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