xref: /linux/fs/proc/base.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/proc/base.c
4  *
5  *  Copyright (C) 1991, 1992 Linus Torvalds
6  *
7  *  proc base directory handling functions
8  *
9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10  *  Instead of using magical inumbers to determine the kind of object
11  *  we allocate and fill in-core inodes upon lookup. They don't even
12  *  go into icache. We cache the reference to task_struct upon lookup too.
13  *  Eventually it should become a filesystem in its own. We don't use the
14  *  rest of procfs anymore.
15  *
16  *
17  *  Changelog:
18  *  17-Jan-2005
19  *  Allan Bezerra
20  *  Bruna Moreira <bruna.moreira@indt.org.br>
21  *  Edjard Mota <edjard.mota@indt.org.br>
22  *  Ilias Biris <ilias.biris@indt.org.br>
23  *  Mauricio Lin <mauricio.lin@indt.org.br>
24  *
25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26  *
27  *  A new process specific entry (smaps) included in /proc. It shows the
28  *  size of rss for each memory area. The maps entry lacks information
29  *  about physical memory size (rss) for each mapped file, i.e.,
30  *  rss information for executables and library files.
31  *  This additional information is useful for any tools that need to know
32  *  about physical memory consumption for a process specific library.
33  *
34  *  Changelog:
35  *  21-Feb-2005
36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37  *  Pud inclusion in the page table walking.
38  *
39  *  ChangeLog:
40  *  10-Mar-2005
41  *  10LE Instituto Nokia de Tecnologia - INdT:
42  *  A better way to walks through the page table as suggested by Hugh Dickins.
43  *
44  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
45  *  Smaps information related to shared, private, clean and dirty pages.
46  *
47  *  Paul Mundt <paul.mundt@nokia.com>:
48  *  Overall revision about smaps.
49  */
50 
51 #include <linux/uaccess.h>
52 
53 #include <linux/errno.h>
54 #include <linux/time.h>
55 #include <linux/proc_fs.h>
56 #include <linux/stat.h>
57 #include <linux/task_io_accounting_ops.h>
58 #include <linux/init.h>
59 #include <linux/capability.h>
60 #include <linux/file.h>
61 #include <linux/fdtable.h>
62 #include <linux/generic-radix-tree.h>
63 #include <linux/string.h>
64 #include <linux/seq_file.h>
65 #include <linux/namei.h>
66 #include <linux/mnt_namespace.h>
67 #include <linux/mm.h>
68 #include <linux/swap.h>
69 #include <linux/rcupdate.h>
70 #include <linux/kallsyms.h>
71 #include <linux/stacktrace.h>
72 #include <linux/resource.h>
73 #include <linux/module.h>
74 #include <linux/mount.h>
75 #include <linux/security.h>
76 #include <linux/ptrace.h>
77 #include <linux/tracehook.h>
78 #include <linux/printk.h>
79 #include <linux/cache.h>
80 #include <linux/cgroup.h>
81 #include <linux/cpuset.h>
82 #include <linux/audit.h>
83 #include <linux/poll.h>
84 #include <linux/nsproxy.h>
85 #include <linux/oom.h>
86 #include <linux/elf.h>
87 #include <linux/pid_namespace.h>
88 #include <linux/user_namespace.h>
89 #include <linux/fs_struct.h>
90 #include <linux/slab.h>
91 #include <linux/sched/autogroup.h>
92 #include <linux/sched/mm.h>
93 #include <linux/sched/coredump.h>
94 #include <linux/sched/debug.h>
95 #include <linux/sched/stat.h>
96 #include <linux/posix-timers.h>
97 #include <trace/events/oom.h>
98 #include "internal.h"
99 #include "fd.h"
100 
101 #include "../../lib/kstrtox.h"
102 
103 /* NOTE:
104  *	Implementing inode permission operations in /proc is almost
105  *	certainly an error.  Permission checks need to happen during
106  *	each system call not at open time.  The reason is that most of
107  *	what we wish to check for permissions in /proc varies at runtime.
108  *
109  *	The classic example of a problem is opening file descriptors
110  *	in /proc for a task before it execs a suid executable.
111  */
112 
113 static u8 nlink_tid __ro_after_init;
114 static u8 nlink_tgid __ro_after_init;
115 
116 struct pid_entry {
117 	const char *name;
118 	unsigned int len;
119 	umode_t mode;
120 	const struct inode_operations *iop;
121 	const struct file_operations *fop;
122 	union proc_op op;
123 };
124 
125 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
126 	.name = (NAME),					\
127 	.len  = sizeof(NAME) - 1,			\
128 	.mode = MODE,					\
129 	.iop  = IOP,					\
130 	.fop  = FOP,					\
131 	.op   = OP,					\
132 }
133 
134 #define DIR(NAME, MODE, iops, fops)	\
135 	NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
136 #define LNK(NAME, get_link)					\
137 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
138 		&proc_pid_link_inode_operations, NULL,		\
139 		{ .proc_get_link = get_link } )
140 #define REG(NAME, MODE, fops)				\
141 	NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
142 #define ONE(NAME, MODE, show)				\
143 	NOD(NAME, (S_IFREG|(MODE)),			\
144 		NULL, &proc_single_file_operations,	\
145 		{ .proc_show = show } )
146 #define ATTR(LSM, NAME, MODE)				\
147 	NOD(NAME, (S_IFREG|(MODE)),			\
148 		NULL, &proc_pid_attr_operations,	\
149 		{ .lsm = LSM })
150 
151 /*
152  * Count the number of hardlinks for the pid_entry table, excluding the .
153  * and .. links.
154  */
155 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
156 	unsigned int n)
157 {
158 	unsigned int i;
159 	unsigned int count;
160 
161 	count = 2;
162 	for (i = 0; i < n; ++i) {
163 		if (S_ISDIR(entries[i].mode))
164 			++count;
165 	}
166 
167 	return count;
168 }
169 
170 static int get_task_root(struct task_struct *task, struct path *root)
171 {
172 	int result = -ENOENT;
173 
174 	task_lock(task);
175 	if (task->fs) {
176 		get_fs_root(task->fs, root);
177 		result = 0;
178 	}
179 	task_unlock(task);
180 	return result;
181 }
182 
183 static int proc_cwd_link(struct dentry *dentry, struct path *path)
184 {
185 	struct task_struct *task = get_proc_task(d_inode(dentry));
186 	int result = -ENOENT;
187 
188 	if (task) {
189 		task_lock(task);
190 		if (task->fs) {
191 			get_fs_pwd(task->fs, path);
192 			result = 0;
193 		}
194 		task_unlock(task);
195 		put_task_struct(task);
196 	}
197 	return result;
198 }
199 
200 static int proc_root_link(struct dentry *dentry, struct path *path)
201 {
202 	struct task_struct *task = get_proc_task(d_inode(dentry));
203 	int result = -ENOENT;
204 
205 	if (task) {
206 		result = get_task_root(task, path);
207 		put_task_struct(task);
208 	}
209 	return result;
210 }
211 
212 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
213 			      size_t count, loff_t *ppos)
214 {
215 	unsigned long arg_start, arg_end, env_start, env_end;
216 	unsigned long pos, len;
217 	char *page;
218 
219 	/* Check if process spawned far enough to have cmdline. */
220 	if (!mm->env_end)
221 		return 0;
222 
223 	spin_lock(&mm->arg_lock);
224 	arg_start = mm->arg_start;
225 	arg_end = mm->arg_end;
226 	env_start = mm->env_start;
227 	env_end = mm->env_end;
228 	spin_unlock(&mm->arg_lock);
229 
230 	if (arg_start >= arg_end)
231 		return 0;
232 
233 	/*
234 	 * We have traditionally allowed the user to re-write
235 	 * the argument strings and overflow the end result
236 	 * into the environment section. But only do that if
237 	 * the environment area is contiguous to the arguments.
238 	 */
239 	if (env_start != arg_end || env_start >= env_end)
240 		env_start = env_end = arg_end;
241 
242 	/* .. and limit it to a maximum of one page of slop */
243 	if (env_end >= arg_end + PAGE_SIZE)
244 		env_end = arg_end + PAGE_SIZE - 1;
245 
246 	/* We're not going to care if "*ppos" has high bits set */
247 	pos = arg_start + *ppos;
248 
249 	/* .. but we do check the result is in the proper range */
250 	if (pos < arg_start || pos >= env_end)
251 		return 0;
252 
253 	/* .. and we never go past env_end */
254 	if (env_end - pos < count)
255 		count = env_end - pos;
256 
257 	page = (char *)__get_free_page(GFP_KERNEL);
258 	if (!page)
259 		return -ENOMEM;
260 
261 	len = 0;
262 	while (count) {
263 		int got;
264 		size_t size = min_t(size_t, PAGE_SIZE, count);
265 		long offset;
266 
267 		/*
268 		 * Are we already starting past the official end?
269 		 * We always include the last byte that is *supposed*
270 		 * to be NUL
271 		 */
272 		offset = (pos >= arg_end) ? pos - arg_end + 1 : 0;
273 
274 		got = access_remote_vm(mm, pos - offset, page, size + offset, FOLL_ANON);
275 		if (got <= offset)
276 			break;
277 		got -= offset;
278 
279 		/* Don't walk past a NUL character once you hit arg_end */
280 		if (pos + got >= arg_end) {
281 			int n = 0;
282 
283 			/*
284 			 * If we started before 'arg_end' but ended up
285 			 * at or after it, we start the NUL character
286 			 * check at arg_end-1 (where we expect the normal
287 			 * EOF to be).
288 			 *
289 			 * NOTE! This is smaller than 'got', because
290 			 * pos + got >= arg_end
291 			 */
292 			if (pos < arg_end)
293 				n = arg_end - pos - 1;
294 
295 			/* Cut off at first NUL after 'n' */
296 			got = n + strnlen(page+n, offset+got-n);
297 			if (got < offset)
298 				break;
299 			got -= offset;
300 
301 			/* Include the NUL if it existed */
302 			if (got < size)
303 				got++;
304 		}
305 
306 		got -= copy_to_user(buf, page+offset, got);
307 		if (unlikely(!got)) {
308 			if (!len)
309 				len = -EFAULT;
310 			break;
311 		}
312 		pos += got;
313 		buf += got;
314 		len += got;
315 		count -= got;
316 	}
317 
318 	free_page((unsigned long)page);
319 	return len;
320 }
321 
322 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
323 				size_t count, loff_t *pos)
324 {
325 	struct mm_struct *mm;
326 	ssize_t ret;
327 
328 	mm = get_task_mm(tsk);
329 	if (!mm)
330 		return 0;
331 
332 	ret = get_mm_cmdline(mm, buf, count, pos);
333 	mmput(mm);
334 	return ret;
335 }
336 
337 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
338 				     size_t count, loff_t *pos)
339 {
340 	struct task_struct *tsk;
341 	ssize_t ret;
342 
343 	BUG_ON(*pos < 0);
344 
345 	tsk = get_proc_task(file_inode(file));
346 	if (!tsk)
347 		return -ESRCH;
348 	ret = get_task_cmdline(tsk, buf, count, pos);
349 	put_task_struct(tsk);
350 	if (ret > 0)
351 		*pos += ret;
352 	return ret;
353 }
354 
355 static const struct file_operations proc_pid_cmdline_ops = {
356 	.read	= proc_pid_cmdline_read,
357 	.llseek	= generic_file_llseek,
358 };
359 
360 #ifdef CONFIG_KALLSYMS
361 /*
362  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
363  * Returns the resolved symbol.  If that fails, simply return the address.
364  */
365 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
366 			  struct pid *pid, struct task_struct *task)
367 {
368 	unsigned long wchan;
369 	char symname[KSYM_NAME_LEN];
370 
371 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
372 		goto print0;
373 
374 	wchan = get_wchan(task);
375 	if (wchan && !lookup_symbol_name(wchan, symname)) {
376 		seq_puts(m, symname);
377 		return 0;
378 	}
379 
380 print0:
381 	seq_putc(m, '0');
382 	return 0;
383 }
384 #endif /* CONFIG_KALLSYMS */
385 
386 static int lock_trace(struct task_struct *task)
387 {
388 	int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
389 	if (err)
390 		return err;
391 	if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
392 		mutex_unlock(&task->signal->cred_guard_mutex);
393 		return -EPERM;
394 	}
395 	return 0;
396 }
397 
398 static void unlock_trace(struct task_struct *task)
399 {
400 	mutex_unlock(&task->signal->cred_guard_mutex);
401 }
402 
403 #ifdef CONFIG_STACKTRACE
404 
405 #define MAX_STACK_TRACE_DEPTH	64
406 
407 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
408 			  struct pid *pid, struct task_struct *task)
409 {
410 	unsigned long *entries;
411 	int err;
412 
413 	/*
414 	 * The ability to racily run the kernel stack unwinder on a running task
415 	 * and then observe the unwinder output is scary; while it is useful for
416 	 * debugging kernel issues, it can also allow an attacker to leak kernel
417 	 * stack contents.
418 	 * Doing this in a manner that is at least safe from races would require
419 	 * some work to ensure that the remote task can not be scheduled; and
420 	 * even then, this would still expose the unwinder as local attack
421 	 * surface.
422 	 * Therefore, this interface is restricted to root.
423 	 */
424 	if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
425 		return -EACCES;
426 
427 	entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
428 				GFP_KERNEL);
429 	if (!entries)
430 		return -ENOMEM;
431 
432 	err = lock_trace(task);
433 	if (!err) {
434 		unsigned int i, nr_entries;
435 
436 		nr_entries = stack_trace_save_tsk(task, entries,
437 						  MAX_STACK_TRACE_DEPTH, 0);
438 
439 		for (i = 0; i < nr_entries; i++) {
440 			seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
441 		}
442 
443 		unlock_trace(task);
444 	}
445 	kfree(entries);
446 
447 	return err;
448 }
449 #endif
450 
451 #ifdef CONFIG_SCHED_INFO
452 /*
453  * Provides /proc/PID/schedstat
454  */
455 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
456 			      struct pid *pid, struct task_struct *task)
457 {
458 	if (unlikely(!sched_info_on()))
459 		seq_puts(m, "0 0 0\n");
460 	else
461 		seq_printf(m, "%llu %llu %lu\n",
462 		   (unsigned long long)task->se.sum_exec_runtime,
463 		   (unsigned long long)task->sched_info.run_delay,
464 		   task->sched_info.pcount);
465 
466 	return 0;
467 }
468 #endif
469 
470 #ifdef CONFIG_LATENCYTOP
471 static int lstats_show_proc(struct seq_file *m, void *v)
472 {
473 	int i;
474 	struct inode *inode = m->private;
475 	struct task_struct *task = get_proc_task(inode);
476 
477 	if (!task)
478 		return -ESRCH;
479 	seq_puts(m, "Latency Top version : v0.1\n");
480 	for (i = 0; i < LT_SAVECOUNT; i++) {
481 		struct latency_record *lr = &task->latency_record[i];
482 		if (lr->backtrace[0]) {
483 			int q;
484 			seq_printf(m, "%i %li %li",
485 				   lr->count, lr->time, lr->max);
486 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
487 				unsigned long bt = lr->backtrace[q];
488 
489 				if (!bt)
490 					break;
491 				seq_printf(m, " %ps", (void *)bt);
492 			}
493 			seq_putc(m, '\n');
494 		}
495 
496 	}
497 	put_task_struct(task);
498 	return 0;
499 }
500 
501 static int lstats_open(struct inode *inode, struct file *file)
502 {
503 	return single_open(file, lstats_show_proc, inode);
504 }
505 
506 static ssize_t lstats_write(struct file *file, const char __user *buf,
507 			    size_t count, loff_t *offs)
508 {
509 	struct task_struct *task = get_proc_task(file_inode(file));
510 
511 	if (!task)
512 		return -ESRCH;
513 	clear_tsk_latency_tracing(task);
514 	put_task_struct(task);
515 
516 	return count;
517 }
518 
519 static const struct file_operations proc_lstats_operations = {
520 	.open		= lstats_open,
521 	.read		= seq_read,
522 	.write		= lstats_write,
523 	.llseek		= seq_lseek,
524 	.release	= single_release,
525 };
526 
527 #endif
528 
529 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
530 			  struct pid *pid, struct task_struct *task)
531 {
532 	unsigned long totalpages = totalram_pages() + total_swap_pages;
533 	unsigned long points = 0;
534 
535 	points = oom_badness(task, totalpages) * 1000 / totalpages;
536 	seq_printf(m, "%lu\n", points);
537 
538 	return 0;
539 }
540 
541 struct limit_names {
542 	const char *name;
543 	const char *unit;
544 };
545 
546 static const struct limit_names lnames[RLIM_NLIMITS] = {
547 	[RLIMIT_CPU] = {"Max cpu time", "seconds"},
548 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
549 	[RLIMIT_DATA] = {"Max data size", "bytes"},
550 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
551 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
552 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
553 	[RLIMIT_NPROC] = {"Max processes", "processes"},
554 	[RLIMIT_NOFILE] = {"Max open files", "files"},
555 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
556 	[RLIMIT_AS] = {"Max address space", "bytes"},
557 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
558 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
559 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
560 	[RLIMIT_NICE] = {"Max nice priority", NULL},
561 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
562 	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
563 };
564 
565 /* Display limits for a process */
566 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
567 			   struct pid *pid, struct task_struct *task)
568 {
569 	unsigned int i;
570 	unsigned long flags;
571 
572 	struct rlimit rlim[RLIM_NLIMITS];
573 
574 	if (!lock_task_sighand(task, &flags))
575 		return 0;
576 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
577 	unlock_task_sighand(task, &flags);
578 
579 	/*
580 	 * print the file header
581 	 */
582 	seq_puts(m, "Limit                     "
583 		"Soft Limit           "
584 		"Hard Limit           "
585 		"Units     \n");
586 
587 	for (i = 0; i < RLIM_NLIMITS; i++) {
588 		if (rlim[i].rlim_cur == RLIM_INFINITY)
589 			seq_printf(m, "%-25s %-20s ",
590 				   lnames[i].name, "unlimited");
591 		else
592 			seq_printf(m, "%-25s %-20lu ",
593 				   lnames[i].name, rlim[i].rlim_cur);
594 
595 		if (rlim[i].rlim_max == RLIM_INFINITY)
596 			seq_printf(m, "%-20s ", "unlimited");
597 		else
598 			seq_printf(m, "%-20lu ", rlim[i].rlim_max);
599 
600 		if (lnames[i].unit)
601 			seq_printf(m, "%-10s\n", lnames[i].unit);
602 		else
603 			seq_putc(m, '\n');
604 	}
605 
606 	return 0;
607 }
608 
609 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
610 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
611 			    struct pid *pid, struct task_struct *task)
612 {
613 	struct syscall_info info;
614 	u64 *args = &info.data.args[0];
615 	int res;
616 
617 	res = lock_trace(task);
618 	if (res)
619 		return res;
620 
621 	if (task_current_syscall(task, &info))
622 		seq_puts(m, "running\n");
623 	else if (info.data.nr < 0)
624 		seq_printf(m, "%d 0x%llx 0x%llx\n",
625 			   info.data.nr, info.sp, info.data.instruction_pointer);
626 	else
627 		seq_printf(m,
628 		       "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
629 		       info.data.nr,
630 		       args[0], args[1], args[2], args[3], args[4], args[5],
631 		       info.sp, info.data.instruction_pointer);
632 	unlock_trace(task);
633 
634 	return 0;
635 }
636 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
637 
638 /************************************************************************/
639 /*                       Here the fs part begins                        */
640 /************************************************************************/
641 
642 /* permission checks */
643 static int proc_fd_access_allowed(struct inode *inode)
644 {
645 	struct task_struct *task;
646 	int allowed = 0;
647 	/* Allow access to a task's file descriptors if it is us or we
648 	 * may use ptrace attach to the process and find out that
649 	 * information.
650 	 */
651 	task = get_proc_task(inode);
652 	if (task) {
653 		allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
654 		put_task_struct(task);
655 	}
656 	return allowed;
657 }
658 
659 int proc_setattr(struct dentry *dentry, struct iattr *attr)
660 {
661 	int error;
662 	struct inode *inode = d_inode(dentry);
663 
664 	if (attr->ia_valid & ATTR_MODE)
665 		return -EPERM;
666 
667 	error = setattr_prepare(dentry, attr);
668 	if (error)
669 		return error;
670 
671 	setattr_copy(inode, attr);
672 	mark_inode_dirty(inode);
673 	return 0;
674 }
675 
676 /*
677  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
678  * or euid/egid (for hide_pid_min=2)?
679  */
680 static bool has_pid_permissions(struct pid_namespace *pid,
681 				 struct task_struct *task,
682 				 int hide_pid_min)
683 {
684 	if (pid->hide_pid < hide_pid_min)
685 		return true;
686 	if (in_group_p(pid->pid_gid))
687 		return true;
688 	return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
689 }
690 
691 
692 static int proc_pid_permission(struct inode *inode, int mask)
693 {
694 	struct pid_namespace *pid = proc_pid_ns(inode);
695 	struct task_struct *task;
696 	bool has_perms;
697 
698 	task = get_proc_task(inode);
699 	if (!task)
700 		return -ESRCH;
701 	has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
702 	put_task_struct(task);
703 
704 	if (!has_perms) {
705 		if (pid->hide_pid == HIDEPID_INVISIBLE) {
706 			/*
707 			 * Let's make getdents(), stat(), and open()
708 			 * consistent with each other.  If a process
709 			 * may not stat() a file, it shouldn't be seen
710 			 * in procfs at all.
711 			 */
712 			return -ENOENT;
713 		}
714 
715 		return -EPERM;
716 	}
717 	return generic_permission(inode, mask);
718 }
719 
720 
721 
722 static const struct inode_operations proc_def_inode_operations = {
723 	.setattr	= proc_setattr,
724 };
725 
726 static int proc_single_show(struct seq_file *m, void *v)
727 {
728 	struct inode *inode = m->private;
729 	struct pid_namespace *ns = proc_pid_ns(inode);
730 	struct pid *pid = proc_pid(inode);
731 	struct task_struct *task;
732 	int ret;
733 
734 	task = get_pid_task(pid, PIDTYPE_PID);
735 	if (!task)
736 		return -ESRCH;
737 
738 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
739 
740 	put_task_struct(task);
741 	return ret;
742 }
743 
744 static int proc_single_open(struct inode *inode, struct file *filp)
745 {
746 	return single_open(filp, proc_single_show, inode);
747 }
748 
749 static const struct file_operations proc_single_file_operations = {
750 	.open		= proc_single_open,
751 	.read		= seq_read,
752 	.llseek		= seq_lseek,
753 	.release	= single_release,
754 };
755 
756 
757 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
758 {
759 	struct task_struct *task = get_proc_task(inode);
760 	struct mm_struct *mm = ERR_PTR(-ESRCH);
761 
762 	if (task) {
763 		mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
764 		put_task_struct(task);
765 
766 		if (!IS_ERR_OR_NULL(mm)) {
767 			/* ensure this mm_struct can't be freed */
768 			mmgrab(mm);
769 			/* but do not pin its memory */
770 			mmput(mm);
771 		}
772 	}
773 
774 	return mm;
775 }
776 
777 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
778 {
779 	struct mm_struct *mm = proc_mem_open(inode, mode);
780 
781 	if (IS_ERR(mm))
782 		return PTR_ERR(mm);
783 
784 	file->private_data = mm;
785 	return 0;
786 }
787 
788 static int mem_open(struct inode *inode, struct file *file)
789 {
790 	int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
791 
792 	/* OK to pass negative loff_t, we can catch out-of-range */
793 	file->f_mode |= FMODE_UNSIGNED_OFFSET;
794 
795 	return ret;
796 }
797 
798 static ssize_t mem_rw(struct file *file, char __user *buf,
799 			size_t count, loff_t *ppos, int write)
800 {
801 	struct mm_struct *mm = file->private_data;
802 	unsigned long addr = *ppos;
803 	ssize_t copied;
804 	char *page;
805 	unsigned int flags;
806 
807 	if (!mm)
808 		return 0;
809 
810 	page = (char *)__get_free_page(GFP_KERNEL);
811 	if (!page)
812 		return -ENOMEM;
813 
814 	copied = 0;
815 	if (!mmget_not_zero(mm))
816 		goto free;
817 
818 	flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
819 
820 	while (count > 0) {
821 		int this_len = min_t(int, count, PAGE_SIZE);
822 
823 		if (write && copy_from_user(page, buf, this_len)) {
824 			copied = -EFAULT;
825 			break;
826 		}
827 
828 		this_len = access_remote_vm(mm, addr, page, this_len, flags);
829 		if (!this_len) {
830 			if (!copied)
831 				copied = -EIO;
832 			break;
833 		}
834 
835 		if (!write && copy_to_user(buf, page, this_len)) {
836 			copied = -EFAULT;
837 			break;
838 		}
839 
840 		buf += this_len;
841 		addr += this_len;
842 		copied += this_len;
843 		count -= this_len;
844 	}
845 	*ppos = addr;
846 
847 	mmput(mm);
848 free:
849 	free_page((unsigned long) page);
850 	return copied;
851 }
852 
853 static ssize_t mem_read(struct file *file, char __user *buf,
854 			size_t count, loff_t *ppos)
855 {
856 	return mem_rw(file, buf, count, ppos, 0);
857 }
858 
859 static ssize_t mem_write(struct file *file, const char __user *buf,
860 			 size_t count, loff_t *ppos)
861 {
862 	return mem_rw(file, (char __user*)buf, count, ppos, 1);
863 }
864 
865 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
866 {
867 	switch (orig) {
868 	case 0:
869 		file->f_pos = offset;
870 		break;
871 	case 1:
872 		file->f_pos += offset;
873 		break;
874 	default:
875 		return -EINVAL;
876 	}
877 	force_successful_syscall_return();
878 	return file->f_pos;
879 }
880 
881 static int mem_release(struct inode *inode, struct file *file)
882 {
883 	struct mm_struct *mm = file->private_data;
884 	if (mm)
885 		mmdrop(mm);
886 	return 0;
887 }
888 
889 static const struct file_operations proc_mem_operations = {
890 	.llseek		= mem_lseek,
891 	.read		= mem_read,
892 	.write		= mem_write,
893 	.open		= mem_open,
894 	.release	= mem_release,
895 };
896 
897 static int environ_open(struct inode *inode, struct file *file)
898 {
899 	return __mem_open(inode, file, PTRACE_MODE_READ);
900 }
901 
902 static ssize_t environ_read(struct file *file, char __user *buf,
903 			size_t count, loff_t *ppos)
904 {
905 	char *page;
906 	unsigned long src = *ppos;
907 	int ret = 0;
908 	struct mm_struct *mm = file->private_data;
909 	unsigned long env_start, env_end;
910 
911 	/* Ensure the process spawned far enough to have an environment. */
912 	if (!mm || !mm->env_end)
913 		return 0;
914 
915 	page = (char *)__get_free_page(GFP_KERNEL);
916 	if (!page)
917 		return -ENOMEM;
918 
919 	ret = 0;
920 	if (!mmget_not_zero(mm))
921 		goto free;
922 
923 	spin_lock(&mm->arg_lock);
924 	env_start = mm->env_start;
925 	env_end = mm->env_end;
926 	spin_unlock(&mm->arg_lock);
927 
928 	while (count > 0) {
929 		size_t this_len, max_len;
930 		int retval;
931 
932 		if (src >= (env_end - env_start))
933 			break;
934 
935 		this_len = env_end - (env_start + src);
936 
937 		max_len = min_t(size_t, PAGE_SIZE, count);
938 		this_len = min(max_len, this_len);
939 
940 		retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
941 
942 		if (retval <= 0) {
943 			ret = retval;
944 			break;
945 		}
946 
947 		if (copy_to_user(buf, page, retval)) {
948 			ret = -EFAULT;
949 			break;
950 		}
951 
952 		ret += retval;
953 		src += retval;
954 		buf += retval;
955 		count -= retval;
956 	}
957 	*ppos = src;
958 	mmput(mm);
959 
960 free:
961 	free_page((unsigned long) page);
962 	return ret;
963 }
964 
965 static const struct file_operations proc_environ_operations = {
966 	.open		= environ_open,
967 	.read		= environ_read,
968 	.llseek		= generic_file_llseek,
969 	.release	= mem_release,
970 };
971 
972 static int auxv_open(struct inode *inode, struct file *file)
973 {
974 	return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
975 }
976 
977 static ssize_t auxv_read(struct file *file, char __user *buf,
978 			size_t count, loff_t *ppos)
979 {
980 	struct mm_struct *mm = file->private_data;
981 	unsigned int nwords = 0;
982 
983 	if (!mm)
984 		return 0;
985 	do {
986 		nwords += 2;
987 	} while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
988 	return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
989 				       nwords * sizeof(mm->saved_auxv[0]));
990 }
991 
992 static const struct file_operations proc_auxv_operations = {
993 	.open		= auxv_open,
994 	.read		= auxv_read,
995 	.llseek		= generic_file_llseek,
996 	.release	= mem_release,
997 };
998 
999 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1000 			    loff_t *ppos)
1001 {
1002 	struct task_struct *task = get_proc_task(file_inode(file));
1003 	char buffer[PROC_NUMBUF];
1004 	int oom_adj = OOM_ADJUST_MIN;
1005 	size_t len;
1006 
1007 	if (!task)
1008 		return -ESRCH;
1009 	if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1010 		oom_adj = OOM_ADJUST_MAX;
1011 	else
1012 		oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1013 			  OOM_SCORE_ADJ_MAX;
1014 	put_task_struct(task);
1015 	len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1016 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1017 }
1018 
1019 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1020 {
1021 	static DEFINE_MUTEX(oom_adj_mutex);
1022 	struct mm_struct *mm = NULL;
1023 	struct task_struct *task;
1024 	int err = 0;
1025 
1026 	task = get_proc_task(file_inode(file));
1027 	if (!task)
1028 		return -ESRCH;
1029 
1030 	mutex_lock(&oom_adj_mutex);
1031 	if (legacy) {
1032 		if (oom_adj < task->signal->oom_score_adj &&
1033 				!capable(CAP_SYS_RESOURCE)) {
1034 			err = -EACCES;
1035 			goto err_unlock;
1036 		}
1037 		/*
1038 		 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1039 		 * /proc/pid/oom_score_adj instead.
1040 		 */
1041 		pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1042 			  current->comm, task_pid_nr(current), task_pid_nr(task),
1043 			  task_pid_nr(task));
1044 	} else {
1045 		if ((short)oom_adj < task->signal->oom_score_adj_min &&
1046 				!capable(CAP_SYS_RESOURCE)) {
1047 			err = -EACCES;
1048 			goto err_unlock;
1049 		}
1050 	}
1051 
1052 	/*
1053 	 * Make sure we will check other processes sharing the mm if this is
1054 	 * not vfrok which wants its own oom_score_adj.
1055 	 * pin the mm so it doesn't go away and get reused after task_unlock
1056 	 */
1057 	if (!task->vfork_done) {
1058 		struct task_struct *p = find_lock_task_mm(task);
1059 
1060 		if (p) {
1061 			if (atomic_read(&p->mm->mm_users) > 1) {
1062 				mm = p->mm;
1063 				mmgrab(mm);
1064 			}
1065 			task_unlock(p);
1066 		}
1067 	}
1068 
1069 	task->signal->oom_score_adj = oom_adj;
1070 	if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1071 		task->signal->oom_score_adj_min = (short)oom_adj;
1072 	trace_oom_score_adj_update(task);
1073 
1074 	if (mm) {
1075 		struct task_struct *p;
1076 
1077 		rcu_read_lock();
1078 		for_each_process(p) {
1079 			if (same_thread_group(task, p))
1080 				continue;
1081 
1082 			/* do not touch kernel threads or the global init */
1083 			if (p->flags & PF_KTHREAD || is_global_init(p))
1084 				continue;
1085 
1086 			task_lock(p);
1087 			if (!p->vfork_done && process_shares_mm(p, mm)) {
1088 				p->signal->oom_score_adj = oom_adj;
1089 				if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1090 					p->signal->oom_score_adj_min = (short)oom_adj;
1091 			}
1092 			task_unlock(p);
1093 		}
1094 		rcu_read_unlock();
1095 		mmdrop(mm);
1096 	}
1097 err_unlock:
1098 	mutex_unlock(&oom_adj_mutex);
1099 	put_task_struct(task);
1100 	return err;
1101 }
1102 
1103 /*
1104  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1105  * kernels.  The effective policy is defined by oom_score_adj, which has a
1106  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1107  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1108  * Processes that become oom disabled via oom_adj will still be oom disabled
1109  * with this implementation.
1110  *
1111  * oom_adj cannot be removed since existing userspace binaries use it.
1112  */
1113 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1114 			     size_t count, loff_t *ppos)
1115 {
1116 	char buffer[PROC_NUMBUF];
1117 	int oom_adj;
1118 	int err;
1119 
1120 	memset(buffer, 0, sizeof(buffer));
1121 	if (count > sizeof(buffer) - 1)
1122 		count = sizeof(buffer) - 1;
1123 	if (copy_from_user(buffer, buf, count)) {
1124 		err = -EFAULT;
1125 		goto out;
1126 	}
1127 
1128 	err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1129 	if (err)
1130 		goto out;
1131 	if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1132 	     oom_adj != OOM_DISABLE) {
1133 		err = -EINVAL;
1134 		goto out;
1135 	}
1136 
1137 	/*
1138 	 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1139 	 * value is always attainable.
1140 	 */
1141 	if (oom_adj == OOM_ADJUST_MAX)
1142 		oom_adj = OOM_SCORE_ADJ_MAX;
1143 	else
1144 		oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1145 
1146 	err = __set_oom_adj(file, oom_adj, true);
1147 out:
1148 	return err < 0 ? err : count;
1149 }
1150 
1151 static const struct file_operations proc_oom_adj_operations = {
1152 	.read		= oom_adj_read,
1153 	.write		= oom_adj_write,
1154 	.llseek		= generic_file_llseek,
1155 };
1156 
1157 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1158 					size_t count, loff_t *ppos)
1159 {
1160 	struct task_struct *task = get_proc_task(file_inode(file));
1161 	char buffer[PROC_NUMBUF];
1162 	short oom_score_adj = OOM_SCORE_ADJ_MIN;
1163 	size_t len;
1164 
1165 	if (!task)
1166 		return -ESRCH;
1167 	oom_score_adj = task->signal->oom_score_adj;
1168 	put_task_struct(task);
1169 	len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1170 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1171 }
1172 
1173 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1174 					size_t count, loff_t *ppos)
1175 {
1176 	char buffer[PROC_NUMBUF];
1177 	int oom_score_adj;
1178 	int err;
1179 
1180 	memset(buffer, 0, sizeof(buffer));
1181 	if (count > sizeof(buffer) - 1)
1182 		count = sizeof(buffer) - 1;
1183 	if (copy_from_user(buffer, buf, count)) {
1184 		err = -EFAULT;
1185 		goto out;
1186 	}
1187 
1188 	err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1189 	if (err)
1190 		goto out;
1191 	if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1192 			oom_score_adj > OOM_SCORE_ADJ_MAX) {
1193 		err = -EINVAL;
1194 		goto out;
1195 	}
1196 
1197 	err = __set_oom_adj(file, oom_score_adj, false);
1198 out:
1199 	return err < 0 ? err : count;
1200 }
1201 
1202 static const struct file_operations proc_oom_score_adj_operations = {
1203 	.read		= oom_score_adj_read,
1204 	.write		= oom_score_adj_write,
1205 	.llseek		= default_llseek,
1206 };
1207 
1208 #ifdef CONFIG_AUDIT
1209 #define TMPBUFLEN 11
1210 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1211 				  size_t count, loff_t *ppos)
1212 {
1213 	struct inode * inode = file_inode(file);
1214 	struct task_struct *task = get_proc_task(inode);
1215 	ssize_t length;
1216 	char tmpbuf[TMPBUFLEN];
1217 
1218 	if (!task)
1219 		return -ESRCH;
1220 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1221 			   from_kuid(file->f_cred->user_ns,
1222 				     audit_get_loginuid(task)));
1223 	put_task_struct(task);
1224 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1225 }
1226 
1227 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1228 				   size_t count, loff_t *ppos)
1229 {
1230 	struct inode * inode = file_inode(file);
1231 	uid_t loginuid;
1232 	kuid_t kloginuid;
1233 	int rv;
1234 
1235 	rcu_read_lock();
1236 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1237 		rcu_read_unlock();
1238 		return -EPERM;
1239 	}
1240 	rcu_read_unlock();
1241 
1242 	if (*ppos != 0) {
1243 		/* No partial writes. */
1244 		return -EINVAL;
1245 	}
1246 
1247 	rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1248 	if (rv < 0)
1249 		return rv;
1250 
1251 	/* is userspace tring to explicitly UNSET the loginuid? */
1252 	if (loginuid == AUDIT_UID_UNSET) {
1253 		kloginuid = INVALID_UID;
1254 	} else {
1255 		kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1256 		if (!uid_valid(kloginuid))
1257 			return -EINVAL;
1258 	}
1259 
1260 	rv = audit_set_loginuid(kloginuid);
1261 	if (rv < 0)
1262 		return rv;
1263 	return count;
1264 }
1265 
1266 static const struct file_operations proc_loginuid_operations = {
1267 	.read		= proc_loginuid_read,
1268 	.write		= proc_loginuid_write,
1269 	.llseek		= generic_file_llseek,
1270 };
1271 
1272 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1273 				  size_t count, loff_t *ppos)
1274 {
1275 	struct inode * inode = file_inode(file);
1276 	struct task_struct *task = get_proc_task(inode);
1277 	ssize_t length;
1278 	char tmpbuf[TMPBUFLEN];
1279 
1280 	if (!task)
1281 		return -ESRCH;
1282 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1283 				audit_get_sessionid(task));
1284 	put_task_struct(task);
1285 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1286 }
1287 
1288 static const struct file_operations proc_sessionid_operations = {
1289 	.read		= proc_sessionid_read,
1290 	.llseek		= generic_file_llseek,
1291 };
1292 #endif
1293 
1294 #ifdef CONFIG_FAULT_INJECTION
1295 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1296 				      size_t count, loff_t *ppos)
1297 {
1298 	struct task_struct *task = get_proc_task(file_inode(file));
1299 	char buffer[PROC_NUMBUF];
1300 	size_t len;
1301 	int make_it_fail;
1302 
1303 	if (!task)
1304 		return -ESRCH;
1305 	make_it_fail = task->make_it_fail;
1306 	put_task_struct(task);
1307 
1308 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1309 
1310 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1311 }
1312 
1313 static ssize_t proc_fault_inject_write(struct file * file,
1314 			const char __user * buf, size_t count, loff_t *ppos)
1315 {
1316 	struct task_struct *task;
1317 	char buffer[PROC_NUMBUF];
1318 	int make_it_fail;
1319 	int rv;
1320 
1321 	if (!capable(CAP_SYS_RESOURCE))
1322 		return -EPERM;
1323 	memset(buffer, 0, sizeof(buffer));
1324 	if (count > sizeof(buffer) - 1)
1325 		count = sizeof(buffer) - 1;
1326 	if (copy_from_user(buffer, buf, count))
1327 		return -EFAULT;
1328 	rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1329 	if (rv < 0)
1330 		return rv;
1331 	if (make_it_fail < 0 || make_it_fail > 1)
1332 		return -EINVAL;
1333 
1334 	task = get_proc_task(file_inode(file));
1335 	if (!task)
1336 		return -ESRCH;
1337 	task->make_it_fail = make_it_fail;
1338 	put_task_struct(task);
1339 
1340 	return count;
1341 }
1342 
1343 static const struct file_operations proc_fault_inject_operations = {
1344 	.read		= proc_fault_inject_read,
1345 	.write		= proc_fault_inject_write,
1346 	.llseek		= generic_file_llseek,
1347 };
1348 
1349 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1350 				   size_t count, loff_t *ppos)
1351 {
1352 	struct task_struct *task;
1353 	int err;
1354 	unsigned int n;
1355 
1356 	err = kstrtouint_from_user(buf, count, 0, &n);
1357 	if (err)
1358 		return err;
1359 
1360 	task = get_proc_task(file_inode(file));
1361 	if (!task)
1362 		return -ESRCH;
1363 	task->fail_nth = n;
1364 	put_task_struct(task);
1365 
1366 	return count;
1367 }
1368 
1369 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1370 				  size_t count, loff_t *ppos)
1371 {
1372 	struct task_struct *task;
1373 	char numbuf[PROC_NUMBUF];
1374 	ssize_t len;
1375 
1376 	task = get_proc_task(file_inode(file));
1377 	if (!task)
1378 		return -ESRCH;
1379 	len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1380 	put_task_struct(task);
1381 	return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1382 }
1383 
1384 static const struct file_operations proc_fail_nth_operations = {
1385 	.read		= proc_fail_nth_read,
1386 	.write		= proc_fail_nth_write,
1387 };
1388 #endif
1389 
1390 
1391 #ifdef CONFIG_SCHED_DEBUG
1392 /*
1393  * Print out various scheduling related per-task fields:
1394  */
1395 static int sched_show(struct seq_file *m, void *v)
1396 {
1397 	struct inode *inode = m->private;
1398 	struct pid_namespace *ns = proc_pid_ns(inode);
1399 	struct task_struct *p;
1400 
1401 	p = get_proc_task(inode);
1402 	if (!p)
1403 		return -ESRCH;
1404 	proc_sched_show_task(p, ns, m);
1405 
1406 	put_task_struct(p);
1407 
1408 	return 0;
1409 }
1410 
1411 static ssize_t
1412 sched_write(struct file *file, const char __user *buf,
1413 	    size_t count, loff_t *offset)
1414 {
1415 	struct inode *inode = file_inode(file);
1416 	struct task_struct *p;
1417 
1418 	p = get_proc_task(inode);
1419 	if (!p)
1420 		return -ESRCH;
1421 	proc_sched_set_task(p);
1422 
1423 	put_task_struct(p);
1424 
1425 	return count;
1426 }
1427 
1428 static int sched_open(struct inode *inode, struct file *filp)
1429 {
1430 	return single_open(filp, sched_show, inode);
1431 }
1432 
1433 static const struct file_operations proc_pid_sched_operations = {
1434 	.open		= sched_open,
1435 	.read		= seq_read,
1436 	.write		= sched_write,
1437 	.llseek		= seq_lseek,
1438 	.release	= single_release,
1439 };
1440 
1441 #endif
1442 
1443 #ifdef CONFIG_SCHED_AUTOGROUP
1444 /*
1445  * Print out autogroup related information:
1446  */
1447 static int sched_autogroup_show(struct seq_file *m, void *v)
1448 {
1449 	struct inode *inode = m->private;
1450 	struct task_struct *p;
1451 
1452 	p = get_proc_task(inode);
1453 	if (!p)
1454 		return -ESRCH;
1455 	proc_sched_autogroup_show_task(p, m);
1456 
1457 	put_task_struct(p);
1458 
1459 	return 0;
1460 }
1461 
1462 static ssize_t
1463 sched_autogroup_write(struct file *file, const char __user *buf,
1464 	    size_t count, loff_t *offset)
1465 {
1466 	struct inode *inode = file_inode(file);
1467 	struct task_struct *p;
1468 	char buffer[PROC_NUMBUF];
1469 	int nice;
1470 	int err;
1471 
1472 	memset(buffer, 0, sizeof(buffer));
1473 	if (count > sizeof(buffer) - 1)
1474 		count = sizeof(buffer) - 1;
1475 	if (copy_from_user(buffer, buf, count))
1476 		return -EFAULT;
1477 
1478 	err = kstrtoint(strstrip(buffer), 0, &nice);
1479 	if (err < 0)
1480 		return err;
1481 
1482 	p = get_proc_task(inode);
1483 	if (!p)
1484 		return -ESRCH;
1485 
1486 	err = proc_sched_autogroup_set_nice(p, nice);
1487 	if (err)
1488 		count = err;
1489 
1490 	put_task_struct(p);
1491 
1492 	return count;
1493 }
1494 
1495 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1496 {
1497 	int ret;
1498 
1499 	ret = single_open(filp, sched_autogroup_show, NULL);
1500 	if (!ret) {
1501 		struct seq_file *m = filp->private_data;
1502 
1503 		m->private = inode;
1504 	}
1505 	return ret;
1506 }
1507 
1508 static const struct file_operations proc_pid_sched_autogroup_operations = {
1509 	.open		= sched_autogroup_open,
1510 	.read		= seq_read,
1511 	.write		= sched_autogroup_write,
1512 	.llseek		= seq_lseek,
1513 	.release	= single_release,
1514 };
1515 
1516 #endif /* CONFIG_SCHED_AUTOGROUP */
1517 
1518 static ssize_t comm_write(struct file *file, const char __user *buf,
1519 				size_t count, loff_t *offset)
1520 {
1521 	struct inode *inode = file_inode(file);
1522 	struct task_struct *p;
1523 	char buffer[TASK_COMM_LEN];
1524 	const size_t maxlen = sizeof(buffer) - 1;
1525 
1526 	memset(buffer, 0, sizeof(buffer));
1527 	if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1528 		return -EFAULT;
1529 
1530 	p = get_proc_task(inode);
1531 	if (!p)
1532 		return -ESRCH;
1533 
1534 	if (same_thread_group(current, p))
1535 		set_task_comm(p, buffer);
1536 	else
1537 		count = -EINVAL;
1538 
1539 	put_task_struct(p);
1540 
1541 	return count;
1542 }
1543 
1544 static int comm_show(struct seq_file *m, void *v)
1545 {
1546 	struct inode *inode = m->private;
1547 	struct task_struct *p;
1548 
1549 	p = get_proc_task(inode);
1550 	if (!p)
1551 		return -ESRCH;
1552 
1553 	proc_task_name(m, p, false);
1554 	seq_putc(m, '\n');
1555 
1556 	put_task_struct(p);
1557 
1558 	return 0;
1559 }
1560 
1561 static int comm_open(struct inode *inode, struct file *filp)
1562 {
1563 	return single_open(filp, comm_show, inode);
1564 }
1565 
1566 static const struct file_operations proc_pid_set_comm_operations = {
1567 	.open		= comm_open,
1568 	.read		= seq_read,
1569 	.write		= comm_write,
1570 	.llseek		= seq_lseek,
1571 	.release	= single_release,
1572 };
1573 
1574 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1575 {
1576 	struct task_struct *task;
1577 	struct file *exe_file;
1578 
1579 	task = get_proc_task(d_inode(dentry));
1580 	if (!task)
1581 		return -ENOENT;
1582 	exe_file = get_task_exe_file(task);
1583 	put_task_struct(task);
1584 	if (exe_file) {
1585 		*exe_path = exe_file->f_path;
1586 		path_get(&exe_file->f_path);
1587 		fput(exe_file);
1588 		return 0;
1589 	} else
1590 		return -ENOENT;
1591 }
1592 
1593 static const char *proc_pid_get_link(struct dentry *dentry,
1594 				     struct inode *inode,
1595 				     struct delayed_call *done)
1596 {
1597 	struct path path;
1598 	int error = -EACCES;
1599 
1600 	if (!dentry)
1601 		return ERR_PTR(-ECHILD);
1602 
1603 	/* Are we allowed to snoop on the tasks file descriptors? */
1604 	if (!proc_fd_access_allowed(inode))
1605 		goto out;
1606 
1607 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1608 	if (error)
1609 		goto out;
1610 
1611 	nd_jump_link(&path);
1612 	return NULL;
1613 out:
1614 	return ERR_PTR(error);
1615 }
1616 
1617 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1618 {
1619 	char *tmp = (char *)__get_free_page(GFP_KERNEL);
1620 	char *pathname;
1621 	int len;
1622 
1623 	if (!tmp)
1624 		return -ENOMEM;
1625 
1626 	pathname = d_path(path, tmp, PAGE_SIZE);
1627 	len = PTR_ERR(pathname);
1628 	if (IS_ERR(pathname))
1629 		goto out;
1630 	len = tmp + PAGE_SIZE - 1 - pathname;
1631 
1632 	if (len > buflen)
1633 		len = buflen;
1634 	if (copy_to_user(buffer, pathname, len))
1635 		len = -EFAULT;
1636  out:
1637 	free_page((unsigned long)tmp);
1638 	return len;
1639 }
1640 
1641 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1642 {
1643 	int error = -EACCES;
1644 	struct inode *inode = d_inode(dentry);
1645 	struct path path;
1646 
1647 	/* Are we allowed to snoop on the tasks file descriptors? */
1648 	if (!proc_fd_access_allowed(inode))
1649 		goto out;
1650 
1651 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1652 	if (error)
1653 		goto out;
1654 
1655 	error = do_proc_readlink(&path, buffer, buflen);
1656 	path_put(&path);
1657 out:
1658 	return error;
1659 }
1660 
1661 const struct inode_operations proc_pid_link_inode_operations = {
1662 	.readlink	= proc_pid_readlink,
1663 	.get_link	= proc_pid_get_link,
1664 	.setattr	= proc_setattr,
1665 };
1666 
1667 
1668 /* building an inode */
1669 
1670 void task_dump_owner(struct task_struct *task, umode_t mode,
1671 		     kuid_t *ruid, kgid_t *rgid)
1672 {
1673 	/* Depending on the state of dumpable compute who should own a
1674 	 * proc file for a task.
1675 	 */
1676 	const struct cred *cred;
1677 	kuid_t uid;
1678 	kgid_t gid;
1679 
1680 	if (unlikely(task->flags & PF_KTHREAD)) {
1681 		*ruid = GLOBAL_ROOT_UID;
1682 		*rgid = GLOBAL_ROOT_GID;
1683 		return;
1684 	}
1685 
1686 	/* Default to the tasks effective ownership */
1687 	rcu_read_lock();
1688 	cred = __task_cred(task);
1689 	uid = cred->euid;
1690 	gid = cred->egid;
1691 	rcu_read_unlock();
1692 
1693 	/*
1694 	 * Before the /proc/pid/status file was created the only way to read
1695 	 * the effective uid of a /process was to stat /proc/pid.  Reading
1696 	 * /proc/pid/status is slow enough that procps and other packages
1697 	 * kept stating /proc/pid.  To keep the rules in /proc simple I have
1698 	 * made this apply to all per process world readable and executable
1699 	 * directories.
1700 	 */
1701 	if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1702 		struct mm_struct *mm;
1703 		task_lock(task);
1704 		mm = task->mm;
1705 		/* Make non-dumpable tasks owned by some root */
1706 		if (mm) {
1707 			if (get_dumpable(mm) != SUID_DUMP_USER) {
1708 				struct user_namespace *user_ns = mm->user_ns;
1709 
1710 				uid = make_kuid(user_ns, 0);
1711 				if (!uid_valid(uid))
1712 					uid = GLOBAL_ROOT_UID;
1713 
1714 				gid = make_kgid(user_ns, 0);
1715 				if (!gid_valid(gid))
1716 					gid = GLOBAL_ROOT_GID;
1717 			}
1718 		} else {
1719 			uid = GLOBAL_ROOT_UID;
1720 			gid = GLOBAL_ROOT_GID;
1721 		}
1722 		task_unlock(task);
1723 	}
1724 	*ruid = uid;
1725 	*rgid = gid;
1726 }
1727 
1728 struct inode *proc_pid_make_inode(struct super_block * sb,
1729 				  struct task_struct *task, umode_t mode)
1730 {
1731 	struct inode * inode;
1732 	struct proc_inode *ei;
1733 
1734 	/* We need a new inode */
1735 
1736 	inode = new_inode(sb);
1737 	if (!inode)
1738 		goto out;
1739 
1740 	/* Common stuff */
1741 	ei = PROC_I(inode);
1742 	inode->i_mode = mode;
1743 	inode->i_ino = get_next_ino();
1744 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1745 	inode->i_op = &proc_def_inode_operations;
1746 
1747 	/*
1748 	 * grab the reference to task.
1749 	 */
1750 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1751 	if (!ei->pid)
1752 		goto out_unlock;
1753 
1754 	task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1755 	security_task_to_inode(task, inode);
1756 
1757 out:
1758 	return inode;
1759 
1760 out_unlock:
1761 	iput(inode);
1762 	return NULL;
1763 }
1764 
1765 int pid_getattr(const struct path *path, struct kstat *stat,
1766 		u32 request_mask, unsigned int query_flags)
1767 {
1768 	struct inode *inode = d_inode(path->dentry);
1769 	struct pid_namespace *pid = proc_pid_ns(inode);
1770 	struct task_struct *task;
1771 
1772 	generic_fillattr(inode, stat);
1773 
1774 	stat->uid = GLOBAL_ROOT_UID;
1775 	stat->gid = GLOBAL_ROOT_GID;
1776 	rcu_read_lock();
1777 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1778 	if (task) {
1779 		if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1780 			rcu_read_unlock();
1781 			/*
1782 			 * This doesn't prevent learning whether PID exists,
1783 			 * it only makes getattr() consistent with readdir().
1784 			 */
1785 			return -ENOENT;
1786 		}
1787 		task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1788 	}
1789 	rcu_read_unlock();
1790 	return 0;
1791 }
1792 
1793 /* dentry stuff */
1794 
1795 /*
1796  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1797  */
1798 void pid_update_inode(struct task_struct *task, struct inode *inode)
1799 {
1800 	task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1801 
1802 	inode->i_mode &= ~(S_ISUID | S_ISGID);
1803 	security_task_to_inode(task, inode);
1804 }
1805 
1806 /*
1807  * Rewrite the inode's ownerships here because the owning task may have
1808  * performed a setuid(), etc.
1809  *
1810  */
1811 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1812 {
1813 	struct inode *inode;
1814 	struct task_struct *task;
1815 
1816 	if (flags & LOOKUP_RCU)
1817 		return -ECHILD;
1818 
1819 	inode = d_inode(dentry);
1820 	task = get_proc_task(inode);
1821 
1822 	if (task) {
1823 		pid_update_inode(task, inode);
1824 		put_task_struct(task);
1825 		return 1;
1826 	}
1827 	return 0;
1828 }
1829 
1830 static inline bool proc_inode_is_dead(struct inode *inode)
1831 {
1832 	return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1833 }
1834 
1835 int pid_delete_dentry(const struct dentry *dentry)
1836 {
1837 	/* Is the task we represent dead?
1838 	 * If so, then don't put the dentry on the lru list,
1839 	 * kill it immediately.
1840 	 */
1841 	return proc_inode_is_dead(d_inode(dentry));
1842 }
1843 
1844 const struct dentry_operations pid_dentry_operations =
1845 {
1846 	.d_revalidate	= pid_revalidate,
1847 	.d_delete	= pid_delete_dentry,
1848 };
1849 
1850 /* Lookups */
1851 
1852 /*
1853  * Fill a directory entry.
1854  *
1855  * If possible create the dcache entry and derive our inode number and
1856  * file type from dcache entry.
1857  *
1858  * Since all of the proc inode numbers are dynamically generated, the inode
1859  * numbers do not exist until the inode is cache.  This means creating the
1860  * the dcache entry in readdir is necessary to keep the inode numbers
1861  * reported by readdir in sync with the inode numbers reported
1862  * by stat.
1863  */
1864 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1865 	const char *name, unsigned int len,
1866 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1867 {
1868 	struct dentry *child, *dir = file->f_path.dentry;
1869 	struct qstr qname = QSTR_INIT(name, len);
1870 	struct inode *inode;
1871 	unsigned type = DT_UNKNOWN;
1872 	ino_t ino = 1;
1873 
1874 	child = d_hash_and_lookup(dir, &qname);
1875 	if (!child) {
1876 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1877 		child = d_alloc_parallel(dir, &qname, &wq);
1878 		if (IS_ERR(child))
1879 			goto end_instantiate;
1880 		if (d_in_lookup(child)) {
1881 			struct dentry *res;
1882 			res = instantiate(child, task, ptr);
1883 			d_lookup_done(child);
1884 			if (unlikely(res)) {
1885 				dput(child);
1886 				child = res;
1887 				if (IS_ERR(child))
1888 					goto end_instantiate;
1889 			}
1890 		}
1891 	}
1892 	inode = d_inode(child);
1893 	ino = inode->i_ino;
1894 	type = inode->i_mode >> 12;
1895 	dput(child);
1896 end_instantiate:
1897 	return dir_emit(ctx, name, len, ino, type);
1898 }
1899 
1900 /*
1901  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1902  * which represent vma start and end addresses.
1903  */
1904 static int dname_to_vma_addr(struct dentry *dentry,
1905 			     unsigned long *start, unsigned long *end)
1906 {
1907 	const char *str = dentry->d_name.name;
1908 	unsigned long long sval, eval;
1909 	unsigned int len;
1910 
1911 	if (str[0] == '0' && str[1] != '-')
1912 		return -EINVAL;
1913 	len = _parse_integer(str, 16, &sval);
1914 	if (len & KSTRTOX_OVERFLOW)
1915 		return -EINVAL;
1916 	if (sval != (unsigned long)sval)
1917 		return -EINVAL;
1918 	str += len;
1919 
1920 	if (*str != '-')
1921 		return -EINVAL;
1922 	str++;
1923 
1924 	if (str[0] == '0' && str[1])
1925 		return -EINVAL;
1926 	len = _parse_integer(str, 16, &eval);
1927 	if (len & KSTRTOX_OVERFLOW)
1928 		return -EINVAL;
1929 	if (eval != (unsigned long)eval)
1930 		return -EINVAL;
1931 	str += len;
1932 
1933 	if (*str != '\0')
1934 		return -EINVAL;
1935 
1936 	*start = sval;
1937 	*end = eval;
1938 
1939 	return 0;
1940 }
1941 
1942 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1943 {
1944 	unsigned long vm_start, vm_end;
1945 	bool exact_vma_exists = false;
1946 	struct mm_struct *mm = NULL;
1947 	struct task_struct *task;
1948 	struct inode *inode;
1949 	int status = 0;
1950 
1951 	if (flags & LOOKUP_RCU)
1952 		return -ECHILD;
1953 
1954 	inode = d_inode(dentry);
1955 	task = get_proc_task(inode);
1956 	if (!task)
1957 		goto out_notask;
1958 
1959 	mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1960 	if (IS_ERR_OR_NULL(mm))
1961 		goto out;
1962 
1963 	if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1964 		status = down_read_killable(&mm->mmap_sem);
1965 		if (!status) {
1966 			exact_vma_exists = !!find_exact_vma(mm, vm_start,
1967 							    vm_end);
1968 			up_read(&mm->mmap_sem);
1969 		}
1970 	}
1971 
1972 	mmput(mm);
1973 
1974 	if (exact_vma_exists) {
1975 		task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1976 
1977 		security_task_to_inode(task, inode);
1978 		status = 1;
1979 	}
1980 
1981 out:
1982 	put_task_struct(task);
1983 
1984 out_notask:
1985 	return status;
1986 }
1987 
1988 static const struct dentry_operations tid_map_files_dentry_operations = {
1989 	.d_revalidate	= map_files_d_revalidate,
1990 	.d_delete	= pid_delete_dentry,
1991 };
1992 
1993 static int map_files_get_link(struct dentry *dentry, struct path *path)
1994 {
1995 	unsigned long vm_start, vm_end;
1996 	struct vm_area_struct *vma;
1997 	struct task_struct *task;
1998 	struct mm_struct *mm;
1999 	int rc;
2000 
2001 	rc = -ENOENT;
2002 	task = get_proc_task(d_inode(dentry));
2003 	if (!task)
2004 		goto out;
2005 
2006 	mm = get_task_mm(task);
2007 	put_task_struct(task);
2008 	if (!mm)
2009 		goto out;
2010 
2011 	rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2012 	if (rc)
2013 		goto out_mmput;
2014 
2015 	rc = down_read_killable(&mm->mmap_sem);
2016 	if (rc)
2017 		goto out_mmput;
2018 
2019 	rc = -ENOENT;
2020 	vma = find_exact_vma(mm, vm_start, vm_end);
2021 	if (vma && vma->vm_file) {
2022 		*path = vma->vm_file->f_path;
2023 		path_get(path);
2024 		rc = 0;
2025 	}
2026 	up_read(&mm->mmap_sem);
2027 
2028 out_mmput:
2029 	mmput(mm);
2030 out:
2031 	return rc;
2032 }
2033 
2034 struct map_files_info {
2035 	unsigned long	start;
2036 	unsigned long	end;
2037 	fmode_t		mode;
2038 };
2039 
2040 /*
2041  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2042  * symlinks may be used to bypass permissions on ancestor directories in the
2043  * path to the file in question.
2044  */
2045 static const char *
2046 proc_map_files_get_link(struct dentry *dentry,
2047 			struct inode *inode,
2048 		        struct delayed_call *done)
2049 {
2050 	if (!capable(CAP_SYS_ADMIN))
2051 		return ERR_PTR(-EPERM);
2052 
2053 	return proc_pid_get_link(dentry, inode, done);
2054 }
2055 
2056 /*
2057  * Identical to proc_pid_link_inode_operations except for get_link()
2058  */
2059 static const struct inode_operations proc_map_files_link_inode_operations = {
2060 	.readlink	= proc_pid_readlink,
2061 	.get_link	= proc_map_files_get_link,
2062 	.setattr	= proc_setattr,
2063 };
2064 
2065 static struct dentry *
2066 proc_map_files_instantiate(struct dentry *dentry,
2067 			   struct task_struct *task, const void *ptr)
2068 {
2069 	fmode_t mode = (fmode_t)(unsigned long)ptr;
2070 	struct proc_inode *ei;
2071 	struct inode *inode;
2072 
2073 	inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2074 				    ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2075 				    ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2076 	if (!inode)
2077 		return ERR_PTR(-ENOENT);
2078 
2079 	ei = PROC_I(inode);
2080 	ei->op.proc_get_link = map_files_get_link;
2081 
2082 	inode->i_op = &proc_map_files_link_inode_operations;
2083 	inode->i_size = 64;
2084 
2085 	d_set_d_op(dentry, &tid_map_files_dentry_operations);
2086 	return d_splice_alias(inode, dentry);
2087 }
2088 
2089 static struct dentry *proc_map_files_lookup(struct inode *dir,
2090 		struct dentry *dentry, unsigned int flags)
2091 {
2092 	unsigned long vm_start, vm_end;
2093 	struct vm_area_struct *vma;
2094 	struct task_struct *task;
2095 	struct dentry *result;
2096 	struct mm_struct *mm;
2097 
2098 	result = ERR_PTR(-ENOENT);
2099 	task = get_proc_task(dir);
2100 	if (!task)
2101 		goto out;
2102 
2103 	result = ERR_PTR(-EACCES);
2104 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2105 		goto out_put_task;
2106 
2107 	result = ERR_PTR(-ENOENT);
2108 	if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2109 		goto out_put_task;
2110 
2111 	mm = get_task_mm(task);
2112 	if (!mm)
2113 		goto out_put_task;
2114 
2115 	result = ERR_PTR(-EINTR);
2116 	if (down_read_killable(&mm->mmap_sem))
2117 		goto out_put_mm;
2118 
2119 	result = ERR_PTR(-ENOENT);
2120 	vma = find_exact_vma(mm, vm_start, vm_end);
2121 	if (!vma)
2122 		goto out_no_vma;
2123 
2124 	if (vma->vm_file)
2125 		result = proc_map_files_instantiate(dentry, task,
2126 				(void *)(unsigned long)vma->vm_file->f_mode);
2127 
2128 out_no_vma:
2129 	up_read(&mm->mmap_sem);
2130 out_put_mm:
2131 	mmput(mm);
2132 out_put_task:
2133 	put_task_struct(task);
2134 out:
2135 	return result;
2136 }
2137 
2138 static const struct inode_operations proc_map_files_inode_operations = {
2139 	.lookup		= proc_map_files_lookup,
2140 	.permission	= proc_fd_permission,
2141 	.setattr	= proc_setattr,
2142 };
2143 
2144 static int
2145 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2146 {
2147 	struct vm_area_struct *vma;
2148 	struct task_struct *task;
2149 	struct mm_struct *mm;
2150 	unsigned long nr_files, pos, i;
2151 	GENRADIX(struct map_files_info) fa;
2152 	struct map_files_info *p;
2153 	int ret;
2154 
2155 	genradix_init(&fa);
2156 
2157 	ret = -ENOENT;
2158 	task = get_proc_task(file_inode(file));
2159 	if (!task)
2160 		goto out;
2161 
2162 	ret = -EACCES;
2163 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2164 		goto out_put_task;
2165 
2166 	ret = 0;
2167 	if (!dir_emit_dots(file, ctx))
2168 		goto out_put_task;
2169 
2170 	mm = get_task_mm(task);
2171 	if (!mm)
2172 		goto out_put_task;
2173 
2174 	ret = down_read_killable(&mm->mmap_sem);
2175 	if (ret) {
2176 		mmput(mm);
2177 		goto out_put_task;
2178 	}
2179 
2180 	nr_files = 0;
2181 
2182 	/*
2183 	 * We need two passes here:
2184 	 *
2185 	 *  1) Collect vmas of mapped files with mmap_sem taken
2186 	 *  2) Release mmap_sem and instantiate entries
2187 	 *
2188 	 * otherwise we get lockdep complained, since filldir()
2189 	 * routine might require mmap_sem taken in might_fault().
2190 	 */
2191 
2192 	for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2193 		if (!vma->vm_file)
2194 			continue;
2195 		if (++pos <= ctx->pos)
2196 			continue;
2197 
2198 		p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2199 		if (!p) {
2200 			ret = -ENOMEM;
2201 			up_read(&mm->mmap_sem);
2202 			mmput(mm);
2203 			goto out_put_task;
2204 		}
2205 
2206 		p->start = vma->vm_start;
2207 		p->end = vma->vm_end;
2208 		p->mode = vma->vm_file->f_mode;
2209 	}
2210 	up_read(&mm->mmap_sem);
2211 	mmput(mm);
2212 
2213 	for (i = 0; i < nr_files; i++) {
2214 		char buf[4 * sizeof(long) + 2];	/* max: %lx-%lx\0 */
2215 		unsigned int len;
2216 
2217 		p = genradix_ptr(&fa, i);
2218 		len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2219 		if (!proc_fill_cache(file, ctx,
2220 				      buf, len,
2221 				      proc_map_files_instantiate,
2222 				      task,
2223 				      (void *)(unsigned long)p->mode))
2224 			break;
2225 		ctx->pos++;
2226 	}
2227 
2228 out_put_task:
2229 	put_task_struct(task);
2230 out:
2231 	genradix_free(&fa);
2232 	return ret;
2233 }
2234 
2235 static const struct file_operations proc_map_files_operations = {
2236 	.read		= generic_read_dir,
2237 	.iterate_shared	= proc_map_files_readdir,
2238 	.llseek		= generic_file_llseek,
2239 };
2240 
2241 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2242 struct timers_private {
2243 	struct pid *pid;
2244 	struct task_struct *task;
2245 	struct sighand_struct *sighand;
2246 	struct pid_namespace *ns;
2247 	unsigned long flags;
2248 };
2249 
2250 static void *timers_start(struct seq_file *m, loff_t *pos)
2251 {
2252 	struct timers_private *tp = m->private;
2253 
2254 	tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2255 	if (!tp->task)
2256 		return ERR_PTR(-ESRCH);
2257 
2258 	tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2259 	if (!tp->sighand)
2260 		return ERR_PTR(-ESRCH);
2261 
2262 	return seq_list_start(&tp->task->signal->posix_timers, *pos);
2263 }
2264 
2265 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2266 {
2267 	struct timers_private *tp = m->private;
2268 	return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2269 }
2270 
2271 static void timers_stop(struct seq_file *m, void *v)
2272 {
2273 	struct timers_private *tp = m->private;
2274 
2275 	if (tp->sighand) {
2276 		unlock_task_sighand(tp->task, &tp->flags);
2277 		tp->sighand = NULL;
2278 	}
2279 
2280 	if (tp->task) {
2281 		put_task_struct(tp->task);
2282 		tp->task = NULL;
2283 	}
2284 }
2285 
2286 static int show_timer(struct seq_file *m, void *v)
2287 {
2288 	struct k_itimer *timer;
2289 	struct timers_private *tp = m->private;
2290 	int notify;
2291 	static const char * const nstr[] = {
2292 		[SIGEV_SIGNAL] = "signal",
2293 		[SIGEV_NONE] = "none",
2294 		[SIGEV_THREAD] = "thread",
2295 	};
2296 
2297 	timer = list_entry((struct list_head *)v, struct k_itimer, list);
2298 	notify = timer->it_sigev_notify;
2299 
2300 	seq_printf(m, "ID: %d\n", timer->it_id);
2301 	seq_printf(m, "signal: %d/%px\n",
2302 		   timer->sigq->info.si_signo,
2303 		   timer->sigq->info.si_value.sival_ptr);
2304 	seq_printf(m, "notify: %s/%s.%d\n",
2305 		   nstr[notify & ~SIGEV_THREAD_ID],
2306 		   (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2307 		   pid_nr_ns(timer->it_pid, tp->ns));
2308 	seq_printf(m, "ClockID: %d\n", timer->it_clock);
2309 
2310 	return 0;
2311 }
2312 
2313 static const struct seq_operations proc_timers_seq_ops = {
2314 	.start	= timers_start,
2315 	.next	= timers_next,
2316 	.stop	= timers_stop,
2317 	.show	= show_timer,
2318 };
2319 
2320 static int proc_timers_open(struct inode *inode, struct file *file)
2321 {
2322 	struct timers_private *tp;
2323 
2324 	tp = __seq_open_private(file, &proc_timers_seq_ops,
2325 			sizeof(struct timers_private));
2326 	if (!tp)
2327 		return -ENOMEM;
2328 
2329 	tp->pid = proc_pid(inode);
2330 	tp->ns = proc_pid_ns(inode);
2331 	return 0;
2332 }
2333 
2334 static const struct file_operations proc_timers_operations = {
2335 	.open		= proc_timers_open,
2336 	.read		= seq_read,
2337 	.llseek		= seq_lseek,
2338 	.release	= seq_release_private,
2339 };
2340 #endif
2341 
2342 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2343 					size_t count, loff_t *offset)
2344 {
2345 	struct inode *inode = file_inode(file);
2346 	struct task_struct *p;
2347 	u64 slack_ns;
2348 	int err;
2349 
2350 	err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2351 	if (err < 0)
2352 		return err;
2353 
2354 	p = get_proc_task(inode);
2355 	if (!p)
2356 		return -ESRCH;
2357 
2358 	if (p != current) {
2359 		rcu_read_lock();
2360 		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2361 			rcu_read_unlock();
2362 			count = -EPERM;
2363 			goto out;
2364 		}
2365 		rcu_read_unlock();
2366 
2367 		err = security_task_setscheduler(p);
2368 		if (err) {
2369 			count = err;
2370 			goto out;
2371 		}
2372 	}
2373 
2374 	task_lock(p);
2375 	if (slack_ns == 0)
2376 		p->timer_slack_ns = p->default_timer_slack_ns;
2377 	else
2378 		p->timer_slack_ns = slack_ns;
2379 	task_unlock(p);
2380 
2381 out:
2382 	put_task_struct(p);
2383 
2384 	return count;
2385 }
2386 
2387 static int timerslack_ns_show(struct seq_file *m, void *v)
2388 {
2389 	struct inode *inode = m->private;
2390 	struct task_struct *p;
2391 	int err = 0;
2392 
2393 	p = get_proc_task(inode);
2394 	if (!p)
2395 		return -ESRCH;
2396 
2397 	if (p != current) {
2398 		rcu_read_lock();
2399 		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2400 			rcu_read_unlock();
2401 			err = -EPERM;
2402 			goto out;
2403 		}
2404 		rcu_read_unlock();
2405 
2406 		err = security_task_getscheduler(p);
2407 		if (err)
2408 			goto out;
2409 	}
2410 
2411 	task_lock(p);
2412 	seq_printf(m, "%llu\n", p->timer_slack_ns);
2413 	task_unlock(p);
2414 
2415 out:
2416 	put_task_struct(p);
2417 
2418 	return err;
2419 }
2420 
2421 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2422 {
2423 	return single_open(filp, timerslack_ns_show, inode);
2424 }
2425 
2426 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2427 	.open		= timerslack_ns_open,
2428 	.read		= seq_read,
2429 	.write		= timerslack_ns_write,
2430 	.llseek		= seq_lseek,
2431 	.release	= single_release,
2432 };
2433 
2434 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2435 	struct task_struct *task, const void *ptr)
2436 {
2437 	const struct pid_entry *p = ptr;
2438 	struct inode *inode;
2439 	struct proc_inode *ei;
2440 
2441 	inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2442 	if (!inode)
2443 		return ERR_PTR(-ENOENT);
2444 
2445 	ei = PROC_I(inode);
2446 	if (S_ISDIR(inode->i_mode))
2447 		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
2448 	if (p->iop)
2449 		inode->i_op = p->iop;
2450 	if (p->fop)
2451 		inode->i_fop = p->fop;
2452 	ei->op = p->op;
2453 	pid_update_inode(task, inode);
2454 	d_set_d_op(dentry, &pid_dentry_operations);
2455 	return d_splice_alias(inode, dentry);
2456 }
2457 
2458 static struct dentry *proc_pident_lookup(struct inode *dir,
2459 					 struct dentry *dentry,
2460 					 const struct pid_entry *p,
2461 					 const struct pid_entry *end)
2462 {
2463 	struct task_struct *task = get_proc_task(dir);
2464 	struct dentry *res = ERR_PTR(-ENOENT);
2465 
2466 	if (!task)
2467 		goto out_no_task;
2468 
2469 	/*
2470 	 * Yes, it does not scale. And it should not. Don't add
2471 	 * new entries into /proc/<tgid>/ without very good reasons.
2472 	 */
2473 	for (; p < end; p++) {
2474 		if (p->len != dentry->d_name.len)
2475 			continue;
2476 		if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2477 			res = proc_pident_instantiate(dentry, task, p);
2478 			break;
2479 		}
2480 	}
2481 	put_task_struct(task);
2482 out_no_task:
2483 	return res;
2484 }
2485 
2486 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2487 		const struct pid_entry *ents, unsigned int nents)
2488 {
2489 	struct task_struct *task = get_proc_task(file_inode(file));
2490 	const struct pid_entry *p;
2491 
2492 	if (!task)
2493 		return -ENOENT;
2494 
2495 	if (!dir_emit_dots(file, ctx))
2496 		goto out;
2497 
2498 	if (ctx->pos >= nents + 2)
2499 		goto out;
2500 
2501 	for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2502 		if (!proc_fill_cache(file, ctx, p->name, p->len,
2503 				proc_pident_instantiate, task, p))
2504 			break;
2505 		ctx->pos++;
2506 	}
2507 out:
2508 	put_task_struct(task);
2509 	return 0;
2510 }
2511 
2512 #ifdef CONFIG_SECURITY
2513 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2514 				  size_t count, loff_t *ppos)
2515 {
2516 	struct inode * inode = file_inode(file);
2517 	char *p = NULL;
2518 	ssize_t length;
2519 	struct task_struct *task = get_proc_task(inode);
2520 
2521 	if (!task)
2522 		return -ESRCH;
2523 
2524 	length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2525 				      (char*)file->f_path.dentry->d_name.name,
2526 				      &p);
2527 	put_task_struct(task);
2528 	if (length > 0)
2529 		length = simple_read_from_buffer(buf, count, ppos, p, length);
2530 	kfree(p);
2531 	return length;
2532 }
2533 
2534 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2535 				   size_t count, loff_t *ppos)
2536 {
2537 	struct inode * inode = file_inode(file);
2538 	struct task_struct *task;
2539 	void *page;
2540 	int rv;
2541 
2542 	rcu_read_lock();
2543 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
2544 	if (!task) {
2545 		rcu_read_unlock();
2546 		return -ESRCH;
2547 	}
2548 	/* A task may only write its own attributes. */
2549 	if (current != task) {
2550 		rcu_read_unlock();
2551 		return -EACCES;
2552 	}
2553 	/* Prevent changes to overridden credentials. */
2554 	if (current_cred() != current_real_cred()) {
2555 		rcu_read_unlock();
2556 		return -EBUSY;
2557 	}
2558 	rcu_read_unlock();
2559 
2560 	if (count > PAGE_SIZE)
2561 		count = PAGE_SIZE;
2562 
2563 	/* No partial writes. */
2564 	if (*ppos != 0)
2565 		return -EINVAL;
2566 
2567 	page = memdup_user(buf, count);
2568 	if (IS_ERR(page)) {
2569 		rv = PTR_ERR(page);
2570 		goto out;
2571 	}
2572 
2573 	/* Guard against adverse ptrace interaction */
2574 	rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2575 	if (rv < 0)
2576 		goto out_free;
2577 
2578 	rv = security_setprocattr(PROC_I(inode)->op.lsm,
2579 				  file->f_path.dentry->d_name.name, page,
2580 				  count);
2581 	mutex_unlock(&current->signal->cred_guard_mutex);
2582 out_free:
2583 	kfree(page);
2584 out:
2585 	return rv;
2586 }
2587 
2588 static const struct file_operations proc_pid_attr_operations = {
2589 	.read		= proc_pid_attr_read,
2590 	.write		= proc_pid_attr_write,
2591 	.llseek		= generic_file_llseek,
2592 };
2593 
2594 #define LSM_DIR_OPS(LSM) \
2595 static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2596 			     struct dir_context *ctx) \
2597 { \
2598 	return proc_pident_readdir(filp, ctx, \
2599 				   LSM##_attr_dir_stuff, \
2600 				   ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2601 } \
2602 \
2603 static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2604 	.read		= generic_read_dir, \
2605 	.iterate	= proc_##LSM##_attr_dir_iterate, \
2606 	.llseek		= default_llseek, \
2607 }; \
2608 \
2609 static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2610 				struct dentry *dentry, unsigned int flags) \
2611 { \
2612 	return proc_pident_lookup(dir, dentry, \
2613 				  LSM##_attr_dir_stuff, \
2614 				  LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2615 } \
2616 \
2617 static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2618 	.lookup		= proc_##LSM##_attr_dir_lookup, \
2619 	.getattr	= pid_getattr, \
2620 	.setattr	= proc_setattr, \
2621 }
2622 
2623 #ifdef CONFIG_SECURITY_SMACK
2624 static const struct pid_entry smack_attr_dir_stuff[] = {
2625 	ATTR("smack", "current",	0666),
2626 };
2627 LSM_DIR_OPS(smack);
2628 #endif
2629 
2630 static const struct pid_entry attr_dir_stuff[] = {
2631 	ATTR(NULL, "current",		0666),
2632 	ATTR(NULL, "prev",		0444),
2633 	ATTR(NULL, "exec",		0666),
2634 	ATTR(NULL, "fscreate",		0666),
2635 	ATTR(NULL, "keycreate",		0666),
2636 	ATTR(NULL, "sockcreate",	0666),
2637 #ifdef CONFIG_SECURITY_SMACK
2638 	DIR("smack",			0555,
2639 	    proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2640 #endif
2641 };
2642 
2643 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2644 {
2645 	return proc_pident_readdir(file, ctx,
2646 				   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2647 }
2648 
2649 static const struct file_operations proc_attr_dir_operations = {
2650 	.read		= generic_read_dir,
2651 	.iterate_shared	= proc_attr_dir_readdir,
2652 	.llseek		= generic_file_llseek,
2653 };
2654 
2655 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2656 				struct dentry *dentry, unsigned int flags)
2657 {
2658 	return proc_pident_lookup(dir, dentry,
2659 				  attr_dir_stuff,
2660 				  attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2661 }
2662 
2663 static const struct inode_operations proc_attr_dir_inode_operations = {
2664 	.lookup		= proc_attr_dir_lookup,
2665 	.getattr	= pid_getattr,
2666 	.setattr	= proc_setattr,
2667 };
2668 
2669 #endif
2670 
2671 #ifdef CONFIG_ELF_CORE
2672 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2673 					 size_t count, loff_t *ppos)
2674 {
2675 	struct task_struct *task = get_proc_task(file_inode(file));
2676 	struct mm_struct *mm;
2677 	char buffer[PROC_NUMBUF];
2678 	size_t len;
2679 	int ret;
2680 
2681 	if (!task)
2682 		return -ESRCH;
2683 
2684 	ret = 0;
2685 	mm = get_task_mm(task);
2686 	if (mm) {
2687 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2688 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2689 				MMF_DUMP_FILTER_SHIFT));
2690 		mmput(mm);
2691 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2692 	}
2693 
2694 	put_task_struct(task);
2695 
2696 	return ret;
2697 }
2698 
2699 static ssize_t proc_coredump_filter_write(struct file *file,
2700 					  const char __user *buf,
2701 					  size_t count,
2702 					  loff_t *ppos)
2703 {
2704 	struct task_struct *task;
2705 	struct mm_struct *mm;
2706 	unsigned int val;
2707 	int ret;
2708 	int i;
2709 	unsigned long mask;
2710 
2711 	ret = kstrtouint_from_user(buf, count, 0, &val);
2712 	if (ret < 0)
2713 		return ret;
2714 
2715 	ret = -ESRCH;
2716 	task = get_proc_task(file_inode(file));
2717 	if (!task)
2718 		goto out_no_task;
2719 
2720 	mm = get_task_mm(task);
2721 	if (!mm)
2722 		goto out_no_mm;
2723 	ret = 0;
2724 
2725 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2726 		if (val & mask)
2727 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2728 		else
2729 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2730 	}
2731 
2732 	mmput(mm);
2733  out_no_mm:
2734 	put_task_struct(task);
2735  out_no_task:
2736 	if (ret < 0)
2737 		return ret;
2738 	return count;
2739 }
2740 
2741 static const struct file_operations proc_coredump_filter_operations = {
2742 	.read		= proc_coredump_filter_read,
2743 	.write		= proc_coredump_filter_write,
2744 	.llseek		= generic_file_llseek,
2745 };
2746 #endif
2747 
2748 #ifdef CONFIG_TASK_IO_ACCOUNTING
2749 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2750 {
2751 	struct task_io_accounting acct = task->ioac;
2752 	unsigned long flags;
2753 	int result;
2754 
2755 	result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2756 	if (result)
2757 		return result;
2758 
2759 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2760 		result = -EACCES;
2761 		goto out_unlock;
2762 	}
2763 
2764 	if (whole && lock_task_sighand(task, &flags)) {
2765 		struct task_struct *t = task;
2766 
2767 		task_io_accounting_add(&acct, &task->signal->ioac);
2768 		while_each_thread(task, t)
2769 			task_io_accounting_add(&acct, &t->ioac);
2770 
2771 		unlock_task_sighand(task, &flags);
2772 	}
2773 	seq_printf(m,
2774 		   "rchar: %llu\n"
2775 		   "wchar: %llu\n"
2776 		   "syscr: %llu\n"
2777 		   "syscw: %llu\n"
2778 		   "read_bytes: %llu\n"
2779 		   "write_bytes: %llu\n"
2780 		   "cancelled_write_bytes: %llu\n",
2781 		   (unsigned long long)acct.rchar,
2782 		   (unsigned long long)acct.wchar,
2783 		   (unsigned long long)acct.syscr,
2784 		   (unsigned long long)acct.syscw,
2785 		   (unsigned long long)acct.read_bytes,
2786 		   (unsigned long long)acct.write_bytes,
2787 		   (unsigned long long)acct.cancelled_write_bytes);
2788 	result = 0;
2789 
2790 out_unlock:
2791 	mutex_unlock(&task->signal->cred_guard_mutex);
2792 	return result;
2793 }
2794 
2795 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2796 				  struct pid *pid, struct task_struct *task)
2797 {
2798 	return do_io_accounting(task, m, 0);
2799 }
2800 
2801 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2802 				   struct pid *pid, struct task_struct *task)
2803 {
2804 	return do_io_accounting(task, m, 1);
2805 }
2806 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2807 
2808 #ifdef CONFIG_USER_NS
2809 static int proc_id_map_open(struct inode *inode, struct file *file,
2810 	const struct seq_operations *seq_ops)
2811 {
2812 	struct user_namespace *ns = NULL;
2813 	struct task_struct *task;
2814 	struct seq_file *seq;
2815 	int ret = -EINVAL;
2816 
2817 	task = get_proc_task(inode);
2818 	if (task) {
2819 		rcu_read_lock();
2820 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2821 		rcu_read_unlock();
2822 		put_task_struct(task);
2823 	}
2824 	if (!ns)
2825 		goto err;
2826 
2827 	ret = seq_open(file, seq_ops);
2828 	if (ret)
2829 		goto err_put_ns;
2830 
2831 	seq = file->private_data;
2832 	seq->private = ns;
2833 
2834 	return 0;
2835 err_put_ns:
2836 	put_user_ns(ns);
2837 err:
2838 	return ret;
2839 }
2840 
2841 static int proc_id_map_release(struct inode *inode, struct file *file)
2842 {
2843 	struct seq_file *seq = file->private_data;
2844 	struct user_namespace *ns = seq->private;
2845 	put_user_ns(ns);
2846 	return seq_release(inode, file);
2847 }
2848 
2849 static int proc_uid_map_open(struct inode *inode, struct file *file)
2850 {
2851 	return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2852 }
2853 
2854 static int proc_gid_map_open(struct inode *inode, struct file *file)
2855 {
2856 	return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2857 }
2858 
2859 static int proc_projid_map_open(struct inode *inode, struct file *file)
2860 {
2861 	return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2862 }
2863 
2864 static const struct file_operations proc_uid_map_operations = {
2865 	.open		= proc_uid_map_open,
2866 	.write		= proc_uid_map_write,
2867 	.read		= seq_read,
2868 	.llseek		= seq_lseek,
2869 	.release	= proc_id_map_release,
2870 };
2871 
2872 static const struct file_operations proc_gid_map_operations = {
2873 	.open		= proc_gid_map_open,
2874 	.write		= proc_gid_map_write,
2875 	.read		= seq_read,
2876 	.llseek		= seq_lseek,
2877 	.release	= proc_id_map_release,
2878 };
2879 
2880 static const struct file_operations proc_projid_map_operations = {
2881 	.open		= proc_projid_map_open,
2882 	.write		= proc_projid_map_write,
2883 	.read		= seq_read,
2884 	.llseek		= seq_lseek,
2885 	.release	= proc_id_map_release,
2886 };
2887 
2888 static int proc_setgroups_open(struct inode *inode, struct file *file)
2889 {
2890 	struct user_namespace *ns = NULL;
2891 	struct task_struct *task;
2892 	int ret;
2893 
2894 	ret = -ESRCH;
2895 	task = get_proc_task(inode);
2896 	if (task) {
2897 		rcu_read_lock();
2898 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2899 		rcu_read_unlock();
2900 		put_task_struct(task);
2901 	}
2902 	if (!ns)
2903 		goto err;
2904 
2905 	if (file->f_mode & FMODE_WRITE) {
2906 		ret = -EACCES;
2907 		if (!ns_capable(ns, CAP_SYS_ADMIN))
2908 			goto err_put_ns;
2909 	}
2910 
2911 	ret = single_open(file, &proc_setgroups_show, ns);
2912 	if (ret)
2913 		goto err_put_ns;
2914 
2915 	return 0;
2916 err_put_ns:
2917 	put_user_ns(ns);
2918 err:
2919 	return ret;
2920 }
2921 
2922 static int proc_setgroups_release(struct inode *inode, struct file *file)
2923 {
2924 	struct seq_file *seq = file->private_data;
2925 	struct user_namespace *ns = seq->private;
2926 	int ret = single_release(inode, file);
2927 	put_user_ns(ns);
2928 	return ret;
2929 }
2930 
2931 static const struct file_operations proc_setgroups_operations = {
2932 	.open		= proc_setgroups_open,
2933 	.write		= proc_setgroups_write,
2934 	.read		= seq_read,
2935 	.llseek		= seq_lseek,
2936 	.release	= proc_setgroups_release,
2937 };
2938 #endif /* CONFIG_USER_NS */
2939 
2940 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2941 				struct pid *pid, struct task_struct *task)
2942 {
2943 	int err = lock_trace(task);
2944 	if (!err) {
2945 		seq_printf(m, "%08x\n", task->personality);
2946 		unlock_trace(task);
2947 	}
2948 	return err;
2949 }
2950 
2951 #ifdef CONFIG_LIVEPATCH
2952 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2953 				struct pid *pid, struct task_struct *task)
2954 {
2955 	seq_printf(m, "%d\n", task->patch_state);
2956 	return 0;
2957 }
2958 #endif /* CONFIG_LIVEPATCH */
2959 
2960 #ifdef CONFIG_STACKLEAK_METRICS
2961 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
2962 				struct pid *pid, struct task_struct *task)
2963 {
2964 	unsigned long prev_depth = THREAD_SIZE -
2965 				(task->prev_lowest_stack & (THREAD_SIZE - 1));
2966 	unsigned long depth = THREAD_SIZE -
2967 				(task->lowest_stack & (THREAD_SIZE - 1));
2968 
2969 	seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
2970 							prev_depth, depth);
2971 	return 0;
2972 }
2973 #endif /* CONFIG_STACKLEAK_METRICS */
2974 
2975 /*
2976  * Thread groups
2977  */
2978 static const struct file_operations proc_task_operations;
2979 static const struct inode_operations proc_task_inode_operations;
2980 
2981 static const struct pid_entry tgid_base_stuff[] = {
2982 	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2983 	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2984 	DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2985 	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2986 	DIR("ns",	  S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2987 #ifdef CONFIG_NET
2988 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2989 #endif
2990 	REG("environ",    S_IRUSR, proc_environ_operations),
2991 	REG("auxv",       S_IRUSR, proc_auxv_operations),
2992 	ONE("status",     S_IRUGO, proc_pid_status),
2993 	ONE("personality", S_IRUSR, proc_pid_personality),
2994 	ONE("limits",	  S_IRUGO, proc_pid_limits),
2995 #ifdef CONFIG_SCHED_DEBUG
2996 	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2997 #endif
2998 #ifdef CONFIG_SCHED_AUTOGROUP
2999 	REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3000 #endif
3001 	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3002 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3003 	ONE("syscall",    S_IRUSR, proc_pid_syscall),
3004 #endif
3005 	REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
3006 	ONE("stat",       S_IRUGO, proc_tgid_stat),
3007 	ONE("statm",      S_IRUGO, proc_pid_statm),
3008 	REG("maps",       S_IRUGO, proc_pid_maps_operations),
3009 #ifdef CONFIG_NUMA
3010 	REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3011 #endif
3012 	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3013 	LNK("cwd",        proc_cwd_link),
3014 	LNK("root",       proc_root_link),
3015 	LNK("exe",        proc_exe_link),
3016 	REG("mounts",     S_IRUGO, proc_mounts_operations),
3017 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3018 	REG("mountstats", S_IRUSR, proc_mountstats_operations),
3019 #ifdef CONFIG_PROC_PAGE_MONITOR
3020 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3021 	REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3022 	REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3023 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3024 #endif
3025 #ifdef CONFIG_SECURITY
3026 	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3027 #endif
3028 #ifdef CONFIG_KALLSYMS
3029 	ONE("wchan",      S_IRUGO, proc_pid_wchan),
3030 #endif
3031 #ifdef CONFIG_STACKTRACE
3032 	ONE("stack",      S_IRUSR, proc_pid_stack),
3033 #endif
3034 #ifdef CONFIG_SCHED_INFO
3035 	ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
3036 #endif
3037 #ifdef CONFIG_LATENCYTOP
3038 	REG("latency",  S_IRUGO, proc_lstats_operations),
3039 #endif
3040 #ifdef CONFIG_PROC_PID_CPUSET
3041 	ONE("cpuset",     S_IRUGO, proc_cpuset_show),
3042 #endif
3043 #ifdef CONFIG_CGROUPS
3044 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3045 #endif
3046 	ONE("oom_score",  S_IRUGO, proc_oom_score),
3047 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3048 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3049 #ifdef CONFIG_AUDIT
3050 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3051 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3052 #endif
3053 #ifdef CONFIG_FAULT_INJECTION
3054 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3055 	REG("fail-nth", 0644, proc_fail_nth_operations),
3056 #endif
3057 #ifdef CONFIG_ELF_CORE
3058 	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3059 #endif
3060 #ifdef CONFIG_TASK_IO_ACCOUNTING
3061 	ONE("io",	S_IRUSR, proc_tgid_io_accounting),
3062 #endif
3063 #ifdef CONFIG_USER_NS
3064 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3065 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3066 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3067 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3068 #endif
3069 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3070 	REG("timers",	  S_IRUGO, proc_timers_operations),
3071 #endif
3072 	REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3073 #ifdef CONFIG_LIVEPATCH
3074 	ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3075 #endif
3076 #ifdef CONFIG_STACKLEAK_METRICS
3077 	ONE("stack_depth", S_IRUGO, proc_stack_depth),
3078 #endif
3079 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3080 	ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3081 #endif
3082 };
3083 
3084 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3085 {
3086 	return proc_pident_readdir(file, ctx,
3087 				   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3088 }
3089 
3090 static const struct file_operations proc_tgid_base_operations = {
3091 	.read		= generic_read_dir,
3092 	.iterate_shared	= proc_tgid_base_readdir,
3093 	.llseek		= generic_file_llseek,
3094 };
3095 
3096 struct pid *tgid_pidfd_to_pid(const struct file *file)
3097 {
3098 	if (file->f_op != &proc_tgid_base_operations)
3099 		return ERR_PTR(-EBADF);
3100 
3101 	return proc_pid(file_inode(file));
3102 }
3103 
3104 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3105 {
3106 	return proc_pident_lookup(dir, dentry,
3107 				  tgid_base_stuff,
3108 				  tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3109 }
3110 
3111 static const struct inode_operations proc_tgid_base_inode_operations = {
3112 	.lookup		= proc_tgid_base_lookup,
3113 	.getattr	= pid_getattr,
3114 	.setattr	= proc_setattr,
3115 	.permission	= proc_pid_permission,
3116 };
3117 
3118 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3119 {
3120 	struct dentry *dentry, *leader, *dir;
3121 	char buf[10 + 1];
3122 	struct qstr name;
3123 
3124 	name.name = buf;
3125 	name.len = snprintf(buf, sizeof(buf), "%u", pid);
3126 	/* no ->d_hash() rejects on procfs */
3127 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3128 	if (dentry) {
3129 		d_invalidate(dentry);
3130 		dput(dentry);
3131 	}
3132 
3133 	if (pid == tgid)
3134 		return;
3135 
3136 	name.name = buf;
3137 	name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3138 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
3139 	if (!leader)
3140 		goto out;
3141 
3142 	name.name = "task";
3143 	name.len = strlen(name.name);
3144 	dir = d_hash_and_lookup(leader, &name);
3145 	if (!dir)
3146 		goto out_put_leader;
3147 
3148 	name.name = buf;
3149 	name.len = snprintf(buf, sizeof(buf), "%u", pid);
3150 	dentry = d_hash_and_lookup(dir, &name);
3151 	if (dentry) {
3152 		d_invalidate(dentry);
3153 		dput(dentry);
3154 	}
3155 
3156 	dput(dir);
3157 out_put_leader:
3158 	dput(leader);
3159 out:
3160 	return;
3161 }
3162 
3163 /**
3164  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3165  * @task: task that should be flushed.
3166  *
3167  * When flushing dentries from proc, one needs to flush them from global
3168  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3169  * in. This call is supposed to do all of this job.
3170  *
3171  * Looks in the dcache for
3172  * /proc/@pid
3173  * /proc/@tgid/task/@pid
3174  * if either directory is present flushes it and all of it'ts children
3175  * from the dcache.
3176  *
3177  * It is safe and reasonable to cache /proc entries for a task until
3178  * that task exits.  After that they just clog up the dcache with
3179  * useless entries, possibly causing useful dcache entries to be
3180  * flushed instead.  This routine is proved to flush those useless
3181  * dcache entries at process exit time.
3182  *
3183  * NOTE: This routine is just an optimization so it does not guarantee
3184  *       that no dcache entries will exist at process exit time it
3185  *       just makes it very unlikely that any will persist.
3186  */
3187 
3188 void proc_flush_task(struct task_struct *task)
3189 {
3190 	int i;
3191 	struct pid *pid, *tgid;
3192 	struct upid *upid;
3193 
3194 	pid = task_pid(task);
3195 	tgid = task_tgid(task);
3196 
3197 	for (i = 0; i <= pid->level; i++) {
3198 		upid = &pid->numbers[i];
3199 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3200 					tgid->numbers[i].nr);
3201 	}
3202 }
3203 
3204 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3205 				   struct task_struct *task, const void *ptr)
3206 {
3207 	struct inode *inode;
3208 
3209 	inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3210 	if (!inode)
3211 		return ERR_PTR(-ENOENT);
3212 
3213 	inode->i_op = &proc_tgid_base_inode_operations;
3214 	inode->i_fop = &proc_tgid_base_operations;
3215 	inode->i_flags|=S_IMMUTABLE;
3216 
3217 	set_nlink(inode, nlink_tgid);
3218 	pid_update_inode(task, inode);
3219 
3220 	d_set_d_op(dentry, &pid_dentry_operations);
3221 	return d_splice_alias(inode, dentry);
3222 }
3223 
3224 struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3225 {
3226 	struct task_struct *task;
3227 	unsigned tgid;
3228 	struct pid_namespace *ns;
3229 	struct dentry *result = ERR_PTR(-ENOENT);
3230 
3231 	tgid = name_to_int(&dentry->d_name);
3232 	if (tgid == ~0U)
3233 		goto out;
3234 
3235 	ns = dentry->d_sb->s_fs_info;
3236 	rcu_read_lock();
3237 	task = find_task_by_pid_ns(tgid, ns);
3238 	if (task)
3239 		get_task_struct(task);
3240 	rcu_read_unlock();
3241 	if (!task)
3242 		goto out;
3243 
3244 	result = proc_pid_instantiate(dentry, task, NULL);
3245 	put_task_struct(task);
3246 out:
3247 	return result;
3248 }
3249 
3250 /*
3251  * Find the first task with tgid >= tgid
3252  *
3253  */
3254 struct tgid_iter {
3255 	unsigned int tgid;
3256 	struct task_struct *task;
3257 };
3258 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3259 {
3260 	struct pid *pid;
3261 
3262 	if (iter.task)
3263 		put_task_struct(iter.task);
3264 	rcu_read_lock();
3265 retry:
3266 	iter.task = NULL;
3267 	pid = find_ge_pid(iter.tgid, ns);
3268 	if (pid) {
3269 		iter.tgid = pid_nr_ns(pid, ns);
3270 		iter.task = pid_task(pid, PIDTYPE_PID);
3271 		/* What we to know is if the pid we have find is the
3272 		 * pid of a thread_group_leader.  Testing for task
3273 		 * being a thread_group_leader is the obvious thing
3274 		 * todo but there is a window when it fails, due to
3275 		 * the pid transfer logic in de_thread.
3276 		 *
3277 		 * So we perform the straight forward test of seeing
3278 		 * if the pid we have found is the pid of a thread
3279 		 * group leader, and don't worry if the task we have
3280 		 * found doesn't happen to be a thread group leader.
3281 		 * As we don't care in the case of readdir.
3282 		 */
3283 		if (!iter.task || !has_group_leader_pid(iter.task)) {
3284 			iter.tgid += 1;
3285 			goto retry;
3286 		}
3287 		get_task_struct(iter.task);
3288 	}
3289 	rcu_read_unlock();
3290 	return iter;
3291 }
3292 
3293 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3294 
3295 /* for the /proc/ directory itself, after non-process stuff has been done */
3296 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3297 {
3298 	struct tgid_iter iter;
3299 	struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3300 	loff_t pos = ctx->pos;
3301 
3302 	if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3303 		return 0;
3304 
3305 	if (pos == TGID_OFFSET - 2) {
3306 		struct inode *inode = d_inode(ns->proc_self);
3307 		if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3308 			return 0;
3309 		ctx->pos = pos = pos + 1;
3310 	}
3311 	if (pos == TGID_OFFSET - 1) {
3312 		struct inode *inode = d_inode(ns->proc_thread_self);
3313 		if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3314 			return 0;
3315 		ctx->pos = pos = pos + 1;
3316 	}
3317 	iter.tgid = pos - TGID_OFFSET;
3318 	iter.task = NULL;
3319 	for (iter = next_tgid(ns, iter);
3320 	     iter.task;
3321 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
3322 		char name[10 + 1];
3323 		unsigned int len;
3324 
3325 		cond_resched();
3326 		if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3327 			continue;
3328 
3329 		len = snprintf(name, sizeof(name), "%u", iter.tgid);
3330 		ctx->pos = iter.tgid + TGID_OFFSET;
3331 		if (!proc_fill_cache(file, ctx, name, len,
3332 				     proc_pid_instantiate, iter.task, NULL)) {
3333 			put_task_struct(iter.task);
3334 			return 0;
3335 		}
3336 	}
3337 	ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3338 	return 0;
3339 }
3340 
3341 /*
3342  * proc_tid_comm_permission is a special permission function exclusively
3343  * used for the node /proc/<pid>/task/<tid>/comm.
3344  * It bypasses generic permission checks in the case where a task of the same
3345  * task group attempts to access the node.
3346  * The rationale behind this is that glibc and bionic access this node for
3347  * cross thread naming (pthread_set/getname_np(!self)). However, if
3348  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3349  * which locks out the cross thread naming implementation.
3350  * This function makes sure that the node is always accessible for members of
3351  * same thread group.
3352  */
3353 static int proc_tid_comm_permission(struct inode *inode, int mask)
3354 {
3355 	bool is_same_tgroup;
3356 	struct task_struct *task;
3357 
3358 	task = get_proc_task(inode);
3359 	if (!task)
3360 		return -ESRCH;
3361 	is_same_tgroup = same_thread_group(current, task);
3362 	put_task_struct(task);
3363 
3364 	if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3365 		/* This file (/proc/<pid>/task/<tid>/comm) can always be
3366 		 * read or written by the members of the corresponding
3367 		 * thread group.
3368 		 */
3369 		return 0;
3370 	}
3371 
3372 	return generic_permission(inode, mask);
3373 }
3374 
3375 static const struct inode_operations proc_tid_comm_inode_operations = {
3376 		.permission = proc_tid_comm_permission,
3377 };
3378 
3379 /*
3380  * Tasks
3381  */
3382 static const struct pid_entry tid_base_stuff[] = {
3383 	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3384 	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3385 	DIR("ns",	 S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3386 #ifdef CONFIG_NET
3387 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3388 #endif
3389 	REG("environ",   S_IRUSR, proc_environ_operations),
3390 	REG("auxv",      S_IRUSR, proc_auxv_operations),
3391 	ONE("status",    S_IRUGO, proc_pid_status),
3392 	ONE("personality", S_IRUSR, proc_pid_personality),
3393 	ONE("limits",	 S_IRUGO, proc_pid_limits),
3394 #ifdef CONFIG_SCHED_DEBUG
3395 	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3396 #endif
3397 	NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3398 			 &proc_tid_comm_inode_operations,
3399 			 &proc_pid_set_comm_operations, {}),
3400 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3401 	ONE("syscall",   S_IRUSR, proc_pid_syscall),
3402 #endif
3403 	REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3404 	ONE("stat",      S_IRUGO, proc_tid_stat),
3405 	ONE("statm",     S_IRUGO, proc_pid_statm),
3406 	REG("maps",      S_IRUGO, proc_pid_maps_operations),
3407 #ifdef CONFIG_PROC_CHILDREN
3408 	REG("children",  S_IRUGO, proc_tid_children_operations),
3409 #endif
3410 #ifdef CONFIG_NUMA
3411 	REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3412 #endif
3413 	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3414 	LNK("cwd",       proc_cwd_link),
3415 	LNK("root",      proc_root_link),
3416 	LNK("exe",       proc_exe_link),
3417 	REG("mounts",    S_IRUGO, proc_mounts_operations),
3418 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3419 #ifdef CONFIG_PROC_PAGE_MONITOR
3420 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3421 	REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3422 	REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3423 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3424 #endif
3425 #ifdef CONFIG_SECURITY
3426 	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3427 #endif
3428 #ifdef CONFIG_KALLSYMS
3429 	ONE("wchan",     S_IRUGO, proc_pid_wchan),
3430 #endif
3431 #ifdef CONFIG_STACKTRACE
3432 	ONE("stack",      S_IRUSR, proc_pid_stack),
3433 #endif
3434 #ifdef CONFIG_SCHED_INFO
3435 	ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3436 #endif
3437 #ifdef CONFIG_LATENCYTOP
3438 	REG("latency",  S_IRUGO, proc_lstats_operations),
3439 #endif
3440 #ifdef CONFIG_PROC_PID_CPUSET
3441 	ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3442 #endif
3443 #ifdef CONFIG_CGROUPS
3444 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3445 #endif
3446 	ONE("oom_score", S_IRUGO, proc_oom_score),
3447 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3448 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3449 #ifdef CONFIG_AUDIT
3450 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3451 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3452 #endif
3453 #ifdef CONFIG_FAULT_INJECTION
3454 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3455 	REG("fail-nth", 0644, proc_fail_nth_operations),
3456 #endif
3457 #ifdef CONFIG_TASK_IO_ACCOUNTING
3458 	ONE("io",	S_IRUSR, proc_tid_io_accounting),
3459 #endif
3460 #ifdef CONFIG_USER_NS
3461 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3462 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3463 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3464 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3465 #endif
3466 #ifdef CONFIG_LIVEPATCH
3467 	ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3468 #endif
3469 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3470 	ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3471 #endif
3472 };
3473 
3474 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3475 {
3476 	return proc_pident_readdir(file, ctx,
3477 				   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3478 }
3479 
3480 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3481 {
3482 	return proc_pident_lookup(dir, dentry,
3483 				  tid_base_stuff,
3484 				  tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3485 }
3486 
3487 static const struct file_operations proc_tid_base_operations = {
3488 	.read		= generic_read_dir,
3489 	.iterate_shared	= proc_tid_base_readdir,
3490 	.llseek		= generic_file_llseek,
3491 };
3492 
3493 static const struct inode_operations proc_tid_base_inode_operations = {
3494 	.lookup		= proc_tid_base_lookup,
3495 	.getattr	= pid_getattr,
3496 	.setattr	= proc_setattr,
3497 };
3498 
3499 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3500 	struct task_struct *task, const void *ptr)
3501 {
3502 	struct inode *inode;
3503 	inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3504 	if (!inode)
3505 		return ERR_PTR(-ENOENT);
3506 
3507 	inode->i_op = &proc_tid_base_inode_operations;
3508 	inode->i_fop = &proc_tid_base_operations;
3509 	inode->i_flags |= S_IMMUTABLE;
3510 
3511 	set_nlink(inode, nlink_tid);
3512 	pid_update_inode(task, inode);
3513 
3514 	d_set_d_op(dentry, &pid_dentry_operations);
3515 	return d_splice_alias(inode, dentry);
3516 }
3517 
3518 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3519 {
3520 	struct task_struct *task;
3521 	struct task_struct *leader = get_proc_task(dir);
3522 	unsigned tid;
3523 	struct pid_namespace *ns;
3524 	struct dentry *result = ERR_PTR(-ENOENT);
3525 
3526 	if (!leader)
3527 		goto out_no_task;
3528 
3529 	tid = name_to_int(&dentry->d_name);
3530 	if (tid == ~0U)
3531 		goto out;
3532 
3533 	ns = dentry->d_sb->s_fs_info;
3534 	rcu_read_lock();
3535 	task = find_task_by_pid_ns(tid, ns);
3536 	if (task)
3537 		get_task_struct(task);
3538 	rcu_read_unlock();
3539 	if (!task)
3540 		goto out;
3541 	if (!same_thread_group(leader, task))
3542 		goto out_drop_task;
3543 
3544 	result = proc_task_instantiate(dentry, task, NULL);
3545 out_drop_task:
3546 	put_task_struct(task);
3547 out:
3548 	put_task_struct(leader);
3549 out_no_task:
3550 	return result;
3551 }
3552 
3553 /*
3554  * Find the first tid of a thread group to return to user space.
3555  *
3556  * Usually this is just the thread group leader, but if the users
3557  * buffer was too small or there was a seek into the middle of the
3558  * directory we have more work todo.
3559  *
3560  * In the case of a short read we start with find_task_by_pid.
3561  *
3562  * In the case of a seek we start with the leader and walk nr
3563  * threads past it.
3564  */
3565 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3566 					struct pid_namespace *ns)
3567 {
3568 	struct task_struct *pos, *task;
3569 	unsigned long nr = f_pos;
3570 
3571 	if (nr != f_pos)	/* 32bit overflow? */
3572 		return NULL;
3573 
3574 	rcu_read_lock();
3575 	task = pid_task(pid, PIDTYPE_PID);
3576 	if (!task)
3577 		goto fail;
3578 
3579 	/* Attempt to start with the tid of a thread */
3580 	if (tid && nr) {
3581 		pos = find_task_by_pid_ns(tid, ns);
3582 		if (pos && same_thread_group(pos, task))
3583 			goto found;
3584 	}
3585 
3586 	/* If nr exceeds the number of threads there is nothing todo */
3587 	if (nr >= get_nr_threads(task))
3588 		goto fail;
3589 
3590 	/* If we haven't found our starting place yet start
3591 	 * with the leader and walk nr threads forward.
3592 	 */
3593 	pos = task = task->group_leader;
3594 	do {
3595 		if (!nr--)
3596 			goto found;
3597 	} while_each_thread(task, pos);
3598 fail:
3599 	pos = NULL;
3600 	goto out;
3601 found:
3602 	get_task_struct(pos);
3603 out:
3604 	rcu_read_unlock();
3605 	return pos;
3606 }
3607 
3608 /*
3609  * Find the next thread in the thread list.
3610  * Return NULL if there is an error or no next thread.
3611  *
3612  * The reference to the input task_struct is released.
3613  */
3614 static struct task_struct *next_tid(struct task_struct *start)
3615 {
3616 	struct task_struct *pos = NULL;
3617 	rcu_read_lock();
3618 	if (pid_alive(start)) {
3619 		pos = next_thread(start);
3620 		if (thread_group_leader(pos))
3621 			pos = NULL;
3622 		else
3623 			get_task_struct(pos);
3624 	}
3625 	rcu_read_unlock();
3626 	put_task_struct(start);
3627 	return pos;
3628 }
3629 
3630 /* for the /proc/TGID/task/ directories */
3631 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3632 {
3633 	struct inode *inode = file_inode(file);
3634 	struct task_struct *task;
3635 	struct pid_namespace *ns;
3636 	int tid;
3637 
3638 	if (proc_inode_is_dead(inode))
3639 		return -ENOENT;
3640 
3641 	if (!dir_emit_dots(file, ctx))
3642 		return 0;
3643 
3644 	/* f_version caches the tgid value that the last readdir call couldn't
3645 	 * return. lseek aka telldir automagically resets f_version to 0.
3646 	 */
3647 	ns = proc_pid_ns(inode);
3648 	tid = (int)file->f_version;
3649 	file->f_version = 0;
3650 	for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3651 	     task;
3652 	     task = next_tid(task), ctx->pos++) {
3653 		char name[10 + 1];
3654 		unsigned int len;
3655 		tid = task_pid_nr_ns(task, ns);
3656 		len = snprintf(name, sizeof(name), "%u", tid);
3657 		if (!proc_fill_cache(file, ctx, name, len,
3658 				proc_task_instantiate, task, NULL)) {
3659 			/* returning this tgid failed, save it as the first
3660 			 * pid for the next readir call */
3661 			file->f_version = (u64)tid;
3662 			put_task_struct(task);
3663 			break;
3664 		}
3665 	}
3666 
3667 	return 0;
3668 }
3669 
3670 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3671 			     u32 request_mask, unsigned int query_flags)
3672 {
3673 	struct inode *inode = d_inode(path->dentry);
3674 	struct task_struct *p = get_proc_task(inode);
3675 	generic_fillattr(inode, stat);
3676 
3677 	if (p) {
3678 		stat->nlink += get_nr_threads(p);
3679 		put_task_struct(p);
3680 	}
3681 
3682 	return 0;
3683 }
3684 
3685 static const struct inode_operations proc_task_inode_operations = {
3686 	.lookup		= proc_task_lookup,
3687 	.getattr	= proc_task_getattr,
3688 	.setattr	= proc_setattr,
3689 	.permission	= proc_pid_permission,
3690 };
3691 
3692 static const struct file_operations proc_task_operations = {
3693 	.read		= generic_read_dir,
3694 	.iterate_shared	= proc_task_readdir,
3695 	.llseek		= generic_file_llseek,
3696 };
3697 
3698 void __init set_proc_pid_nlink(void)
3699 {
3700 	nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3701 	nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3702 }
3703