = (char *)__get_free_page(GFP_TEMPORARY);
231 if (!page) {
232 rv = -ENOMEM;
233 goto out_mmput;
234 }
235
236 down_read(&mm->mmap_sem);
237 arg_start = mm->arg_start;
238 arg_end = mm->arg_end;
239 env_start = mm->env_start;
240 env_end = mm->env_end;
241 up_read(&mm->mmap_sem);
242
243 BUG_ON(arg_start > arg_end);
244 BUG_ON(env_start > env_end);
245
246 len1 = arg_end - arg_start;
247 len2 = env_end - env_start;
248
249 /* Empty ARGV. */
250 if (len1 == 0) {
251 rv = 0;
252 goto out_free_page;
253 }
254 /*
255 * Inherently racy -- command line shares address space
256 * with code and data.
257 */
258 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
259 if (rv <= 0)
260 goto out_free_page;
261
262 rv = 0;
263
264 if (c == '\0') {
265 /* Command line (set of strings) occupies whole ARGV. */
266 if (len1 <= *pos)
267 goto out_free_page;
268
269 p = arg_start + *pos;
270 len = len1 - *pos;
271 while (count > 0 && len > 0) {
272 unsigned int _count;
273 int nr_read;
274
275 _count = min3(count, len, PAGE_SIZE);
276 nr_read = access_remote_vm(mm, p, page, _count, 0);
277 if (nr_read < 0)
278 rv = nr_read;
279 if (nr_read <= 0)
280 goto out_free_page;
281
282 if (copy_to_user(buf, page, nr_read)) {
283 rv = -EFAULT;
284 goto out_free_page;
285 }
286
287 p += nr_read;
288 len -= nr_read;
289 buf += nr_read;
290 count -= nr_read;
291 rv += nr_read;
292 }
293 } else {
294 /*
295 * Command line (1 string) occupies ARGV and maybe
296 * extends into ENVP.
297 */
298 if (len1 + len2 <= *pos)
299 goto skip_argv_envp;
300 if (len1 <= *pos)
301 goto skip_argv;
302
303 p = arg_start + *pos;
304 len = len1 - *pos;
305 while (count > 0 && len > 0) {
306 unsigned int _count, l;
307 int nr_read;
308 bool final;
309
310 _count = min3(count, len, PAGE_SIZE);
311 nr_read = access_remote_vm(mm, p, page, _count, 0);
312 if (nr_read < 0)
313 rv = nr_read;
314 if (nr_read <= 0)
315 goto out_free_page;
316
317 /*
318 * Command line can be shorter than whole ARGV
319 * even if last "marker" byte says it is not.
320 */
321 final = false;
322 l = strnlen(page, nr_read);
323 if (l < nr_read) {
324 nr_read = l;
325 final = true;
326 }
327
328 if (copy_to_user(buf, page, nr_read)) {
329 rv = -EFAULT;
330 goto out_free_page;
331 }
332
333 p += nr_read;
334 len -= nr_read;
335 buf += nr_read;
336 count -= nr_read;
337 rv += nr_read;
338
339 if (final)
340 goto out_free_page;
341 }
342 skip_argv:
343 /*
344 * Command line (1 string) occupies ARGV and
345 * extends into ENVP.
346 */
347 if (len1 <= *pos) {
348 p = env_start + *pos - len1;
349 len = len1 + len2 - *pos;
350 } else {
351 p = env_start;
352 len = len2;
353 }
354 while (count > 0 && len > 0) {
355 unsigned int _count, l;
356 int nr_read;
357 bool final;
358
359 _count = min3(count, len, PAGE_SIZE);
360 nr_read = access_remote_vm(mm, p, page, _count, 0);
361 if (nr_read < 0)
362 rv = nr_read;
363 if (nr_read <= 0)
364 goto out_free_page;
365
366 /* Find EOS. */
367 final = false;
368 l = strnlen(page, nr_read);
369 if (l < nr_read) {
370 nr_read = l;
371 final = true;
372 }
373
374 if (copy_to_user(buf, page, nr_read)) {
375 rv = -EFAULT;
376 goto out_free_page;
377 }
378
379 p += nr_read;
380 len -= nr_read;
381 buf += nr_read;
382 count -= nr_read;
383 rv += nr_read;
384
385 if (final)
386 goto out_free_page;
387 }
388 skip_argv_envp:
389 ;
390 }
391
392 out_free_page:
393 free_page((unsigned long)page);
394 out_mmput:
395 mmput(mm);
396 if (rv > 0)
397 *pos += rv;
398 return rv;
399 }
400
401 static const struct file_operations proc_pid_cmdline_ops = {
402 .read = proc_pid_cmdline_read,
403 .llseek = generic_file_llseek,
404 };
405
406 #ifdef CONFIG_KALLSYMS
407 /*
408 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
409 * Returns the resolved symbol. If that fails, simply return the address.
410 */
411 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
412 struct pid *pid, struct task_struct *task)
413 {
414 unsigned long wchan;
415 char symname[KSYM_NAME_LEN];
416
417 wchan = get_wchan(task);
418
419 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
420 && !lookup_symbol_name(wchan, symname))
421 seq_printf(m, "%s", symname);
422 else
423 seq_putc(m, '0');
424
425 return 0;
426 }
427 #endif /* CONFIG_KALLSYMS */
428
429 static int lock_trace(struct task_struct *task)
430 {
431 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
432 if (err)
433 return err;
434 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
435 mutex_unlock(&task->signal->cred_guard_mutex);
436 return -EPERM;
437 }
438 return 0;
439 }
440
441 static void unlock_trace(struct task_struct *task)
442 {
443 mutex_unlock(&task->signal->cred_guard_mutex);
444 }
445
446 #ifdef CONFIG_STACKTRACE
447
448 #define MAX_STACK_TRACE_DEPTH 64
449
450 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
451 struct pid *pid, struct task_struct *task)
452 {
453 struct stack_trace trace;
454 unsigned long *entries;
455 int err;
456 int i;
457
458 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
459 if (!entries)
460 return -ENOMEM;
461
462 trace.nr_entries = 0;
463 trace.max_entries = MAX_STACK_TRACE_DEPTH;
464 trace.entries = entries;
465 trace.skip = 0;
466
467 err = lock_trace(task);
468 if (!err) {
469 save_stack_trace_tsk(task, &trace);
470
471 for (i = 0; i < trace.nr_entries; i++) {
472 seq_printf(m, "[<%pK>] %pB\n",
473 (void *)entries[i], (void *)entries[i]);
474 }
475 unlock_trace(task);
476 }
477 kfree(entries);
478
479 return err;
480 }
481 #endif
482
483 #ifdef CONFIG_SCHED_INFO
484 /*
485 * Provides /proc/PID/schedstat
486 */
487 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
488 struct pid *pid, struct task_struct *task)
489 {
490 if (unlikely(!sched_info_on()))
491 seq_printf(m, "0 0 0\n");
492 else
493 seq_printf(m, "%llu %llu %lu\n",
494 (unsigned long long)task->se.sum_exec_runtime,
495 (unsigned long long)task->sched_info.run_delay,
496 task->sched_info.pcount);
497
498 return 0;
499 }
500 #endif
501
502 #ifdef CONFIG_LATENCYTOP
503 static int lstats_show_proc(struct seq_file *m, void *v)
504 {
505 int i;
506 struct inode *inode = m->private;
507 struct task_struct *task = get_proc_task(inode);
508
509 if (!task)
510 return -ESRCH;
511 seq_puts(m, "Latency Top version : v0.1\n");
512 for (i = 0; i < 32; i++) {
513 struct latency_record *lr = &task->latency_record[i];
514 if (lr->backtrace[0]) {
515 int q;
516 seq_printf(m, "%i %li %li",
517 lr->count, lr->time, lr->max);
518 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
519 unsigned long bt = lr->backtrace[q];
520 if (!bt)
521 break;
522 if (bt == ULONG_MAX)
523 break;
524 seq_printf(m, " %ps", (void *)bt);
525 }
526 seq_putc(m, '\n');
527 }
528
529 }
530 put_task_struct(task);
531 return 0;
532 }
533
534 static int lstats_open(struct inode *inode, struct file *file)
535 {
536 return single_open(file, lstats_show_proc, inode);
537 }
538
539 static ssize_t lstats_write(struct file *file, const char __user *buf,
540 size_t count, loff_t *offs)
541 {
542 struct task_struct *task = get_proc_task(file_inode(file));
543
544 if (!task)
545 return -ESRCH;
546 clear_all_latency_tracing(task);
547 put_task_struct(task);
548
549 return count;
550 }
551
552 static const struct file_operations proc_lstats_operations = {
553 .open = lstats_open,
554 .read = seq_read,
555 .write = lstats_write,
556 .llseek = seq_lseek,
557 .release = single_release,
558 };
559
560 #endif
561
562 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
563 struct pid *pid, struct task_struct *task)
564 {
565 unsigned long totalpages = totalram_pages + total_swap_pages;
566 unsigned long points = 0;
567
568 points = oom_badness(task, NULL, NULL, totalpages) *
569 1000 / totalpages;
570 seq_printf(m, "%lu\n", points);
571
572 return 0;
573 }
574
575 struct limit_names {
576 const char *name;
577 const char *unit;
578 };
579
580 static const struct limit_names lnames[RLIM_NLIMITS] = {
581 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
582 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
583 [RLIMIT_DATA] = {"Max data size", "bytes"},
584 [RLIMIT_STACK] = {"Max stack size", "bytes"},
585 [RLIMIT_CORE] = {"Max core file size", "bytes"},
586 [RLIMIT_RSS] = {"Max resident set", "bytes"},
587 [RLIMIT_NPROC] = {"Max processes", "processes"},
588 [RLIMIT_NOFILE] = {"Max open files", "files"},
589 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
590 [RLIMIT_AS] = {"Max address space", "bytes"},
591 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
592 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
593 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
594 [RLIMIT_NICE] = {"Max nice priority", NULL},
595 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
596 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
597 };
598
599 /* Display limits for a process */
600 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
601 struct pid *pid, struct task_struct *task)
602 {
603 unsigned int i;
604 unsigned long flags;
605
606 struct rlimit rlim[RLIM_NLIMITS];
607
608 if (!lock_task_sighand(task, &flags))
609 return 0;
610 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
611 unlock_task_sighand(task, &flags);
612
613 /*
614 * print the file header
615 */
616 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
617 "Limit", "Soft Limit", "Hard Limit", "Units");
618
619 for (i = 0; i < RLIM_NLIMITS; i++) {
620 if (rlim[i].rlim_cur == RLIM_INFINITY)
621 seq_printf(m, "%-25s %-20s ",
622 lnames[i].name, "unlimited");
623 else
624 seq_printf(m, "%-25s %-20lu ",
625 lnames[i].name, rlim[i].rlim_cur);
626
627 if (rlim[i].rlim_max == RLIM_INFINITY)
628 seq_printf(m, "%-20s ", "unlimited");
629 else
630 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
631
632 if (lnames[i].unit)
633 seq_printf(m, "%-10s\n", lnames[i].unit);
634 else
635 seq_putc(m, '\n');
636 }
637
638 return 0;
639 }
640
641 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
642 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
643 struct pid *pid, struct task_struct *task)
644 {
645 long nr;
646 unsigned long args[6], sp, pc;
647 int res;
648
649 res = lock_trace(task);
650 if (res)
651 return res;
652
653 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
654 seq_puts(m, "running\n");
655 else if (nr < 0)
656 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
657 else
658 seq_printf(m,
659 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
660 nr,
661 args[0], args[1], args[2], args[3], args[4], args[5],
662 sp, pc);
663 unlock_trace(task);
664
665 return 0;
666 }
667 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
668
669 /************************************************************************/
670 /* Here the fs part begins */
671 /************************************************************************/
672
673 /* permission checks */
674 static int proc_fd_access_allowed(struct inode *inode)
675 {
676 struct task_struct *task;
677 int allowed = 0;
678 /* Allow access to a task's file descriptors if it is us or we
679 * may use ptrace attach to the process and find out that
680 * information.
681 */
682 task = get_proc_task(inode);
683 if (task) {
684 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
685 put_task_struct(task);
686 }
687 return allowed;
688 }
689
690 int proc_setattr(struct dentry *dentry, struct iattr *attr)
691 {
692 int error;
693 struct inode *inode = d_inode(dentry);
694
695 if (attr->ia_valid & ATTR_MODE)
696 return -EPERM;
697
698 error = setattr_prepare(dentry, attr);
699 if (error)
700 return error;
701
702 setattr_copy(inode, attr);
703 mark_inode_dirty(inode);
704 return 0;
705 }
706
707 /*
708 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
709 * or euid/egid (for hide_pid_min=2)?
710 */
711 static bool has_pid_permissions(struct pid_namespace *pid,
712 struct task_struct *task,
713 int hide_pid_min)
714 {
715 if (pid->hide_pid < hide_pid_min)
716 return true;
717 if (in_group_p(pid->pid_gid))
718 return true;
719 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
720 }
721
722
723 static int proc_pid_permission(struct inode *inode, int mask)
724 {
725 struct pid_namespace *pid = inode->i_sb->s_fs_info;
726 struct task_struct *task;
727 bool has_perms;
728
729 task = get_proc_task(inode);
730 if (!task)
731 return -ESRCH;
732 has_perms = has_pid_permissions(pid, task, 1);
733 put_task_struct(task);
734
735 if (!has_perms) {
736 if (pid->hide_pid == 2) {
737 /*
738 * Let's make getdents(), stat(), and open()
739 * consistent with each other. If a process
740 * may not stat() a file, it shouldn't be seen
741 * in procfs at all.
742 */
743 return -ENOENT;
744 }
745
746 return -EPERM;
747 }
748 return generic_permission(inode, mask);
749 }
750
751
752
753 static const struct inode_operations proc_def_inode_operations = {
754 .setattr = proc_setattr,
755 };
756
757 static int proc_single_show(struct seq_file *m, void *v)
758 {
759 struct inode *inode = m->private;
760 struct pid_namespace *ns;
761 struct pid *pid;
762 struct task_struct *task;
763 int ret;
764
765 ns = inode->i_sb->s_fs_info;
766 pid = proc_pid(inode);
767 task = get_pid_task(pid, PIDTYPE_PID);
768 if (!task)
769 return -ESRCH;
770
771 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
772
773 put_task_struct(task);
774 return ret;
775 }
776
777 static int proc_single_open(struct inode *inode, struct file *filp)
778 {
779 return single_open(filp, proc_single_show, inode);
780 }
781
782 static const struct file_operations proc_single_file_operations = {
783 .open = proc_single_open,
784 .read = seq_read,
785 .llseek = seq_lseek,
786 .release = single_release,
787 };
788
789
790 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
791 {
792 struct task_struct *task = get_proc_task(inode);
793 struct mm_struct *mm = ERR_PTR(-ESRCH);
794
795 if (task) {
796 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
797 put_task_struct(task);
798
799 if (!IS_ERR_OR_NULL(mm)) {
800 /* ensure this mm_struct can't be freed */
801 atomic_inc(&mm->mm_count);
802 /* but do not pin its memory */
803 mmput(mm);
804 }
805 }
806
807 return mm;
808 }
809
810 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
811 {
812 struct mm_struct *mm = proc_mem_open(inode, mode);
813
814 if (IS_ERR(mm))
815 return PTR_ERR(mm);
816
817 file->private_data = mm;
818 return 0;
819 }
820
821 static int mem_open(struct inode *inode, struct file *file)
822 {
823 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
824
825 /* OK to pass negative loff_t, we can catch out-of-range */
826 file->f_mode |= FMODE_UNSIGNED_OFFSET;
827
828 return ret;
829 }
830
831 static ssize_t mem_rw(struct file *file, char __user *buf,
832 size_t count, loff_t *ppos, int write)
833 {
834 struct mm_struct *mm = file->private_data;
835 unsigned long addr = *ppos;
836 ssize_t copied;
837 char *page;
838 unsigned int flags;
839
840 if (!mm)
841 return 0;
842
843 page = (char *)__get_free_page(GFP_TEMPORARY);
844 if (!page)
845 return -ENOMEM;
846
847 copied = 0;
848 if (!atomic_inc_not_zero(&mm->mm_users))
849 goto free;
850
851 /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
852 flags = FOLL_FORCE;
853 if (write)
854 flags |= FOLL_WRITE;
855
856 while (count > 0) {
857 int this_len = min_t(int, count, PAGE_SIZE);
858
859 if (write && copy_from_user(page, buf, this_len)) {
860 copied = -EFAULT;
861 break;
862 }
863
864 this_len = access_remote_vm(mm, addr, page, this_len, flags);
865 if (!this_len) {
866 if (!copied)
867 copied = -EIO;
868 break;
869 }
870
871 if (!write && copy_to_user(buf, page, this_len)) {
872 copied = -EFAULT;
873 break;
874 }
875
876 buf += this_len;
877 addr += this_len;
878 copied += this_len;
879 count -= this_len;
880 }
881 *ppos = addr;
882
883 mmput(mm);
884 free:
885 free_page((unsigned long) page);
886 return copied;
887 }
888
889 static ssize_t mem_read(struct file *file, char __user *buf,
890 size_t count, loff_t *ppos)
891 {
892 return mem_rw(file, buf, count, ppos, 0);
893 }
894
895 static ssize_t mem_write(struct file *file, const char __user *buf,
896 size_t count, loff_t *ppos)
897 {
898 return mem_rw(file, (char __user*)buf, count, ppos, 1);
899 }
900
901 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
902 {
903 switch (orig) {
904 case 0:
905 file->f_pos = offset;
906 break;
907 case 1:
908 file->f_pos += offset;
909 break;
910 default:
911 return -EINVAL;
912 }
913 force_successful_syscall_return();
914 return file->f_pos;
915 }
916
917 static int mem_release(struct inode *inode, struct file *file)
918 {
919 struct mm_struct *mm = file->private_data;
920 if (mm)
921 mmdrop(mm);
922 return 0;
923 }
924
925 static const struct file_operations proc_mem_operations = {
926 .llseek = mem_lseek,
927 .read = mem_read,
928 .write = mem_write,
929 .open = mem_open,
930 .release = mem_release,
931 };
932
933 static int environ_open(struct inode *inode, struct file *file)
934 {
935 return __mem_open(inode, file, PTRACE_MODE_READ);
936 }
937
938 static ssize_t environ_read(struct file *file, char __user *buf,
939 size_t count, loff_t *ppos)
940 {
941 char *page;
942 unsigned long src = *ppos;
943 int ret = 0;
944 struct mm_struct *mm = file->private_data;
945 unsigned long env_start, env_end;
946
947 /* Ensure the process spawned far enough to have an environment. */
948 if (!mm || !mm->env_end)
949 return 0;
950
951 page = (char *)__get_free_page(GFP_TEMPORARY);
952 if (!page)
953 return -ENOMEM;
954
955 ret = 0;
956 if (!atomic_inc_not_zero(&mm->mm_users))
957 goto free;
958
959 down_read(&mm->mmap_sem);
960 env_start = mm->env_start;
961 env_end = mm->env_end;
962 up_read(&mm->mmap_sem);
963
964 while (count > 0) {
965 size_t this_len, max_len;
966 int retval;
967
968 if (src >= (env_end - env_start))
969 break;
970
971 this_len = env_end - (env_start + src);
972
973 max_len = min_t(size_t, PAGE_SIZE, count);
974 this_len = min(max_len, this_len);
975
976 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
977
978 if (retval <= 0) {
979 ret = retval;
980 break;
981 }
982
983 if (copy_to_user(buf, page, retval)) {
984 ret = -EFAULT;
985 break;
986 }
987
988 ret += retval;
989 src += retval;
990 buf += retval;
991 count -= retval;
992 }
993 *ppos = src;
994 mmput(mm);
995
996 free:
997 free_page((unsigned long) page);
998 return ret;
999 }
1000
1001 static const struct file_operations proc_environ_operations = {
1002 .open = environ_open,
1003 .read = environ_read,
1004 .llseek = generic_file_llseek,
1005 .release = mem_release,
1006 };
1007
1008 static int auxv_open(struct inode *inode, struct file *file)
1009 {
1010 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1011 }
1012
1013 static ssize_t auxv_read(struct file *file, char __user *buf,
1014 size_t count, loff_t *ppos)
1015 {
1016 struct mm_struct *mm = file->private_data;
1017 unsigned int nwords = 0;
1018
1019 if (!mm)
1020 return 0;
1021 do {
1022 nwords += 2;
1023 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1024 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1025 nwords * sizeof(mm->saved_auxv[0]));
1026 }
1027
1028 static const struct file_operations proc_auxv_operations = {
1029 .open = auxv_open,
1030 .read = auxv_read,
1031 .llseek = generic_file_llseek,
1032 .release = mem_release,
1033 };
1034
1035 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1036 loff_t *ppos)
1037 {
1038 struct task_struct *task = get_proc_task(file_inode(file));
1039 char buffer[PROC_NUMBUF];
1040 int oom_adj = OOM_ADJUST_MIN;
1041 size_t len;
1042
1043 if (!task)
1044 return -ESRCH;
1045 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1046 oom_adj = OOM_ADJUST_MAX;
1047 else
1048 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1049 OOM_SCORE_ADJ_MAX;
1050 put_task_struct(task);
1051 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1052 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1053 }
1054
1055 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1056 {
1057 static DEFINE_MUTEX(oom_adj_mutex);
1058 struct mm_struct *mm = NULL;
1059 struct task_struct *task;
1060 int err = 0;
1061
1062 task = get_proc_task(file_inode(file));
1063 if (!task)
1064 return -ESRCH;
1065
1066 mutex_lock(&oom_adj_mutex);
1067 if (legacy) {
1068 if (oom_adj < task->signal->oom_score_adj &&
1069 !capable(CAP_SYS_RESOURCE)) {
1070 err = -EACCES;
1071 goto err_unlock;
1072 }
1073 /*
1074 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1075 * /proc/pid/oom_score_adj instead.
1076 */
1077 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1078 current->comm, task_pid_nr(current), task_pid_nr(task),
1079 task_pid_nr(task));
1080 } else {
1081 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1082 !capable(CAP_SYS_RESOURCE)) {
1083 err = -EACCES;
1084 goto err_unlock;
1085 }
1086 }
1087
1088 /*
1089 * Make sure we will check other processes sharing the mm if this is
1090 * not vfrok which wants its own oom_score_adj.
1091 * pin the mm so it doesn't go away and get reused after task_unlock
1092 */
1093 if (!task->vfork_done) {
1094 struct task_struct *p = find_lock_task_mm(task);
1095
1096 if (p) {
1097 if (atomic_read(&p->mm->mm_users) > 1) {
1098 mm = p->mm;
1099 atomic_inc(&mm->mm_count);
1100 }
1101 task_unlock(p);
1102 }
1103 }
1104
1105 task->signal->oom_score_adj = oom_adj;
1106 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1107 task->signal->oom_score_adj_min = (short)oom_adj;
1108 trace_oom_score_adj_update(task);
1109
1110 if (mm) {
1111 struct task_struct *p;
1112
1113 rcu_read_lock();
1114 for_each_process(p) {
1115 if (same_thread_group(task, p))
1116 continue;
1117
1118 /* do not touch kernel threads or the global init */
1119 if (p->flags & PF_KTHREAD || is_global_init(p))
1120 continue;
1121
1122 task_lock(p);
1123 if (!p->vfork_done && process_shares_mm(p, mm)) {
1124 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1125 task_pid_nr(p), p->comm,
1126 p->signal->oom_score_adj, oom_adj,
1127 task_pid_nr(task), task->comm);
1128 p->signal->oom_score_adj = oom_adj;
1129 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1130 p->signal->oom_score_adj_min = (short)oom_adj;
1131 }
1132 task_unlock(p);
1133 }
1134 rcu_read_unlock();
1135 mmdrop(mm);
1136 }
1137 err_unlock:
1138 mutex_unlock(&oom_adj_mutex);
1139 put_task_struct(task);
1140 return err;
1141 }
1142
1143 /*
1144 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1145 * kernels. The effective policy is defined by oom_score_adj, which has a
1146 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1147 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1148 * Processes that become oom disabled via oom_adj will still be oom disabled
1149 * with this implementation.
1150 *
1151 * oom_adj cannot be removed since existing userspace binaries use it.
1152 */
1153 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1154 size_t count, loff_t *ppos)
1155 {
1156 char buffer[PROC_NUMBUF];
1157 int oom_adj;
1158 int err;
1159
1160 memset(buffer, 0, sizeof(buffer));
1161 if (count > sizeof(buffer) - 1)
1162 count = sizeof(buffer) - 1;
1163 if (copy_from_user(buffer, buf, count)) {
1164 err = -EFAULT;
1165 goto out;
1166 }
1167
1168 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1169 if (err)
1170 goto out;
1171 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1172 oom_adj != OOM_DISABLE) {
1173 err = -EINVAL;
1174 goto out;
1175 }
1176
1177 /*
1178 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1179 * value is always attainable.
1180 */
1181 if (oom_adj == OOM_ADJUST_MAX)
1182 oom_adj = OOM_SCORE_ADJ_MAX;
1183 else
1184 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1185
1186 err = __set_oom_adj(file, oom_adj, true);
1187 out:
1188 return err < 0 ? err : count;
1189 }
1190
1191 static const struct file_operations proc_oom_adj_operations = {
1192 .read = oom_adj_read,
1193 .write = oom_adj_write,
1194 .llseek = generic_file_llseek,
1195 };
1196
1197 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1198 size_t count, loff_t *ppos)
1199 {
1200 struct task_struct *task = get_proc_task(file_inode(file));
1201 char buffer[PROC_NUMBUF];
1202 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1203 size_t len;
1204
1205 if (!task)
1206 return -ESRCH;
1207 oom_score_adj = task->signal->oom_score_adj;
1208 put_task_struct(task);
1209 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1210 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1211 }
1212
1213 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1214 size_t count, loff_t *ppos)
1215 {
1216 char buffer[PROC_NUMBUF];
1217 int oom_score_adj;
1218 int err;
1219
1220 memset(buffer, 0, sizeof(buffer));
1221 if (count > sizeof(buffer) - 1)
1222 count = sizeof(buffer) - 1;
1223 if (copy_from_user(buffer, buf, count)) {
1224 err = -EFAULT;
1225 goto out;
1226 }
1227
1228 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1229 if (err)
1230 goto out;
1231 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1232 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1233 err = -EINVAL;
1234 goto out;
1235 }
1236
1237 err = __set_oom_adj(file, oom_score_adj, false);
1238 out:
1239 return err < 0 ? err : count;
1240 }
1241
1242 static const struct file_operations proc_oom_score_adj_operations = {
1243 .read = oom_score_adj_read,
1244 .write = oom_score_adj_write,
1245 .llseek = default_llseek,
1246 };
1247
1248 #ifdef CONFIG_AUDITSYSCALL
1249 #define TMPBUFLEN 21
1250 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1251 size_t count, loff_t *ppos)
1252 {
1253 struct inode * inode = file_inode(file);
1254 struct task_struct *task = get_proc_task(inode);
1255 ssize_t length;
1256 char tmpbuf[TMPBUFLEN];
1257
1258 if (!task)
1259 return -ESRCH;
1260 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1261 from_kuid(file->f_cred->user_ns,
1262 audit_get_loginuid(task)));
1263 put_task_struct(task);
1264 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1265 }
1266
1267 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1268 size_t count, loff_t *ppos)
1269 {
1270 struct inode * inode = file_inode(file);
1271 uid_t loginuid;
1272 kuid_t kloginuid;
1273 int rv;
1274
1275 rcu_read_lock();
1276 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1277 rcu_read_unlock();
1278 return -EPERM;
1279 }
1280 rcu_read_unlock();
1281
1282 if (*ppos != 0) {
1283 /* No partial writes. */
1284 return -EINVAL;
1285 }
1286
1287 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1288 if (rv < 0)
1289 return rv;
1290
1291 /* is userspace tring to explicitly UNSET the loginuid? */
1292 if (loginuid == AUDIT_UID_UNSET) {
1293 kloginuid = INVALID_UID;
1294 } else {
1295 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1296 if (!uid_valid(kloginuid))
1297 return -EINVAL;
1298 }
1299
1300 rv = audit_set_loginuid(kloginuid);
1301 if (rv < 0)
1302 return rv;
1303 return count;
1304 }
1305
1306 static const struct file_operations proc_loginuid_operations = {
1307 .read = proc_loginuid_read,
1308 .write = proc_loginuid_write,
1309 .llseek = generic_file_llseek,
1310 };
1311
1312 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1313 size_t count, loff_t *ppos)
1314 {
1315 struct inode * inode = file_inode(file);
1316 struct task_struct *task = get_proc_task(inode);
1317 ssize_t length;
1318 char tmpbuf[TMPBUFLEN];
1319
1320 if (!task)
1321 return -ESRCH;
1322 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1323 audit_get_sessionid(task));
1324 put_task_struct(task);
1325 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1326 }
1327
1328 static const struct file_operations proc_sessionid_operations = {
1329 .read = proc_sessionid_read,
1330 .llseek = generic_file_llseek,
1331 };
1332 #endif
1333
1334 #ifdef CONFIG_FAULT_INJECTION
1335 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1336 size_t count, loff_t *ppos)
1337 {
1338 struct task_struct *task = get_proc_task(file_inode(file));
1339 char buffer[PROC_NUMBUF];
1340 size_t len;
1341 int make_it_fail;
1342
1343 if (!task)
1344 return -ESRCH;
1345 make_it_fail = task->make_it_fail;
1346 put_task_struct(task);
1347
1348 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1349
1350 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1351 }
1352
1353 static ssize_t proc_fault_inject_write(struct file * file,
1354 const char __user * buf, size_t count, loff_t *ppos)
1355 {
1356 struct task_struct *task;
1357 char buffer[PROC_NUMBUF];
1358 int make_it_fail;
1359 int rv;
1360
1361 if (!capable(CAP_SYS_RESOURCE))
1362 return -EPERM;
1363 memset(buffer, 0, sizeof(buffer));
1364 if (count > sizeof(buffer) - 1)
1365 count = sizeof(buffer) - 1;
1366 if (copy_from_user(buffer, buf, count))
1367 return -EFAULT;
1368 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1369 if (rv < 0)
1370 return rv;
1371 if (make_it_fail < 0 || make_it_fail > 1)
1372 return -EINVAL;
1373
1374 task = get_proc_task(file_inode(file));
1375 if (!task)
1376 return -ESRCH;
1377 task->make_it_fail = make_it_fail;
1378 put_task_struct(task);
1379
1380 return count;
1381 }
1382
1383 static const struct file_operations proc_fault_inject_operations = {
1384 .read = proc_fault_inject_read,
1385 .write = proc_fault_inject_write,
1386 .llseek = generic_file_llseek,
1387 };
1388 #endif
1389
1390
1391 #ifdef CONFIG_SCHED_DEBUG
1392 /*
1393 * Print out various scheduling related per-task fields:
1394 */
1395 static int sched_show(struct seq_file *m, void *v)
1396 {
1397 struct inode *inode = m->private;
1398 struct task_struct *p;
1399
1400 p = get_proc_task(inode);
1401 if (!p)
1402 return -ESRCH;
1403 proc_sched_show_task(p, m);
1404
1405 put_task_struct(p);
1406
1407 return 0;
1408 }
1409
1410 static ssize_t
1411 sched_write(struct file *file, const char __user *buf,
1412 size_t count, loff_t *offset)
1413 {
1414 struct inode *inode = file_inode(file);
1415 struct task_struct *p;
1416
1417 p = get_proc_task(inode);
1418 if (!p)
1419 return -ESRCH;
1420 proc_sched_set_task(p);
1421
1422 put_task_struct(p);
1423
1424 return count;
1425 }
1426
1427 static int sched_open(struct inode *inode, struct file *filp)
1428 {
1429 return single_open(filp, sched_show, inode);
1430 }
1431
1432 static const struct file_operations proc_pid_sched_operations = {
1433 .open = sched_open,
1434 .read = seq_read,
1435 .write = sched_write,
1436 .llseek = seq_lseek,
1437 .release = single_release,
1438 };
1439
1440 #endif
1441
1442 #ifdef CONFIG_SCHED_AUTOGROUP
1443 /*
1444 * Print out autogroup related information:
1445 */
1446 static int sched_autogroup_show(struct seq_file *m, void *v)
1447 {
1448 struct inode *inode = m->private;
1449 struct task_struct *p;
1450
1451 p = get_proc_task(inode);
1452 if (!p)
1453 return -ESRCH;
1454 proc_sched_autogroup_show_task(p, m);
1455
1456 put_task_struct(p);
1457
1458 return 0;
1459 }
1460
1461 static ssize_t
1462 sched_autogroup_write(struct file *file, const char __user *buf,
1463 size_t count, loff_t *offset)
1464 {
1465 struct inode *inode = file_inode(file);
1466 struct task_struct *p;
1467 char buffer[PROC_NUMBUF];
1468 int nice;
1469 int err;
1470
1471 memset(buffer, 0, sizeof(buffer));
1472 if (count > sizeof(buffer) - 1)
1473 count = sizeof(buffer) - 1;
1474 if (copy_from_user(buffer, buf, count))
1475 return -EFAULT;
1476
1477 err = kstrtoint(strstrip(buffer), 0, &nice);
1478 if (err < 0)
1479 return err;
1480
1481 p = get_proc_task(inode);
1482 if (!p)
1483 return -ESRCH;
1484
1485 err = proc_sched_autogroup_set_nice(p, nice);
1486 if (err)
1487 count = err;
1488
1489 put_task_struct(p);
1490
1491 return count;
1492 }
1493
1494 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1495 {
1496 int ret;
1497
1498 ret = single_open(filp, sched_autogroup_show, NULL);
1499 if (!ret) {
1500 struct seq_file *m = filp->private_data;
1501
1502 m->private = inode;
1503 }
1504 return ret;
1505 }
1506
1507 static const struct file_operations proc_pid_sched_autogroup_operations = {
1508 .open = sched_autogroup_open,
1509 .read = seq_read,
1510 .write = sched_autogroup_write,
1511 .llseek = seq_lseek,
1512 .release = single_release,
1513 };
1514
1515 #endif /* CONFIG_SCHED_AUTOGROUP */
1516
1517 static ssize_t comm_write(struct file *file, const char __user *buf,
1518 size_t count, loff_t *offset)
1519 {
1520 struct inode *inode = file_inode(file);
1521 struct task_struct *p;
1522 char buffer[TASK_COMM_LEN];
1523 const size_t maxlen = sizeof(buffer) - 1;
1524
1525 memset(buffer, 0, sizeof(buffer));
1526 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1527 return -EFAULT;
1528
1529 p = get_proc_task(inode);
1530 if (!p)
1531 return -ESRCH;
1532
1533 if (same_thread_group(current, p))
1534 set_task_comm(p, buffer);
1535 else
1536 count = -EINVAL;
1537
1538 put_task_struct(p);
1539
1540 return count;
1541 }
1542
1543 static int comm_show(struct seq_file *m, void *v)
1544 {
1545 struct inode *inode = m->private;
1546 struct task_struct *p;
1547
1548 p = get_proc_task(inode);
1549 if (!p)
1550 return -ESRCH;
1551
1552 task_lock(p);
1553 seq_printf(m, "%s\n", p->comm);
1554 task_unlock(p);
1555
1556 put_task_struct(p);
1557
1558 return 0;
1559 }
1560
1561 static int comm_open(struct inode *inode, struct file *filp)
1562 {
1563 return single_open(filp, comm_show, inode);
1564 }
1565
1566 static const struct file_operations proc_pid_set_comm_operations = {
1567 .open = comm_open,
1568 .read = seq_read,
1569 .write = comm_write,
1570 .llseek = seq_lseek,
1571 .release = single_release,
1572 };
1573
1574 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1575 {
1576 struct task_struct *task;
1577 struct file *exe_file;
1578
1579 task = get_proc_task(d_inode(dentry));
1580 if (!task)
1581 return -ENOENT;
1582 exe_file = get_task_exe_file(task);
1583 put_task_struct(task);
1584 if (exe_file) {
1585 *exe_path = exe_file->f_path;
1586 path_get(&exe_file->f_path);
1587 fput(exe_file);
1588 return 0;
1589 } else
1590 return -ENOENT;
1591 }
1592
1593 static const char *proc_pid_get_link(struct dentry *dentry,
1594 struct inode *inode,
1595 struct delayed_call *done)
1596 {
1597 struct path path;
1598 int error = -EACCES;
1599
1600 if (!dentry)
1601 return ERR_PTR(-ECHILD);
1602
1603 /* Are we allowed to snoop on the tasks file descriptors? */
1604 if (!proc_fd_access_allowed(inode))
1605 goto out;
1606
1607 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1608 if (error)
1609 goto out;
1610
1611 nd_jump_link(&path);
1612 return NULL;
1613 out:
1614 return ERR_PTR(error);
1615 }
1616
1617 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1618 {
1619 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1620 char *pathname;
1621 int len;
1622
1623 if (!tmp)
1624 return -ENOMEM;
1625
1626 pathname = d_path(path, tmp, PAGE_SIZE);
1627 len = PTR_ERR(pathname);
1628 if (IS_ERR(pathname))
1629 goto out;
1630 len = tmp + PAGE_SIZE - 1 - pathname;
1631
1632 if (len > buflen)
1633 len = buflen;
1634 if (copy_to_user(buffer, pathname, len))
1635 len = -EFAULT;
1636 out:
1637 free_page((unsigned long)tmp);
1638 return len;
1639 }
1640
1641 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1642 {
1643 int error = -EACCES;
1644 struct inode *inode = d_inode(dentry);
1645 struct path path;
1646
1647 /* Are we allowed to snoop on the tasks file descriptors? */
1648 if (!proc_fd_access_allowed(inode))
1649 goto out;
1650
1651 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1652 if (error)
1653 goto out;
1654
1655 error = do_proc_readlink(&path, buffer, buflen);
1656 path_put(&path);
1657 out:
1658 return error;
1659 }
1660
1661 const struct inode_operations proc_pid_link_inode_operations = {
1662 .readlink = proc_pid_readlink,
1663 .get_link = proc_pid_get_link,
1664 .setattr = proc_setattr,
1665 };
1666
1667
1668 /* building an inode */
1669
1670 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1671 {
1672 struct inode * inode;
1673 struct proc_inode *ei;
1674 const struct cred *cred;
1675
1676 /* We need a new inode */
1677
1678 inode = new_inode(sb);
1679 if (!inode)
1680 goto out;
1681
1682 /* Common stuff */
1683 ei = PROC_I(inode);
1684 inode->i_ino = get_next_ino();
1685 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1686 inode->i_op = &proc_def_inode_operations;
1687
1688 /*
1689 * grab the reference to task.
1690 */
1691 ei->pid = get_task_pid(task, PIDTYPE_PID);
1692 if (!ei->pid)
1693 goto out_unlock;
1694
1695 if (task_dumpable(task)) {
1696 rcu_read_lock();
1697 cred = __task_cred(task);
1698 inode->i_uid = cred->euid;
1699 inode->i_gid = cred->egid;
1700 rcu_read_unlock();
1701 }
1702 security_task_to_inode(task, inode);
1703
1704 out:
1705 return inode;
1706
1707 out_unlock:
1708 iput(inode);
1709 return NULL;
1710 }
1711
1712 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1713 {
1714 struct inode *inode = d_inode(dentry);
1715 struct task_struct *task;
1716 const struct cred *cred;
1717 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1718
1719 generic_fillattr(inode, stat);
1720
1721 rcu_read_lock();
1722 stat->uid = GLOBAL_ROOT_UID;
1723 stat->gid = GLOBAL_ROOT_GID;
1724 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1725 if (task) {
1726 if (!has_pid_permissions(pid, task, 2)) {
1727 rcu_read_unlock();
1728 /*
1729 * This doesn't prevent learning whether PID exists,
1730 * it only makes getattr() consistent with readdir().
1731 */
1732 return -ENOENT;
1733 }
1734 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1735 task_dumpable(task)) {
1736 cred = __task_cred(task);
1737 stat->uid = cred->euid;
1738 stat->gid = cred->egid;
1739 }
1740 }
1741 rcu_read_unlock();
1742 return 0;
1743 }
1744
1745 /* dentry stuff */
1746
1747 /*
1748 * Exceptional case: normally we are not allowed to unhash a busy
1749 * directory. In this case, however, we can do it - no aliasing problems
1750 * due to the way we treat inodes.
1751 *
1752 * Rewrite the inode's ownerships here because the owning task may have
1753 * performed a setuid(), etc.
1754 *
1755 * Before the /proc/pid/status file was created the only way to read
1756 * the effective uid of a /process was to stat /proc/pid. Reading
1757 * /proc/pid/status is slow enough that procps and other packages
1758 * kept stating /proc/pid. To keep the rules in /proc simple I have
1759 * made this apply to all per process world readable and executable
1760 * directories.
1761 */
1762 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1763 {
1764 struct inode *inode;
1765 struct task_struct *task;
1766 const struct cred *cred;
1767
1768 if (flags & LOOKUP_RCU)
1769 return -ECHILD;
1770
1771 inode = d_inode(dentry);
1772 task = get_proc_task(inode);
1773
1774 if (task) {
1775 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1776 task_dumpable(task)) {
1777 rcu_read_lock();
1778 cred = __task_cred(task);
1779 inode->i_uid = cred->euid;
1780 inode->i_gid = cred->egid;
1781 rcu_read_unlock();
1782 } else {
1783 inode->i_uid = GLOBAL_ROOT_UID;
1784 inode->i_gid = GLOBAL_ROOT_GID;
1785 }
1786 inode->i_mode &= ~(S_ISUID | S_ISGID);
1787 security_task_to_inode(task, inode);
1788 put_task_struct(task);
1789 return 1;
1790 }
1791 return 0;
1792 }
1793
1794 static inline bool proc_inode_is_dead(struct inode *inode)
1795 {
1796 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1797 }
1798
1799 int pid_delete_dentry(const struct dentry *dentry)
1800 {
1801 /* Is the task we represent dead?
1802 * If so, then don't put the dentry on the lru list,
1803 * kill it immediately.
1804 */
1805 return proc_inode_is_dead(d_inode(dentry));
1806 }
1807
1808 const struct dentry_operations pid_dentry_operations =
1809 {
1810 .d_revalidate = pid_revalidate,
1811 .d_delete = pid_delete_dentry,
1812 };
1813
1814 /* Lookups */
1815
1816 /*
1817 * Fill a directory entry.
1818 *
1819 * If possible create the dcache entry and derive our inode number and
1820 * file type from dcache entry.
1821 *
1822 * Since all of the proc inode numbers are dynamically generated, the inode
1823 * numbers do not exist until the inode is cache. This means creating the
1824 * the dcache entry in readdir is necessary to keep the inode numbers
1825 * reported by readdir in sync with the inode numbers reported
1826 * by stat.
1827 */
1828 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1829 const char *name, int len,
1830 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1831 {
1832 struct dentry *child, *dir = file->f_path.dentry;
1833 struct qstr qname = QSTR_INIT(name, len);
1834 struct inode *inode;
1835 unsigned type;
1836 ino_t ino;
1837
1838 child = d_hash_and_lookup(dir, &qname);
1839 if (!child) {
1840 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1841 child = d_alloc_parallel(dir, &qname, &wq);
1842 if (IS_ERR(child))
1843 goto end_instantiate;
1844 if (d_in_lookup(child)) {
1845 int err = instantiate(d_inode(dir), child, task, ptr);
1846 d_lookup_done(child);
1847 if (err < 0) {
1848 dput(child);
1849 goto end_instantiate;
1850 }
1851 }
1852 }
1853 inode = d_inode(child);
1854 ino = inode->i_ino;
1855 type = inode->i_mode >> 12;
1856 dput(child);
1857 return dir_emit(ctx, name, len, ino, type);
1858
1859 end_instantiate:
1860 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1861 }
1862
1863 /*
1864 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1865 * which represent vma start and end addresses.
1866 */
1867 static int dname_to_vma_addr(struct dentry *dentry,
1868 unsigned long *start, unsigned long *end)
1869 {
1870 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1871 return -EINVAL;
1872
1873 return 0;
1874 }
1875
1876 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1877 {
1878 unsigned long vm_start, vm_end;
1879 bool exact_vma_exists = false;
1880 struct mm_struct *mm = NULL;
1881 struct task_struct *task;
1882 const struct cred *cred;
1883 struct inode *inode;
1884 int status = 0;
1885
1886 if (flags & LOOKUP_RCU)
1887 return -ECHILD;
1888
1889 inode = d_inode(dentry);
1890 task = get_proc_task(inode);
1891 if (!task)
1892 goto out_notask;
1893
1894 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1895 if (IS_ERR_OR_NULL(mm))
1896 goto out;
1897
1898 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1899 down_read(&mm->mmap_sem);
1900 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1901 up_read(&mm->mmap_sem);
1902 }
1903
1904 mmput(mm);
1905
1906 if (exact_vma_exists) {
1907 if (task_dumpable(task)) {
1908 rcu_read_lock();
1909 cred = __task_cred(task);
1910 inode->i_uid = cred->euid;
1911 inode->i_gid = cred->egid;
1912 rcu_read_unlock();
1913 } else {
1914 inode->i_uid = GLOBAL_ROOT_UID;
1915 inode->i_gid = GLOBAL_ROOT_GID;
1916 }
1917 security_task_to_inode(task, inode);
1918 status = 1;
1919 }
1920
1921 out:
1922 put_task_struct(task);
1923
1924 out_notask:
1925 return status;
1926 }
1927
1928 static const struct dentry_operations tid_map_files_dentry_operations = {
1929 .d_revalidate = map_files_d_revalidate,
1930 .d_delete = pid_delete_dentry,
1931 };
1932
1933 static int map_files_get_link(struct dentry *dentry, struct path *path)
1934 {
1935 unsigned long vm_start, vm_end;
1936 struct vm_area_struct *vma;
1937 struct task_struct *task;
1938 struct mm_struct *mm;
1939 int rc;
1940
1941 rc = -ENOENT;
1942 task = get_proc_task(d_inode(dentry));
1943 if (!task)
1944 goto out;
1945
1946 mm = get_task_mm(task);
1947 put_task_struct(task);
1948 if (!mm)
1949 goto out;
1950
1951 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1952 if (rc)
1953 goto out_mmput;
1954
1955 rc = -ENOENT;
1956 down_read(&mm->mmap_sem);
1957 vma = find_exact_vma(mm, vm_start, vm_end);
1958 if (vma && vma->vm_file) {
1959 *path = vma->vm_file->f_path;
1960 path_get(path);
1961 rc = 0;
1962 }
1963 up_read(&mm->mmap_sem);
1964
1965 out_mmput:
1966 mmput(mm);
1967 out:
1968 return rc;
1969 }
1970
1971 struct map_files_info {
1972 fmode_t mode;
1973 unsigned int len;
1974 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1975 };
1976
1977 /*
1978 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1979 * symlinks may be used to bypass permissions on ancestor directories in the
1980 * path to the file in question.
1981 */
1982 static const char *
1983 proc_map_files_get_link(struct dentry *dentry,
1984 struct inode *inode,
1985 struct delayed_call *done)
1986 {
1987 if (!capable(CAP_SYS_ADMIN))
1988 return ERR_PTR(-EPERM);
1989
1990 return proc_pid_get_link(dentry, inode, done);
1991 }
1992
1993 /*
1994 * Identical to proc_pid_link_inode_operations except for get_link()
1995 */
1996 static const struct inode_operations proc_map_files_link_inode_operations = {
1997 .readlink = proc_pid_readlink,
1998 .get_link = proc_map_files_get_link,
1999 .setattr = proc_setattr,
2000 };
2001
2002 static int
2003 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2004 struct task_struct *task, const void *ptr)
2005 {
2006 fmode_t mode = (fmode_t)(unsigned long)ptr;
2007 struct proc_inode *ei;
2008 struct inode *inode;
2009
2010 inode = proc_pid_make_inode(dir->i_sb, task);
2011 if (!inode)
2012 return -ENOENT;
2013
2014 ei = PROC_I(inode);
2015 ei->op.proc_get_link = map_files_get_link;
2016
2017 inode->i_op = &proc_map_files_link_inode_operations;
2018 inode->i_size = 64;
2019 inode->i_mode = S_IFLNK;
2020
2021 if (mode & FMODE_READ)
2022 inode->i_mode |= S_IRUSR;
2023 if (mode & FMODE_WRITE)
2024 inode->i_mode |= S_IWUSR;
2025
2026 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2027 d_add(dentry, inode);
2028
2029 return 0;
2030 }
2031
2032 static struct dentry *proc_map_files_lookup(struct inode *dir,
2033 struct dentry *dentry, unsigned int flags)
2034 {
2035 unsigned long vm_start, vm_end;
2036 struct vm_area_struct *vma;
2037 struct task_struct *task;
2038 int result;
2039 struct mm_struct *mm;
2040
2041 result = -ENOENT;
2042 task = get_proc_task(dir);
2043 if (!task)
2044 goto out;
2045
2046 result = -EACCES;
2047 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2048 goto out_put_task;
2049
2050 result = -ENOENT;
2051 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2052 goto out_put_task;
2053
2054 mm = get_task_mm(task);
2055 if (!mm)
2056 goto out_put_task;
2057
2058 down_read(&mm->mmap_sem);
2059 vma = find_exact_vma(mm, vm_start, vm_end);
2060 if (!vma)
2061 goto out_no_vma;
2062
2063 if (vma->vm_file)
2064 result = proc_map_files_instantiate(dir, dentry, task,
2065 (void *)(unsigned long)vma->vm_file->f_mode);
2066
2067 out_no_vma:
2068 up_read(&mm->mmap_sem);
2069 mmput(mm);
2070 out_put_task:
2071 put_task_struct(task);
2072 out:
2073 return ERR_PTR(result);
2074 }
2075
2076 static const struct inode_operations proc_map_files_inode_operations = {
2077 .lookup = proc_map_files_lookup,
2078 .permission = proc_fd_permission,
2079 .setattr = proc_setattr,
2080 };
2081
2082 static int
2083 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2084 {
2085 struct vm_area_struct *vma;
2086 struct task_struct *task;
2087 struct mm_struct *mm;
2088 unsigned long nr_files, pos, i;
2089 struct flex_array *fa = NULL;
2090 struct map_files_info info;
2091 struct map_files_info *p;
2092 int ret;
2093
2094 ret = -ENOENT;
2095 task = get_proc_task(file_inode(file));
2096 if (!task)
2097 goto out;
2098
2099 ret = -EACCES;
2100 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2101 goto out_put_task;
2102
2103 ret = 0;
2104 if (!dir_emit_dots(file, ctx))
2105 goto out_put_task;
2106
2107 mm = get_task_mm(task);
2108 if (!mm)
2109 goto out_put_task;
2110 down_read(&mm->mmap_sem);
2111
2112 nr_files = 0;
2113
2114 /*
2115 * We need two passes here:
2116 *
2117 * 1) Collect vmas of mapped files with mmap_sem taken
2118 * 2) Release mmap_sem and instantiate entries
2119 *
2120 * otherwise we get lockdep complained, since filldir()
2121 * routine might require mmap_sem taken in might_fault().
2122 */
2123
2124 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2125 if (vma->vm_file && ++pos > ctx->pos)
2126 nr_files++;
2127 }
2128
2129 if (nr_files) {
2130 fa = flex_array_alloc(sizeof(info), nr_files,
2131 GFP_KERNEL);
2132 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2133 GFP_KERNEL)) {
2134 ret = -ENOMEM;
2135 if (fa)
2136 flex_array_free(fa);
2137 up_read(&mm->mmap_sem);
2138 mmput(mm);
2139 goto out_put_task;
2140 }
2141 for (i = 0, vma = mm->mmap, pos = 2; vma;
2142 vma = vma->vm_next) {
2143 if (!vma->vm_file)
2144 continue;
2145 if (++pos <= ctx->pos)
2146 continue;
2147
2148 info.mode = vma->vm_file->f_mode;
2149 info.len = snprintf(info.name,
2150 sizeof(info.name), "%lx-%lx",
2151 vma->vm_start, vma->vm_end);
2152 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2153 BUG();
2154 }
2155 }
2156 up_read(&mm->mmap_sem);
2157
2158 for (i = 0; i < nr_files; i++) {
2159 p = flex_array_get(fa, i);
2160 if (!proc_fill_cache(file, ctx,
2161 p->name, p->len,
2162 proc_map_files_instantiate,
2163 task,
2164 (void *)(unsigned long)p->mode))
2165 break;
2166 ctx->pos++;
2167 }
2168 if (fa)
2169 flex_array_free(fa);
2170 mmput(mm);
2171
2172 out_put_task:
2173 put_task_struct(task);
2174 out:
2175 return ret;
2176 }
2177
2178 static const struct file_operations proc_map_files_operations = {
2179 .read = generic_read_dir,
2180 .iterate_shared = proc_map_files_readdir,
2181 .llseek = generic_file_llseek,
2182 };
2183
2184 #ifdef CONFIG_CHECKPOINT_RESTORE
2185 struct timers_private {
2186 struct pid *pid;
2187 struct task_struct *task;
2188 struct sighand_struct *sighand;
2189 struct pid_namespace *ns;
2190 unsigned long flags;
2191 };
2192
2193 static void *timers_start(struct seq_file *m, loff_t *pos)
2194 {
2195 struct timers_private *tp = m->private;
2196
2197 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2198 if (!tp->task)
2199 return ERR_PTR(-ESRCH);
2200
2201 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2202 if (!tp->sighand)
2203 return ERR_PTR(-ESRCH);
2204
2205 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2206 }
2207
2208 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2209 {
2210 struct timers_private *tp = m->private;
2211 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2212 }
2213
2214 static void timers_stop(struct seq_file *m, void *v)
2215 {
2216 struct timers_private *tp = m->private;
2217
2218 if (tp->sighand) {
2219 unlock_task_sighand(tp->task, &tp->flags);
2220 tp->sighand = NULL;
2221 }
2222
2223 if (tp->task) {
2224 put_task_struct(tp->task);
2225 tp->task = NULL;
2226 }
2227 }
2228
2229 static int show_timer(struct seq_file *m, void *v)
2230 {
2231 struct k_itimer *timer;
2232 struct timers_private *tp = m->private;
2233 int notify;
2234 static const char * const nstr[] = {
2235 [SIGEV_SIGNAL] = "signal",
2236 [SIGEV_NONE] = "none",
2237 [SIGEV_THREAD] = "thread",
2238 };
2239
2240 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2241 notify = timer->it_sigev_notify;
2242
2243 seq_printf(m, "ID: %d\n", timer->it_id);
2244 seq_printf(m, "signal: %d/%p\n",
2245 timer->sigq->info.si_signo,
2246 timer->sigq->info.si_value.sival_ptr);
2247 seq_printf(m, "notify: %s/%s.%d\n",
2248 nstr[notify & ~SIGEV_THREAD_ID],
2249 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2250 pid_nr_ns(timer->it_pid, tp->ns));
2251 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2252
2253 return 0;
2254 }
2255
2256 static const struct seq_operations proc_timers_seq_ops = {
2257 .start = timers_start,
2258 .next = timers_next,
2259 .stop = timers_stop,
2260 .show = show_timer,
2261 };
2262
2263 static int proc_timers_open(struct inode *inode, struct file *file)
2264 {
2265 struct timers_private *tp;
2266
2267 tp = __seq_open_private(file, &proc_timers_seq_ops,
2268 sizeof(struct timers_private));
2269 if (!tp)
2270 return -ENOMEM;
2271
2272 tp->pid = proc_pid(inode);
2273 tp->ns = inode->i_sb->s_fs_info;
2274 return 0;
2275 }
2276
2277 static const struct file_operations proc_timers_operations = {
2278 .open = proc_timers_open,
2279 .read = seq_read,
2280 .llseek = seq_lseek,
2281 .release = seq_release_private,
2282 };
2283 #endif
2284
2285 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2286 size_t count, loff_t *offset)
2287 {
2288 struct inode *inode = file_inode(file);
2289 struct task_struct *p;
2290 u64 slack_ns;
2291 int err;
2292
2293 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2294 if (err < 0)
2295 return err;
2296
2297 p = get_proc_task(inode);
2298 if (!p)
2299 return -ESRCH;
2300
2301 if (p != current) {
2302 if (!capable(CAP_SYS_NICE)) {
2303 count = -EPERM;
2304 goto out;
2305 }
2306
2307 err = security_task_setscheduler(p);
2308 if (err) {
2309 count = err;
2310 goto out;
2311 }
2312 }
2313
2314 task_lock(p);
2315 if (slack_ns == 0)
2316 p->timer_slack_ns = p->default_timer_slack_ns;
2317 else
2318 p->timer_slack_ns = slack_ns;
2319 task_unlock(p);
2320
2321 out:
2322 put_task_struct(p);
2323
2324 return count;
2325 }
2326
2327 static int timerslack_ns_show(struct seq_file *m, void *v)
2328 {
2329 struct inode *inode = m->private;
2330 struct task_struct *p;
2331 int err = 0;
2332
2333 p = get_proc_task(inode);
2334 if (!p)
2335 return -ESRCH;
2336
2337 if (p != current) {
2338
2339 if (!capable(CAP_SYS_NICE)) {
2340 err = -EPERM;
2341 goto out;
2342 }
2343 err = security_task_getscheduler(p);
2344 if (err)
2345 goto out;
2346 }
2347
2348 task_lock(p);
2349 seq_printf(m, "%llu\n", p->timer_slack_ns);
2350 task_unlock(p);
2351
2352 out:
2353 put_task_struct(p);
2354
2355 return err;
2356 }
2357
2358 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2359 {
2360 return single_open(filp, timerslack_ns_show, inode);
2361 }
2362
2363 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2364 .open = timerslack_ns_open,
2365 .read = seq_read,
2366 .write = timerslack_ns_write,
2367 .llseek = seq_lseek,
2368 .release = single_release,
2369 };
2370
2371 static int proc_pident_instantiate(struct inode *dir,
2372 struct dentry *dentry, struct task_struct *task, const void *ptr)
2373 {
2374 const struct pid_entry *p = ptr;
2375 struct inode *inode;
2376 struct proc_inode *ei;
2377
2378 inode = proc_pid_make_inode(dir->i_sb, task);
2379 if (!inode)
2380 goto out;
2381
2382 ei = PROC_I(inode);
2383 inode->i_mode = p->mode;
2384 if (S_ISDIR(inode->i_mode))
2385 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2386 if (p->iop)
2387 inode->i_op = p->iop;
2388 if (p->fop)
2389 inode->i_fop = p->fop;
2390 ei->op = p->op;
2391 d_set_d_op(dentry, &pid_dentry_operations);
2392 d_add(dentry, inode);
2393 /* Close the race of the process dying before we return the dentry */
2394 if (pid_revalidate(dentry, 0))
2395 return 0;
2396 out:
2397 return -ENOENT;
2398 }
2399
2400 static struct dentry *proc_pident_lookup(struct inode *dir,
2401 struct dentry *dentry,
2402 const struct pid_entry *ents,
2403 unsigned int nents)
2404 {
2405 int error;
2406 struct task_struct *task = get_proc_task(dir);
2407 const struct pid_entry *p, *last;
2408
2409 error = -ENOENT;
2410
2411 if (!task)
2412 goto out_no_task;
2413
2414 /*
2415 * Yes, it does not scale. And it should not. Don't add
2416 * new entries into /proc/<tgid>/ without very good reasons.
2417 */
2418 last = &ents[nents];
2419 for (p = ents; p < last; p++) {
2420 if (p->len != dentry->d_name.len)
2421 continue;
2422 if (!memcmp(dentry->d_name.name, p->name, p->len))
2423 break;
2424 }
2425 if (p >= last)
2426 goto out;
2427
2428 error = proc_pident_instantiate(dir, dentry, task, p);
2429 out:
2430 put_task_struct(task);
2431 out_no_task:
2432 return ERR_PTR(error);
2433 }
2434
2435 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2436 const struct pid_entry *ents, unsigned int nents)
2437 {
2438 struct task_struct *task = get_proc_task(file_inode(file));
2439 const struct pid_entry *p;
2440
2441 if (!task)
2442 return -ENOENT;
2443
2444 if (!dir_emit_dots(file, ctx))
2445 goto out;
2446
2447 if (ctx->pos >= nents + 2)
2448 goto out;
2449
2450 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2451 if (!proc_fill_cache(file, ctx, p->name, p->len,
2452 proc_pident_instantiate, task, p))
2453 break;
2454 ctx->pos++;
2455 }
2456 out:
2457 put_task_struct(task);
2458 return 0;
2459 }
2460
2461 #ifdef CONFIG_SECURITY
2462 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2463 size_t count, loff_t *ppos)
2464 {
2465 struct inode * inode = file_inode(file);
2466 char *p = NULL;
2467 ssize_t length;
2468 struct task_struct *task = get_proc_task(inode);
2469
2470 if (!task)
2471 return -ESRCH;
2472
2473 length = security_getprocattr(task,
2474 (char*)file->f_path.dentry->d_name.name,
2475 &p);
2476 put_task_struct(task);
2477 if (length > 0)
2478 length = simple_read_from_buffer(buf, count, ppos, p, length);
2479 kfree(p);
2480 return length;
2481 }
2482
2483 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2484 size_t count, loff_t *ppos)
2485 {
2486 struct inode * inode = file_inode(file);
2487 void *page;
2488 ssize_t length;
2489 struct task_struct *task = get_proc_task(inode);
2490
2491 length = -ESRCH;
2492 if (!task)
2493 goto out_no_task;
2494 if (count > PAGE_SIZE)
2495 count = PAGE_SIZE;
2496
2497 /* No partial writes. */
2498 length = -EINVAL;
2499 if (*ppos != 0)
2500 goto out;
2501
2502 page = memdup_user(buf, count);
2503 if (IS_ERR(page)) {
2504 length = PTR_ERR(page);
2505 goto out;
2506 }
2507
2508 /* Guard against adverse ptrace interaction */
2509 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2510 if (length < 0)
2511 goto out_free;
2512
2513 length = security_setprocattr(task,
2514 (char*)file->f_path.dentry->d_name.name,
2515 page, count);
2516 mutex_unlock(&task->signal->cred_guard_mutex);
2517 out_free:
2518 kfree(page);
2519 out:
2520 put_task_struct(task);
2521 out_no_task:
2522 return length;
2523 }
2524
2525 static const struct file_operations proc_pid_attr_operations = {
2526 .read = proc_pid_attr_read,
2527 .write = proc_pid_attr_write,
2528 .llseek = generic_file_llseek,
2529 };
2530
2531 static const struct pid_entry attr_dir_stuff[] = {
2532 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2533 REG("prev", S_IRUGO, proc_pid_attr_operations),
2534 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2535 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2536 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2537 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2538 };
2539
2540 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2541 {
2542 return proc_pident_readdir(file, ctx,
2543 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2544 }
2545
2546 static const struct file_operations proc_attr_dir_operations = {
2547 .read = generic_read_dir,
2548 .iterate_shared = proc_attr_dir_readdir,
2549 .llseek = generic_file_llseek,
2550 };
2551
2552 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2553 struct dentry *dentry, unsigned int flags)
2554 {
2555 return proc_pident_lookup(dir, dentry,
2556 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2557 }
2558
2559 static const struct inode_operations proc_attr_dir_inode_operations = {
2560 .lookup = proc_attr_dir_lookup,
2561 .getattr = pid_getattr,
2562 .setattr = proc_setattr,
2563 };
2564
2565 #endif
2566
2567 #ifdef CONFIG_ELF_CORE
2568 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2569 size_t count, loff_t *ppos)
2570 {
2571 struct task_struct *task = get_proc_task(file_inode(file));
2572 struct mm_struct *mm;
2573 char buffer[PROC_NUMBUF];
2574 size_t len;
2575 int ret;
2576
2577 if (!task)
2578 return -ESRCH;
2579
2580 ret = 0;
2581 mm = get_task_mm(task);
2582 if (mm) {
2583 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2584 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2585 MMF_DUMP_FILTER_SHIFT));
2586 mmput(mm);
2587 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2588 }
2589
2590 put_task_struct(task);
2591
2592 return ret;
2593 }
2594
2595 static ssize_t proc_coredump_filter_write(struct file *file,
2596 const char __user *buf,
2597 size_t count,
2598 loff_t *ppos)
2599 {
2600 struct task_struct *task;
2601 struct mm_struct *mm;
2602 unsigned int val;
2603 int ret;
2604 int i;
2605 unsigned long mask;
2606
2607 ret = kstrtouint_from_user(buf, count, 0, &val);
2608 if (ret < 0)
2609 return ret;
2610
2611 ret = -ESRCH;
2612 task = get_proc_task(file_inode(file));
2613 if (!task)
2614 goto out_no_task;
2615
2616 mm = get_task_mm(task);
2617 if (!mm)
2618 goto out_no_mm;
2619 ret = 0;
2620
2621 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2622 if (val & mask)
2623 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2624 else
2625 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2626 }
2627
2628 mmput(mm);
2629 out_no_mm:
2630 put_task_struct(task);
2631 out_no_task:
2632 if (ret < 0)
2633 return ret;
2634 return count;
2635 }
2636
2637 static const struct file_operations proc_coredump_filter_operations = {
2638 .read = proc_coredump_filter_read,
2639 .write = proc_coredump_filter_write,
2640 .llseek = generic_file_llseek,
2641 };
2642 #endif
2643
2644 #ifdef CONFIG_TASK_IO_ACCOUNTING
2645 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2646 {
2647 struct task_io_accounting acct = task->ioac;
2648 unsigned long flags;
2649 int result;
2650
2651 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2652 if (result)
2653 return result;
2654
2655 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2656 result = -EACCES;
2657 goto out_unlock;
2658 }
2659
2660 if (whole && lock_task_sighand(task, &flags)) {
2661 struct task_struct *t = task;
2662
2663 task_io_accounting_add(&acct, &task->signal->ioac);
2664 while_each_thread(task, t)
2665 task_io_accounting_add(&acct, &t->ioac);
2666
2667 unlock_task_sighand(task, &flags);
2668 }
2669 seq_printf(m,
2670 "rchar: %llu\n"
2671 "wchar: %llu\n"
2672 "syscr: %llu\n"
2673 "syscw: %llu\n"
2674 "read_bytes: %llu\n"
2675 "write_bytes: %llu\n"
2676 "cancelled_write_bytes: %llu\n",
2677 (unsigned long long)acct.rchar,
2678 (unsigned long long)acct.wchar,
2679 (unsigned long long)acct.syscr,
2680 (unsigned long long)acct.syscw,
2681 (unsigned long long)acct.read_bytes,
2682 (unsigned long long)acct.write_bytes,
2683 (unsigned long long)acct.cancelled_write_bytes);
2684 result = 0;
2685
2686 out_unlock:
2687 mutex_unlock(&task->signal->cred_guard_mutex);
2688 return result;
2689 }
2690
2691 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2692 struct pid *pid, struct task_struct *task)
2693 {
2694 return do_io_accounting(task, m, 0);
2695 }
2696
2697 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2698 struct pid *pid, struct task_struct *task)
2699 {
2700 return do_io_accounting(task, m, 1);
2701 }
2702 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2703
2704 #ifdef CONFIG_USER_NS
2705 static int proc_id_map_open(struct inode *inode, struct file *file,
2706 const struct seq_operations *seq_ops)
2707 {
2708 struct user_namespace *ns = NULL;
2709 struct task_struct *task;
2710 struct seq_file *seq;
2711 int ret = -EINVAL;
2712
2713 task = get_proc_task(inode);
2714 if (task) {
2715 rcu_read_lock();
2716 ns = get_user_ns(task_cred_xxx(task, user_ns));
2717 rcu_read_unlock();
2718 put_task_struct(task);
2719 }
2720 if (!ns)
2721 goto err;
2722
2723 ret = seq_open(file, seq_ops);
2724 if (ret)
2725 goto err_put_ns;
2726
2727 seq = file->private_data;
2728 seq->private = ns;
2729
2730 return 0;
2731 err_put_ns:
2732 put_user_ns(ns);
2733 err:
2734 return ret;
2735 }
2736
2737 static int proc_id_map_release(struct inode *inode, struct file *file)
2738 {
2739 struct seq_file *seq = file->private_data;
2740 struct user_namespace *ns = seq->private;
2741 put_user_ns(ns);
2742 return seq_release(inode, file);
2743 }
2744
2745 static int proc_uid_map_open(struct inode *inode, struct file *file)
2746 {
2747 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2748 }
2749
2750 static int proc_gid_map_open(struct inode *inode, struct file *file)
2751 {
2752 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2753 }
2754
2755 static int proc_projid_map_open(struct inode *inode, struct file *file)
2756 {
2757 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2758 }
2759
2760 static const struct file_operations proc_uid_map_operations = {
2761 .open = proc_uid_map_open,
2762 .write = proc_uid_map_write,
2763 .read = seq_read,
2764 .llseek = seq_lseek,
2765 .release = proc_id_map_release,
2766 };
2767
2768 static const struct file_operations proc_gid_map_operations = {
2769 .open = proc_gid_map_open,
2770 .write = proc_gid_map_write,
2771 .read = seq_read,
2772 .llseek = seq_lseek,
2773 .release = proc_id_map_release,
2774 };
2775
2776 static const struct file_operations proc_projid_map_operations = {
2777 .open = proc_projid_map_open,
2778 .write = proc_projid_map_write,
2779 .read = seq_read,
2780 .llseek = seq_lseek,
2781 .release = proc_id_map_release,
2782 };
2783
2784 static int proc_setgroups_open(struct inode *inode, struct file *file)
2785 {
2786 struct user_namespace *ns = NULL;
2787 struct task_struct *task;
2788 int ret;
2789
2790 ret = -ESRCH;
2791 task = get_proc_task(inode);
2792 if (task) {
2793 rcu_read_lock();
2794 ns = get_user_ns(task_cred_xxx(task, user_ns));
2795 rcu_read_unlock();
2796 put_task_struct(task);
2797 }
2798 if (!ns)
2799 goto err;
2800
2801 if (file->f_mode & FMODE_WRITE) {
2802 ret = -EACCES;
2803 if (!ns_capable(ns, CAP_SYS_ADMIN))
2804 goto err_put_ns;
2805 }
2806
2807 ret = single_open(file, &proc_setgroups_show, ns);
2808 if (ret)
2809 goto err_put_ns;
2810
2811 return 0;
2812 err_put_ns:
2813 put_user_ns(ns);
2814 err:
2815 return ret;
2816 }
2817
2818 static int proc_setgroups_release(struct inode *inode, struct file *file)
2819 {
2820 struct seq_file *seq = file->private_data;
2821 struct user_namespace *ns = seq->private;
2822 int ret = single_release(inode, file);
2823 put_user_ns(ns);
2824 return ret;
2825 }
2826
2827 static const struct file_operations proc_setgroups_operations = {
2828 .open = proc_setgroups_open,
2829 .write = proc_setgroups_write,
2830 .read = seq_read,
2831 .llseek = seq_lseek,
2832 .release = proc_setgroups_release,
2833 };
2834 #endif /* CONFIG_USER_NS */
2835
2836 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2837 struct pid *pid, struct task_struct *task)
2838 {
2839 int err = lock_trace(task);
2840 if (!err) {
2841 seq_printf(m, "%08x\n", task->personality);
2842 unlock_trace(task);
2843 }
2844 return err;
2845 }
2846
2847 /*
2848 * Thread groups
2849 */
2850 static const struct file_operations proc_task_operations;
2851 static const struct inode_operations proc_task_inode_operations;
2852
2853 static const struct pid_entry tgid_base_stuff[] = {
2854 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2855 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2856 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2857 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2858 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2859 #ifdef CONFIG_NET
2860 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2861 #endif
2862 REG("environ", S_IRUSR, proc_environ_operations),
2863 REG("auxv", S_IRUSR, proc_auxv_operations),
2864 ONE("status", S_IRUGO, proc_pid_status),
2865 ONE("personality", S_IRUSR, proc_pid_personality),
2866 ONE("limits", S_IRUGO, proc_pid_limits),
2867 #ifdef CONFIG_SCHED_DEBUG
2868 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2869 #endif
2870 #ifdef CONFIG_SCHED_AUTOGROUP
2871 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2872 #endif
2873 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2874 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2875 ONE("syscall", S_IRUSR, proc_pid_syscall),
2876 #endif
2877 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2878 ONE("stat", S_IRUGO, proc_tgid_stat),
2879 ONE("statm", S_IRUGO, proc_pid_statm),
2880 REG("maps", S_IRUGO, proc_pid_maps_operations),
2881 #ifdef CONFIG_NUMA
2882 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2883 #endif
2884 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2885 LNK("cwd", proc_cwd_link),
2886 LNK("root", proc_root_link),
2887 LNK("exe", proc_exe_link),
2888 REG("mounts", S_IRUGO, proc_mounts_operations),
2889 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2890 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2891 #ifdef CONFIG_PROC_PAGE_MONITOR
2892 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2893 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2894 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2895 #endif
2896 #ifdef CONFIG_SECURITY
2897 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2898 #endif
2899 #ifdef CONFIG_KALLSYMS
2900 ONE("wchan", S_IRUGO, proc_pid_wchan),
2901 #endif
2902 #ifdef CONFIG_STACKTRACE
2903 ONE("stack", S_IRUSR, proc_pid_stack),
2904 #endif
2905 #ifdef CONFIG_SCHED_INFO
2906 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2907 #endif
2908 #ifdef CONFIG_LATENCYTOP
2909 REG("latency", S_IRUGO, proc_lstats_operations),
2910 #endif
2911 #ifdef CONFIG_PROC_PID_CPUSET
2912 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2913 #endif
2914 #ifdef CONFIG_CGROUPS
2915 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2916 #endif
2917 ONE("oom_score", S_IRUGO, proc_oom_score),
2918 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2919 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2920 #ifdef CONFIG_AUDITSYSCALL
2921 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2922 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2923 #endif
2924 #ifdef CONFIG_FAULT_INJECTION
2925 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2926 #endif
2927 #ifdef CONFIG_ELF_CORE
2928 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2929 #endif
2930 #ifdef CONFIG_TASK_IO_ACCOUNTING
2931 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2932 #endif
2933 #ifdef CONFIG_HARDWALL
2934 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2935 #endif
2936 #ifdef CONFIG_USER_NS
2937 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2938 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2939 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2940 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2941 #endif
2942 #ifdef CONFIG_CHECKPOINT_RESTORE
2943 REG("timers", S_IRUGO, proc_timers_operations),
2944 #endif
2945 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2946 };
2947
2948 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2949 {
2950 return proc_pident_readdir(file, ctx,
2951 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2952 }
2953
2954 static const struct file_operations proc_tgid_base_operations = {
2955 .read = generic_read_dir,
2956 .iterate_shared = proc_tgid_base_readdir,
2957 .llseek = generic_file_llseek,
2958 };
2959
2960 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2961 {
2962 return proc_pident_lookup(dir, dentry,
2963 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2964 }
2965
2966 static const struct inode_operations proc_tgid_base_inode_operations = {
2967 .lookup = proc_tgid_base_lookup,
2968 .getattr = pid_getattr,
2969 .setattr = proc_setattr,
2970 .permission = proc_pid_permission,
2971 };
2972
2973 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2974 {
2975 struct dentry *dentry, *leader, *dir;
2976 char buf[PROC_NUMBUF];
2977 struct qstr name;
2978
2979 name.name = buf;
2980 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2981 /* no ->d_hash() rejects on procfs */
2982 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2983 if (dentry) {
2984 d_invalidate(dentry);
2985 dput(dentry);
2986 }
2987
2988 if (pid == tgid)
2989 return;
2990
2991 name.name = buf;
2992 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2993 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2994 if (!leader)
2995 goto out;
2996
2997 name.name = "task";
2998 name.len = strlen(name.name);
2999 dir = d_hash_and_lookup(leader, &name);
3000 if (!dir)
3001 goto out_put_leader;
3002
3003 name.name = buf;
3004 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3005 dentry = d_hash_and_lookup(dir, &name);
3006 if (dentry) {
3007 d_invalidate(dentry);
3008 dput(dentry);
3009 }
3010
3011 dput(dir);
3012 out_put_leader:
3013 dput(leader);
3014 out:
3015 return;
3016 }
3017
3018 /**
3019 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3020 * @task: task that should be flushed.
3021 *
3022 * When flushing dentries from proc, one needs to flush them from global
3023 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3024 * in. This call is supposed to do all of this job.
3025 *
3026 * Looks in the dcache for
3027 * /proc/@pid
3028 * /proc/@tgid/task/@pid
3029 * if either directory is present flushes it and all of it'ts children
3030 * from the dcache.
3031 *
3032 * It is safe and reasonable to cache /proc entries for a task until
3033 * that task exits. After that they just clog up the dcache with
3034 * useless entries, possibly causing useful dcache entries to be
3035 * flushed instead. This routine is proved to flush those useless
3036 * dcache entries at process exit time.
3037 *
3038 * NOTE: This routine is just an optimization so it does not guarantee
3039 * that no dcache entries will exist at process exit time it
3040 * just makes it very unlikely that any will persist.
3041 */
3042
3043 void proc_flush_task(struct task_struct *task)
3044 {
3045 int i;
3046 struct pid *pid, *tgid;
3047 struct upid *upid;
3048
3049 pid = task_pid(task);
3050 tgid = task_tgid(task);
3051
3052 for (i = 0; i <= pid->level; i++) {
3053 upid = &pid->numbers[i];
3054 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3055 tgid->numbers[i].nr);
3056 }
3057 }
3058
3059 static int proc_pid_instantiate(struct inode *dir,
3060 struct dentry * dentry,
3061 struct task_struct *task, const void *ptr)
3062 {
3063 struct inode *inode;
3064
3065 inode = proc_pid_make_inode(dir->i_sb, task);
3066 if (!inode)
3067 goto out;
3068
3069 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3070 inode->i_op = &proc_tgid_base_inode_operations;
3071 inode->i_fop = &proc_tgid_base_operations;
3072 inode->i_flags|=S_IMMUTABLE;
3073
3074 set_nlink(inode, nlink_tgid);
3075
3076 d_set_d_op(dentry, &pid_dentry_operations);
3077
3078 d_add(dentry, inode);
3079 /* Close the race of the process dying before we return the dentry */
3080 if (pid_revalidate(dentry, 0))
3081 return 0;
3082 out:
3083 return -ENOENT;
3084 }
3085
3086 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3087 {
3088 int result = -ENOENT;
3089 struct task_struct *task;
3090 unsigned tgid;
3091 struct pid_namespace *ns;
3092
3093 tgid = name_to_int(&dentry->d_name);
3094 if (tgid == ~0U)
3095 goto out;
3096
3097 ns = dentry->d_sb->s_fs_info;
3098 rcu_read_lock();
3099 task = find_task_by_pid_ns(tgid, ns);
3100 if (task)
3101 get_task_struct(task);
3102 rcu_read_unlock();
3103 if (!task)
3104 goto out;
3105
3106 result = proc_pid_instantiate(dir, dentry, task, NULL);
3107 put_task_struct(task);
3108 out:
3109 return ERR_PTR(result);
3110 }
3111
3112 /*
3113 * Find the first task with tgid >= tgid
3114 *
3115 */
3116 struct tgid_iter {
3117 unsigned int tgid;
3118 struct task_struct *task;
3119 };
3120 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3121 {
3122 struct pid *pid;
3123
3124 if (iter.task)
3125 put_task_struct(iter.task);
3126 rcu_read_lock();
3127 retry:
3128 iter.task = NULL;
3129 pid = find_ge_pid(iter.tgid, ns);
3130 if (pid) {
3131 iter.tgid = pid_nr_ns(pid, ns);
3132 iter.task = pid_task(pid, PIDTYPE_PID);
3133 /* What we to know is if the pid we have find is the
3134 * pid of a thread_group_leader. Testing for task
3135 * being a thread_group_leader is the obvious thing
3136 * todo but there is a window when it fails, due to
3137 * the pid transfer logic in de_thread.
3138 *
3139 * So we perform the straight forward test of seeing
3140 * if the pid we have found is the pid of a thread
3141 * group leader, and don't worry if the task we have
3142 * found doesn't happen to be a thread group leader.
3143 * As we don't care in the case of readdir.
3144 */
3145 if (!iter.task || !has_group_leader_pid(iter.task)) {
3146 iter.tgid += 1;
3147 goto retry;
3148 }
3149 get_task_struct(iter.task);
3150 }
3151 rcu_read_unlock();
3152 return iter;
3153 }
3154
3155 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3156
3157 /* for the /proc/ directory itself, after non-process stuff has been done */
3158 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3159 {
3160 struct tgid_iter iter;
3161 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3162 loff_t pos = ctx->pos;
3163
3164 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3165 return 0;
3166
3167 if (pos == TGID_OFFSET - 2) {
3168 struct inode *inode = d_inode(ns->proc_self);
3169 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3170 return 0;
3171 ctx->pos = pos = pos + 1;
3172 }
3173 if (pos == TGID_OFFSET - 1) {
3174 struct inode *inode = d_inode(ns->proc_thread_self);
3175 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3176 return 0;
3177 ctx->pos = pos = pos + 1;
3178 }
3179 iter.tgid = pos - TGID_OFFSET;
3180 iter.task = NULL;
3181 for (iter = next_tgid(ns, iter);
3182 iter.task;
3183 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3184 char name[PROC_NUMBUF];
3185 int len;
3186 if (!has_pid_permissions(ns, iter.task, 2))
3187 continue;
3188
3189 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3190 ctx->pos = iter.tgid + TGID_OFFSET;
3191 if (!proc_fill_cache(file, ctx, name, len,
3192 proc_pid_instantiate, iter.task, NULL)) {
3193 put_task_struct(iter.task);
3194 return 0;
3195 }
3196 }
3197 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3198 return 0;
3199 }
3200
3201 /*
3202 * proc_tid_comm_permission is a special permission function exclusively
3203 * used for the node /proc/<pid>/task/<tid>/comm.
3204 * It bypasses generic permission checks in the case where a task of the same
3205 * task group attempts to access the node.
3206 * The rationale behind this is that glibc and bionic access this node for
3207 * cross thread naming (pthread_set/getname_np(!self)). However, if
3208 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3209 * which locks out the cross thread naming implementation.
3210 * This function makes sure that the node is always accessible for members of
3211 * same thread group.
3212 */
3213 static int proc_tid_comm_permission(struct inode *inode, int mask)
3214 {
3215 bool is_same_tgroup;
3216 struct task_struct *task;
3217
3218 task = get_proc_task(inode);
3219 if (!task)
3220 return -ESRCH;
3221 is_same_tgroup = same_thread_group(current, task);
3222 put_task_struct(task);
3223
3224 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3225 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3226 * read or written by the members of the corresponding
3227 * thread group.
3228 */
3229 return 0;
3230 }
3231
3232 return generic_permission(inode, mask);
3233 }
3234
3235 static const struct inode_operations proc_tid_comm_inode_operations = {
3236 .permission = proc_tid_comm_permission,
3237 };
3238
3239 /*
3240 * Tasks
3241 */
3242 static const struct pid_entry tid_base_stuff[] = {
3243 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3244 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3245 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3246 #ifdef CONFIG_NET
3247 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3248 #endif
3249 REG("environ", S_IRUSR, proc_environ_operations),
3250 REG("auxv", S_IRUSR, proc_auxv_operations),
3251 ONE("status", S_IRUGO, proc_pid_status),
3252 ONE("personality", S_IRUSR, proc_pid_personality),
3253 ONE("limits", S_IRUGO, proc_pid_limits),
3254 #ifdef CONFIG_SCHED_DEBUG
3255 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3256 #endif
3257 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3258 &proc_tid_comm_inode_operations,
3259 &proc_pid_set_comm_operations, {}),
3260 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3261 ONE("syscall", S_IRUSR, proc_pid_syscall),
3262 #endif
3263 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3264 ONE("stat", S_IRUGO, proc_tid_stat),
3265 ONE("statm", S_IRUGO, proc_pid_statm),
3266 REG("maps", S_IRUGO, proc_tid_maps_operations),
3267 #ifdef CONFIG_PROC_CHILDREN
3268 REG("children", S_IRUGO, proc_tid_children_operations),
3269 #endif
3270 #ifdef CONFIG_NUMA
3271 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3272 #endif
3273 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3274 LNK("cwd", proc_cwd_link),
3275 LNK("root", proc_root_link),
3276 LNK("exe", proc_exe_link),
3277 REG("mounts", S_IRUGO, proc_mounts_operations),
3278 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3279 #ifdef CONFIG_PROC_PAGE_MONITOR
3280 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3281 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3282 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3283 #endif
3284 #ifdef CONFIG_SECURITY
3285 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3286 #endif
3287 #ifdef CONFIG_KALLSYMS
3288 ONE("wchan", S_IRUGO, proc_pid_wchan),
3289 #endif
3290 #ifdef CONFIG_STACKTRACE
3291 ONE("stack", S_IRUSR, proc_pid_stack),
3292 #endif
3293 #ifdef CONFIG_SCHED_INFO
3294 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3295 #endif
3296 #ifdef CONFIG_LATENCYTOP
3297 REG("latency", S_IRUGO, proc_lstats_operations),
3298 #endif
3299 #ifdef CONFIG_PROC_PID_CPUSET
3300 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3301 #endif
3302 #ifdef CONFIG_CGROUPS
3303 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3304 #endif
3305 ONE("oom_score", S_IRUGO, proc_oom_score),
3306 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3307 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3308 #ifdef CONFIG_AUDITSYSCALL
3309 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3310 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3311 #endif
3312 #ifdef CONFIG_FAULT_INJECTION
3313 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3314 #endif
3315 #ifdef CONFIG_TASK_IO_ACCOUNTING
3316 ONE("io", S_IRUSR, proc_tid_io_accounting),
3317 #endif
3318 #ifdef CONFIG_HARDWALL
3319 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3320 #endif
3321 #ifdef CONFIG_USER_NS
3322 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3323 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3324 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3325 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3326 #endif
3327 };
3328
3329 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3330 {
3331 return proc_pident_readdir(file, ctx,
3332 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3333 }
3334
3335 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3336 {
3337 return proc_pident_lookup(dir, dentry,
3338 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3339 }
3340
3341 static const struct file_operations proc_tid_base_operations = {
3342 .read = generic_read_dir,
3343 .iterate_shared = proc_tid_base_readdir,
3344 .llseek = generic_file_llseek,
3345 };
3346
3347 static const struct inode_operations proc_tid_base_inode_operations = {
3348 .lookup = proc_tid_base_lookup,
3349 .getattr = pid_getattr,
3350 .setattr = proc_setattr,
3351 };
3352
3353 static int proc_task_instantiate(struct inode *dir,
3354 struct dentry *dentry, struct task_struct *task, const void *ptr)
3355 {
3356 struct inode *inode;
3357 inode = proc_pid_make_inode(dir->i_sb, task);
3358
3359 if (!inode)
3360 goto out;
3361 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3362 inode->i_op = &proc_tid_base_inode_operations;
3363 inode->i_fop = &proc_tid_base_operations;
3364 inode->i_flags|=S_IMMUTABLE;
3365
3366 set_nlink(inode, nlink_tid);
3367
3368 d_set_d_op(dentry, &pid_dentry_operations);
3369
3370 d_add(dentry, inode);
3371 /* Close the race of the process dying before we return the dentry */
3372 if (pid_revalidate(dentry, 0))
3373 return 0;
3374 out:
3375 return -ENOENT;
3376 }
3377
3378 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3379 {
3380 int result = -ENOENT;
3381 struct task_struct *task;
3382 struct task_struct *leader = get_proc_task(dir);
3383 unsigned tid;
3384 struct pid_namespace *ns;
3385
3386 if (!leader)
3387 goto out_no_task;
3388
3389 tid = name_to_int(&dentry->d_name);
3390 if (tid == ~0U)
3391 goto out;
3392
3393 ns = dentry->d_sb->s_fs_info;
3394 rcu_read_lock();
3395 task = find_task_by_pid_ns(tid, ns);
3396 if (task)
3397 get_task_struct(task);
3398 rcu_read_unlock();
3399 if (!task)
3400 goto out;
3401 if (!same_thread_group(leader, task))
3402 goto out_drop_task;
3403
3404 result = proc_task_instantiate(dir, dentry, task, NULL);
3405 out_drop_task:
3406 put_task_struct(task);
3407 out:
3408 put_task_struct(leader);
3409 out_no_task:
3410 return ERR_PTR(result);
3411 }
3412
3413 /*
3414 * Find the first tid of a thread group to return to user space.
3415 *
3416 * Usually this is just the thread group leader, but if the users
3417 * buffer was too small or there was a seek into the middle of the
3418 * directory we have more work todo.
3419 *
3420 * In the case of a short read we start with find_task_by_pid.
3421 *
3422 * In the case of a seek we start with the leader and walk nr
3423 * threads past it.
3424 */
3425 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3426 struct pid_namespace *ns)
3427 {
3428 struct task_struct *pos, *task;
3429 unsigned long nr = f_pos;
3430
3431 if (nr != f_pos) /* 32bit overflow? */
3432 return NULL;
3433
3434 rcu_read_lock();
3435 task = pid_task(pid, PIDTYPE_PID);
3436 if (!task)
3437 goto fail;
3438
3439 /* Attempt to start with the tid of a thread */
3440 if (tid && nr) {
3441 pos = find_task_by_pid_ns(tid, ns);
3442 if (pos && same_thread_group(pos, task))
3443 goto found;
3444 }
3445
3446 /* If nr exceeds the number of threads there is nothing todo */
3447 if (nr >= get_nr_threads(task))
3448 goto fail;
3449
3450 /* If we haven't found our starting place yet start
3451 * with the leader and walk nr threads forward.
3452 */
3453 pos = task = task->group_leader;
3454 do {
3455 if (!nr--)
3456 goto found;
3457 } while_each_thread(task, pos);
3458 fail:
3459 pos = NULL;
3460 goto out;
3461 found:
3462 get_task_struct(pos);
3463 out:
3464 rcu_read_unlock();
3465 return pos;
3466 }
3467
3468 /*
3469 * Find the next thread in the thread list.
3470 * Return NULL if there is an error or no next thread.
3471 *
3472 * The reference to the input task_struct is released.
3473 */
3474 static struct task_struct *next_tid(struct task_struct *start)
3475 {
3476 struct task_struct *pos = NULL;
3477 rcu_read_lock();
3478 if (pid_alive(start)) {
3479 pos = next_thread(start);
3480 if (thread_group_leader(pos))
3481 pos = NULL;
3482 else
3483 get_task_struct(pos);
3484 }
3485 rcu_read_unlock();
3486 put_task_struct(start);
3487 return pos;
3488 }
3489
3490 /* for the /proc/TGID/task/ directories */
3491 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3492 {
3493 struct inode *inode = file_inode(file);
3494 struct task_struct *task;
3495 struct pid_namespace *ns;
3496 int tid;
3497
3498 if (proc_inode_is_dead(inode))
3499 return -ENOENT;
3500
3501 if (!dir_emit_dots(file, ctx))
3502 return 0;
3503
3504 /* f_version caches the tgid value that the last readdir call couldn't
3505 * return. lseek aka telldir automagically resets f_version to 0.
3506 */
3507 ns = inode->i_sb->s_fs_info;
3508 tid = (int)file->f_version;
3509 file->f_version = 0;
3510 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3511 task;
3512 task = next_tid(task), ctx->pos++) {
3513 char name[PROC_NUMBUF];
3514 int len;
3515 tid = task_pid_nr_ns(task, ns);
3516 len = snprintf(name, sizeof(name), "%d", tid);
3517 if (!proc_fill_cache(file, ctx, name, len,
3518 proc_task_instantiate, task, NULL)) {
3519 /* returning this tgid failed, save it as the first
3520 * pid for the next readir call */
3521 file->f_version = (u64)tid;
3522 put_task_struct(task);
3523 break;
3524 }
3525 }
3526
3527 return 0;
3528 }
3529
3530 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3531 {
3532 struct inode *inode = d_inode(dentry);
3533 struct task_struct *p = get_proc_task(inode);
3534 generic_fillattr(inode, stat);
3535
3536 if (p) {
3537 stat->nlink += get_nr_threads(p);
3538 put_task_struct(p);
3539 }
3540
3541 return 0;
3542 }
3543
3544 static const struct inode_operations proc_task_inode_operations = {
3545 .lookup = proc_task_lookup,
3546 .getattr = proc_task_getattr,
3547 .setattr = proc_setattr,
3548 .permission = proc_pid_permission,
3549 };
3550
3551 static const struct file_operations proc_task_operations = {
3552 .read = generic_read_dir,
3553 .iterate_shared = proc_task_readdir,
3554 .llseek = generic_file_llseek,
3555 };
3556
3557 void __init set_proc_pid_nlink(void)
3558 {
3559 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3560 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3561 }
(deprecated) Btrfs devel tree