projects / GitHub / LineageOS / android_kernel_motorola_exynos9610.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mm: per-process reclaim
[GitHub/LineageOS/android_kernel_motorola_exynos9610.git] / fs / proc / base.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/proc/base.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * proc base directory handling functions
8 *
9 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10 * Instead of using magical inumbers to determine the kind of object
11 * we allocate and fill in-core inodes upon lookup. They don't even
12 * go into icache. We cache the reference to task_struct upon lookup too.
13 * Eventually it should become a filesystem in its own. We don't use the
14 * rest of procfs anymore.
15 *
16 *
17 * Changelog:
18 * 17-Jan-2005
19 * Allan Bezerra
20 * Bruna Moreira <bruna.moreira@indt.org.br>
21 * Edjard Mota <edjard.mota@indt.org.br>
22 * Ilias Biris <ilias.biris@indt.org.br>
23 * Mauricio Lin <mauricio.lin@indt.org.br>
24 *
25 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 *
27 * A new process specific entry (smaps) included in /proc. It shows the
28 * size of rss for each memory area. The maps entry lacks information
29 * about physical memory size (rss) for each mapped file, i.e.,
30 * rss information for executables and library files.
31 * This additional information is useful for any tools that need to know
32 * about physical memory consumption for a process specific library.
33 *
34 * Changelog:
35 * 21-Feb-2005
36 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37 * Pud inclusion in the page table walking.
38 *
39 * ChangeLog:
40 * 10-Mar-2005
41 * 10LE Instituto Nokia de Tecnologia - INdT:
42 * A better way to walks through the page table as suggested by Hugh Dickins.
43 *
44 * Simo Piiroinen <simo.piiroinen@nokia.com>:
45 * Smaps information related to shared, private, clean and dirty pages.
46 *
47 * Paul Mundt <paul.mundt@nokia.com>:
48 * Overall revision about smaps.
49 */
50
51 #include <linux/uaccess.h>
52
53 #include <linux/errno.h>
54 #include <linux/time.h>
55 #include <linux/proc_fs.h>
56 #include <linux/stat.h>
57 #include <linux/task_io_accounting_ops.h>
58 #include <linux/init.h>
59 #include <linux/capability.h>
60 #include <linux/file.h>
61 #include <linux/fdtable.h>
62 #include <linux/string.h>
63 #include <linux/seq_file.h>
64 #include <linux/namei.h>
65 #include <linux/mnt_namespace.h>
66 #include <linux/mm.h>
67 #include <linux/swap.h>
68 #include <linux/rcupdate.h>
69 #include <linux/kallsyms.h>
70 #include <linux/stacktrace.h>
71 #include <linux/resource.h>
72 #include <linux/module.h>
73 #include <linux/mount.h>
74 #include <linux/security.h>
75 #include <linux/ptrace.h>
76 #include <linux/tracehook.h>
77 #include <linux/printk.h>
78 #include <linux/cgroup.h>
79 #include <linux/cpuset.h>
80 #include <linux/audit.h>
81 #include <linux/poll.h>
82 #include <linux/nsproxy.h>
83 #include <linux/oom.h>
84 #include <linux/elf.h>
85 #include <linux/pid_namespace.h>
86 #include <linux/user_namespace.h>
87 #include <linux/fs_struct.h>
88 #include <linux/slab.h>
89 #include <linux/sched/autogroup.h>
90 #include <linux/sched/mm.h>
91 #include <linux/sched/coredump.h>
92 #include <linux/sched/debug.h>
93 #include <linux/sched/stat.h>
94 #include <linux/flex_array.h>
95 #include <linux/posix-timers.h>
96 #include <linux/cpufreq_times.h>
97 #ifdef CONFIG_HARDWALL
98 #include <asm/hardwall.h>
99 #endif
100 #include <trace/events/oom.h>
101 #include "internal.h"
102 #include "fd.h"
103
104 #include "../../lib/kstrtox.h"
105
106 /* NOTE:
107 * Implementing inode permission operations in /proc is almost
108 * certainly an error. Permission checks need to happen during
109 * each system call not at open time. The reason is that most of
110 * what we wish to check for permissions in /proc varies at runtime.
111 *
112 * The classic example of a problem is opening file descriptors
113 * in /proc for a task before it execs a suid executable.
114 */
115
116 static u8 nlink_tid;
117 static u8 nlink_tgid;
118
119 struct pid_entry {
120 const char *name;
121 unsigned int len;
122 umode_t mode;
123 const struct inode_operations *iop;
124 const struct file_operations *fop;
125 union proc_op op;
126 };
127
128 #define NOD(NAME, MODE, IOP, FOP, OP) { \
129 .name = (NAME), \
130 .len = sizeof(NAME) - 1, \
131 .mode = MODE, \
132 .iop = IOP, \
133 .fop = FOP, \
134 .op = OP, \
135 }
136
137 #define DIR(NAME, MODE, iops, fops) \
138 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
139 #define LNK(NAME, get_link) \
140 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
141 &proc_pid_link_inode_operations, NULL, \
142 { .proc_get_link = get_link } )
143 #define REG(NAME, MODE, fops) \
144 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
145 #define ONE(NAME, MODE, show) \
146 NOD(NAME, (S_IFREG|(MODE)), \
147 NULL, &proc_single_file_operations, \
148 { .proc_show = show } )
149
150 /*
151 * Count the number of hardlinks for the pid_entry table, excluding the .
152 * and .. links.
153 */
154 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
155 unsigned int n)
156 {
157 unsigned int i;
158 unsigned int count;
159
160 count = 2;
161 for (i = 0; i < n; ++i) {
162 if (S_ISDIR(entries[i].mode))
163 ++count;
164 }
165
166 return count;
167 }
168
169 static int get_task_root(struct task_struct *task, struct path *root)
170 {
171 int result = -ENOENT;
172
173 task_lock(task);
174 if (task->fs) {
175 get_fs_root(task->fs, root);
176 result = 0;
177 }
178 task_unlock(task);
179 return result;
180 }
181
182 static int proc_cwd_link(struct dentry *dentry, struct path *path)
183 {
184 struct task_struct *task = get_proc_task(d_inode(dentry));
185 int result = -ENOENT;
186
187 if (task) {
188 task_lock(task);
189 if (task->fs) {
190 get_fs_pwd(task->fs, path);
191 result = 0;
192 }
193 task_unlock(task);
194 put_task_struct(task);
195 }
196 return result;
197 }
198
199 static int proc_root_link(struct dentry *dentry, struct path *path)
200 {
201 struct task_struct *task = get_proc_task(d_inode(dentry));
202 int result = -ENOENT;
203
204 if (task) {
205 result = get_task_root(task, path);
206 put_task_struct(task);
207 }
208 return result;
209 }
210
211 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
212 size_t _count, loff_t *pos)
213 {
214 struct task_struct *tsk;
215 struct mm_struct *mm;
216 char *page;
217 unsigned long count = _count;
218 unsigned long arg_start, arg_end, env_start, env_end;
219 unsigned long len1, len2, len;
220 unsigned long p;
221 char c;
222 ssize_t rv;
223
224 BUG_ON(*pos < 0);
225
226 tsk = get_proc_task(file_inode(file));
227 if (!tsk)
228 return -ESRCH;
229 mm = get_task_mm(tsk);
230 put_task_struct(tsk);
231 if (!mm)
232 return 0;
233 /* Check if process spawned far enough to have cmdline. */
234 if (!mm->env_end) {
235 rv = 0;
236 goto out_mmput;
237 }
238
239 page = (char *)__get_free_page(GFP_KERNEL);
240 if (!page) {
241 rv = -ENOMEM;
242 goto out_mmput;
243 }
244
245 down_read(&mm->mmap_sem);
246 arg_start = mm->arg_start;
247 arg_end = mm->arg_end;
248 env_start = mm->env_start;
249 env_end = mm->env_end;
250 up_read(&mm->mmap_sem);
251
252 BUG_ON(arg_start > arg_end);
253 BUG_ON(env_start > env_end);
254
255 len1 = arg_end - arg_start;
256 len2 = env_end - env_start;
257
258 /* Empty ARGV. */
259 if (len1 == 0) {
260 rv = 0;
261 goto out_free_page;
262 }
263 /*
264 * Inherently racy -- command line shares address space
265 * with code and data.
266 */
267 rv = access_remote_vm(mm, arg_end - 1, &c, 1, FOLL_ANON);
268 if (rv <= 0)
269 goto out_free_page;
270
271 rv = 0;
272
273 if (c == '\0') {
274 /* Command line (set of strings) occupies whole ARGV. */
275 if (len1 <= *pos)
276 goto out_free_page;
277
278 p = arg_start + *pos;
279 len = len1 - *pos;
280 while (count > 0 && len > 0) {
281 unsigned int _count;
282 int nr_read;
283
284 _count = min3(count, len, PAGE_SIZE);
285 nr_read = access_remote_vm(mm, p, page, _count, FOLL_ANON);
286 if (nr_read < 0)
287 rv = nr_read;
288 if (nr_read <= 0)
289 goto out_free_page;
290
291 if (copy_to_user(buf, page, nr_read)) {
292 rv = -EFAULT;
293 goto out_free_page;
294 }
295
296 p += nr_read;
297 len -= nr_read;
298 buf += nr_read;
299 count -= nr_read;
300 rv += nr_read;
301 }
302 } else {
303 /*
304 * Command line (1 string) occupies ARGV and
305 * extends into ENVP.
306 */
307 struct {
308 unsigned long p;
309 unsigned long len;
310 } cmdline[2] = {
311 { .p = arg_start, .len = len1 },
312 { .p = env_start, .len = len2 },
313 };
314 loff_t pos1 = *pos;
315 unsigned int i;
316
317 i = 0;
318 while (i < 2 && pos1 >= cmdline[i].len) {
319 pos1 -= cmdline[i].len;
320 i++;
321 }
322 while (i < 2) {
323 p = cmdline[i].p + pos1;
324 len = cmdline[i].len - pos1;
325 while (count > 0 && len > 0) {
326 unsigned int _count, l;
327 int nr_read;
328 bool final;
329
330 _count = min3(count, len, PAGE_SIZE);
331 nr_read = access_remote_vm(mm, p, page, _count, FOLL_ANON);
332 if (nr_read < 0)
333 rv = nr_read;
334 if (nr_read <= 0)
335 goto out_free_page;
336
337 /*
338 * Command line can be shorter than whole ARGV
339 * even if last "marker" byte says it is not.
340 */
341 final = false;
342 l = strnlen(page, nr_read);
343 if (l < nr_read) {
344 nr_read = l;
345 final = true;
346 }
347
348 if (copy_to_user(buf, page, nr_read)) {
349 rv = -EFAULT;
350 goto out_free_page;
351 }
352
353 p += nr_read;
354 len -= nr_read;
355 buf += nr_read;
356 count -= nr_read;
357 rv += nr_read;
358
359 if (final)
360 goto out_free_page;
361 }
362
363 /* Only first chunk can be read partially. */
364 pos1 = 0;
365 i++;
366 }
367 }
368
369 out_free_page:
370 free_page((unsigned long)page);
371 out_mmput:
372 mmput(mm);
373 if (rv > 0)
374 *pos += rv;
375 return rv;
376 }
377
378 static const struct file_operations proc_pid_cmdline_ops = {
379 .read = proc_pid_cmdline_read,
380 .llseek = generic_file_llseek,
381 };
382
383 #ifdef CONFIG_KALLSYMS
384 /*
385 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
386 * Returns the resolved symbol. If that fails, simply return the address.
387 */
388 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
389 struct pid *pid, struct task_struct *task)
390 {
391 unsigned long wchan;
392 char symname[KSYM_NAME_LEN];
393
394 wchan = get_wchan(task);
395
396 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
397 && !lookup_symbol_name(wchan, symname))
398 seq_printf(m, "%s", symname);
399 else
400 seq_putc(m, '0');
401
402 return 0;
403 }
404 #endif /* CONFIG_KALLSYMS */
405
406 static int lock_trace(struct task_struct *task)
407 {
408 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
409 if (err)
410 return err;
411 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
412 mutex_unlock(&task->signal->cred_guard_mutex);
413 return -EPERM;
414 }
415 return 0;
416 }
417
418 static void unlock_trace(struct task_struct *task)
419 {
420 mutex_unlock(&task->signal->cred_guard_mutex);
421 }
422
423 #ifdef CONFIG_STACKTRACE
424
425 #define MAX_STACK_TRACE_DEPTH 64
426
427 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
428 struct pid *pid, struct task_struct *task)
429 {
430 struct stack_trace trace;
431 unsigned long *entries;
432 int err;
433 int i;
434
435 /*
436 * The ability to racily run the kernel stack unwinder on a running task
437 * and then observe the unwinder output is scary; while it is useful for
438 * debugging kernel issues, it can also allow an attacker to leak kernel
439 * stack contents.
440 * Doing this in a manner that is at least safe from races would require
441 * some work to ensure that the remote task can not be scheduled; and
442 * even then, this would still expose the unwinder as local attack
443 * surface.
444 * Therefore, this interface is restricted to root.
445 */
446 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
447 return -EACCES;
448
449 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
450 if (!entries)
451 return -ENOMEM;
452
453 trace.nr_entries = 0;
454 trace.max_entries = MAX_STACK_TRACE_DEPTH;
455 trace.entries = entries;
456 trace.skip = 0;
457
458 err = lock_trace(task);
459 if (!err) {
460 save_stack_trace_tsk(task, &trace);
461
462 for (i = 0; i < trace.nr_entries; i++) {
463 seq_printf(m, "[<%pK>] %pB\n",
464 (void *)entries[i], (void *)entries[i]);
465 }
466 unlock_trace(task);
467 }
468 kfree(entries);
469
470 return err;
471 }
472 #endif
473
474 #ifdef CONFIG_SCHED_INFO
475 /*
476 * Provides /proc/PID/schedstat
477 */
478 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
479 struct pid *pid, struct task_struct *task)
480 {
481 if (unlikely(!sched_info_on()))
482 seq_printf(m, "0 0 0\n");
483 else
484 seq_printf(m, "%llu %llu %lu\n",
485 (unsigned long long)task->se.sum_exec_runtime,
486 (unsigned long long)task->sched_info.run_delay,
487 task->sched_info.pcount);
488
489 return 0;
490 }
491 #endif
492
493 #ifdef CONFIG_LATENCYTOP
494 static int lstats_show_proc(struct seq_file *m, void *v)
495 {
496 int i;
497 struct inode *inode = m->private;
498 struct task_struct *task = get_proc_task(inode);
499
500 if (!task)
501 return -ESRCH;
502 seq_puts(m, "Latency Top version : v0.1\n");
503 for (i = 0; i < 32; i++) {
504 struct latency_record *lr = &task->latency_record[i];
505 if (lr->backtrace[0]) {
506 int q;
507 seq_printf(m, "%i %li %li",
508 lr->count, lr->time, lr->max);
509 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
510 unsigned long bt = lr->backtrace[q];
511 if (!bt)
512 break;
513 if (bt == ULONG_MAX)
514 break;
515 seq_printf(m, " %ps", (void *)bt);
516 }
517 seq_putc(m, '\n');
518 }
519
520 }
521 put_task_struct(task);
522 return 0;
523 }
524
525 static int lstats_open(struct inode *inode, struct file *file)
526 {
527 return single_open(file, lstats_show_proc, inode);
528 }
529
530 static ssize_t lstats_write(struct file *file, const char __user *buf,
531 size_t count, loff_t *offs)
532 {
533 struct task_struct *task = get_proc_task(file_inode(file));
534
535 if (!task)
536 return -ESRCH;
537 clear_all_latency_tracing(task);
538 put_task_struct(task);
539
540 return count;
541 }
542
543 static const struct file_operations proc_lstats_operations = {
544 .open = lstats_open,
545 .read = seq_read,
546 .write = lstats_write,
547 .llseek = seq_lseek,
548 .release = single_release,
549 };
550
551 #endif
552
553 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
554 struct pid *pid, struct task_struct *task)
555 {
556 unsigned long totalpages = totalram_pages + total_swap_pages;
557 unsigned long points = 0;
558
559 points = oom_badness(task, NULL, NULL, totalpages) *
560 1000 / totalpages;
561 seq_printf(m, "%lu\n", points);
562
563 return 0;
564 }
565
566 struct limit_names {
567 const char *name;
568 const char *unit;
569 };
570
571 static const struct limit_names lnames[RLIM_NLIMITS] = {
572 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
573 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
574 [RLIMIT_DATA] = {"Max data size", "bytes"},
575 [RLIMIT_STACK] = {"Max stack size", "bytes"},
576 [RLIMIT_CORE] = {"Max core file size", "bytes"},
577 [RLIMIT_RSS] = {"Max resident set", "bytes"},
578 [RLIMIT_NPROC] = {"Max processes", "processes"},
579 [RLIMIT_NOFILE] = {"Max open files", "files"},
580 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
581 [RLIMIT_AS] = {"Max address space", "bytes"},
582 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
583 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
584 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
585 [RLIMIT_NICE] = {"Max nice priority", NULL},
586 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
587 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
588 };
589
590 /* Display limits for a process */
591 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
592 struct pid *pid, struct task_struct *task)
593 {
594 unsigned int i;
595 unsigned long flags;
596
597 struct rlimit rlim[RLIM_NLIMITS];
598
599 if (!lock_task_sighand(task, &flags))
600 return 0;
601 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
602 unlock_task_sighand(task, &flags);
603
604 /*
605 * print the file header
606 */
607 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
608 "Limit", "Soft Limit", "Hard Limit", "Units");
609
610 for (i = 0; i < RLIM_NLIMITS; i++) {
611 if (rlim[i].rlim_cur == RLIM_INFINITY)
612 seq_printf(m, "%-25s %-20s ",
613 lnames[i].name, "unlimited");
614 else
615 seq_printf(m, "%-25s %-20lu ",
616 lnames[i].name, rlim[i].rlim_cur);
617
618 if (rlim[i].rlim_max == RLIM_INFINITY)
619 seq_printf(m, "%-20s ", "unlimited");
620 else
621 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
622
623 if (lnames[i].unit)
624 seq_printf(m, "%-10s\n", lnames[i].unit);
625 else
626 seq_putc(m, '\n');
627 }
628
629 return 0;
630 }
631
632 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
633 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
634 struct pid *pid, struct task_struct *task)
635 {
636 long nr;
637 unsigned long args[6], sp, pc;
638 int res;
639
640 res = lock_trace(task);
641 if (res)
642 return res;
643
644 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
645 seq_puts(m, "running\n");
646 else if (nr < 0)
647 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
648 else
649 seq_printf(m,
650 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
651 nr,
652 args[0], args[1], args[2], args[3], args[4], args[5],
653 sp, pc);
654 unlock_trace(task);
655
656 return 0;
657 }
658 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
659
660 /************************************************************************/
661 /* Here the fs part begins */
662 /************************************************************************/
663
664 /* permission checks */
665 static int proc_fd_access_allowed(struct inode *inode)
666 {
667 struct task_struct *task;
668 int allowed = 0;
669 /* Allow access to a task's file descriptors if it is us or we
670 * may use ptrace attach to the process and find out that
671 * information.
672 */
673 task = get_proc_task(inode);
674 if (task) {
675 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
676 put_task_struct(task);
677 }
678 return allowed;
679 }
680
681 int proc_setattr(struct dentry *dentry, struct iattr *attr)
682 {
683 int error;
684 struct inode *inode = d_inode(dentry);
685
686 if (attr->ia_valid & ATTR_MODE)
687 return -EPERM;
688
689 error = setattr_prepare(dentry, attr);
690 if (error)
691 return error;
692
693 setattr_copy(inode, attr);
694 mark_inode_dirty(inode);
695 return 0;
696 }
697
698 /*
699 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
700 * or euid/egid (for hide_pid_min=2)?
701 */
702 static bool has_pid_permissions(struct pid_namespace *pid,
703 struct task_struct *task,
704 int hide_pid_min)
705 {
706 if (pid->hide_pid < hide_pid_min)
707 return true;
708 if (in_group_p(pid->pid_gid))
709 return true;
710 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
711 }
712
713
714 static int proc_pid_permission(struct inode *inode, int mask)
715 {
716 struct pid_namespace *pid = inode->i_sb->s_fs_info;
717 struct task_struct *task;
718 bool has_perms;
719
720 task = get_proc_task(inode);
721 if (!task)
722 return -ESRCH;
723 has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
724 put_task_struct(task);
725
726 if (!has_perms) {
727 if (pid->hide_pid == HIDEPID_INVISIBLE) {
728 /*
729 * Let's make getdents(), stat(), and open()
730 * consistent with each other. If a process
731 * may not stat() a file, it shouldn't be seen
732 * in procfs at all.
733 */
734 return -ENOENT;
735 }
736
737 return -EPERM;
738 }
739 return generic_permission(inode, mask);
740 }
741
742
743
744 static const struct inode_operations proc_def_inode_operations = {
745 .setattr = proc_setattr,
746 };
747
748 static int proc_single_show(struct seq_file *m, void *v)
749 {
750 struct inode *inode = m->private;
751 struct pid_namespace *ns;
752 struct pid *pid;
753 struct task_struct *task;
754 int ret;
755
756 ns = inode->i_sb->s_fs_info;
757 pid = proc_pid(inode);
758 task = get_pid_task(pid, PIDTYPE_PID);
759 if (!task)
760 return -ESRCH;
761
762 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
763
764 put_task_struct(task);
765 return ret;
766 }
767
768 static int proc_single_open(struct inode *inode, struct file *filp)
769 {
770 return single_open(filp, proc_single_show, inode);
771 }
772
773 static const struct file_operations proc_single_file_operations = {
774 .open = proc_single_open,
775 .read = seq_read,
776 .llseek = seq_lseek,
777 .release = single_release,
778 };
779
780
781 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
782 {
783 struct task_struct *task = get_proc_task(inode);
784 struct mm_struct *mm = ERR_PTR(-ESRCH);
785
786 if (task) {
787 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
788 put_task_struct(task);
789
790 if (!IS_ERR_OR_NULL(mm)) {
791 /* ensure this mm_struct can't be freed */
792 mmgrab(mm);
793 /* but do not pin its memory */
794 mmput(mm);
795 }
796 }
797
798 return mm;
799 }
800
801 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
802 {
803 struct mm_struct *mm = proc_mem_open(inode, mode);
804
805 if (IS_ERR(mm))
806 return PTR_ERR(mm);
807
808 file->private_data = mm;
809 return 0;
810 }
811
812 static int mem_open(struct inode *inode, struct file *file)
813 {
814 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
815
816 /* OK to pass negative loff_t, we can catch out-of-range */
817 file->f_mode |= FMODE_UNSIGNED_OFFSET;
818
819 return ret;
820 }
821
822 static ssize_t mem_rw(struct file *file, char __user *buf,
823 size_t count, loff_t *ppos, int write)
824 {
825 struct mm_struct *mm = file->private_data;
826 unsigned long addr = *ppos;
827 ssize_t copied;
828 char *page;
829 unsigned int flags;
830
831 if (!mm)
832 return 0;
833
834 page = (char *)__get_free_page(GFP_KERNEL);
835 if (!page)
836 return -ENOMEM;
837
838 copied = 0;
839 if (!mmget_not_zero(mm))
840 goto free;
841
842 flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
843
844 while (count > 0) {
845 int this_len = min_t(int, count, PAGE_SIZE);
846
847 if (write && copy_from_user(page, buf, this_len)) {
848 copied = -EFAULT;
849 break;
850 }
851
852 this_len = access_remote_vm(mm, addr, page, this_len, flags);
853 if (!this_len) {
854 if (!copied)
855 copied = -EIO;
856 break;
857 }
858
859 if (!write && copy_to_user(buf, page, this_len)) {
860 copied = -EFAULT;
861 break;
862 }
863
864 buf += this_len;
865 addr += this_len;
866 copied += this_len;
867 count -= this_len;
868 }
869 *ppos = addr;
870
871 mmput(mm);
872 free:
873 free_page((unsigned long) page);
874 return copied;
875 }
876
877 static ssize_t mem_read(struct file *file, char __user *buf,
878 size_t count, loff_t *ppos)
879 {
880 return mem_rw(file, buf, count, ppos, 0);
881 }
882
883 static ssize_t mem_write(struct file *file, const char __user *buf,
884 size_t count, loff_t *ppos)
885 {
886 return mem_rw(file, (char __user*)buf, count, ppos, 1);
887 }
888
889 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
890 {
891 switch (orig) {
892 case 0:
893 file->f_pos = offset;
894 break;
895 case 1:
896 file->f_pos += offset;
897 break;
898 default:
899 return -EINVAL;
900 }
901 force_successful_syscall_return();
902 return file->f_pos;
903 }
904
905 static int mem_release(struct inode *inode, struct file *file)
906 {
907 struct mm_struct *mm = file->private_data;
908 if (mm)
909 mmdrop(mm);
910 return 0;
911 }
912
913 static const struct file_operations proc_mem_operations = {
914 .llseek = mem_lseek,
915 .read = mem_read,
916 .write = mem_write,
917 .open = mem_open,
918 .release = mem_release,
919 };
920
921 static int environ_open(struct inode *inode, struct file *file)
922 {
923 return __mem_open(inode, file, PTRACE_MODE_READ);
924 }
925
926 static ssize_t environ_read(struct file *file, char __user *buf,
927 size_t count, loff_t *ppos)
928 {
929 char *page;
930 unsigned long src = *ppos;
931 int ret = 0;
932 struct mm_struct *mm = file->private_data;
933 unsigned long env_start, env_end;
934
935 /* Ensure the process spawned far enough to have an environment. */
936 if (!mm || !mm->env_end)
937 return 0;
938
939 page = (char *)__get_free_page(GFP_KERNEL);
940 if (!page)
941 return -ENOMEM;
942
943 ret = 0;
944 if (!mmget_not_zero(mm))
945 goto free;
946
947 down_read(&mm->mmap_sem);
948 env_start = mm->env_start;
949 env_end = mm->env_end;
950 up_read(&mm->mmap_sem);
951
952 while (count > 0) {
953 size_t this_len, max_len;
954 int retval;
955
956 if (src >= (env_end - env_start))
957 break;
958
959 this_len = env_end - (env_start + src);
960
961 max_len = min_t(size_t, PAGE_SIZE, count);
962 this_len = min(max_len, this_len);
963
964 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
965
966 if (retval <= 0) {
967 ret = retval;
968 break;
969 }
970
971 if (copy_to_user(buf, page, retval)) {
972 ret = -EFAULT;
973 break;
974 }
975
976 ret += retval;
977 src += retval;
978 buf += retval;
979 count -= retval;
980 }
981 *ppos = src;
982 mmput(mm);
983
984 free:
985 free_page((unsigned long) page);
986 return ret;
987 }
988
989 static const struct file_operations proc_environ_operations = {
990 .open = environ_open,
991 .read = environ_read,
992 .llseek = generic_file_llseek,
993 .release = mem_release,
994 };
995
996 static int auxv_open(struct inode *inode, struct file *file)
997 {
998 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
999 }
1000
1001 static ssize_t auxv_read(struct file *file, char __user *buf,
1002 size_t count, loff_t *ppos)
1003 {
1004 struct mm_struct *mm = file->private_data;
1005 unsigned int nwords = 0;
1006
1007 if (!mm)
1008 return 0;
1009 do {
1010 nwords += 2;
1011 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1012 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1013 nwords * sizeof(mm->saved_auxv[0]));
1014 }
1015
1016 static const struct file_operations proc_auxv_operations = {
1017 .open = auxv_open,
1018 .read = auxv_read,
1019 .llseek = generic_file_llseek,
1020 .release = mem_release,
1021 };
1022
1023 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1024 loff_t *ppos)
1025 {
1026 struct task_struct *task = get_proc_task(file_inode(file));
1027 char buffer[PROC_NUMBUF];
1028 int oom_adj = OOM_ADJUST_MIN;
1029 size_t len;
1030
1031 if (!task)
1032 return -ESRCH;
1033 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1034 oom_adj = OOM_ADJUST_MAX;
1035 else
1036 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1037 OOM_SCORE_ADJ_MAX;
1038 put_task_struct(task);
1039 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1040 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1041 }
1042
1043 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1044 {
1045 static DEFINE_MUTEX(oom_adj_mutex);
1046 struct mm_struct *mm = NULL;
1047 struct task_struct *task;
1048 int err = 0;
1049
1050 task = get_proc_task(file_inode(file));
1051 if (!task)
1052 return -ESRCH;
1053
1054 mutex_lock(&oom_adj_mutex);
1055 if (legacy) {
1056 if (oom_adj < task->signal->oom_score_adj &&
1057 !capable(CAP_SYS_RESOURCE)) {
1058 err = -EACCES;
1059 goto err_unlock;
1060 }
1061 /*
1062 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1063 * /proc/pid/oom_score_adj instead.
1064 */
1065 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1066 current->comm, task_pid_nr(current), task_pid_nr(task),
1067 task_pid_nr(task));
1068 } else {
1069 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1070 !capable(CAP_SYS_RESOURCE)) {
1071 err = -EACCES;
1072 goto err_unlock;
1073 }
1074 }
1075
1076 /*
1077 * Make sure we will check other processes sharing the mm if this is
1078 * not vfrok which wants its own oom_score_adj.
1079 * pin the mm so it doesn't go away and get reused after task_unlock
1080 */
1081 if (!task->vfork_done) {
1082 struct task_struct *p = find_lock_task_mm(task);
1083
1084 if (p) {
1085 if (atomic_read(&p->mm->mm_users) > 1) {
1086 mm = p->mm;
1087 mmgrab(mm);
1088 }
1089 task_unlock(p);
1090 }
1091 }
1092
1093 task->signal->oom_score_adj = oom_adj;
1094 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1095 task->signal->oom_score_adj_min = (short)oom_adj;
1096 trace_oom_score_adj_update(task);
1097
1098 if (mm) {
1099 struct task_struct *p;
1100
1101 rcu_read_lock();
1102 for_each_process(p) {
1103 if (same_thread_group(task, p))
1104 continue;
1105
1106 /* do not touch kernel threads or the global init */
1107 if (p->flags & PF_KTHREAD || is_global_init(p))
1108 continue;
1109
1110 task_lock(p);
1111 if (!p->vfork_done && process_shares_mm(p, mm)) {
1112 p->signal->oom_score_adj = oom_adj;
1113 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1114 p->signal->oom_score_adj_min = (short)oom_adj;
1115 }
1116 task_unlock(p);
1117 }
1118 rcu_read_unlock();
1119 mmdrop(mm);
1120 }
1121 err_unlock:
1122 mutex_unlock(&oom_adj_mutex);
1123 put_task_struct(task);
1124 return err;
1125 }
1126
1127 /*
1128 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1129 * kernels. The effective policy is defined by oom_score_adj, which has a
1130 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1131 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1132 * Processes that become oom disabled via oom_adj will still be oom disabled
1133 * with this implementation.
1134 *
1135 * oom_adj cannot be removed since existing userspace binaries use it.
1136 */
1137 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1138 size_t count, loff_t *ppos)
1139 {
1140 char buffer[PROC_NUMBUF];
1141 int oom_adj;
1142 int err;
1143
1144 memset(buffer, 0, sizeof(buffer));
1145 if (count > sizeof(buffer) - 1)
1146 count = sizeof(buffer) - 1;
1147 if (copy_from_user(buffer, buf, count)) {
1148 err = -EFAULT;
1149 goto out;
1150 }
1151
1152 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1153 if (err)
1154 goto out;
1155 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1156 oom_adj != OOM_DISABLE) {
1157 err = -EINVAL;
1158 goto out;
1159 }
1160
1161 /*
1162 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1163 * value is always attainable.
1164 */
1165 if (oom_adj == OOM_ADJUST_MAX)
1166 oom_adj = OOM_SCORE_ADJ_MAX;
1167 else
1168 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1169
1170 err = __set_oom_adj(file, oom_adj, true);
1171 out:
1172 return err < 0 ? err : count;
1173 }
1174
1175 static const struct file_operations proc_oom_adj_operations = {
1176 .read = oom_adj_read,
1177 .write = oom_adj_write,
1178 .llseek = generic_file_llseek,
1179 };
1180
1181 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1182 size_t count, loff_t *ppos)
1183 {
1184 struct task_struct *task = get_proc_task(file_inode(file));
1185 char buffer[PROC_NUMBUF];
1186 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1187 size_t len;
1188
1189 if (!task)
1190 return -ESRCH;
1191 oom_score_adj = task->signal->oom_score_adj;
1192 put_task_struct(task);
1193 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1194 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1195 }
1196
1197 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1198 size_t count, loff_t *ppos)
1199 {
1200 char buffer[PROC_NUMBUF];
1201 int oom_score_adj;
1202 int err;
1203
1204 memset(buffer, 0, sizeof(buffer));
1205 if (count > sizeof(buffer) - 1)
1206 count = sizeof(buffer) - 1;
1207 if (copy_from_user(buffer, buf, count)) {
1208 err = -EFAULT;
1209 goto out;
1210 }
1211
1212 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1213 if (err)
1214 goto out;
1215 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1216 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1217 err = -EINVAL;
1218 goto out;
1219 }
1220
1221 err = __set_oom_adj(file, oom_score_adj, false);
1222 out:
1223 return err < 0 ? err : count;
1224 }
1225
1226 static const struct file_operations proc_oom_score_adj_operations = {
1227 .read = oom_score_adj_read,
1228 .write = oom_score_adj_write,
1229 .llseek = default_llseek,
1230 };
1231
1232 #ifdef CONFIG_AUDITSYSCALL
1233 #define TMPBUFLEN 11
1234 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1235 size_t count, loff_t *ppos)
1236 {
1237 struct inode * inode = file_inode(file);
1238 struct task_struct *task = get_proc_task(inode);
1239 ssize_t length;
1240 char tmpbuf[TMPBUFLEN];
1241
1242 if (!task)
1243 return -ESRCH;
1244 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1245 from_kuid(file->f_cred->user_ns,
1246 audit_get_loginuid(task)));
1247 put_task_struct(task);
1248 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1249 }
1250
1251 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1252 size_t count, loff_t *ppos)
1253 {
1254 struct inode * inode = file_inode(file);
1255 uid_t loginuid;
1256 kuid_t kloginuid;
1257 int rv;
1258
1259 rcu_read_lock();
1260 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1261 rcu_read_unlock();
1262 return -EPERM;
1263 }
1264 rcu_read_unlock();
1265
1266 if (*ppos != 0) {
1267 /* No partial writes. */
1268 return -EINVAL;
1269 }
1270
1271 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1272 if (rv < 0)
1273 return rv;
1274
1275 /* is userspace tring to explicitly UNSET the loginuid? */
1276 if (loginuid == AUDIT_UID_UNSET) {
1277 kloginuid = INVALID_UID;
1278 } else {
1279 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1280 if (!uid_valid(kloginuid))
1281 return -EINVAL;
1282 }
1283
1284 rv = audit_set_loginuid(kloginuid);
1285 if (rv < 0)
1286 return rv;
1287 return count;
1288 }
1289
1290 static const struct file_operations proc_loginuid_operations = {
1291 .read = proc_loginuid_read,
1292 .write = proc_loginuid_write,
1293 .llseek = generic_file_llseek,
1294 };
1295
1296 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1297 size_t count, loff_t *ppos)
1298 {
1299 struct inode * inode = file_inode(file);
1300 struct task_struct *task = get_proc_task(inode);
1301 ssize_t length;
1302 char tmpbuf[TMPBUFLEN];
1303
1304 if (!task)
1305 return -ESRCH;
1306 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1307 audit_get_sessionid(task));
1308 put_task_struct(task);
1309 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1310 }
1311
1312 static const struct file_operations proc_sessionid_operations = {
1313 .read = proc_sessionid_read,
1314 .llseek = generic_file_llseek,
1315 };
1316 #endif
1317
1318 #ifdef CONFIG_FAULT_INJECTION
1319 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1320 size_t count, loff_t *ppos)
1321 {
1322 struct task_struct *task = get_proc_task(file_inode(file));
1323 char buffer[PROC_NUMBUF];
1324 size_t len;
1325 int make_it_fail;
1326
1327 if (!task)
1328 return -ESRCH;
1329 make_it_fail = task->make_it_fail;
1330 put_task_struct(task);
1331
1332 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1333
1334 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1335 }
1336
1337 static ssize_t proc_fault_inject_write(struct file * file,
1338 const char __user * buf, size_t count, loff_t *ppos)
1339 {
1340 struct task_struct *task;
1341 char buffer[PROC_NUMBUF];
1342 int make_it_fail;
1343 int rv;
1344
1345 if (!capable(CAP_SYS_RESOURCE))
1346 return -EPERM;
1347 memset(buffer, 0, sizeof(buffer));
1348 if (count > sizeof(buffer) - 1)
1349 count = sizeof(buffer) - 1;
1350 if (copy_from_user(buffer, buf, count))
1351 return -EFAULT;
1352 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1353 if (rv < 0)
1354 return rv;
1355 if (make_it_fail < 0 || make_it_fail > 1)
1356 return -EINVAL;
1357
1358 task = get_proc_task(file_inode(file));
1359 if (!task)
1360 return -ESRCH;
1361 task->make_it_fail = make_it_fail;
1362 put_task_struct(task);
1363
1364 return count;
1365 }
1366
1367 static const struct file_operations proc_fault_inject_operations = {
1368 .read = proc_fault_inject_read,
1369 .write = proc_fault_inject_write,
1370 .llseek = generic_file_llseek,
1371 };
1372
1373 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1374 size_t count, loff_t *ppos)
1375 {
1376 struct task_struct *task;
1377 int err;
1378 unsigned int n;
1379
1380 err = kstrtouint_from_user(buf, count, 0, &n);
1381 if (err)
1382 return err;
1383
1384 task = get_proc_task(file_inode(file));
1385 if (!task)
1386 return -ESRCH;
1387 WRITE_ONCE(task->fail_nth, n);
1388 put_task_struct(task);
1389
1390 return count;
1391 }
1392
1393 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1394 size_t count, loff_t *ppos)
1395 {
1396 struct task_struct *task;
1397 char numbuf[PROC_NUMBUF];
1398 ssize_t len;
1399
1400 task = get_proc_task(file_inode(file));
1401 if (!task)
1402 return -ESRCH;
1403 len = snprintf(numbuf, sizeof(numbuf), "%u\n",
1404 READ_ONCE(task->fail_nth));
1405 len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1406 put_task_struct(task);
1407
1408 return len;
1409 }
1410
1411 static const struct file_operations proc_fail_nth_operations = {
1412 .read = proc_fail_nth_read,
1413 .write = proc_fail_nth_write,
1414 };
1415 #endif
1416
1417
1418 #ifdef CONFIG_SCHED_DEBUG
1419 /*
1420 * Print out various scheduling related per-task fields:
1421 */
1422 static int sched_show(struct seq_file *m, void *v)
1423 {
1424 struct inode *inode = m->private;
1425 struct pid_namespace *ns = inode->i_sb->s_fs_info;
1426 struct task_struct *p;
1427
1428 p = get_proc_task(inode);
1429 if (!p)
1430 return -ESRCH;
1431 proc_sched_show_task(p, ns, m);
1432
1433 put_task_struct(p);
1434
1435 return 0;
1436 }
1437
1438 static ssize_t
1439 sched_write(struct file *file, const char __user *buf,
1440 size_t count, loff_t *offset)
1441 {
1442 struct inode *inode = file_inode(file);
1443 struct task_struct *p;
1444
1445 p = get_proc_task(inode);
1446 if (!p)
1447 return -ESRCH;
1448 proc_sched_set_task(p);
1449
1450 put_task_struct(p);
1451
1452 return count;
1453 }
1454
1455 static int sched_open(struct inode *inode, struct file *filp)
1456 {
1457 return single_open(filp, sched_show, inode);
1458 }
1459
1460 static const struct file_operations proc_pid_sched_operations = {
1461 .open = sched_open,
1462 .read = seq_read,
1463 .write = sched_write,
1464 .llseek = seq_lseek,
1465 .release = single_release,
1466 };
1467
1468 #endif
1469
1470 #ifdef CONFIG_SCHED_AUTOGROUP
1471 /*
1472 * Print out autogroup related information:
1473 */
1474 static int sched_autogroup_show(struct seq_file *m, void *v)
1475 {
1476 struct inode *inode = m->private;
1477 struct task_struct *p;
1478
1479 p = get_proc_task(inode);
1480 if (!p)
1481 return -ESRCH;
1482 proc_sched_autogroup_show_task(p, m);
1483
1484 put_task_struct(p);
1485
1486 return 0;
1487 }
1488
1489 static ssize_t
1490 sched_autogroup_write(struct file *file, const char __user *buf,
1491 size_t count, loff_t *offset)
1492 {
1493 struct inode *inode = file_inode(file);
1494 struct task_struct *p;
1495 char buffer[PROC_NUMBUF];
1496 int nice;
1497 int err;
1498
1499 memset(buffer, 0, sizeof(buffer));
1500 if (count > sizeof(buffer) - 1)
1501 count = sizeof(buffer) - 1;
1502 if (copy_from_user(buffer, buf, count))
1503 return -EFAULT;
1504
1505 err = kstrtoint(strstrip(buffer), 0, &nice);
1506 if (err < 0)
1507 return err;
1508
1509 p = get_proc_task(inode);
1510 if (!p)
1511 return -ESRCH;
1512
1513 err = proc_sched_autogroup_set_nice(p, nice);
1514 if (err)
1515 count = err;
1516
1517 put_task_struct(p);
1518
1519 return count;
1520 }
1521
1522 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1523 {
1524 int ret;
1525
1526 ret = single_open(filp, sched_autogroup_show, NULL);
1527 if (!ret) {
1528 struct seq_file *m = filp->private_data;
1529
1530 m->private = inode;
1531 }
1532 return ret;
1533 }
1534
1535 static const struct file_operations proc_pid_sched_autogroup_operations = {
1536 .open = sched_autogroup_open,
1537 .read = seq_read,
1538 .write = sched_autogroup_write,
1539 .llseek = seq_lseek,
1540 .release = single_release,
1541 };
1542
1543 #endif /* CONFIG_SCHED_AUTOGROUP */
1544
1545 static ssize_t comm_write(struct file *file, const char __user *buf,
1546 size_t count, loff_t *offset)
1547 {
1548 struct inode *inode = file_inode(file);
1549 struct task_struct *p;
1550 char buffer[TASK_COMM_LEN];
1551 const size_t maxlen = sizeof(buffer) - 1;
1552
1553 memset(buffer, 0, sizeof(buffer));
1554 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1555 return -EFAULT;
1556
1557 p = get_proc_task(inode);
1558 if (!p)
1559 return -ESRCH;
1560
1561 if (same_thread_group(current, p))
1562 set_task_comm(p, buffer);
1563 else
1564 count = -EINVAL;
1565
1566 put_task_struct(p);
1567
1568 return count;
1569 }
1570
1571 static int comm_show(struct seq_file *m, void *v)
1572 {
1573 struct inode *inode = m->private;
1574 struct task_struct *p;
1575
1576 p = get_proc_task(inode);
1577 if (!p)
1578 return -ESRCH;
1579
1580 task_lock(p);
1581 seq_printf(m, "%s\n", p->comm);
1582 task_unlock(p);
1583
1584 put_task_struct(p);
1585
1586 return 0;
1587 }
1588
1589 static int comm_open(struct inode *inode, struct file *filp)
1590 {
1591 return single_open(filp, comm_show, inode);
1592 }
1593
1594 static const struct file_operations proc_pid_set_comm_operations = {
1595 .open = comm_open,
1596 .read = seq_read,
1597 .write = comm_write,
1598 .llseek = seq_lseek,
1599 .release = single_release,
1600 };
1601
1602 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1603 {
1604 struct task_struct *task;
1605 struct file *exe_file;
1606
1607 task = get_proc_task(d_inode(dentry));
1608 if (!task)
1609 return -ENOENT;
1610 exe_file = get_task_exe_file(task);
1611 put_task_struct(task);
1612 if (exe_file) {
1613 *exe_path = exe_file->f_path;
1614 path_get(&exe_file->f_path);
1615 fput(exe_file);
1616 return 0;
1617 } else
1618 return -ENOENT;
1619 }
1620
1621 static const char *proc_pid_get_link(struct dentry *dentry,
1622 struct inode *inode,
1623 struct delayed_call *done)
1624 {
1625 struct path path;
1626 int error = -EACCES;
1627
1628 if (!dentry)
1629 return ERR_PTR(-ECHILD);
1630
1631 /* Are we allowed to snoop on the tasks file descriptors? */
1632 if (!proc_fd_access_allowed(inode))
1633 goto out;
1634
1635 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1636 if (error)
1637 goto out;
1638
1639 nd_jump_link(&path);
1640 return NULL;
1641 out:
1642 return ERR_PTR(error);
1643 }
1644
1645 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1646 {
1647 char *tmp = (char *)__get_free_page(GFP_KERNEL);
1648 char *pathname;
1649 int len;
1650
1651 if (!tmp)
1652 return -ENOMEM;
1653
1654 pathname = d_path(path, tmp, PAGE_SIZE);
1655 len = PTR_ERR(pathname);
1656 if (IS_ERR(pathname))
1657 goto out;
1658 len = tmp + PAGE_SIZE - 1 - pathname;
1659
1660 if (len > buflen)
1661 len = buflen;
1662 if (copy_to_user(buffer, pathname, len))
1663 len = -EFAULT;
1664 out:
1665 free_page((unsigned long)tmp);
1666 return len;
1667 }
1668
1669 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1670 {
1671 int error = -EACCES;
1672 struct inode *inode = d_inode(dentry);
1673 struct path path;
1674
1675 /* Are we allowed to snoop on the tasks file descriptors? */
1676 if (!proc_fd_access_allowed(inode))
1677 goto out;
1678
1679 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1680 if (error)
1681 goto out;
1682
1683 error = do_proc_readlink(&path, buffer, buflen);
1684 path_put(&path);
1685 out:
1686 return error;
1687 }
1688
1689 const struct inode_operations proc_pid_link_inode_operations = {
1690 .readlink = proc_pid_readlink,
1691 .get_link = proc_pid_get_link,
1692 .setattr = proc_setattr,
1693 };
1694
1695
1696 /* building an inode */
1697
1698 void task_dump_owner(struct task_struct *task, mode_t mode,
1699 kuid_t *ruid, kgid_t *rgid)
1700 {
1701 /* Depending on the state of dumpable compute who should own a
1702 * proc file for a task.
1703 */
1704 const struct cred *cred;
1705 kuid_t uid;
1706 kgid_t gid;
1707
1708 if (unlikely(task->flags & PF_KTHREAD)) {
1709 *ruid = GLOBAL_ROOT_UID;
1710 *rgid = GLOBAL_ROOT_GID;
1711 return;
1712 }
1713
1714 /* Default to the tasks effective ownership */
1715 rcu_read_lock();
1716 cred = __task_cred(task);
1717 uid = cred->euid;
1718 gid = cred->egid;
1719 rcu_read_unlock();
1720
1721 /*
1722 * Before the /proc/pid/status file was created the only way to read
1723 * the effective uid of a /process was to stat /proc/pid. Reading
1724 * /proc/pid/status is slow enough that procps and other packages
1725 * kept stating /proc/pid. To keep the rules in /proc simple I have
1726 * made this apply to all per process world readable and executable
1727 * directories.
1728 */
1729 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1730 struct mm_struct *mm;
1731 task_lock(task);
1732 mm = task->mm;
1733 /* Make non-dumpable tasks owned by some root */
1734 if (mm) {
1735 if (get_dumpable(mm) != SUID_DUMP_USER) {
1736 struct user_namespace *user_ns = mm->user_ns;
1737
1738 uid = make_kuid(user_ns, 0);
1739 if (!uid_valid(uid))
1740 uid = GLOBAL_ROOT_UID;
1741
1742 gid = make_kgid(user_ns, 0);
1743 if (!gid_valid(gid))
1744 gid = GLOBAL_ROOT_GID;
1745 }
1746 } else {
1747 uid = GLOBAL_ROOT_UID;
1748 gid = GLOBAL_ROOT_GID;
1749 }
1750 task_unlock(task);
1751 }
1752 *ruid = uid;
1753 *rgid = gid;
1754 }
1755
1756 struct inode *proc_pid_make_inode(struct super_block * sb,
1757 struct task_struct *task, umode_t mode)
1758 {
1759 struct inode * inode;
1760 struct proc_inode *ei;
1761
1762 /* We need a new inode */
1763
1764 inode = new_inode(sb);
1765 if (!inode)
1766 goto out;
1767
1768 /* Common stuff */
1769 ei = PROC_I(inode);
1770 inode->i_mode = mode;
1771 inode->i_ino = get_next_ino();
1772 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1773 inode->i_op = &proc_def_inode_operations;
1774
1775 /*
1776 * grab the reference to task.
1777 */
1778 ei->pid = get_task_pid(task, PIDTYPE_PID);
1779 if (!ei->pid)
1780 goto out_unlock;
1781
1782 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1783 security_task_to_inode(task, inode);
1784
1785 out:
1786 return inode;
1787
1788 out_unlock:
1789 iput(inode);
1790 return NULL;
1791 }
1792
1793 int pid_getattr(const struct path *path, struct kstat *stat,
1794 u32 request_mask, unsigned int query_flags)
1795 {
1796 struct inode *inode = d_inode(path->dentry);
1797 struct task_struct *task;
1798 struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1799
1800 generic_fillattr(inode, stat);
1801
1802 rcu_read_lock();
1803 stat->uid = GLOBAL_ROOT_UID;
1804 stat->gid = GLOBAL_ROOT_GID;
1805 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1806 if (task) {
1807 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1808 rcu_read_unlock();
1809 /*
1810 * This doesn't prevent learning whether PID exists,
1811 * it only makes getattr() consistent with readdir().
1812 */
1813 return -ENOENT;
1814 }
1815 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1816 }
1817 rcu_read_unlock();
1818 return 0;
1819 }
1820
1821 /* dentry stuff */
1822
1823 /*
1824 * Exceptional case: normally we are not allowed to unhash a busy
1825 * directory. In this case, however, we can do it - no aliasing problems
1826 * due to the way we treat inodes.
1827 *
1828 * Rewrite the inode's ownerships here because the owning task may have
1829 * performed a setuid(), etc.
1830 *
1831 */
1832 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1833 {
1834 struct inode *inode;
1835 struct task_struct *task;
1836
1837 if (flags & LOOKUP_RCU)
1838 return -ECHILD;
1839
1840 inode = d_inode(dentry);
1841 task = get_proc_task(inode);
1842
1843 if (task) {
1844 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1845
1846 inode->i_mode &= ~(S_ISUID | S_ISGID);
1847 security_task_to_inode(task, inode);
1848 put_task_struct(task);
1849 return 1;
1850 }
1851 return 0;
1852 }
1853
1854 static inline bool proc_inode_is_dead(struct inode *inode)
1855 {
1856 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1857 }
1858
1859 int pid_delete_dentry(const struct dentry *dentry)
1860 {
1861 /* Is the task we represent dead?
1862 * If so, then don't put the dentry on the lru list,
1863 * kill it immediately.
1864 */
1865 return proc_inode_is_dead(d_inode(dentry));
1866 }
1867
1868 const struct dentry_operations pid_dentry_operations =
1869 {
1870 .d_revalidate = pid_revalidate,
1871 .d_delete = pid_delete_dentry,
1872 };
1873
1874 /* Lookups */
1875
1876 /*
1877 * Fill a directory entry.
1878 *
1879 * If possible create the dcache entry and derive our inode number and
1880 * file type from dcache entry.
1881 *
1882 * Since all of the proc inode numbers are dynamically generated, the inode
1883 * numbers do not exist until the inode is cache. This means creating the
1884 * the dcache entry in readdir is necessary to keep the inode numbers
1885 * reported by readdir in sync with the inode numbers reported
1886 * by stat.
1887 */
1888 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1889 const char *name, int len,
1890 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1891 {
1892 struct dentry *child, *dir = file->f_path.dentry;
1893 struct qstr qname = QSTR_INIT(name, len);
1894 struct inode *inode;
1895 unsigned type;
1896 ino_t ino;
1897
1898 child = d_hash_and_lookup(dir, &qname);
1899 if (!child) {
1900 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1901 child = d_alloc_parallel(dir, &qname, &wq);
1902 if (IS_ERR(child))
1903 goto end_instantiate;
1904 if (d_in_lookup(child)) {
1905 int err = instantiate(d_inode(dir), child, task, ptr);
1906 d_lookup_done(child);
1907 if (err < 0) {
1908 dput(child);
1909 goto end_instantiate;
1910 }
1911 }
1912 }
1913 inode = d_inode(child);
1914 ino = inode->i_ino;
1915 type = inode->i_mode >> 12;
1916 dput(child);
1917 return dir_emit(ctx, name, len, ino, type);
1918
1919 end_instantiate:
1920 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1921 }
1922
1923 /*
1924 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1925 * which represent vma start and end addresses.
1926 */
1927 static int dname_to_vma_addr(struct dentry *dentry,
1928 unsigned long *start, unsigned long *end)
1929 {
1930 const char *str = dentry->d_name.name;
1931 unsigned long long sval, eval;
1932 unsigned int len;
1933
1934 len = _parse_integer(str, 16, &sval);
1935 if (len & KSTRTOX_OVERFLOW)
1936 return -EINVAL;
1937 if (sval != (unsigned long)sval)
1938 return -EINVAL;
1939 str += len;
1940
1941 if (*str != '-')
1942 return -EINVAL;
1943 str++;
1944
1945 len = _parse_integer(str, 16, &eval);
1946 if (len & KSTRTOX_OVERFLOW)
1947 return -EINVAL;
1948 if (eval != (unsigned long)eval)
1949 return -EINVAL;
1950 str += len;
1951
1952 if (*str != '\0')
1953 return -EINVAL;
1954
1955 *start = sval;
1956 *end = eval;
1957
1958 return 0;
1959 }
1960
1961 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1962 {
1963 unsigned long vm_start, vm_end;
1964 bool exact_vma_exists = false;
1965 struct mm_struct *mm = NULL;
1966 struct task_struct *task;
1967 struct inode *inode;
1968 int status = 0;
1969
1970 if (flags & LOOKUP_RCU)
1971 return -ECHILD;
1972
1973 inode = d_inode(dentry);
1974 task = get_proc_task(inode);
1975 if (!task)
1976 goto out_notask;
1977
1978 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1979 if (IS_ERR_OR_NULL(mm))
1980 goto out;
1981
1982 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1983 down_read(&mm->mmap_sem);
1984 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1985 up_read(&mm->mmap_sem);
1986 }
1987
1988 mmput(mm);
1989
1990 if (exact_vma_exists) {
1991 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1992
1993 security_task_to_inode(task, inode);
1994 status = 1;
1995 }
1996
1997 out:
1998 put_task_struct(task);
1999
2000 out_notask:
2001 return status;
2002 }
2003
2004 static const struct dentry_operations tid_map_files_dentry_operations = {
2005 .d_revalidate = map_files_d_revalidate,
2006 .d_delete = pid_delete_dentry,
2007 };
2008
2009 static int map_files_get_link(struct dentry *dentry, struct path *path)
2010 {
2011 unsigned long vm_start, vm_end;
2012 struct vm_area_struct *vma;
2013 struct task_struct *task;
2014 struct mm_struct *mm;
2015 int rc;
2016
2017 rc = -ENOENT;
2018 task = get_proc_task(d_inode(dentry));
2019 if (!task)
2020 goto out;
2021
2022 mm = get_task_mm(task);
2023 put_task_struct(task);
2024 if (!mm)
2025 goto out;
2026
2027 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2028 if (rc)
2029 goto out_mmput;
2030
2031 rc = -ENOENT;
2032 down_read(&mm->mmap_sem);
2033 vma = find_exact_vma(mm, vm_start, vm_end);
2034 if (vma && vma->vm_file) {
2035 *path = vma->vm_file->f_path;
2036 path_get(path);
2037 rc = 0;
2038 }
2039 up_read(&mm->mmap_sem);
2040
2041 out_mmput:
2042 mmput(mm);
2043 out:
2044 return rc;
2045 }
2046
2047 struct map_files_info {
2048 fmode_t mode;
2049 unsigned int len;
2050 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2051 };
2052
2053 /*
2054 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2055 * symlinks may be used to bypass permissions on ancestor directories in the
2056 * path to the file in question.
2057 */
2058 static const char *
2059 proc_map_files_get_link(struct dentry *dentry,
2060 struct inode *inode,
2061 struct delayed_call *done)
2062 {
2063 if (!capable(CAP_SYS_ADMIN))
2064 return ERR_PTR(-EPERM);
2065
2066 return proc_pid_get_link(dentry, inode, done);
2067 }
2068
2069 /*
2070 * Identical to proc_pid_link_inode_operations except for get_link()
2071 */
2072 static const struct inode_operations proc_map_files_link_inode_operations = {
2073 .readlink = proc_pid_readlink,
2074 .get_link = proc_map_files_get_link,
2075 .setattr = proc_setattr,
2076 };
2077
2078 static int
2079 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2080 struct task_struct *task, const void *ptr)
2081 {
2082 fmode_t mode = (fmode_t)(unsigned long)ptr;
2083 struct proc_inode *ei;
2084 struct inode *inode;
2085
2086 inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2087 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2088 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2089 if (!inode)
2090 return -ENOENT;
2091
2092 ei = PROC_I(inode);
2093 ei->op.proc_get_link = map_files_get_link;
2094
2095 inode->i_op = &proc_map_files_link_inode_operations;
2096 inode->i_size = 64;
2097
2098 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2099 d_add(dentry, inode);
2100
2101 return 0;
2102 }
2103
2104 static struct dentry *proc_map_files_lookup(struct inode *dir,
2105 struct dentry *dentry, unsigned int flags)
2106 {
2107 unsigned long vm_start, vm_end;
2108 struct vm_area_struct *vma;
2109 struct task_struct *task;
2110 int result;
2111 struct mm_struct *mm;
2112
2113 result = -ENOENT;
2114 task = get_proc_task(dir);
2115 if (!task)
2116 goto out;
2117
2118 result = -EACCES;
2119 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2120 goto out_put_task;
2121
2122 result = -ENOENT;
2123 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2124 goto out_put_task;
2125
2126 mm = get_task_mm(task);
2127 if (!mm)
2128 goto out_put_task;
2129
2130 down_read(&mm->mmap_sem);
2131 vma = find_exact_vma(mm, vm_start, vm_end);
2132 if (!vma)
2133 goto out_no_vma;
2134
2135 if (vma->vm_file)
2136 result = proc_map_files_instantiate(dir, dentry, task,
2137 (void *)(unsigned long)vma->vm_file->f_mode);
2138
2139 out_no_vma:
2140 up_read(&mm->mmap_sem);
2141 mmput(mm);
2142 out_put_task:
2143 put_task_struct(task);
2144 out:
2145 return ERR_PTR(result);
2146 }
2147
2148 static const struct inode_operations proc_map_files_inode_operations = {
2149 .lookup = proc_map_files_lookup,
2150 .permission = proc_fd_permission,
2151 .setattr = proc_setattr,
2152 };
2153
2154 static int
2155 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2156 {
2157 struct vm_area_struct *vma;
2158 struct task_struct *task;
2159 struct mm_struct *mm;
2160 unsigned long nr_files, pos, i;
2161 struct flex_array *fa = NULL;
2162 struct map_files_info info;
2163 struct map_files_info *p;
2164 int ret;
2165
2166 ret = -ENOENT;
2167 task = get_proc_task(file_inode(file));
2168 if (!task)
2169 goto out;
2170
2171 ret = -EACCES;
2172 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2173 goto out_put_task;
2174
2175 ret = 0;
2176 if (!dir_emit_dots(file, ctx))
2177 goto out_put_task;
2178
2179 mm = get_task_mm(task);
2180 if (!mm)
2181 goto out_put_task;
2182 down_read(&mm->mmap_sem);
2183
2184 nr_files = 0;
2185
2186 /*
2187 * We need two passes here:
2188 *
2189 * 1) Collect vmas of mapped files with mmap_sem taken
2190 * 2) Release mmap_sem and instantiate entries
2191 *
2192 * otherwise we get lockdep complained, since filldir()
2193 * routine might require mmap_sem taken in might_fault().
2194 */
2195
2196 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2197 if (vma->vm_file && ++pos > ctx->pos)
2198 nr_files++;
2199 }
2200
2201 if (nr_files) {
2202 fa = flex_array_alloc(sizeof(info), nr_files,
2203 GFP_KERNEL);
2204 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2205 GFP_KERNEL)) {
2206 ret = -ENOMEM;
2207 if (fa)
2208 flex_array_free(fa);
2209 up_read(&mm->mmap_sem);
2210 mmput(mm);
2211 goto out_put_task;
2212 }
2213 for (i = 0, vma = mm->mmap, pos = 2; vma;
2214 vma = vma->vm_next) {
2215 if (!vma->vm_file)
2216 continue;
2217 if (++pos <= ctx->pos)
2218 continue;
2219
2220 info.mode = vma->vm_file->f_mode;
2221 info.len = snprintf(info.name,
2222 sizeof(info.name), "%lx-%lx",
2223 vma->vm_start, vma->vm_end);
2224 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2225 BUG();
2226 }
2227 }
2228 up_read(&mm->mmap_sem);
2229
2230 for (i = 0; i < nr_files; i++) {
2231 p = flex_array_get(fa, i);
2232 if (!proc_fill_cache(file, ctx,
2233 p->name, p->len,
2234 proc_map_files_instantiate,
2235 task,
2236 (void *)(unsigned long)p->mode))
2237 break;
2238 ctx->pos++;
2239 }
2240 if (fa)
2241 flex_array_free(fa);
2242 mmput(mm);
2243
2244 out_put_task:
2245 put_task_struct(task);
2246 out:
2247 return ret;
2248 }
2249
2250 static const struct file_operations proc_map_files_operations = {
2251 .read = generic_read_dir,
2252 .iterate_shared = proc_map_files_readdir,
2253 .llseek = generic_file_llseek,
2254 };
2255
2256 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2257 struct timers_private {
2258 struct pid *pid;
2259 struct task_struct *task;
2260 struct sighand_struct *sighand;
2261 struct pid_namespace *ns;
2262 unsigned long flags;
2263 };
2264
2265 static void *timers_start(struct seq_file *m, loff_t *pos)
2266 {
2267 struct timers_private *tp = m->private;
2268
2269 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2270 if (!tp->task)
2271 return ERR_PTR(-ESRCH);
2272
2273 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2274 if (!tp->sighand)
2275 return ERR_PTR(-ESRCH);
2276
2277 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2278 }
2279
2280 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2281 {
2282 struct timers_private *tp = m->private;
2283 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2284 }
2285
2286 static void timers_stop(struct seq_file *m, void *v)
2287 {
2288 struct timers_private *tp = m->private;
2289
2290 if (tp->sighand) {
2291 unlock_task_sighand(tp->task, &tp->flags);
2292 tp->sighand = NULL;
2293 }
2294
2295 if (tp->task) {
2296 put_task_struct(tp->task);
2297 tp->task = NULL;
2298 }
2299 }
2300
2301 static int show_timer(struct seq_file *m, void *v)
2302 {
2303 struct k_itimer *timer;
2304 struct timers_private *tp = m->private;
2305 int notify;
2306 static const char * const nstr[] = {
2307 [SIGEV_SIGNAL] = "signal",
2308 [SIGEV_NONE] = "none",
2309 [SIGEV_THREAD] = "thread",
2310 };
2311
2312 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2313 notify = timer->it_sigev_notify;
2314
2315 seq_printf(m, "ID: %d\n", timer->it_id);
2316 seq_printf(m, "signal: %d/%p\n",
2317 timer->sigq->info.si_signo,
2318 timer->sigq->info.si_value.sival_ptr);
2319 seq_printf(m, "notify: %s/%s.%d\n",
2320 nstr[notify & ~SIGEV_THREAD_ID],
2321 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2322 pid_nr_ns(timer->it_pid, tp->ns));
2323 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2324
2325 return 0;
2326 }
2327
2328 static const struct seq_operations proc_timers_seq_ops = {
2329 .start = timers_start,
2330 .next = timers_next,
2331 .stop = timers_stop,
2332 .show = show_timer,
2333 };
2334
2335 static int proc_timers_open(struct inode *inode, struct file *file)
2336 {
2337 struct timers_private *tp;
2338
2339 tp = __seq_open_private(file, &proc_timers_seq_ops,
2340 sizeof(struct timers_private));
2341 if (!tp)
2342 return -ENOMEM;
2343
2344 tp->pid = proc_pid(inode);
2345 tp->ns = inode->i_sb->s_fs_info;
2346 return 0;
2347 }
2348
2349 static const struct file_operations proc_timers_operations = {
2350 .open = proc_timers_open,
2351 .read = seq_read,
2352 .llseek = seq_lseek,
2353 .release = seq_release_private,
2354 };
2355 #endif
2356
2357 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2358 size_t count, loff_t *offset)
2359 {
2360 struct inode *inode = file_inode(file);
2361 struct task_struct *p;
2362 u64 slack_ns;
2363 int err;
2364
2365 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2366 if (err < 0)
2367 return err;
2368
2369 p = get_proc_task(inode);
2370 if (!p)
2371 return -ESRCH;
2372
2373 if (p != current) {
2374 if (!capable(CAP_SYS_NICE)) {
2375 count = -EPERM;
2376 goto out;
2377 }
2378
2379 err = security_task_setscheduler(p);
2380 if (err) {
2381 count = err;
2382 goto out;
2383 }
2384 }
2385
2386 task_lock(p);
2387 if (slack_ns == 0)
2388 p->timer_slack_ns = p->default_timer_slack_ns;
2389 else
2390 p->timer_slack_ns = slack_ns;
2391 task_unlock(p);
2392
2393 out:
2394 put_task_struct(p);
2395
2396 return count;
2397 }
2398
2399 static int timerslack_ns_show(struct seq_file *m, void *v)
2400 {
2401 struct inode *inode = m->private;
2402 struct task_struct *p;
2403 int err = 0;
2404
2405 p = get_proc_task(inode);
2406 if (!p)
2407 return -ESRCH;
2408
2409 if (p != current) {
2410
2411 if (!capable(CAP_SYS_NICE)) {
2412 err = -EPERM;
2413 goto out;
2414 }
2415 err = security_task_getscheduler(p);
2416 if (err)
2417 goto out;
2418 }
2419
2420 task_lock(p);
2421 seq_printf(m, "%llu\n", p->timer_slack_ns);
2422 task_unlock(p);
2423
2424 out:
2425 put_task_struct(p);
2426
2427 return err;
2428 }
2429
2430 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2431 {
2432 return single_open(filp, timerslack_ns_show, inode);
2433 }
2434
2435 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2436 .open = timerslack_ns_open,
2437 .read = seq_read,
2438 .write = timerslack_ns_write,
2439 .llseek = seq_lseek,
2440 .release = single_release,
2441 };
2442
2443 static int proc_pident_instantiate(struct inode *dir,
2444 struct dentry *dentry, struct task_struct *task, const void *ptr)
2445 {
2446 const struct pid_entry *p = ptr;
2447 struct inode *inode;
2448 struct proc_inode *ei;
2449
2450 inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2451 if (!inode)
2452 goto out;
2453
2454 ei = PROC_I(inode);
2455 if (S_ISDIR(inode->i_mode))
2456 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2457 if (p->iop)
2458 inode->i_op = p->iop;
2459 if (p->fop)
2460 inode->i_fop = p->fop;
2461 ei->op = p->op;
2462 d_set_d_op(dentry, &pid_dentry_operations);
2463 d_add(dentry, inode);
2464 /* Close the race of the process dying before we return the dentry */
2465 if (pid_revalidate(dentry, 0))
2466 return 0;
2467 out:
2468 return -ENOENT;
2469 }
2470
2471 static struct dentry *proc_pident_lookup(struct inode *dir,
2472 struct dentry *dentry,
2473 const struct pid_entry *ents,
2474 unsigned int nents)
2475 {
2476 int error;
2477 struct task_struct *task = get_proc_task(dir);
2478 const struct pid_entry *p, *last;
2479
2480 error = -ENOENT;
2481
2482 if (!task)
2483 goto out_no_task;
2484
2485 /*
2486 * Yes, it does not scale. And it should not. Don't add
2487 * new entries into /proc/<tgid>/ without very good reasons.
2488 */
2489 last = &ents[nents];
2490 for (p = ents; p < last; p++) {
2491 if (p->len != dentry->d_name.len)
2492 continue;
2493 if (!memcmp(dentry->d_name.name, p->name, p->len))
2494 break;
2495 }
2496 if (p >= last)
2497 goto out;
2498
2499 error = proc_pident_instantiate(dir, dentry, task, p);
2500 out:
2501 put_task_struct(task);
2502 out_no_task:
2503 return ERR_PTR(error);
2504 }
2505
2506 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2507 const struct pid_entry *ents, unsigned int nents)
2508 {
2509 struct task_struct *task = get_proc_task(file_inode(file));
2510 const struct pid_entry *p;
2511
2512 if (!task)
2513 return -ENOENT;
2514
2515 if (!dir_emit_dots(file, ctx))
2516 goto out;
2517
2518 if (ctx->pos >= nents + 2)
2519 goto out;
2520
2521 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2522 if (!proc_fill_cache(file, ctx, p->name, p->len,
2523 proc_pident_instantiate, task, p))
2524 break;
2525 ctx->pos++;
2526 }
2527 out:
2528 put_task_struct(task);
2529 return 0;
2530 }
2531
2532 #ifdef CONFIG_SECURITY
2533 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2534 size_t count, loff_t *ppos)
2535 {
2536 struct inode * inode = file_inode(file);
2537 char *p = NULL;
2538 ssize_t length;
2539 struct task_struct *task = get_proc_task(inode);
2540
2541 if (!task)
2542 return -ESRCH;
2543
2544 length = security_getprocattr(task,
2545 (char*)file->f_path.dentry->d_name.name,
2546 &p);
2547 put_task_struct(task);
2548 if (length > 0)
2549 length = simple_read_from_buffer(buf, count, ppos, p, length);
2550 kfree(p);
2551 return length;
2552 }
2553
2554 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2555 size_t count, loff_t *ppos)
2556 {
2557 struct inode * inode = file_inode(file);
2558 void *page;
2559 ssize_t length;
2560 struct task_struct *task = get_proc_task(inode);
2561
2562 length = -ESRCH;
2563 if (!task)
2564 goto out_no_task;
2565
2566 /* A task may only write its own attributes. */
2567 length = -EACCES;
2568 if (current != task)
2569 goto out;
2570
2571 if (count > PAGE_SIZE)
2572 count = PAGE_SIZE;
2573
2574 /* No partial writes. */
2575 length = -EINVAL;
2576 if (*ppos != 0)
2577 goto out;
2578
2579 page = memdup_user(buf, count);
2580 if (IS_ERR(page)) {
2581 length = PTR_ERR(page);
2582 goto out;
2583 }
2584
2585 /* Guard against adverse ptrace interaction */
2586 length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2587 if (length < 0)
2588 goto out_free;
2589
2590 length = security_setprocattr(file->f_path.dentry->d_name.name,
2591 page, count);
2592 mutex_unlock(&current->signal->cred_guard_mutex);
2593 out_free:
2594 kfree(page);
2595 out:
2596 put_task_struct(task);
2597 out_no_task:
2598 return length;
2599 }
2600
2601 static const struct file_operations proc_pid_attr_operations = {
2602 .read = proc_pid_attr_read,
2603 .write = proc_pid_attr_write,
2604 .llseek = generic_file_llseek,
2605 };
2606
2607 static const struct pid_entry attr_dir_stuff[] = {
2608 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2609 REG("prev", S_IRUGO, proc_pid_attr_operations),
2610 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2611 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2612 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2613 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2614 };
2615
2616 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2617 {
2618 return proc_pident_readdir(file, ctx,
2619 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2620 }
2621
2622 static const struct file_operations proc_attr_dir_operations = {
2623 .read = generic_read_dir,
2624 .iterate_shared = proc_attr_dir_readdir,
2625 .llseek = generic_file_llseek,
2626 };
2627
2628 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2629 struct dentry *dentry, unsigned int flags)
2630 {
2631 return proc_pident_lookup(dir, dentry,
2632 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2633 }
2634
2635 static const struct inode_operations proc_attr_dir_inode_operations = {
2636 .lookup = proc_attr_dir_lookup,
2637 .getattr = pid_getattr,
2638 .setattr = proc_setattr,
2639 };
2640
2641 #endif
2642
2643 #ifdef CONFIG_ELF_CORE
2644 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2645 size_t count, loff_t *ppos)
2646 {
2647 struct task_struct *task = get_proc_task(file_inode(file));
2648 struct mm_struct *mm;
2649 char buffer[PROC_NUMBUF];
2650 size_t len;
2651 int ret;
2652
2653 if (!task)
2654 return -ESRCH;
2655
2656 ret = 0;
2657 mm = get_task_mm(task);
2658 if (mm) {
2659 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2660 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2661 MMF_DUMP_FILTER_SHIFT));
2662 mmput(mm);
2663 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2664 }
2665
2666 put_task_struct(task);
2667
2668 return ret;
2669 }
2670
2671 static ssize_t proc_coredump_filter_write(struct file *file,
2672 const char __user *buf,
2673 size_t count,
2674 loff_t *ppos)
2675 {
2676 struct task_struct *task;
2677 struct mm_struct *mm;
2678 unsigned int val;
2679 int ret;
2680 int i;
2681 unsigned long mask;
2682
2683 ret = kstrtouint_from_user(buf, count, 0, &val);
2684 if (ret < 0)
2685 return ret;
2686
2687 ret = -ESRCH;
2688 task = get_proc_task(file_inode(file));
2689 if (!task)
2690 goto out_no_task;
2691
2692 mm = get_task_mm(task);
2693 if (!mm)
2694 goto out_no_mm;
2695 ret = 0;
2696
2697 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2698 if (val & mask)
2699 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2700 else
2701 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2702 }
2703
2704 mmput(mm);
2705 out_no_mm:
2706 put_task_struct(task);
2707 out_no_task:
2708 if (ret < 0)
2709 return ret;
2710 return count;
2711 }
2712
2713 static const struct file_operations proc_coredump_filter_operations = {
2714 .read = proc_coredump_filter_read,
2715 .write = proc_coredump_filter_write,
2716 .llseek = generic_file_llseek,
2717 };
2718 #endif
2719
2720 #ifdef CONFIG_TASK_IO_ACCOUNTING
2721 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2722 {
2723 struct task_io_accounting acct = task->ioac;
2724 unsigned long flags;
2725 int result;
2726
2727 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2728 if (result)
2729 return result;
2730
2731 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2732 result = -EACCES;
2733 goto out_unlock;
2734 }
2735
2736 if (whole && lock_task_sighand(task, &flags)) {
2737 struct task_struct *t = task;
2738
2739 task_io_accounting_add(&acct, &task->signal->ioac);
2740 while_each_thread(task, t)
2741 task_io_accounting_add(&acct, &t->ioac);
2742
2743 unlock_task_sighand(task, &flags);
2744 }
2745 seq_printf(m,
2746 "rchar: %llu\n"
2747 "wchar: %llu\n"
2748 "syscr: %llu\n"
2749 "syscw: %llu\n"
2750 "read_bytes: %llu\n"
2751 "write_bytes: %llu\n"
2752 "cancelled_write_bytes: %llu\n",
2753 (unsigned long long)acct.rchar,
2754 (unsigned long long)acct.wchar,
2755 (unsigned long long)acct.syscr,
2756 (unsigned long long)acct.syscw,
2757 (unsigned long long)acct.read_bytes,
2758 (unsigned long long)acct.write_bytes,
2759 (unsigned long long)acct.cancelled_write_bytes);
2760 result = 0;
2761
2762 out_unlock:
2763 mutex_unlock(&task->signal->cred_guard_mutex);
2764 return result;
2765 }
2766
2767 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2768 struct pid *pid, struct task_struct *task)
2769 {
2770 return do_io_accounting(task, m, 0);
2771 }
2772
2773 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2774 struct pid *pid, struct task_struct *task)
2775 {
2776 return do_io_accounting(task, m, 1);
2777 }
2778 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2779
2780 #ifdef CONFIG_USER_NS
2781 static int proc_id_map_open(struct inode *inode, struct file *file,
2782 const struct seq_operations *seq_ops)
2783 {
2784 struct user_namespace *ns = NULL;
2785 struct task_struct *task;
2786 struct seq_file *seq;
2787 int ret = -EINVAL;
2788
2789 task = get_proc_task(inode);
2790 if (task) {
2791 rcu_read_lock();
2792 ns = get_user_ns(task_cred_xxx(task, user_ns));
2793 rcu_read_unlock();
2794 put_task_struct(task);
2795 }
2796 if (!ns)
2797 goto err;
2798
2799 ret = seq_open(file, seq_ops);
2800 if (ret)
2801 goto err_put_ns;
2802
2803 seq = file->private_data;
2804 seq->private = ns;
2805
2806 return 0;
2807 err_put_ns:
2808 put_user_ns(ns);
2809 err:
2810 return ret;
2811 }
2812
2813 static int proc_id_map_release(struct inode *inode, struct file *file)
2814 {
2815 struct seq_file *seq = file->private_data;
2816 struct user_namespace *ns = seq->private;
2817 put_user_ns(ns);
2818 return seq_release(inode, file);
2819 }
2820
2821 static int proc_uid_map_open(struct inode *inode, struct file *file)
2822 {
2823 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2824 }
2825
2826 static int proc_gid_map_open(struct inode *inode, struct file *file)
2827 {
2828 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2829 }
2830
2831 static int proc_projid_map_open(struct inode *inode, struct file *file)
2832 {
2833 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2834 }
2835
2836 static const struct file_operations proc_uid_map_operations = {
2837 .open = proc_uid_map_open,
2838 .write = proc_uid_map_write,
2839 .read = seq_read,
2840 .llseek = seq_lseek,
2841 .release = proc_id_map_release,
2842 };
2843
2844 static const struct file_operations proc_gid_map_operations = {
2845 .open = proc_gid_map_open,
2846 .write = proc_gid_map_write,
2847 .read = seq_read,
2848 .llseek = seq_lseek,
2849 .release = proc_id_map_release,
2850 };
2851
2852 static const struct file_operations proc_projid_map_operations = {
2853 .open = proc_projid_map_open,
2854 .write = proc_projid_map_write,
2855 .read = seq_read,
2856 .llseek = seq_lseek,
2857 .release = proc_id_map_release,
2858 };
2859
2860 static int proc_setgroups_open(struct inode *inode, struct file *file)
2861 {
2862 struct user_namespace *ns = NULL;
2863 struct task_struct *task;
2864 int ret;
2865
2866 ret = -ESRCH;
2867 task = get_proc_task(inode);
2868 if (task) {
2869 rcu_read_lock();
2870 ns = get_user_ns(task_cred_xxx(task, user_ns));
2871 rcu_read_unlock();
2872 put_task_struct(task);
2873 }
2874 if (!ns)
2875 goto err;
2876
2877 if (file->f_mode & FMODE_WRITE) {
2878 ret = -EACCES;
2879 if (!ns_capable(ns, CAP_SYS_ADMIN))
2880 goto err_put_ns;
2881 }
2882
2883 ret = single_open(file, &proc_setgroups_show, ns);
2884 if (ret)
2885 goto err_put_ns;
2886
2887 return 0;
2888 err_put_ns:
2889 put_user_ns(ns);
2890 err:
2891 return ret;
2892 }
2893
2894 static int proc_setgroups_release(struct inode *inode, struct file *file)
2895 {
2896 struct seq_file *seq = file->private_data;
2897 struct user_namespace *ns = seq->private;
2898 int ret = single_release(inode, file);
2899 put_user_ns(ns);
2900 return ret;
2901 }
2902
2903 static const struct file_operations proc_setgroups_operations = {
2904 .open = proc_setgroups_open,
2905 .write = proc_setgroups_write,
2906 .read = seq_read,
2907 .llseek = seq_lseek,
2908 .release = proc_setgroups_release,
2909 };
2910 #endif /* CONFIG_USER_NS */
2911
2912 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2913 struct pid *pid, struct task_struct *task)
2914 {
2915 int err = lock_trace(task);
2916 if (!err) {
2917 seq_printf(m, "%08x\n", task->personality);
2918 unlock_trace(task);
2919 }
2920 return err;
2921 }
2922
2923 #ifdef CONFIG_LIVEPATCH
2924 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2925 struct pid *pid, struct task_struct *task)
2926 {
2927 seq_printf(m, "%d\n", task->patch_state);
2928 return 0;
2929 }
2930 #endif /* CONFIG_LIVEPATCH */
2931
2932 /*
2933 * Thread groups
2934 */
2935 static const struct file_operations proc_task_operations;
2936 static const struct inode_operations proc_task_inode_operations;
2937
2938 static const struct pid_entry tgid_base_stuff[] = {
2939 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2940 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2941 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2942 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2943 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2944 #ifdef CONFIG_NET
2945 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2946 #endif
2947 REG("environ", S_IRUSR, proc_environ_operations),
2948 REG("auxv", S_IRUSR, proc_auxv_operations),
2949 ONE("status", S_IRUGO, proc_pid_status),
2950 ONE("personality", S_IRUSR, proc_pid_personality),
2951 ONE("limits", S_IRUGO, proc_pid_limits),
2952 #ifdef CONFIG_SCHED_DEBUG
2953 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2954 #endif
2955 #ifdef CONFIG_SCHED_AUTOGROUP
2956 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2957 #endif
2958 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2959 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2960 ONE("syscall", S_IRUSR, proc_pid_syscall),
2961 #endif
2962 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2963 ONE("stat", S_IRUGO, proc_tgid_stat),
2964 ONE("statm", S_IRUGO, proc_pid_statm),
2965 REG("maps", S_IRUGO, proc_pid_maps_operations),
2966 #ifdef CONFIG_NUMA
2967 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2968 #endif
2969 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2970 LNK("cwd", proc_cwd_link),
2971 LNK("root", proc_root_link),
2972 LNK("exe", proc_exe_link),
2973 REG("mounts", S_IRUGO, proc_mounts_operations),
2974 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2975 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2976 #ifdef CONFIG_PROCESS_RECLAIM
2977 REG("reclaim", S_IWUGO, proc_reclaim_operations),
2978 #endif
2979 #ifdef CONFIG_PROC_PAGE_MONITOR
2980 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2981 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2982 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2983 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2984 #endif
2985 #ifdef CONFIG_SECURITY
2986 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2987 #endif
2988 #ifdef CONFIG_KALLSYMS
2989 ONE("wchan", S_IRUGO, proc_pid_wchan),
2990 #endif
2991 #ifdef CONFIG_STACKTRACE
2992 ONE("stack", S_IRUSR, proc_pid_stack),
2993 #endif
2994 #ifdef CONFIG_SCHED_INFO
2995 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2996 #endif
2997 #ifdef CONFIG_LATENCYTOP
2998 REG("latency", S_IRUGO, proc_lstats_operations),
2999 #endif
3000 #ifdef CONFIG_PROC_PID_CPUSET
3001 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3002 #endif
3003 #ifdef CONFIG_CGROUPS
3004 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3005 #endif
3006 ONE("oom_score", S_IRUGO, proc_oom_score),
3007 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3008 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3009 #ifdef CONFIG_AUDITSYSCALL
3010 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3011 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3012 #endif
3013 #ifdef CONFIG_FAULT_INJECTION
3014 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3015 REG("fail-nth", 0644, proc_fail_nth_operations),
3016 #endif
3017 #ifdef CONFIG_ELF_CORE
3018 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3019 #endif
3020 #ifdef CONFIG_TASK_IO_ACCOUNTING
3021 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3022 #endif
3023 #ifdef CONFIG_HARDWALL
3024 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3025 #endif
3026 #ifdef CONFIG_USER_NS
3027 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3028 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3029 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3030 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3031 #endif
3032 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3033 REG("timers", S_IRUGO, proc_timers_operations),
3034 #endif
3035 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3036 #ifdef CONFIG_LIVEPATCH
3037 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3038 #endif
3039 #ifdef CONFIG_CPU_FREQ_TIMES
3040 ONE("time_in_state", 0444, proc_time_in_state_show),
3041 #endif
3042 };
3043
3044 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3045 {
3046 return proc_pident_readdir(file, ctx,
3047 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3048 }
3049
3050 static const struct file_operations proc_tgid_base_operations = {
3051 .read = generic_read_dir,
3052 .iterate_shared = proc_tgid_base_readdir,
3053 .llseek = generic_file_llseek,
3054 };
3055
3056 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3057 {
3058 return proc_pident_lookup(dir, dentry,
3059 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3060 }
3061
3062 static const struct inode_operations proc_tgid_base_inode_operations = {
3063 .lookup = proc_tgid_base_lookup,
3064 .getattr = pid_getattr,
3065 .setattr = proc_setattr,
3066 .permission = proc_pid_permission,
3067 };
3068
3069 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3070 {
3071 struct dentry *dentry, *leader, *dir;
3072 char buf[PROC_NUMBUF];
3073 struct qstr name;
3074
3075 name.name = buf;
3076 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3077 /* no ->d_hash() rejects on procfs */
3078 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3079 if (dentry) {
3080 d_invalidate(dentry);
3081 dput(dentry);
3082 }
3083
3084 if (pid == tgid)
3085 return;
3086
3087 name.name = buf;
3088 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3089 leader = d_hash_and_lookup(mnt->mnt_root, &name);
3090 if (!leader)
3091 goto out;
3092
3093 name.name = "task";
3094 name.len = strlen(name.name);
3095 dir = d_hash_and_lookup(leader, &name);
3096 if (!dir)
3097 goto out_put_leader;
3098
3099 name.name = buf;
3100 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3101 dentry = d_hash_and_lookup(dir, &name);
3102 if (dentry) {
3103 d_invalidate(dentry);
3104 dput(dentry);
3105 }
3106
3107 dput(dir);
3108 out_put_leader:
3109 dput(leader);
3110 out:
3111 return;
3112 }
3113
3114 /**
3115 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3116 * @task: task that should be flushed.
3117 *
3118 * When flushing dentries from proc, one needs to flush them from global
3119 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3120 * in. This call is supposed to do all of this job.
3121 *
3122 * Looks in the dcache for
3123 * /proc/@pid
3124 * /proc/@tgid/task/@pid
3125 * if either directory is present flushes it and all of it'ts children
3126 * from the dcache.
3127 *
3128 * It is safe and reasonable to cache /proc entries for a task until
3129 * that task exits. After that they just clog up the dcache with
3130 * useless entries, possibly causing useful dcache entries to be
3131 * flushed instead. This routine is proved to flush those useless
3132 * dcache entries at process exit time.
3133 *
3134 * NOTE: This routine is just an optimization so it does not guarantee
3135 * that no dcache entries will exist at process exit time it
3136 * just makes it very unlikely that any will persist.
3137 */
3138
3139 void proc_flush_task(struct task_struct *task)
3140 {
3141 int i;
3142 struct pid *pid, *tgid;
3143 struct upid *upid;
3144
3145 pid = task_pid(task);
3146 tgid = task_tgid(task);
3147
3148 for (i = 0; i <= pid->level; i++) {
3149 upid = &pid->numbers[i];
3150 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3151 tgid->numbers[i].nr);
3152 }
3153 }
3154
3155 static int proc_pid_instantiate(struct inode *dir,
3156 struct dentry * dentry,
3157 struct task_struct *task, const void *ptr)
3158 {
3159 struct inode *inode;
3160
3161 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3162 if (!inode)
3163 goto out;
3164
3165 inode->i_op = &proc_tgid_base_inode_operations;
3166 inode->i_fop = &proc_tgid_base_operations;
3167 inode->i_flags|=S_IMMUTABLE;
3168
3169 set_nlink(inode, nlink_tgid);
3170
3171 d_set_d_op(dentry, &pid_dentry_operations);
3172
3173 d_add(dentry, inode);
3174 /* Close the race of the process dying before we return the dentry */
3175 if (pid_revalidate(dentry, 0))
3176 return 0;
3177 out:
3178 return -ENOENT;
3179 }
3180
3181 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3182 {
3183 int result = -ENOENT;
3184 struct task_struct *task;
3185 unsigned tgid;
3186 struct pid_namespace *ns;
3187
3188 tgid = name_to_int(&dentry->d_name);
3189 if (tgid == ~0U)
3190 goto out;
3191
3192 ns = dentry->d_sb->s_fs_info;
3193 rcu_read_lock();
3194 task = find_task_by_pid_ns(tgid, ns);
3195 if (task)
3196 get_task_struct(task);
3197 rcu_read_unlock();
3198 if (!task)
3199 goto out;
3200
3201 result = proc_pid_instantiate(dir, dentry, task, NULL);
3202 put_task_struct(task);
3203 out:
3204 return ERR_PTR(result);
3205 }
3206
3207 /*
3208 * Find the first task with tgid >= tgid
3209 *
3210 */
3211 struct tgid_iter {
3212 unsigned int tgid;
3213 struct task_struct *task;
3214 };
3215 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3216 {
3217 struct pid *pid;
3218
3219 if (iter.task)
3220 put_task_struct(iter.task);
3221 rcu_read_lock();
3222 retry:
3223 iter.task = NULL;
3224 pid = find_ge_pid(iter.tgid, ns);
3225 if (pid) {
3226 iter.tgid = pid_nr_ns(pid, ns);
3227 iter.task = pid_task(pid, PIDTYPE_PID);
3228 /* What we to know is if the pid we have find is the
3229 * pid of a thread_group_leader. Testing for task
3230 * being a thread_group_leader is the obvious thing
3231 * todo but there is a window when it fails, due to
3232 * the pid transfer logic in de_thread.
3233 *
3234 * So we perform the straight forward test of seeing
3235 * if the pid we have found is the pid of a thread
3236 * group leader, and don't worry if the task we have
3237 * found doesn't happen to be a thread group leader.
3238 * As we don't care in the case of readdir.
3239 */
3240 if (!iter.task || !has_group_leader_pid(iter.task)) {
3241 iter.tgid += 1;
3242 goto retry;
3243 }
3244 get_task_struct(iter.task);
3245 }
3246 rcu_read_unlock();
3247 return iter;
3248 }
3249
3250 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3251
3252 /* for the /proc/ directory itself, after non-process stuff has been done */
3253 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3254 {
3255 struct tgid_iter iter;
3256 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3257 loff_t pos = ctx->pos;
3258
3259 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3260 return 0;
3261
3262 if (pos == TGID_OFFSET - 2) {
3263 struct inode *inode = d_inode(ns->proc_self);
3264 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3265 return 0;
3266 ctx->pos = pos = pos + 1;
3267 }
3268 if (pos == TGID_OFFSET - 1) {
3269 struct inode *inode = d_inode(ns->proc_thread_self);
3270 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3271 return 0;
3272 ctx->pos = pos = pos + 1;
3273 }
3274 iter.tgid = pos - TGID_OFFSET;
3275 iter.task = NULL;
3276 for (iter = next_tgid(ns, iter);
3277 iter.task;
3278 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3279 char name[PROC_NUMBUF];
3280 int len;
3281
3282 cond_resched();
3283 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3284 continue;
3285
3286 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3287 ctx->pos = iter.tgid + TGID_OFFSET;
3288 if (!proc_fill_cache(file, ctx, name, len,
3289 proc_pid_instantiate, iter.task, NULL)) {
3290 put_task_struct(iter.task);
3291 return 0;
3292 }
3293 }
3294 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3295 return 0;
3296 }
3297
3298 /*
3299 * proc_tid_comm_permission is a special permission function exclusively
3300 * used for the node /proc/<pid>/task/<tid>/comm.
3301 * It bypasses generic permission checks in the case where a task of the same
3302 * task group attempts to access the node.
3303 * The rationale behind this is that glibc and bionic access this node for
3304 * cross thread naming (pthread_set/getname_np(!self)). However, if
3305 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3306 * which locks out the cross thread naming implementation.
3307 * This function makes sure that the node is always accessible for members of
3308 * same thread group.
3309 */
3310 static int proc_tid_comm_permission(struct inode *inode, int mask)
3311 {
3312 bool is_same_tgroup;
3313 struct task_struct *task;
3314
3315 task = get_proc_task(inode);
3316 if (!task)
3317 return -ESRCH;
3318 is_same_tgroup = same_thread_group(current, task);
3319 put_task_struct(task);
3320
3321 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3322 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3323 * read or written by the members of the corresponding
3324 * thread group.
3325 */
3326 return 0;
3327 }
3328
3329 return generic_permission(inode, mask);
3330 }
3331
3332 static const struct inode_operations proc_tid_comm_inode_operations = {
3333 .permission = proc_tid_comm_permission,
3334 };
3335
3336 /*
3337 * Tasks
3338 */
3339 static const struct pid_entry tid_base_stuff[] = {
3340 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3341 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3342 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3343 #ifdef CONFIG_NET
3344 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3345 #endif
3346 REG("environ", S_IRUSR, proc_environ_operations),
3347 REG("auxv", S_IRUSR, proc_auxv_operations),
3348 ONE("status", S_IRUGO, proc_pid_status),
3349 ONE("personality", S_IRUSR, proc_pid_personality),
3350 ONE("limits", S_IRUGO, proc_pid_limits),
3351 #ifdef CONFIG_SCHED_DEBUG
3352 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3353 #endif
3354 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3355 &proc_tid_comm_inode_operations,
3356 &proc_pid_set_comm_operations, {}),
3357 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3358 ONE("syscall", S_IRUSR, proc_pid_syscall),
3359 #endif
3360 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3361 ONE("stat", S_IRUGO, proc_tid_stat),
3362 ONE("statm", S_IRUGO, proc_pid_statm),
3363 REG("maps", S_IRUGO, proc_tid_maps_operations),
3364 #ifdef CONFIG_PROC_CHILDREN
3365 REG("children", S_IRUGO, proc_tid_children_operations),
3366 #endif
3367 #ifdef CONFIG_NUMA
3368 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3369 #endif
3370 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3371 LNK("cwd", proc_cwd_link),
3372 LNK("root", proc_root_link),
3373 LNK("exe", proc_exe_link),
3374 REG("mounts", S_IRUGO, proc_mounts_operations),
3375 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3376 #ifdef CONFIG_PROC_PAGE_MONITOR
3377 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3378 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3379 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3380 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3381 #endif
3382 #ifdef CONFIG_SECURITY
3383 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3384 #endif
3385 #ifdef CONFIG_KALLSYMS
3386 ONE("wchan", S_IRUGO, proc_pid_wchan),
3387 #endif
3388 #ifdef CONFIG_STACKTRACE
3389 ONE("stack", S_IRUSR, proc_pid_stack),
3390 #endif
3391 #ifdef CONFIG_SCHED_INFO
3392 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3393 #endif
3394 #ifdef CONFIG_LATENCYTOP
3395 REG("latency", S_IRUGO, proc_lstats_operations),
3396 #endif
3397 #ifdef CONFIG_PROC_PID_CPUSET
3398 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3399 #endif
3400 #ifdef CONFIG_CGROUPS
3401 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3402 #endif
3403 ONE("oom_score", S_IRUGO, proc_oom_score),
3404 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3405 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3406 #ifdef CONFIG_AUDITSYSCALL
3407 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3408 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3409 #endif
3410 #ifdef CONFIG_FAULT_INJECTION
3411 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3412 REG("fail-nth", 0644, proc_fail_nth_operations),
3413 #endif
3414 #ifdef CONFIG_TASK_IO_ACCOUNTING
3415 ONE("io", S_IRUSR, proc_tid_io_accounting),
3416 #endif
3417 #ifdef CONFIG_HARDWALL
3418 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3419 #endif
3420 #ifdef CONFIG_USER_NS
3421 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3422 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3423 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3424 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3425 #endif
3426 #ifdef CONFIG_LIVEPATCH
3427 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3428 #endif
3429 #ifdef CONFIG_CPU_FREQ_TIMES
3430 ONE("time_in_state", 0444, proc_time_in_state_show),
3431 #endif
3432 };
3433
3434 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3435 {
3436 return proc_pident_readdir(file, ctx,
3437 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3438 }
3439
3440 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3441 {
3442 return proc_pident_lookup(dir, dentry,
3443 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3444 }
3445
3446 static const struct file_operations proc_tid_base_operations = {
3447 .read = generic_read_dir,
3448 .iterate_shared = proc_tid_base_readdir,
3449 .llseek = generic_file_llseek,
3450 };
3451
3452 static const struct inode_operations proc_tid_base_inode_operations = {
3453 .lookup = proc_tid_base_lookup,
3454 .getattr = pid_getattr,
3455 .setattr = proc_setattr,
3456 };
3457
3458 static int proc_task_instantiate(struct inode *dir,
3459 struct dentry *dentry, struct task_struct *task, const void *ptr)
3460 {
3461 struct inode *inode;
3462 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3463
3464 if (!inode)
3465 goto out;
3466 inode->i_op = &proc_tid_base_inode_operations;
3467 inode->i_fop = &proc_tid_base_operations;
3468 inode->i_flags|=S_IMMUTABLE;
3469
3470 set_nlink(inode, nlink_tid);
3471
3472 d_set_d_op(dentry, &pid_dentry_operations);
3473
3474 d_add(dentry, inode);
3475 /* Close the race of the process dying before we return the dentry */
3476 if (pid_revalidate(dentry, 0))
3477 return 0;
3478 out:
3479 return -ENOENT;
3480 }
3481
3482 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3483 {
3484 int result = -ENOENT;
3485 struct task_struct *task;
3486 struct task_struct *leader = get_proc_task(dir);
3487 unsigned tid;
3488 struct pid_namespace *ns;
3489
3490 if (!leader)
3491 goto out_no_task;
3492
3493 tid = name_to_int(&dentry->d_name);
3494 if (tid == ~0U)
3495 goto out;
3496
3497 ns = dentry->d_sb->s_fs_info;
3498 rcu_read_lock();
3499 task = find_task_by_pid_ns(tid, ns);
3500 if (task)
3501 get_task_struct(task);
3502 rcu_read_unlock();
3503 if (!task)
3504 goto out;
3505 if (!same_thread_group(leader, task))
3506 goto out_drop_task;
3507
3508 result = proc_task_instantiate(dir, dentry, task, NULL);
3509 out_drop_task:
3510 put_task_struct(task);
3511 out:
3512 put_task_struct(leader);
3513 out_no_task:
3514 return ERR_PTR(result);
3515 }
3516
3517 /*
3518 * Find the first tid of a thread group to return to user space.
3519 *
3520 * Usually this is just the thread group leader, but if the users
3521 * buffer was too small or there was a seek into the middle of the
3522 * directory we have more work todo.
3523 *
3524 * In the case of a short read we start with find_task_by_pid.
3525 *
3526 * In the case of a seek we start with the leader and walk nr
3527 * threads past it.
3528 */
3529 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3530 struct pid_namespace *ns)
3531 {
3532 struct task_struct *pos, *task;
3533 unsigned long nr = f_pos;
3534
3535 if (nr != f_pos) /* 32bit overflow? */
3536 return NULL;
3537
3538 rcu_read_lock();
3539 task = pid_task(pid, PIDTYPE_PID);
3540 if (!task)
3541 goto fail;
3542
3543 /* Attempt to start with the tid of a thread */
3544 if (tid && nr) {
3545 pos = find_task_by_pid_ns(tid, ns);
3546 if (pos && same_thread_group(pos, task))
3547 goto found;
3548 }
3549
3550 /* If nr exceeds the number of threads there is nothing todo */
3551 if (nr >= get_nr_threads(task))
3552 goto fail;
3553
3554 /* If we haven't found our starting place yet start
3555 * with the leader and walk nr threads forward.
3556 */
3557 pos = task = task->group_leader;
3558 do {
3559 if (!nr--)
3560 goto found;
3561 } while_each_thread(task, pos);
3562 fail:
3563 pos = NULL;
3564 goto out;
3565 found:
3566 get_task_struct(pos);
3567 out:
3568 rcu_read_unlock();
3569 return pos;
3570 }
3571
3572 /*
3573 * Find the next thread in the thread list.
3574 * Return NULL if there is an error or no next thread.
3575 *
3576 * The reference to the input task_struct is released.
3577 */
3578 static struct task_struct *next_tid(struct task_struct *start)
3579 {
3580 struct task_struct *pos = NULL;
3581 rcu_read_lock();
3582 if (pid_alive(start)) {
3583 pos = next_thread(start);
3584 if (thread_group_leader(pos))
3585 pos = NULL;
3586 else
3587 get_task_struct(pos);
3588 }
3589 rcu_read_unlock();
3590 put_task_struct(start);
3591 return pos;
3592 }
3593
3594 /* for the /proc/TGID/task/ directories */
3595 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3596 {
3597 struct inode *inode = file_inode(file);
3598 struct task_struct *task;
3599 struct pid_namespace *ns;
3600 int tid;
3601
3602 if (proc_inode_is_dead(inode))
3603 return -ENOENT;
3604
3605 if (!dir_emit_dots(file, ctx))
3606 return 0;
3607
3608 /* f_version caches the tgid value that the last readdir call couldn't
3609 * return. lseek aka telldir automagically resets f_version to 0.
3610 */
3611 ns = inode->i_sb->s_fs_info;
3612 tid = (int)file->f_version;
3613 file->f_version = 0;
3614 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3615 task;
3616 task = next_tid(task), ctx->pos++) {
3617 char name[PROC_NUMBUF];
3618 int len;
3619 tid = task_pid_nr_ns(task, ns);
3620 len = snprintf(name, sizeof(name), "%d", tid);
3621 if (!proc_fill_cache(file, ctx, name, len,
3622 proc_task_instantiate, task, NULL)) {
3623 /* returning this tgid failed, save it as the first
3624 * pid for the next readir call */
3625 file->f_version = (u64)tid;
3626 put_task_struct(task);
3627 break;
3628 }
3629 }
3630
3631 return 0;
3632 }
3633
3634 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3635 u32 request_mask, unsigned int query_flags)
3636 {
3637 struct inode *inode = d_inode(path->dentry);
3638 struct task_struct *p = get_proc_task(inode);
3639 generic_fillattr(inode, stat);
3640
3641 if (p) {
3642 stat->nlink += get_nr_threads(p);
3643 put_task_struct(p);
3644 }
3645
3646 return 0;
3647 }
3648
3649 static const struct inode_operations proc_task_inode_operations = {
3650 .lookup = proc_task_lookup,
3651 .getattr = proc_task_getattr,
3652 .setattr = proc_setattr,
3653 .permission = proc_pid_permission,
3654 };
3655
3656 static const struct file_operations proc_task_operations = {
3657 .read = generic_read_dir,
3658 .iterate_shared = proc_task_readdir,
3659 .llseek = generic_file_llseek,
3660 };
3661
3662 void __init set_proc_pid_nlink(void)
3663 {
3664 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3665 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3666 }