xref: /openbmc/linux/fs/proc/base.c (revision 239480ab)
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49 
50 #include <linux/uaccess.h>
51 
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/printk.h>
77 #include <linux/cgroup.h>
78 #include <linux/cpuset.h>
79 #include <linux/audit.h>
80 #include <linux/poll.h>
81 #include <linux/nsproxy.h>
82 #include <linux/oom.h>
83 #include <linux/elf.h>
84 #include <linux/pid_namespace.h>
85 #include <linux/user_namespace.h>
86 #include <linux/fs_struct.h>
87 #include <linux/slab.h>
88 #include <linux/sched/autogroup.h>
89 #include <linux/sched/mm.h>
90 #include <linux/sched/coredump.h>
91 #include <linux/sched/debug.h>
92 #include <linux/sched/stat.h>
93 #include <linux/flex_array.h>
94 #include <linux/posix-timers.h>
95 #ifdef CONFIG_HARDWALL
96 #include <asm/hardwall.h>
97 #endif
98 #include <trace/events/oom.h>
99 #include "internal.h"
100 #include "fd.h"
101 
102 /* NOTE:
103  *	Implementing inode permission operations in /proc is almost
104  *	certainly an error.  Permission checks need to happen during
105  *	each system call not at open time.  The reason is that most of
106  *	what we wish to check for permissions in /proc varies at runtime.
107  *
108  *	The classic example of a problem is opening file descriptors
109  *	in /proc for a task before it execs a suid executable.
110  */
111 
112 static u8 nlink_tid;
113 static u8 nlink_tgid;
114 
115 struct pid_entry {
116 	const char *name;
117 	unsigned int len;
118 	umode_t mode;
119 	const struct inode_operations *iop;
120 	const struct file_operations *fop;
121 	union proc_op op;
122 };
123 
124 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
125 	.name = (NAME),					\
126 	.len  = sizeof(NAME) - 1,			\
127 	.mode = MODE,					\
128 	.iop  = IOP,					\
129 	.fop  = FOP,					\
130 	.op   = OP,					\
131 }
132 
133 #define DIR(NAME, MODE, iops, fops)	\
134 	NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
135 #define LNK(NAME, get_link)					\
136 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
137 		&proc_pid_link_inode_operations, NULL,		\
138 		{ .proc_get_link = get_link } )
139 #define REG(NAME, MODE, fops)				\
140 	NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
141 #define ONE(NAME, MODE, show)				\
142 	NOD(NAME, (S_IFREG|(MODE)), 			\
143 		NULL, &proc_single_file_operations,	\
144 		{ .proc_show = show } )
145 
146 /*
147  * Count the number of hardlinks for the pid_entry table, excluding the .
148  * and .. links.
149  */
150 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
151 	unsigned int n)
152 {
153 	unsigned int i;
154 	unsigned int count;
155 
156 	count = 2;
157 	for (i = 0; i < n; ++i) {
158 		if (S_ISDIR(entries[i].mode))
159 			++count;
160 	}
161 
162 	return count;
163 }
164 
165 static int get_task_root(struct task_struct *task, struct path *root)
166 {
167 	int result = -ENOENT;
168 
169 	task_lock(task);
170 	if (task->fs) {
171 		get_fs_root(task->fs, root);
172 		result = 0;
173 	}
174 	task_unlock(task);
175 	return result;
176 }
177 
178 static int proc_cwd_link(struct dentry *dentry, struct path *path)
179 {
180 	struct task_struct *task = get_proc_task(d_inode(dentry));
181 	int result = -ENOENT;
182 
183 	if (task) {
184 		task_lock(task);
185 		if (task->fs) {
186 			get_fs_pwd(task->fs, path);
187 			result = 0;
188 		}
189 		task_unlock(task);
190 		put_task_struct(task);
191 	}
192 	return result;
193 }
194 
195 static int proc_root_link(struct dentry *dentry, struct path *path)
196 {
197 	struct task_struct *task = get_proc_task(d_inode(dentry));
198 	int result = -ENOENT;
199 
200 	if (task) {
201 		result = get_task_root(task, path);
202 		put_task_struct(task);
203 	}
204 	return result;
205 }
206 
207 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
208 				     size_t _count, loff_t *pos)
209 {
210 	struct task_struct *tsk;
211 	struct mm_struct *mm;
212 	char *page;
213 	unsigned long count = _count;
214 	unsigned long arg_start, arg_end, env_start, env_end;
215 	unsigned long len1, len2, len;
216 	unsigned long p;
217 	char c;
218 	ssize_t rv;
219 
220 	BUG_ON(*pos < 0);
221 
222 	tsk = get_proc_task(file_inode(file));
223 	if (!tsk)
224 		return -ESRCH;
225 	mm = get_task_mm(tsk);
226 	put_task_struct(tsk);
227 	if (!mm)
228 		return 0;
229 	/* Check if process spawned far enough to have cmdline. */
230 	if (!mm->env_end) {
231 		rv = 0;
232 		goto out_mmput;
233 	}
234 
235 	page = (char *)__get_free_page(GFP_TEMPORARY);
236 	if (!page) {
237 		rv = -ENOMEM;
238 		goto out_mmput;
239 	}
240 
241 	down_read(&mm->mmap_sem);
242 	arg_start = mm->arg_start;
243 	arg_end = mm->arg_end;
244 	env_start = mm->env_start;
245 	env_end = mm->env_end;
246 	up_read(&mm->mmap_sem);
247 
248 	BUG_ON(arg_start > arg_end);
249 	BUG_ON(env_start > env_end);
250 
251 	len1 = arg_end - arg_start;
252 	len2 = env_end - env_start;
253 
254 	/* Empty ARGV. */
255 	if (len1 == 0) {
256 		rv = 0;
257 		goto out_free_page;
258 	}
259 	/*
260 	 * Inherently racy -- command line shares address space
261 	 * with code and data.
262 	 */
263 	rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
264 	if (rv <= 0)
265 		goto out_free_page;
266 
267 	rv = 0;
268 
269 	if (c == '\0') {
270 		/* Command line (set of strings) occupies whole ARGV. */
271 		if (len1 <= *pos)
272 			goto out_free_page;
273 
274 		p = arg_start + *pos;
275 		len = len1 - *pos;
276 		while (count > 0 && len > 0) {
277 			unsigned int _count;
278 			int nr_read;
279 
280 			_count = min3(count, len, PAGE_SIZE);
281 			nr_read = access_remote_vm(mm, p, page, _count, 0);
282 			if (nr_read < 0)
283 				rv = nr_read;
284 			if (nr_read <= 0)
285 				goto out_free_page;
286 
287 			if (copy_to_user(buf, page, nr_read)) {
288 				rv = -EFAULT;
289 				goto out_free_page;
290 			}
291 
292 			p	+= nr_read;
293 			len	-= nr_read;
294 			buf	+= nr_read;
295 			count	-= nr_read;
296 			rv	+= nr_read;
297 		}
298 	} else {
299 		/*
300 		 * Command line (1 string) occupies ARGV and
301 		 * extends into ENVP.
302 		 */
303 		struct {
304 			unsigned long p;
305 			unsigned long len;
306 		} cmdline[2] = {
307 			{ .p = arg_start, .len = len1 },
308 			{ .p = env_start, .len = len2 },
309 		};
310 		loff_t pos1 = *pos;
311 		unsigned int i;
312 
313 		i = 0;
314 		while (i < 2 && pos1 >= cmdline[i].len) {
315 			pos1 -= cmdline[i].len;
316 			i++;
317 		}
318 		while (i < 2) {
319 			p = cmdline[i].p + pos1;
320 			len = cmdline[i].len - pos1;
321 			while (count > 0 && len > 0) {
322 				unsigned int _count, l;
323 				int nr_read;
324 				bool final;
325 
326 				_count = min3(count, len, PAGE_SIZE);
327 				nr_read = access_remote_vm(mm, p, page, _count, 0);
328 				if (nr_read < 0)
329 					rv = nr_read;
330 				if (nr_read <= 0)
331 					goto out_free_page;
332 
333 				/*
334 				 * Command line can be shorter than whole ARGV
335 				 * even if last "marker" byte says it is not.
336 				 */
337 				final = false;
338 				l = strnlen(page, nr_read);
339 				if (l < nr_read) {
340 					nr_read = l;
341 					final = true;
342 				}
343 
344 				if (copy_to_user(buf, page, nr_read)) {
345 					rv = -EFAULT;
346 					goto out_free_page;
347 				}
348 
349 				p	+= nr_read;
350 				len	-= nr_read;
351 				buf	+= nr_read;
352 				count	-= nr_read;
353 				rv	+= nr_read;
354 
355 				if (final)
356 					goto out_free_page;
357 			}
358 
359 			/* Only first chunk can be read partially. */
360 			pos1 = 0;
361 			i++;
362 		}
363 	}
364 
365 out_free_page:
366 	free_page((unsigned long)page);
367 out_mmput:
368 	mmput(mm);
369 	if (rv > 0)
370 		*pos += rv;
371 	return rv;
372 }
373 
374 static const struct file_operations proc_pid_cmdline_ops = {
375 	.read	= proc_pid_cmdline_read,
376 	.llseek	= generic_file_llseek,
377 };
378 
379 #ifdef CONFIG_KALLSYMS
380 /*
381  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
382  * Returns the resolved symbol.  If that fails, simply return the address.
383  */
384 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
385 			  struct pid *pid, struct task_struct *task)
386 {
387 	unsigned long wchan;
388 	char symname[KSYM_NAME_LEN];
389 
390 	wchan = get_wchan(task);
391 
392 	if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
393 			&& !lookup_symbol_name(wchan, symname))
394 		seq_printf(m, "%s", symname);
395 	else
396 		seq_putc(m, '0');
397 
398 	return 0;
399 }
400 #endif /* CONFIG_KALLSYMS */
401 
402 static int lock_trace(struct task_struct *task)
403 {
404 	int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
405 	if (err)
406 		return err;
407 	if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
408 		mutex_unlock(&task->signal->cred_guard_mutex);
409 		return -EPERM;
410 	}
411 	return 0;
412 }
413 
414 static void unlock_trace(struct task_struct *task)
415 {
416 	mutex_unlock(&task->signal->cred_guard_mutex);
417 }
418 
419 #ifdef CONFIG_STACKTRACE
420 
421 #define MAX_STACK_TRACE_DEPTH	64
422 
423 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
424 			  struct pid *pid, struct task_struct *task)
425 {
426 	struct stack_trace trace;
427 	unsigned long *entries;
428 	int err;
429 	int i;
430 
431 	entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
432 	if (!entries)
433 		return -ENOMEM;
434 
435 	trace.nr_entries	= 0;
436 	trace.max_entries	= MAX_STACK_TRACE_DEPTH;
437 	trace.entries		= entries;
438 	trace.skip		= 0;
439 
440 	err = lock_trace(task);
441 	if (!err) {
442 		save_stack_trace_tsk(task, &trace);
443 
444 		for (i = 0; i < trace.nr_entries; i++) {
445 			seq_printf(m, "[<%pK>] %pB\n",
446 				   (void *)entries[i], (void *)entries[i]);
447 		}
448 		unlock_trace(task);
449 	}
450 	kfree(entries);
451 
452 	return err;
453 }
454 #endif
455 
456 #ifdef CONFIG_SCHED_INFO
457 /*
458  * Provides /proc/PID/schedstat
459  */
460 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
461 			      struct pid *pid, struct task_struct *task)
462 {
463 	if (unlikely(!sched_info_on()))
464 		seq_printf(m, "0 0 0\n");
465 	else
466 		seq_printf(m, "%llu %llu %lu\n",
467 		   (unsigned long long)task->se.sum_exec_runtime,
468 		   (unsigned long long)task->sched_info.run_delay,
469 		   task->sched_info.pcount);
470 
471 	return 0;
472 }
473 #endif
474 
475 #ifdef CONFIG_LATENCYTOP
476 static int lstats_show_proc(struct seq_file *m, void *v)
477 {
478 	int i;
479 	struct inode *inode = m->private;
480 	struct task_struct *task = get_proc_task(inode);
481 
482 	if (!task)
483 		return -ESRCH;
484 	seq_puts(m, "Latency Top version : v0.1\n");
485 	for (i = 0; i < 32; i++) {
486 		struct latency_record *lr = &task->latency_record[i];
487 		if (lr->backtrace[0]) {
488 			int q;
489 			seq_printf(m, "%i %li %li",
490 				   lr->count, lr->time, lr->max);
491 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
492 				unsigned long bt = lr->backtrace[q];
493 				if (!bt)
494 					break;
495 				if (bt == ULONG_MAX)
496 					break;
497 				seq_printf(m, " %ps", (void *)bt);
498 			}
499 			seq_putc(m, '\n');
500 		}
501 
502 	}
503 	put_task_struct(task);
504 	return 0;
505 }
506 
507 static int lstats_open(struct inode *inode, struct file *file)
508 {
509 	return single_open(file, lstats_show_proc, inode);
510 }
511 
512 static ssize_t lstats_write(struct file *file, const char __user *buf,
513 			    size_t count, loff_t *offs)
514 {
515 	struct task_struct *task = get_proc_task(file_inode(file));
516 
517 	if (!task)
518 		return -ESRCH;
519 	clear_all_latency_tracing(task);
520 	put_task_struct(task);
521 
522 	return count;
523 }
524 
525 static const struct file_operations proc_lstats_operations = {
526 	.open		= lstats_open,
527 	.read		= seq_read,
528 	.write		= lstats_write,
529 	.llseek		= seq_lseek,
530 	.release	= single_release,
531 };
532 
533 #endif
534 
535 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
536 			  struct pid *pid, struct task_struct *task)
537 {
538 	unsigned long totalpages = totalram_pages + total_swap_pages;
539 	unsigned long points = 0;
540 
541 	points = oom_badness(task, NULL, NULL, totalpages) *
542 					1000 / totalpages;
543 	seq_printf(m, "%lu\n", points);
544 
545 	return 0;
546 }
547 
548 struct limit_names {
549 	const char *name;
550 	const char *unit;
551 };
552 
553 static const struct limit_names lnames[RLIM_NLIMITS] = {
554 	[RLIMIT_CPU] = {"Max cpu time", "seconds"},
555 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
556 	[RLIMIT_DATA] = {"Max data size", "bytes"},
557 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
558 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
559 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
560 	[RLIMIT_NPROC] = {"Max processes", "processes"},
561 	[RLIMIT_NOFILE] = {"Max open files", "files"},
562 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
563 	[RLIMIT_AS] = {"Max address space", "bytes"},
564 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
565 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
566 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
567 	[RLIMIT_NICE] = {"Max nice priority", NULL},
568 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
569 	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
570 };
571 
572 /* Display limits for a process */
573 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
574 			   struct pid *pid, struct task_struct *task)
575 {
576 	unsigned int i;
577 	unsigned long flags;
578 
579 	struct rlimit rlim[RLIM_NLIMITS];
580 
581 	if (!lock_task_sighand(task, &flags))
582 		return 0;
583 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
584 	unlock_task_sighand(task, &flags);
585 
586 	/*
587 	 * print the file header
588 	 */
589        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
590 		  "Limit", "Soft Limit", "Hard Limit", "Units");
591 
592 	for (i = 0; i < RLIM_NLIMITS; i++) {
593 		if (rlim[i].rlim_cur == RLIM_INFINITY)
594 			seq_printf(m, "%-25s %-20s ",
595 				   lnames[i].name, "unlimited");
596 		else
597 			seq_printf(m, "%-25s %-20lu ",
598 				   lnames[i].name, rlim[i].rlim_cur);
599 
600 		if (rlim[i].rlim_max == RLIM_INFINITY)
601 			seq_printf(m, "%-20s ", "unlimited");
602 		else
603 			seq_printf(m, "%-20lu ", rlim[i].rlim_max);
604 
605 		if (lnames[i].unit)
606 			seq_printf(m, "%-10s\n", lnames[i].unit);
607 		else
608 			seq_putc(m, '\n');
609 	}
610 
611 	return 0;
612 }
613 
614 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
615 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
616 			    struct pid *pid, struct task_struct *task)
617 {
618 	long nr;
619 	unsigned long args[6], sp, pc;
620 	int res;
621 
622 	res = lock_trace(task);
623 	if (res)
624 		return res;
625 
626 	if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
627 		seq_puts(m, "running\n");
628 	else if (nr < 0)
629 		seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
630 	else
631 		seq_printf(m,
632 		       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
633 		       nr,
634 		       args[0], args[1], args[2], args[3], args[4], args[5],
635 		       sp, pc);
636 	unlock_trace(task);
637 
638 	return 0;
639 }
640 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
641 
642 /************************************************************************/
643 /*                       Here the fs part begins                        */
644 /************************************************************************/
645 
646 /* permission checks */
647 static int proc_fd_access_allowed(struct inode *inode)
648 {
649 	struct task_struct *task;
650 	int allowed = 0;
651 	/* Allow access to a task's file descriptors if it is us or we
652 	 * may use ptrace attach to the process and find out that
653 	 * information.
654 	 */
655 	task = get_proc_task(inode);
656 	if (task) {
657 		allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
658 		put_task_struct(task);
659 	}
660 	return allowed;
661 }
662 
663 int proc_setattr(struct dentry *dentry, struct iattr *attr)
664 {
665 	int error;
666 	struct inode *inode = d_inode(dentry);
667 
668 	if (attr->ia_valid & ATTR_MODE)
669 		return -EPERM;
670 
671 	error = setattr_prepare(dentry, attr);
672 	if (error)
673 		return error;
674 
675 	setattr_copy(inode, attr);
676 	mark_inode_dirty(inode);
677 	return 0;
678 }
679 
680 /*
681  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
682  * or euid/egid (for hide_pid_min=2)?
683  */
684 static bool has_pid_permissions(struct pid_namespace *pid,
685 				 struct task_struct *task,
686 				 int hide_pid_min)
687 {
688 	if (pid->hide_pid < hide_pid_min)
689 		return true;
690 	if (in_group_p(pid->pid_gid))
691 		return true;
692 	return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
693 }
694 
695 
696 static int proc_pid_permission(struct inode *inode, int mask)
697 {
698 	struct pid_namespace *pid = inode->i_sb->s_fs_info;
699 	struct task_struct *task;
700 	bool has_perms;
701 
702 	task = get_proc_task(inode);
703 	if (!task)
704 		return -ESRCH;
705 	has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
706 	put_task_struct(task);
707 
708 	if (!has_perms) {
709 		if (pid->hide_pid == HIDEPID_INVISIBLE) {
710 			/*
711 			 * Let's make getdents(), stat(), and open()
712 			 * consistent with each other.  If a process
713 			 * may not stat() a file, it shouldn't be seen
714 			 * in procfs at all.
715 			 */
716 			return -ENOENT;
717 		}
718 
719 		return -EPERM;
720 	}
721 	return generic_permission(inode, mask);
722 }
723 
724 
725 
726 static const struct inode_operations proc_def_inode_operations = {
727 	.setattr	= proc_setattr,
728 };
729 
730 static int proc_single_show(struct seq_file *m, void *v)
731 {
732 	struct inode *inode = m->private;
733 	struct pid_namespace *ns;
734 	struct pid *pid;
735 	struct task_struct *task;
736 	int ret;
737 
738 	ns = inode->i_sb->s_fs_info;
739 	pid = proc_pid(inode);
740 	task = get_pid_task(pid, PIDTYPE_PID);
741 	if (!task)
742 		return -ESRCH;
743 
744 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
745 
746 	put_task_struct(task);
747 	return ret;
748 }
749 
750 static int proc_single_open(struct inode *inode, struct file *filp)
751 {
752 	return single_open(filp, proc_single_show, inode);
753 }
754 
755 static const struct file_operations proc_single_file_operations = {
756 	.open		= proc_single_open,
757 	.read		= seq_read,
758 	.llseek		= seq_lseek,
759 	.release	= single_release,
760 };
761 
762 
763 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
764 {
765 	struct task_struct *task = get_proc_task(inode);
766 	struct mm_struct *mm = ERR_PTR(-ESRCH);
767 
768 	if (task) {
769 		mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
770 		put_task_struct(task);
771 
772 		if (!IS_ERR_OR_NULL(mm)) {
773 			/* ensure this mm_struct can't be freed */
774 			mmgrab(mm);
775 			/* but do not pin its memory */
776 			mmput(mm);
777 		}
778 	}
779 
780 	return mm;
781 }
782 
783 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
784 {
785 	struct mm_struct *mm = proc_mem_open(inode, mode);
786 
787 	if (IS_ERR(mm))
788 		return PTR_ERR(mm);
789 
790 	file->private_data = mm;
791 	return 0;
792 }
793 
794 static int mem_open(struct inode *inode, struct file *file)
795 {
796 	int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
797 
798 	/* OK to pass negative loff_t, we can catch out-of-range */
799 	file->f_mode |= FMODE_UNSIGNED_OFFSET;
800 
801 	return ret;
802 }
803 
804 static ssize_t mem_rw(struct file *file, char __user *buf,
805 			size_t count, loff_t *ppos, int write)
806 {
807 	struct mm_struct *mm = file->private_data;
808 	unsigned long addr = *ppos;
809 	ssize_t copied;
810 	char *page;
811 	unsigned int flags;
812 
813 	if (!mm)
814 		return 0;
815 
816 	page = (char *)__get_free_page(GFP_TEMPORARY);
817 	if (!page)
818 		return -ENOMEM;
819 
820 	copied = 0;
821 	if (!mmget_not_zero(mm))
822 		goto free;
823 
824 	flags = write ? FOLL_WRITE : 0;
825 
826 	while (count > 0) {
827 		int this_len = min_t(int, count, PAGE_SIZE);
828 
829 		if (write && copy_from_user(page, buf, this_len)) {
830 			copied = -EFAULT;
831 			break;
832 		}
833 
834 		this_len = access_remote_vm(mm, addr, page, this_len, flags);
835 		if (!this_len) {
836 			if (!copied)
837 				copied = -EIO;
838 			break;
839 		}
840 
841 		if (!write && copy_to_user(buf, page, this_len)) {
842 			copied = -EFAULT;
843 			break;
844 		}
845 
846 		buf += this_len;
847 		addr += this_len;
848 		copied += this_len;
849 		count -= this_len;
850 	}
851 	*ppos = addr;
852 
853 	mmput(mm);
854 free:
855 	free_page((unsigned long) page);
856 	return copied;
857 }
858 
859 static ssize_t mem_read(struct file *file, char __user *buf,
860 			size_t count, loff_t *ppos)
861 {
862 	return mem_rw(file, buf, count, ppos, 0);
863 }
864 
865 static ssize_t mem_write(struct file *file, const char __user *buf,
866 			 size_t count, loff_t *ppos)
867 {
868 	return mem_rw(file, (char __user*)buf, count, ppos, 1);
869 }
870 
871 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
872 {
873 	switch (orig) {
874 	case 0:
875 		file->f_pos = offset;
876 		break;
877 	case 1:
878 		file->f_pos += offset;
879 		break;
880 	default:
881 		return -EINVAL;
882 	}
883 	force_successful_syscall_return();
884 	return file->f_pos;
885 }
886 
887 static int mem_release(struct inode *inode, struct file *file)
888 {
889 	struct mm_struct *mm = file->private_data;
890 	if (mm)
891 		mmdrop(mm);
892 	return 0;
893 }
894 
895 static const struct file_operations proc_mem_operations = {
896 	.llseek		= mem_lseek,
897 	.read		= mem_read,
898 	.write		= mem_write,
899 	.open		= mem_open,
900 	.release	= mem_release,
901 };
902 
903 static int environ_open(struct inode *inode, struct file *file)
904 {
905 	return __mem_open(inode, file, PTRACE_MODE_READ);
906 }
907 
908 static ssize_t environ_read(struct file *file, char __user *buf,
909 			size_t count, loff_t *ppos)
910 {
911 	char *page;
912 	unsigned long src = *ppos;
913 	int ret = 0;
914 	struct mm_struct *mm = file->private_data;
915 	unsigned long env_start, env_end;
916 
917 	/* Ensure the process spawned far enough to have an environment. */
918 	if (!mm || !mm->env_end)
919 		return 0;
920 
921 	page = (char *)__get_free_page(GFP_TEMPORARY);
922 	if (!page)
923 		return -ENOMEM;
924 
925 	ret = 0;
926 	if (!mmget_not_zero(mm))
927 		goto free;
928 
929 	down_read(&mm->mmap_sem);
930 	env_start = mm->env_start;
931 	env_end = mm->env_end;
932 	up_read(&mm->mmap_sem);
933 
934 	while (count > 0) {
935 		size_t this_len, max_len;
936 		int retval;
937 
938 		if (src >= (env_end - env_start))
939 			break;
940 
941 		this_len = env_end - (env_start + src);
942 
943 		max_len = min_t(size_t, PAGE_SIZE, count);
944 		this_len = min(max_len, this_len);
945 
946 		retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
947 
948 		if (retval <= 0) {
949 			ret = retval;
950 			break;
951 		}
952 
953 		if (copy_to_user(buf, page, retval)) {
954 			ret = -EFAULT;
955 			break;
956 		}
957 
958 		ret += retval;
959 		src += retval;
960 		buf += retval;
961 		count -= retval;
962 	}
963 	*ppos = src;
964 	mmput(mm);
965 
966 free:
967 	free_page((unsigned long) page);
968 	return ret;
969 }
970 
971 static const struct file_operations proc_environ_operations = {
972 	.open		= environ_open,
973 	.read		= environ_read,
974 	.llseek		= generic_file_llseek,
975 	.release	= mem_release,
976 };
977 
978 static int auxv_open(struct inode *inode, struct file *file)
979 {
980 	return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
981 }
982 
983 static ssize_t auxv_read(struct file *file, char __user *buf,
984 			size_t count, loff_t *ppos)
985 {
986 	struct mm_struct *mm = file->private_data;
987 	unsigned int nwords = 0;
988 
989 	if (!mm)
990 		return 0;
991 	do {
992 		nwords += 2;
993 	} while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
994 	return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
995 				       nwords * sizeof(mm->saved_auxv[0]));
996 }
997 
998 static const struct file_operations proc_auxv_operations = {
999 	.open		= auxv_open,
1000 	.read		= auxv_read,
1001 	.llseek		= generic_file_llseek,
1002 	.release	= mem_release,
1003 };
1004 
1005 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1006 			    loff_t *ppos)
1007 {
1008 	struct task_struct *task = get_proc_task(file_inode(file));
1009 	char buffer[PROC_NUMBUF];
1010 	int oom_adj = OOM_ADJUST_MIN;
1011 	size_t len;
1012 
1013 	if (!task)
1014 		return -ESRCH;
1015 	if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1016 		oom_adj = OOM_ADJUST_MAX;
1017 	else
1018 		oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1019 			  OOM_SCORE_ADJ_MAX;
1020 	put_task_struct(task);
1021 	len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1022 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1023 }
1024 
1025 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1026 {
1027 	static DEFINE_MUTEX(oom_adj_mutex);
1028 	struct mm_struct *mm = NULL;
1029 	struct task_struct *task;
1030 	int err = 0;
1031 
1032 	task = get_proc_task(file_inode(file));
1033 	if (!task)
1034 		return -ESRCH;
1035 
1036 	mutex_lock(&oom_adj_mutex);
1037 	if (legacy) {
1038 		if (oom_adj < task->signal->oom_score_adj &&
1039 				!capable(CAP_SYS_RESOURCE)) {
1040 			err = -EACCES;
1041 			goto err_unlock;
1042 		}
1043 		/*
1044 		 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1045 		 * /proc/pid/oom_score_adj instead.
1046 		 */
1047 		pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1048 			  current->comm, task_pid_nr(current), task_pid_nr(task),
1049 			  task_pid_nr(task));
1050 	} else {
1051 		if ((short)oom_adj < task->signal->oom_score_adj_min &&
1052 				!capable(CAP_SYS_RESOURCE)) {
1053 			err = -EACCES;
1054 			goto err_unlock;
1055 		}
1056 	}
1057 
1058 	/*
1059 	 * Make sure we will check other processes sharing the mm if this is
1060 	 * not vfrok which wants its own oom_score_adj.
1061 	 * pin the mm so it doesn't go away and get reused after task_unlock
1062 	 */
1063 	if (!task->vfork_done) {
1064 		struct task_struct *p = find_lock_task_mm(task);
1065 
1066 		if (p) {
1067 			if (atomic_read(&p->mm->mm_users) > 1) {
1068 				mm = p->mm;
1069 				mmgrab(mm);
1070 			}
1071 			task_unlock(p);
1072 		}
1073 	}
1074 
1075 	task->signal->oom_score_adj = oom_adj;
1076 	if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1077 		task->signal->oom_score_adj_min = (short)oom_adj;
1078 	trace_oom_score_adj_update(task);
1079 
1080 	if (mm) {
1081 		struct task_struct *p;
1082 
1083 		rcu_read_lock();
1084 		for_each_process(p) {
1085 			if (same_thread_group(task, p))
1086 				continue;
1087 
1088 			/* do not touch kernel threads or the global init */
1089 			if (p->flags & PF_KTHREAD || is_global_init(p))
1090 				continue;
1091 
1092 			task_lock(p);
1093 			if (!p->vfork_done && process_shares_mm(p, mm)) {
1094 				pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1095 						task_pid_nr(p), p->comm,
1096 						p->signal->oom_score_adj, oom_adj,
1097 						task_pid_nr(task), task->comm);
1098 				p->signal->oom_score_adj = oom_adj;
1099 				if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1100 					p->signal->oom_score_adj_min = (short)oom_adj;
1101 			}
1102 			task_unlock(p);
1103 		}
1104 		rcu_read_unlock();
1105 		mmdrop(mm);
1106 	}
1107 err_unlock:
1108 	mutex_unlock(&oom_adj_mutex);
1109 	put_task_struct(task);
1110 	return err;
1111 }
1112 
1113 /*
1114  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1115  * kernels.  The effective policy is defined by oom_score_adj, which has a
1116  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1117  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1118  * Processes that become oom disabled via oom_adj will still be oom disabled
1119  * with this implementation.
1120  *
1121  * oom_adj cannot be removed since existing userspace binaries use it.
1122  */
1123 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1124 			     size_t count, loff_t *ppos)
1125 {
1126 	char buffer[PROC_NUMBUF];
1127 	int oom_adj;
1128 	int err;
1129 
1130 	memset(buffer, 0, sizeof(buffer));
1131 	if (count > sizeof(buffer) - 1)
1132 		count = sizeof(buffer) - 1;
1133 	if (copy_from_user(buffer, buf, count)) {
1134 		err = -EFAULT;
1135 		goto out;
1136 	}
1137 
1138 	err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1139 	if (err)
1140 		goto out;
1141 	if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1142 	     oom_adj != OOM_DISABLE) {
1143 		err = -EINVAL;
1144 		goto out;
1145 	}
1146 
1147 	/*
1148 	 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1149 	 * value is always attainable.
1150 	 */
1151 	if (oom_adj == OOM_ADJUST_MAX)
1152 		oom_adj = OOM_SCORE_ADJ_MAX;
1153 	else
1154 		oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1155 
1156 	err = __set_oom_adj(file, oom_adj, true);
1157 out:
1158 	return err < 0 ? err : count;
1159 }
1160 
1161 static const struct file_operations proc_oom_adj_operations = {
1162 	.read		= oom_adj_read,
1163 	.write		= oom_adj_write,
1164 	.llseek		= generic_file_llseek,
1165 };
1166 
1167 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1168 					size_t count, loff_t *ppos)
1169 {
1170 	struct task_struct *task = get_proc_task(file_inode(file));
1171 	char buffer[PROC_NUMBUF];
1172 	short oom_score_adj = OOM_SCORE_ADJ_MIN;
1173 	size_t len;
1174 
1175 	if (!task)
1176 		return -ESRCH;
1177 	oom_score_adj = task->signal->oom_score_adj;
1178 	put_task_struct(task);
1179 	len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1180 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1181 }
1182 
1183 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1184 					size_t count, loff_t *ppos)
1185 {
1186 	char buffer[PROC_NUMBUF];
1187 	int oom_score_adj;
1188 	int err;
1189 
1190 	memset(buffer, 0, sizeof(buffer));
1191 	if (count > sizeof(buffer) - 1)
1192 		count = sizeof(buffer) - 1;
1193 	if (copy_from_user(buffer, buf, count)) {
1194 		err = -EFAULT;
1195 		goto out;
1196 	}
1197 
1198 	err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1199 	if (err)
1200 		goto out;
1201 	if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1202 			oom_score_adj > OOM_SCORE_ADJ_MAX) {
1203 		err = -EINVAL;
1204 		goto out;
1205 	}
1206 
1207 	err = __set_oom_adj(file, oom_score_adj, false);
1208 out:
1209 	return err < 0 ? err : count;
1210 }
1211 
1212 static const struct file_operations proc_oom_score_adj_operations = {
1213 	.read		= oom_score_adj_read,
1214 	.write		= oom_score_adj_write,
1215 	.llseek		= default_llseek,
1216 };
1217 
1218 #ifdef CONFIG_AUDITSYSCALL
1219 #define TMPBUFLEN 11
1220 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1221 				  size_t count, loff_t *ppos)
1222 {
1223 	struct inode * inode = file_inode(file);
1224 	struct task_struct *task = get_proc_task(inode);
1225 	ssize_t length;
1226 	char tmpbuf[TMPBUFLEN];
1227 
1228 	if (!task)
1229 		return -ESRCH;
1230 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1231 			   from_kuid(file->f_cred->user_ns,
1232 				     audit_get_loginuid(task)));
1233 	put_task_struct(task);
1234 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1235 }
1236 
1237 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1238 				   size_t count, loff_t *ppos)
1239 {
1240 	struct inode * inode = file_inode(file);
1241 	uid_t loginuid;
1242 	kuid_t kloginuid;
1243 	int rv;
1244 
1245 	rcu_read_lock();
1246 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1247 		rcu_read_unlock();
1248 		return -EPERM;
1249 	}
1250 	rcu_read_unlock();
1251 
1252 	if (*ppos != 0) {
1253 		/* No partial writes. */
1254 		return -EINVAL;
1255 	}
1256 
1257 	rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1258 	if (rv < 0)
1259 		return rv;
1260 
1261 	/* is userspace tring to explicitly UNSET the loginuid? */
1262 	if (loginuid == AUDIT_UID_UNSET) {
1263 		kloginuid = INVALID_UID;
1264 	} else {
1265 		kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1266 		if (!uid_valid(kloginuid))
1267 			return -EINVAL;
1268 	}
1269 
1270 	rv = audit_set_loginuid(kloginuid);
1271 	if (rv < 0)
1272 		return rv;
1273 	return count;
1274 }
1275 
1276 static const struct file_operations proc_loginuid_operations = {
1277 	.read		= proc_loginuid_read,
1278 	.write		= proc_loginuid_write,
1279 	.llseek		= generic_file_llseek,
1280 };
1281 
1282 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1283 				  size_t count, loff_t *ppos)
1284 {
1285 	struct inode * inode = file_inode(file);
1286 	struct task_struct *task = get_proc_task(inode);
1287 	ssize_t length;
1288 	char tmpbuf[TMPBUFLEN];
1289 
1290 	if (!task)
1291 		return -ESRCH;
1292 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1293 				audit_get_sessionid(task));
1294 	put_task_struct(task);
1295 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1296 }
1297 
1298 static const struct file_operations proc_sessionid_operations = {
1299 	.read		= proc_sessionid_read,
1300 	.llseek		= generic_file_llseek,
1301 };
1302 #endif
1303 
1304 #ifdef CONFIG_FAULT_INJECTION
1305 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1306 				      size_t count, loff_t *ppos)
1307 {
1308 	struct task_struct *task = get_proc_task(file_inode(file));
1309 	char buffer[PROC_NUMBUF];
1310 	size_t len;
1311 	int make_it_fail;
1312 
1313 	if (!task)
1314 		return -ESRCH;
1315 	make_it_fail = task->make_it_fail;
1316 	put_task_struct(task);
1317 
1318 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1319 
1320 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1321 }
1322 
1323 static ssize_t proc_fault_inject_write(struct file * file,
1324 			const char __user * buf, size_t count, loff_t *ppos)
1325 {
1326 	struct task_struct *task;
1327 	char buffer[PROC_NUMBUF];
1328 	int make_it_fail;
1329 	int rv;
1330 
1331 	if (!capable(CAP_SYS_RESOURCE))
1332 		return -EPERM;
1333 	memset(buffer, 0, sizeof(buffer));
1334 	if (count > sizeof(buffer) - 1)
1335 		count = sizeof(buffer) - 1;
1336 	if (copy_from_user(buffer, buf, count))
1337 		return -EFAULT;
1338 	rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1339 	if (rv < 0)
1340 		return rv;
1341 	if (make_it_fail < 0 || make_it_fail > 1)
1342 		return -EINVAL;
1343 
1344 	task = get_proc_task(file_inode(file));
1345 	if (!task)
1346 		return -ESRCH;
1347 	task->make_it_fail = make_it_fail;
1348 	put_task_struct(task);
1349 
1350 	return count;
1351 }
1352 
1353 static const struct file_operations proc_fault_inject_operations = {
1354 	.read		= proc_fault_inject_read,
1355 	.write		= proc_fault_inject_write,
1356 	.llseek		= generic_file_llseek,
1357 };
1358 #endif
1359 
1360 
1361 #ifdef CONFIG_SCHED_DEBUG
1362 /*
1363  * Print out various scheduling related per-task fields:
1364  */
1365 static int sched_show(struct seq_file *m, void *v)
1366 {
1367 	struct inode *inode = m->private;
1368 	struct task_struct *p;
1369 
1370 	p = get_proc_task(inode);
1371 	if (!p)
1372 		return -ESRCH;
1373 	proc_sched_show_task(p, m);
1374 
1375 	put_task_struct(p);
1376 
1377 	return 0;
1378 }
1379 
1380 static ssize_t
1381 sched_write(struct file *file, const char __user *buf,
1382 	    size_t count, loff_t *offset)
1383 {
1384 	struct inode *inode = file_inode(file);
1385 	struct task_struct *p;
1386 
1387 	p = get_proc_task(inode);
1388 	if (!p)
1389 		return -ESRCH;
1390 	proc_sched_set_task(p);
1391 
1392 	put_task_struct(p);
1393 
1394 	return count;
1395 }
1396 
1397 static int sched_open(struct inode *inode, struct file *filp)
1398 {
1399 	return single_open(filp, sched_show, inode);
1400 }
1401 
1402 static const struct file_operations proc_pid_sched_operations = {
1403 	.open		= sched_open,
1404 	.read		= seq_read,
1405 	.write		= sched_write,
1406 	.llseek		= seq_lseek,
1407 	.release	= single_release,
1408 };
1409 
1410 #endif
1411 
1412 #ifdef CONFIG_SCHED_AUTOGROUP
1413 /*
1414  * Print out autogroup related information:
1415  */
1416 static int sched_autogroup_show(struct seq_file *m, void *v)
1417 {
1418 	struct inode *inode = m->private;
1419 	struct task_struct *p;
1420 
1421 	p = get_proc_task(inode);
1422 	if (!p)
1423 		return -ESRCH;
1424 	proc_sched_autogroup_show_task(p, m);
1425 
1426 	put_task_struct(p);
1427 
1428 	return 0;
1429 }
1430 
1431 static ssize_t
1432 sched_autogroup_write(struct file *file, const char __user *buf,
1433 	    size_t count, loff_t *offset)
1434 {
1435 	struct inode *inode = file_inode(file);
1436 	struct task_struct *p;
1437 	char buffer[PROC_NUMBUF];
1438 	int nice;
1439 	int err;
1440 
1441 	memset(buffer, 0, sizeof(buffer));
1442 	if (count > sizeof(buffer) - 1)
1443 		count = sizeof(buffer) - 1;
1444 	if (copy_from_user(buffer, buf, count))
1445 		return -EFAULT;
1446 
1447 	err = kstrtoint(strstrip(buffer), 0, &nice);
1448 	if (err < 0)
1449 		return err;
1450 
1451 	p = get_proc_task(inode);
1452 	if (!p)
1453 		return -ESRCH;
1454 
1455 	err = proc_sched_autogroup_set_nice(p, nice);
1456 	if (err)
1457 		count = err;
1458 
1459 	put_task_struct(p);
1460 
1461 	return count;
1462 }
1463 
1464 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1465 {
1466 	int ret;
1467 
1468 	ret = single_open(filp, sched_autogroup_show, NULL);
1469 	if (!ret) {
1470 		struct seq_file *m = filp->private_data;
1471 
1472 		m->private = inode;
1473 	}
1474 	return ret;
1475 }
1476 
1477 static const struct file_operations proc_pid_sched_autogroup_operations = {
1478 	.open		= sched_autogroup_open,
1479 	.read		= seq_read,
1480 	.write		= sched_autogroup_write,
1481 	.llseek		= seq_lseek,
1482 	.release	= single_release,
1483 };
1484 
1485 #endif /* CONFIG_SCHED_AUTOGROUP */
1486 
1487 static ssize_t comm_write(struct file *file, const char __user *buf,
1488 				size_t count, loff_t *offset)
1489 {
1490 	struct inode *inode = file_inode(file);
1491 	struct task_struct *p;
1492 	char buffer[TASK_COMM_LEN];
1493 	const size_t maxlen = sizeof(buffer) - 1;
1494 
1495 	memset(buffer, 0, sizeof(buffer));
1496 	if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1497 		return -EFAULT;
1498 
1499 	p = get_proc_task(inode);
1500 	if (!p)
1501 		return -ESRCH;
1502 
1503 	if (same_thread_group(current, p))
1504 		set_task_comm(p, buffer);
1505 	else
1506 		count = -EINVAL;
1507 
1508 	put_task_struct(p);
1509 
1510 	return count;
1511 }
1512 
1513 static int comm_show(struct seq_file *m, void *v)
1514 {
1515 	struct inode *inode = m->private;
1516 	struct task_struct *p;
1517 
1518 	p = get_proc_task(inode);
1519 	if (!p)
1520 		return -ESRCH;
1521 
1522 	task_lock(p);
1523 	seq_printf(m, "%s\n", p->comm);
1524 	task_unlock(p);
1525 
1526 	put_task_struct(p);
1527 
1528 	return 0;
1529 }
1530 
1531 static int comm_open(struct inode *inode, struct file *filp)
1532 {
1533 	return single_open(filp, comm_show, inode);
1534 }
1535 
1536 static const struct file_operations proc_pid_set_comm_operations = {
1537 	.open		= comm_open,
1538 	.read		= seq_read,
1539 	.write		= comm_write,
1540 	.llseek		= seq_lseek,
1541 	.release	= single_release,
1542 };
1543 
1544 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1545 {
1546 	struct task_struct *task;
1547 	struct file *exe_file;
1548 
1549 	task = get_proc_task(d_inode(dentry));
1550 	if (!task)
1551 		return -ENOENT;
1552 	exe_file = get_task_exe_file(task);
1553 	put_task_struct(task);
1554 	if (exe_file) {
1555 		*exe_path = exe_file->f_path;
1556 		path_get(&exe_file->f_path);
1557 		fput(exe_file);
1558 		return 0;
1559 	} else
1560 		return -ENOENT;
1561 }
1562 
1563 static const char *proc_pid_get_link(struct dentry *dentry,
1564 				     struct inode *inode,
1565 				     struct delayed_call *done)
1566 {
1567 	struct path path;
1568 	int error = -EACCES;
1569 
1570 	if (!dentry)
1571 		return ERR_PTR(-ECHILD);
1572 
1573 	/* Are we allowed to snoop on the tasks file descriptors? */
1574 	if (!proc_fd_access_allowed(inode))
1575 		goto out;
1576 
1577 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1578 	if (error)
1579 		goto out;
1580 
1581 	nd_jump_link(&path);
1582 	return NULL;
1583 out:
1584 	return ERR_PTR(error);
1585 }
1586 
1587 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1588 {
1589 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1590 	char *pathname;
1591 	int len;
1592 
1593 	if (!tmp)
1594 		return -ENOMEM;
1595 
1596 	pathname = d_path(path, tmp, PAGE_SIZE);
1597 	len = PTR_ERR(pathname);
1598 	if (IS_ERR(pathname))
1599 		goto out;
1600 	len = tmp + PAGE_SIZE - 1 - pathname;
1601 
1602 	if (len > buflen)
1603 		len = buflen;
1604 	if (copy_to_user(buffer, pathname, len))
1605 		len = -EFAULT;
1606  out:
1607 	free_page((unsigned long)tmp);
1608 	return len;
1609 }
1610 
1611 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1612 {
1613 	int error = -EACCES;
1614 	struct inode *inode = d_inode(dentry);
1615 	struct path path;
1616 
1617 	/* Are we allowed to snoop on the tasks file descriptors? */
1618 	if (!proc_fd_access_allowed(inode))
1619 		goto out;
1620 
1621 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1622 	if (error)
1623 		goto out;
1624 
1625 	error = do_proc_readlink(&path, buffer, buflen);
1626 	path_put(&path);
1627 out:
1628 	return error;
1629 }
1630 
1631 const struct inode_operations proc_pid_link_inode_operations = {
1632 	.readlink	= proc_pid_readlink,
1633 	.get_link	= proc_pid_get_link,
1634 	.setattr	= proc_setattr,
1635 };
1636 
1637 
1638 /* building an inode */
1639 
1640 void task_dump_owner(struct task_struct *task, mode_t mode,
1641 		     kuid_t *ruid, kgid_t *rgid)
1642 {
1643 	/* Depending on the state of dumpable compute who should own a
1644 	 * proc file for a task.
1645 	 */
1646 	const struct cred *cred;
1647 	kuid_t uid;
1648 	kgid_t gid;
1649 
1650 	/* Default to the tasks effective ownership */
1651 	rcu_read_lock();
1652 	cred = __task_cred(task);
1653 	uid = cred->euid;
1654 	gid = cred->egid;
1655 	rcu_read_unlock();
1656 
1657 	/*
1658 	 * Before the /proc/pid/status file was created the only way to read
1659 	 * the effective uid of a /process was to stat /proc/pid.  Reading
1660 	 * /proc/pid/status is slow enough that procps and other packages
1661 	 * kept stating /proc/pid.  To keep the rules in /proc simple I have
1662 	 * made this apply to all per process world readable and executable
1663 	 * directories.
1664 	 */
1665 	if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1666 		struct mm_struct *mm;
1667 		task_lock(task);
1668 		mm = task->mm;
1669 		/* Make non-dumpable tasks owned by some root */
1670 		if (mm) {
1671 			if (get_dumpable(mm) != SUID_DUMP_USER) {
1672 				struct user_namespace *user_ns = mm->user_ns;
1673 
1674 				uid = make_kuid(user_ns, 0);
1675 				if (!uid_valid(uid))
1676 					uid = GLOBAL_ROOT_UID;
1677 
1678 				gid = make_kgid(user_ns, 0);
1679 				if (!gid_valid(gid))
1680 					gid = GLOBAL_ROOT_GID;
1681 			}
1682 		} else {
1683 			uid = GLOBAL_ROOT_UID;
1684 			gid = GLOBAL_ROOT_GID;
1685 		}
1686 		task_unlock(task);
1687 	}
1688 	*ruid = uid;
1689 	*rgid = gid;
1690 }
1691 
1692 struct inode *proc_pid_make_inode(struct super_block * sb,
1693 				  struct task_struct *task, umode_t mode)
1694 {
1695 	struct inode * inode;
1696 	struct proc_inode *ei;
1697 
1698 	/* We need a new inode */
1699 
1700 	inode = new_inode(sb);
1701 	if (!inode)
1702 		goto out;
1703 
1704 	/* Common stuff */
1705 	ei = PROC_I(inode);
1706 	inode->i_mode = mode;
1707 	inode->i_ino = get_next_ino();
1708 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1709 	inode->i_op = &proc_def_inode_operations;
1710 
1711 	/*
1712 	 * grab the reference to task.
1713 	 */
1714 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1715 	if (!ei->pid)
1716 		goto out_unlock;
1717 
1718 	task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1719 	security_task_to_inode(task, inode);
1720 
1721 out:
1722 	return inode;
1723 
1724 out_unlock:
1725 	iput(inode);
1726 	return NULL;
1727 }
1728 
1729 int pid_getattr(const struct path *path, struct kstat *stat,
1730 		u32 request_mask, unsigned int query_flags)
1731 {
1732 	struct inode *inode = d_inode(path->dentry);
1733 	struct task_struct *task;
1734 	struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1735 
1736 	generic_fillattr(inode, stat);
1737 
1738 	rcu_read_lock();
1739 	stat->uid = GLOBAL_ROOT_UID;
1740 	stat->gid = GLOBAL_ROOT_GID;
1741 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1742 	if (task) {
1743 		if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1744 			rcu_read_unlock();
1745 			/*
1746 			 * This doesn't prevent learning whether PID exists,
1747 			 * it only makes getattr() consistent with readdir().
1748 			 */
1749 			return -ENOENT;
1750 		}
1751 		task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1752 	}
1753 	rcu_read_unlock();
1754 	return 0;
1755 }
1756 
1757 /* dentry stuff */
1758 
1759 /*
1760  *	Exceptional case: normally we are not allowed to unhash a busy
1761  * directory. In this case, however, we can do it - no aliasing problems
1762  * due to the way we treat inodes.
1763  *
1764  * Rewrite the inode's ownerships here because the owning task may have
1765  * performed a setuid(), etc.
1766  *
1767  */
1768 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1769 {
1770 	struct inode *inode;
1771 	struct task_struct *task;
1772 
1773 	if (flags & LOOKUP_RCU)
1774 		return -ECHILD;
1775 
1776 	inode = d_inode(dentry);
1777 	task = get_proc_task(inode);
1778 
1779 	if (task) {
1780 		task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1781 
1782 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1783 		security_task_to_inode(task, inode);
1784 		put_task_struct(task);
1785 		return 1;
1786 	}
1787 	return 0;
1788 }
1789 
1790 static inline bool proc_inode_is_dead(struct inode *inode)
1791 {
1792 	return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1793 }
1794 
1795 int pid_delete_dentry(const struct dentry *dentry)
1796 {
1797 	/* Is the task we represent dead?
1798 	 * If so, then don't put the dentry on the lru list,
1799 	 * kill it immediately.
1800 	 */
1801 	return proc_inode_is_dead(d_inode(dentry));
1802 }
1803 
1804 const struct dentry_operations pid_dentry_operations =
1805 {
1806 	.d_revalidate	= pid_revalidate,
1807 	.d_delete	= pid_delete_dentry,
1808 };
1809 
1810 /* Lookups */
1811 
1812 /*
1813  * Fill a directory entry.
1814  *
1815  * If possible create the dcache entry and derive our inode number and
1816  * file type from dcache entry.
1817  *
1818  * Since all of the proc inode numbers are dynamically generated, the inode
1819  * numbers do not exist until the inode is cache.  This means creating the
1820  * the dcache entry in readdir is necessary to keep the inode numbers
1821  * reported by readdir in sync with the inode numbers reported
1822  * by stat.
1823  */
1824 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1825 	const char *name, int len,
1826 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1827 {
1828 	struct dentry *child, *dir = file->f_path.dentry;
1829 	struct qstr qname = QSTR_INIT(name, len);
1830 	struct inode *inode;
1831 	unsigned type;
1832 	ino_t ino;
1833 
1834 	child = d_hash_and_lookup(dir, &qname);
1835 	if (!child) {
1836 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1837 		child = d_alloc_parallel(dir, &qname, &wq);
1838 		if (IS_ERR(child))
1839 			goto end_instantiate;
1840 		if (d_in_lookup(child)) {
1841 			int err = instantiate(d_inode(dir), child, task, ptr);
1842 			d_lookup_done(child);
1843 			if (err < 0) {
1844 				dput(child);
1845 				goto end_instantiate;
1846 			}
1847 		}
1848 	}
1849 	inode = d_inode(child);
1850 	ino = inode->i_ino;
1851 	type = inode->i_mode >> 12;
1852 	dput(child);
1853 	return dir_emit(ctx, name, len, ino, type);
1854 
1855 end_instantiate:
1856 	return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1857 }
1858 
1859 /*
1860  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1861  * which represent vma start and end addresses.
1862  */
1863 static int dname_to_vma_addr(struct dentry *dentry,
1864 			     unsigned long *start, unsigned long *end)
1865 {
1866 	if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1867 		return -EINVAL;
1868 
1869 	return 0;
1870 }
1871 
1872 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1873 {
1874 	unsigned long vm_start, vm_end;
1875 	bool exact_vma_exists = false;
1876 	struct mm_struct *mm = NULL;
1877 	struct task_struct *task;
1878 	struct inode *inode;
1879 	int status = 0;
1880 
1881 	if (flags & LOOKUP_RCU)
1882 		return -ECHILD;
1883 
1884 	inode = d_inode(dentry);
1885 	task = get_proc_task(inode);
1886 	if (!task)
1887 		goto out_notask;
1888 
1889 	mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1890 	if (IS_ERR_OR_NULL(mm))
1891 		goto out;
1892 
1893 	if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1894 		down_read(&mm->mmap_sem);
1895 		exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1896 		up_read(&mm->mmap_sem);
1897 	}
1898 
1899 	mmput(mm);
1900 
1901 	if (exact_vma_exists) {
1902 		task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1903 
1904 		security_task_to_inode(task, inode);
1905 		status = 1;
1906 	}
1907 
1908 out:
1909 	put_task_struct(task);
1910 
1911 out_notask:
1912 	return status;
1913 }
1914 
1915 static const struct dentry_operations tid_map_files_dentry_operations = {
1916 	.d_revalidate	= map_files_d_revalidate,
1917 	.d_delete	= pid_delete_dentry,
1918 };
1919 
1920 static int map_files_get_link(struct dentry *dentry, struct path *path)
1921 {
1922 	unsigned long vm_start, vm_end;
1923 	struct vm_area_struct *vma;
1924 	struct task_struct *task;
1925 	struct mm_struct *mm;
1926 	int rc;
1927 
1928 	rc = -ENOENT;
1929 	task = get_proc_task(d_inode(dentry));
1930 	if (!task)
1931 		goto out;
1932 
1933 	mm = get_task_mm(task);
1934 	put_task_struct(task);
1935 	if (!mm)
1936 		goto out;
1937 
1938 	rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1939 	if (rc)
1940 		goto out_mmput;
1941 
1942 	rc = -ENOENT;
1943 	down_read(&mm->mmap_sem);
1944 	vma = find_exact_vma(mm, vm_start, vm_end);
1945 	if (vma && vma->vm_file) {
1946 		*path = vma->vm_file->f_path;
1947 		path_get(path);
1948 		rc = 0;
1949 	}
1950 	up_read(&mm->mmap_sem);
1951 
1952 out_mmput:
1953 	mmput(mm);
1954 out:
1955 	return rc;
1956 }
1957 
1958 struct map_files_info {
1959 	fmode_t		mode;
1960 	unsigned int	len;
1961 	unsigned char	name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1962 };
1963 
1964 /*
1965  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1966  * symlinks may be used to bypass permissions on ancestor directories in the
1967  * path to the file in question.
1968  */
1969 static const char *
1970 proc_map_files_get_link(struct dentry *dentry,
1971 			struct inode *inode,
1972 		        struct delayed_call *done)
1973 {
1974 	if (!capable(CAP_SYS_ADMIN))
1975 		return ERR_PTR(-EPERM);
1976 
1977 	return proc_pid_get_link(dentry, inode, done);
1978 }
1979 
1980 /*
1981  * Identical to proc_pid_link_inode_operations except for get_link()
1982  */
1983 static const struct inode_operations proc_map_files_link_inode_operations = {
1984 	.readlink	= proc_pid_readlink,
1985 	.get_link	= proc_map_files_get_link,
1986 	.setattr	= proc_setattr,
1987 };
1988 
1989 static int
1990 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1991 			   struct task_struct *task, const void *ptr)
1992 {
1993 	fmode_t mode = (fmode_t)(unsigned long)ptr;
1994 	struct proc_inode *ei;
1995 	struct inode *inode;
1996 
1997 	inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
1998 				    ((mode & FMODE_READ ) ? S_IRUSR : 0) |
1999 				    ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2000 	if (!inode)
2001 		return -ENOENT;
2002 
2003 	ei = PROC_I(inode);
2004 	ei->op.proc_get_link = map_files_get_link;
2005 
2006 	inode->i_op = &proc_map_files_link_inode_operations;
2007 	inode->i_size = 64;
2008 
2009 	d_set_d_op(dentry, &tid_map_files_dentry_operations);
2010 	d_add(dentry, inode);
2011 
2012 	return 0;
2013 }
2014 
2015 static struct dentry *proc_map_files_lookup(struct inode *dir,
2016 		struct dentry *dentry, unsigned int flags)
2017 {
2018 	unsigned long vm_start, vm_end;
2019 	struct vm_area_struct *vma;
2020 	struct task_struct *task;
2021 	int result;
2022 	struct mm_struct *mm;
2023 
2024 	result = -ENOENT;
2025 	task = get_proc_task(dir);
2026 	if (!task)
2027 		goto out;
2028 
2029 	result = -EACCES;
2030 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2031 		goto out_put_task;
2032 
2033 	result = -ENOENT;
2034 	if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2035 		goto out_put_task;
2036 
2037 	mm = get_task_mm(task);
2038 	if (!mm)
2039 		goto out_put_task;
2040 
2041 	down_read(&mm->mmap_sem);
2042 	vma = find_exact_vma(mm, vm_start, vm_end);
2043 	if (!vma)
2044 		goto out_no_vma;
2045 
2046 	if (vma->vm_file)
2047 		result = proc_map_files_instantiate(dir, dentry, task,
2048 				(void *)(unsigned long)vma->vm_file->f_mode);
2049 
2050 out_no_vma:
2051 	up_read(&mm->mmap_sem);
2052 	mmput(mm);
2053 out_put_task:
2054 	put_task_struct(task);
2055 out:
2056 	return ERR_PTR(result);
2057 }
2058 
2059 static const struct inode_operations proc_map_files_inode_operations = {
2060 	.lookup		= proc_map_files_lookup,
2061 	.permission	= proc_fd_permission,
2062 	.setattr	= proc_setattr,
2063 };
2064 
2065 static int
2066 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2067 {
2068 	struct vm_area_struct *vma;
2069 	struct task_struct *task;
2070 	struct mm_struct *mm;
2071 	unsigned long nr_files, pos, i;
2072 	struct flex_array *fa = NULL;
2073 	struct map_files_info info;
2074 	struct map_files_info *p;
2075 	int ret;
2076 
2077 	ret = -ENOENT;
2078 	task = get_proc_task(file_inode(file));
2079 	if (!task)
2080 		goto out;
2081 
2082 	ret = -EACCES;
2083 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2084 		goto out_put_task;
2085 
2086 	ret = 0;
2087 	if (!dir_emit_dots(file, ctx))
2088 		goto out_put_task;
2089 
2090 	mm = get_task_mm(task);
2091 	if (!mm)
2092 		goto out_put_task;
2093 	down_read(&mm->mmap_sem);
2094 
2095 	nr_files = 0;
2096 
2097 	/*
2098 	 * We need two passes here:
2099 	 *
2100 	 *  1) Collect vmas of mapped files with mmap_sem taken
2101 	 *  2) Release mmap_sem and instantiate entries
2102 	 *
2103 	 * otherwise we get lockdep complained, since filldir()
2104 	 * routine might require mmap_sem taken in might_fault().
2105 	 */
2106 
2107 	for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2108 		if (vma->vm_file && ++pos > ctx->pos)
2109 			nr_files++;
2110 	}
2111 
2112 	if (nr_files) {
2113 		fa = flex_array_alloc(sizeof(info), nr_files,
2114 					GFP_KERNEL);
2115 		if (!fa || flex_array_prealloc(fa, 0, nr_files,
2116 						GFP_KERNEL)) {
2117 			ret = -ENOMEM;
2118 			if (fa)
2119 				flex_array_free(fa);
2120 			up_read(&mm->mmap_sem);
2121 			mmput(mm);
2122 			goto out_put_task;
2123 		}
2124 		for (i = 0, vma = mm->mmap, pos = 2; vma;
2125 				vma = vma->vm_next) {
2126 			if (!vma->vm_file)
2127 				continue;
2128 			if (++pos <= ctx->pos)
2129 				continue;
2130 
2131 			info.mode = vma->vm_file->f_mode;
2132 			info.len = snprintf(info.name,
2133 					sizeof(info.name), "%lx-%lx",
2134 					vma->vm_start, vma->vm_end);
2135 			if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2136 				BUG();
2137 		}
2138 	}
2139 	up_read(&mm->mmap_sem);
2140 
2141 	for (i = 0; i < nr_files; i++) {
2142 		p = flex_array_get(fa, i);
2143 		if (!proc_fill_cache(file, ctx,
2144 				      p->name, p->len,
2145 				      proc_map_files_instantiate,
2146 				      task,
2147 				      (void *)(unsigned long)p->mode))
2148 			break;
2149 		ctx->pos++;
2150 	}
2151 	if (fa)
2152 		flex_array_free(fa);
2153 	mmput(mm);
2154 
2155 out_put_task:
2156 	put_task_struct(task);
2157 out:
2158 	return ret;
2159 }
2160 
2161 static const struct file_operations proc_map_files_operations = {
2162 	.read		= generic_read_dir,
2163 	.iterate_shared	= proc_map_files_readdir,
2164 	.llseek		= generic_file_llseek,
2165 };
2166 
2167 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2168 struct timers_private {
2169 	struct pid *pid;
2170 	struct task_struct *task;
2171 	struct sighand_struct *sighand;
2172 	struct pid_namespace *ns;
2173 	unsigned long flags;
2174 };
2175 
2176 static void *timers_start(struct seq_file *m, loff_t *pos)
2177 {
2178 	struct timers_private *tp = m->private;
2179 
2180 	tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2181 	if (!tp->task)
2182 		return ERR_PTR(-ESRCH);
2183 
2184 	tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2185 	if (!tp->sighand)
2186 		return ERR_PTR(-ESRCH);
2187 
2188 	return seq_list_start(&tp->task->signal->posix_timers, *pos);
2189 }
2190 
2191 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2192 {
2193 	struct timers_private *tp = m->private;
2194 	return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2195 }
2196 
2197 static void timers_stop(struct seq_file *m, void *v)
2198 {
2199 	struct timers_private *tp = m->private;
2200 
2201 	if (tp->sighand) {
2202 		unlock_task_sighand(tp->task, &tp->flags);
2203 		tp->sighand = NULL;
2204 	}
2205 
2206 	if (tp->task) {
2207 		put_task_struct(tp->task);
2208 		tp->task = NULL;
2209 	}
2210 }
2211 
2212 static int show_timer(struct seq_file *m, void *v)
2213 {
2214 	struct k_itimer *timer;
2215 	struct timers_private *tp = m->private;
2216 	int notify;
2217 	static const char * const nstr[] = {
2218 		[SIGEV_SIGNAL] = "signal",
2219 		[SIGEV_NONE] = "none",
2220 		[SIGEV_THREAD] = "thread",
2221 	};
2222 
2223 	timer = list_entry((struct list_head *)v, struct k_itimer, list);
2224 	notify = timer->it_sigev_notify;
2225 
2226 	seq_printf(m, "ID: %d\n", timer->it_id);
2227 	seq_printf(m, "signal: %d/%p\n",
2228 		   timer->sigq->info.si_signo,
2229 		   timer->sigq->info.si_value.sival_ptr);
2230 	seq_printf(m, "notify: %s/%s.%d\n",
2231 		   nstr[notify & ~SIGEV_THREAD_ID],
2232 		   (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2233 		   pid_nr_ns(timer->it_pid, tp->ns));
2234 	seq_printf(m, "ClockID: %d\n", timer->it_clock);
2235 
2236 	return 0;
2237 }
2238 
2239 static const struct seq_operations proc_timers_seq_ops = {
2240 	.start	= timers_start,
2241 	.next	= timers_next,
2242 	.stop	= timers_stop,
2243 	.show	= show_timer,
2244 };
2245 
2246 static int proc_timers_open(struct inode *inode, struct file *file)
2247 {
2248 	struct timers_private *tp;
2249 
2250 	tp = __seq_open_private(file, &proc_timers_seq_ops,
2251 			sizeof(struct timers_private));
2252 	if (!tp)
2253 		return -ENOMEM;
2254 
2255 	tp->pid = proc_pid(inode);
2256 	tp->ns = inode->i_sb->s_fs_info;
2257 	return 0;
2258 }
2259 
2260 static const struct file_operations proc_timers_operations = {
2261 	.open		= proc_timers_open,
2262 	.read		= seq_read,
2263 	.llseek		= seq_lseek,
2264 	.release	= seq_release_private,
2265 };
2266 #endif
2267 
2268 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2269 					size_t count, loff_t *offset)
2270 {
2271 	struct inode *inode = file_inode(file);
2272 	struct task_struct *p;
2273 	u64 slack_ns;
2274 	int err;
2275 
2276 	err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2277 	if (err < 0)
2278 		return err;
2279 
2280 	p = get_proc_task(inode);
2281 	if (!p)
2282 		return -ESRCH;
2283 
2284 	if (p != current) {
2285 		if (!capable(CAP_SYS_NICE)) {
2286 			count = -EPERM;
2287 			goto out;
2288 		}
2289 
2290 		err = security_task_setscheduler(p);
2291 		if (err) {
2292 			count = err;
2293 			goto out;
2294 		}
2295 	}
2296 
2297 	task_lock(p);
2298 	if (slack_ns == 0)
2299 		p->timer_slack_ns = p->default_timer_slack_ns;
2300 	else
2301 		p->timer_slack_ns = slack_ns;
2302 	task_unlock(p);
2303 
2304 out:
2305 	put_task_struct(p);
2306 
2307 	return count;
2308 }
2309 
2310 static int timerslack_ns_show(struct seq_file *m, void *v)
2311 {
2312 	struct inode *inode = m->private;
2313 	struct task_struct *p;
2314 	int err = 0;
2315 
2316 	p = get_proc_task(inode);
2317 	if (!p)
2318 		return -ESRCH;
2319 
2320 	if (p != current) {
2321 
2322 		if (!capable(CAP_SYS_NICE)) {
2323 			err = -EPERM;
2324 			goto out;
2325 		}
2326 		err = security_task_getscheduler(p);
2327 		if (err)
2328 			goto out;
2329 	}
2330 
2331 	task_lock(p);
2332 	seq_printf(m, "%llu\n", p->timer_slack_ns);
2333 	task_unlock(p);
2334 
2335 out:
2336 	put_task_struct(p);
2337 
2338 	return err;
2339 }
2340 
2341 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2342 {
2343 	return single_open(filp, timerslack_ns_show, inode);
2344 }
2345 
2346 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2347 	.open		= timerslack_ns_open,
2348 	.read		= seq_read,
2349 	.write		= timerslack_ns_write,
2350 	.llseek		= seq_lseek,
2351 	.release	= single_release,
2352 };
2353 
2354 static int proc_pident_instantiate(struct inode *dir,
2355 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2356 {
2357 	const struct pid_entry *p = ptr;
2358 	struct inode *inode;
2359 	struct proc_inode *ei;
2360 
2361 	inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2362 	if (!inode)
2363 		goto out;
2364 
2365 	ei = PROC_I(inode);
2366 	if (S_ISDIR(inode->i_mode))
2367 		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
2368 	if (p->iop)
2369 		inode->i_op = p->iop;
2370 	if (p->fop)
2371 		inode->i_fop = p->fop;
2372 	ei->op = p->op;
2373 	d_set_d_op(dentry, &pid_dentry_operations);
2374 	d_add(dentry, inode);
2375 	/* Close the race of the process dying before we return the dentry */
2376 	if (pid_revalidate(dentry, 0))
2377 		return 0;
2378 out:
2379 	return -ENOENT;
2380 }
2381 
2382 static struct dentry *proc_pident_lookup(struct inode *dir,
2383 					 struct dentry *dentry,
2384 					 const struct pid_entry *ents,
2385 					 unsigned int nents)
2386 {
2387 	int error;
2388 	struct task_struct *task = get_proc_task(dir);
2389 	const struct pid_entry *p, *last;
2390 
2391 	error = -ENOENT;
2392 
2393 	if (!task)
2394 		goto out_no_task;
2395 
2396 	/*
2397 	 * Yes, it does not scale. And it should not. Don't add
2398 	 * new entries into /proc/<tgid>/ without very good reasons.
2399 	 */
2400 	last = &ents[nents];
2401 	for (p = ents; p < last; p++) {
2402 		if (p->len != dentry->d_name.len)
2403 			continue;
2404 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2405 			break;
2406 	}
2407 	if (p >= last)
2408 		goto out;
2409 
2410 	error = proc_pident_instantiate(dir, dentry, task, p);
2411 out:
2412 	put_task_struct(task);
2413 out_no_task:
2414 	return ERR_PTR(error);
2415 }
2416 
2417 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2418 		const struct pid_entry *ents, unsigned int nents)
2419 {
2420 	struct task_struct *task = get_proc_task(file_inode(file));
2421 	const struct pid_entry *p;
2422 
2423 	if (!task)
2424 		return -ENOENT;
2425 
2426 	if (!dir_emit_dots(file, ctx))
2427 		goto out;
2428 
2429 	if (ctx->pos >= nents + 2)
2430 		goto out;
2431 
2432 	for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2433 		if (!proc_fill_cache(file, ctx, p->name, p->len,
2434 				proc_pident_instantiate, task, p))
2435 			break;
2436 		ctx->pos++;
2437 	}
2438 out:
2439 	put_task_struct(task);
2440 	return 0;
2441 }
2442 
2443 #ifdef CONFIG_SECURITY
2444 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2445 				  size_t count, loff_t *ppos)
2446 {
2447 	struct inode * inode = file_inode(file);
2448 	char *p = NULL;
2449 	ssize_t length;
2450 	struct task_struct *task = get_proc_task(inode);
2451 
2452 	if (!task)
2453 		return -ESRCH;
2454 
2455 	length = security_getprocattr(task,
2456 				      (char*)file->f_path.dentry->d_name.name,
2457 				      &p);
2458 	put_task_struct(task);
2459 	if (length > 0)
2460 		length = simple_read_from_buffer(buf, count, ppos, p, length);
2461 	kfree(p);
2462 	return length;
2463 }
2464 
2465 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2466 				   size_t count, loff_t *ppos)
2467 {
2468 	struct inode * inode = file_inode(file);
2469 	void *page;
2470 	ssize_t length;
2471 	struct task_struct *task = get_proc_task(inode);
2472 
2473 	length = -ESRCH;
2474 	if (!task)
2475 		goto out_no_task;
2476 
2477 	/* A task may only write its own attributes. */
2478 	length = -EACCES;
2479 	if (current != task)
2480 		goto out;
2481 
2482 	if (count > PAGE_SIZE)
2483 		count = PAGE_SIZE;
2484 
2485 	/* No partial writes. */
2486 	length = -EINVAL;
2487 	if (*ppos != 0)
2488 		goto out;
2489 
2490 	page = memdup_user(buf, count);
2491 	if (IS_ERR(page)) {
2492 		length = PTR_ERR(page);
2493 		goto out;
2494 	}
2495 
2496 	/* Guard against adverse ptrace interaction */
2497 	length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2498 	if (length < 0)
2499 		goto out_free;
2500 
2501 	length = security_setprocattr(file->f_path.dentry->d_name.name,
2502 				      page, count);
2503 	mutex_unlock(&current->signal->cred_guard_mutex);
2504 out_free:
2505 	kfree(page);
2506 out:
2507 	put_task_struct(task);
2508 out_no_task:
2509 	return length;
2510 }
2511 
2512 static const struct file_operations proc_pid_attr_operations = {
2513 	.read		= proc_pid_attr_read,
2514 	.write		= proc_pid_attr_write,
2515 	.llseek		= generic_file_llseek,
2516 };
2517 
2518 static const struct pid_entry attr_dir_stuff[] = {
2519 	REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2520 	REG("prev",       S_IRUGO,	   proc_pid_attr_operations),
2521 	REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2522 	REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2523 	REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2524 	REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2525 };
2526 
2527 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2528 {
2529 	return proc_pident_readdir(file, ctx,
2530 				   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2531 }
2532 
2533 static const struct file_operations proc_attr_dir_operations = {
2534 	.read		= generic_read_dir,
2535 	.iterate_shared	= proc_attr_dir_readdir,
2536 	.llseek		= generic_file_llseek,
2537 };
2538 
2539 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2540 				struct dentry *dentry, unsigned int flags)
2541 {
2542 	return proc_pident_lookup(dir, dentry,
2543 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2544 }
2545 
2546 static const struct inode_operations proc_attr_dir_inode_operations = {
2547 	.lookup		= proc_attr_dir_lookup,
2548 	.getattr	= pid_getattr,
2549 	.setattr	= proc_setattr,
2550 };
2551 
2552 #endif
2553 
2554 #ifdef CONFIG_ELF_CORE
2555 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2556 					 size_t count, loff_t *ppos)
2557 {
2558 	struct task_struct *task = get_proc_task(file_inode(file));
2559 	struct mm_struct *mm;
2560 	char buffer[PROC_NUMBUF];
2561 	size_t len;
2562 	int ret;
2563 
2564 	if (!task)
2565 		return -ESRCH;
2566 
2567 	ret = 0;
2568 	mm = get_task_mm(task);
2569 	if (mm) {
2570 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2571 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2572 				MMF_DUMP_FILTER_SHIFT));
2573 		mmput(mm);
2574 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2575 	}
2576 
2577 	put_task_struct(task);
2578 
2579 	return ret;
2580 }
2581 
2582 static ssize_t proc_coredump_filter_write(struct file *file,
2583 					  const char __user *buf,
2584 					  size_t count,
2585 					  loff_t *ppos)
2586 {
2587 	struct task_struct *task;
2588 	struct mm_struct *mm;
2589 	unsigned int val;
2590 	int ret;
2591 	int i;
2592 	unsigned long mask;
2593 
2594 	ret = kstrtouint_from_user(buf, count, 0, &val);
2595 	if (ret < 0)
2596 		return ret;
2597 
2598 	ret = -ESRCH;
2599 	task = get_proc_task(file_inode(file));
2600 	if (!task)
2601 		goto out_no_task;
2602 
2603 	mm = get_task_mm(task);
2604 	if (!mm)
2605 		goto out_no_mm;
2606 	ret = 0;
2607 
2608 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2609 		if (val & mask)
2610 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2611 		else
2612 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2613 	}
2614 
2615 	mmput(mm);
2616  out_no_mm:
2617 	put_task_struct(task);
2618  out_no_task:
2619 	if (ret < 0)
2620 		return ret;
2621 	return count;
2622 }
2623 
2624 static const struct file_operations proc_coredump_filter_operations = {
2625 	.read		= proc_coredump_filter_read,
2626 	.write		= proc_coredump_filter_write,
2627 	.llseek		= generic_file_llseek,
2628 };
2629 #endif
2630 
2631 #ifdef CONFIG_TASK_IO_ACCOUNTING
2632 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2633 {
2634 	struct task_io_accounting acct = task->ioac;
2635 	unsigned long flags;
2636 	int result;
2637 
2638 	result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2639 	if (result)
2640 		return result;
2641 
2642 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2643 		result = -EACCES;
2644 		goto out_unlock;
2645 	}
2646 
2647 	if (whole && lock_task_sighand(task, &flags)) {
2648 		struct task_struct *t = task;
2649 
2650 		task_io_accounting_add(&acct, &task->signal->ioac);
2651 		while_each_thread(task, t)
2652 			task_io_accounting_add(&acct, &t->ioac);
2653 
2654 		unlock_task_sighand(task, &flags);
2655 	}
2656 	seq_printf(m,
2657 		   "rchar: %llu\n"
2658 		   "wchar: %llu\n"
2659 		   "syscr: %llu\n"
2660 		   "syscw: %llu\n"
2661 		   "read_bytes: %llu\n"
2662 		   "write_bytes: %llu\n"
2663 		   "cancelled_write_bytes: %llu\n",
2664 		   (unsigned long long)acct.rchar,
2665 		   (unsigned long long)acct.wchar,
2666 		   (unsigned long long)acct.syscr,
2667 		   (unsigned long long)acct.syscw,
2668 		   (unsigned long long)acct.read_bytes,
2669 		   (unsigned long long)acct.write_bytes,
2670 		   (unsigned long long)acct.cancelled_write_bytes);
2671 	result = 0;
2672 
2673 out_unlock:
2674 	mutex_unlock(&task->signal->cred_guard_mutex);
2675 	return result;
2676 }
2677 
2678 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2679 				  struct pid *pid, struct task_struct *task)
2680 {
2681 	return do_io_accounting(task, m, 0);
2682 }
2683 
2684 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2685 				   struct pid *pid, struct task_struct *task)
2686 {
2687 	return do_io_accounting(task, m, 1);
2688 }
2689 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2690 
2691 #ifdef CONFIG_USER_NS
2692 static int proc_id_map_open(struct inode *inode, struct file *file,
2693 	const struct seq_operations *seq_ops)
2694 {
2695 	struct user_namespace *ns = NULL;
2696 	struct task_struct *task;
2697 	struct seq_file *seq;
2698 	int ret = -EINVAL;
2699 
2700 	task = get_proc_task(inode);
2701 	if (task) {
2702 		rcu_read_lock();
2703 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2704 		rcu_read_unlock();
2705 		put_task_struct(task);
2706 	}
2707 	if (!ns)
2708 		goto err;
2709 
2710 	ret = seq_open(file, seq_ops);
2711 	if (ret)
2712 		goto err_put_ns;
2713 
2714 	seq = file->private_data;
2715 	seq->private = ns;
2716 
2717 	return 0;
2718 err_put_ns:
2719 	put_user_ns(ns);
2720 err:
2721 	return ret;
2722 }
2723 
2724 static int proc_id_map_release(struct inode *inode, struct file *file)
2725 {
2726 	struct seq_file *seq = file->private_data;
2727 	struct user_namespace *ns = seq->private;
2728 	put_user_ns(ns);
2729 	return seq_release(inode, file);
2730 }
2731 
2732 static int proc_uid_map_open(struct inode *inode, struct file *file)
2733 {
2734 	return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2735 }
2736 
2737 static int proc_gid_map_open(struct inode *inode, struct file *file)
2738 {
2739 	return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2740 }
2741 
2742 static int proc_projid_map_open(struct inode *inode, struct file *file)
2743 {
2744 	return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2745 }
2746 
2747 static const struct file_operations proc_uid_map_operations = {
2748 	.open		= proc_uid_map_open,
2749 	.write		= proc_uid_map_write,
2750 	.read		= seq_read,
2751 	.llseek		= seq_lseek,
2752 	.release	= proc_id_map_release,
2753 };
2754 
2755 static const struct file_operations proc_gid_map_operations = {
2756 	.open		= proc_gid_map_open,
2757 	.write		= proc_gid_map_write,
2758 	.read		= seq_read,
2759 	.llseek		= seq_lseek,
2760 	.release	= proc_id_map_release,
2761 };
2762 
2763 static const struct file_operations proc_projid_map_operations = {
2764 	.open		= proc_projid_map_open,
2765 	.write		= proc_projid_map_write,
2766 	.read		= seq_read,
2767 	.llseek		= seq_lseek,
2768 	.release	= proc_id_map_release,
2769 };
2770 
2771 static int proc_setgroups_open(struct inode *inode, struct file *file)
2772 {
2773 	struct user_namespace *ns = NULL;
2774 	struct task_struct *task;
2775 	int ret;
2776 
2777 	ret = -ESRCH;
2778 	task = get_proc_task(inode);
2779 	if (task) {
2780 		rcu_read_lock();
2781 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2782 		rcu_read_unlock();
2783 		put_task_struct(task);
2784 	}
2785 	if (!ns)
2786 		goto err;
2787 
2788 	if (file->f_mode & FMODE_WRITE) {
2789 		ret = -EACCES;
2790 		if (!ns_capable(ns, CAP_SYS_ADMIN))
2791 			goto err_put_ns;
2792 	}
2793 
2794 	ret = single_open(file, &proc_setgroups_show, ns);
2795 	if (ret)
2796 		goto err_put_ns;
2797 
2798 	return 0;
2799 err_put_ns:
2800 	put_user_ns(ns);
2801 err:
2802 	return ret;
2803 }
2804 
2805 static int proc_setgroups_release(struct inode *inode, struct file *file)
2806 {
2807 	struct seq_file *seq = file->private_data;
2808 	struct user_namespace *ns = seq->private;
2809 	int ret = single_release(inode, file);
2810 	put_user_ns(ns);
2811 	return ret;
2812 }
2813 
2814 static const struct file_operations proc_setgroups_operations = {
2815 	.open		= proc_setgroups_open,
2816 	.write		= proc_setgroups_write,
2817 	.read		= seq_read,
2818 	.llseek		= seq_lseek,
2819 	.release	= proc_setgroups_release,
2820 };
2821 #endif /* CONFIG_USER_NS */
2822 
2823 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2824 				struct pid *pid, struct task_struct *task)
2825 {
2826 	int err = lock_trace(task);
2827 	if (!err) {
2828 		seq_printf(m, "%08x\n", task->personality);
2829 		unlock_trace(task);
2830 	}
2831 	return err;
2832 }
2833 
2834 #ifdef CONFIG_LIVEPATCH
2835 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2836 				struct pid *pid, struct task_struct *task)
2837 {
2838 	seq_printf(m, "%d\n", task->patch_state);
2839 	return 0;
2840 }
2841 #endif /* CONFIG_LIVEPATCH */
2842 
2843 /*
2844  * Thread groups
2845  */
2846 static const struct file_operations proc_task_operations;
2847 static const struct inode_operations proc_task_inode_operations;
2848 
2849 static const struct pid_entry tgid_base_stuff[] = {
2850 	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2851 	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2852 	DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2853 	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2854 	DIR("ns",	  S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2855 #ifdef CONFIG_NET
2856 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2857 #endif
2858 	REG("environ",    S_IRUSR, proc_environ_operations),
2859 	REG("auxv",       S_IRUSR, proc_auxv_operations),
2860 	ONE("status",     S_IRUGO, proc_pid_status),
2861 	ONE("personality", S_IRUSR, proc_pid_personality),
2862 	ONE("limits",	  S_IRUGO, proc_pid_limits),
2863 #ifdef CONFIG_SCHED_DEBUG
2864 	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2865 #endif
2866 #ifdef CONFIG_SCHED_AUTOGROUP
2867 	REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2868 #endif
2869 	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2870 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2871 	ONE("syscall",    S_IRUSR, proc_pid_syscall),
2872 #endif
2873 	REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2874 	ONE("stat",       S_IRUGO, proc_tgid_stat),
2875 	ONE("statm",      S_IRUGO, proc_pid_statm),
2876 	REG("maps",       S_IRUGO, proc_pid_maps_operations),
2877 #ifdef CONFIG_NUMA
2878 	REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2879 #endif
2880 	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2881 	LNK("cwd",        proc_cwd_link),
2882 	LNK("root",       proc_root_link),
2883 	LNK("exe",        proc_exe_link),
2884 	REG("mounts",     S_IRUGO, proc_mounts_operations),
2885 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2886 	REG("mountstats", S_IRUSR, proc_mountstats_operations),
2887 #ifdef CONFIG_PROC_PAGE_MONITOR
2888 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2889 	REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2890 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2891 #endif
2892 #ifdef CONFIG_SECURITY
2893 	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2894 #endif
2895 #ifdef CONFIG_KALLSYMS
2896 	ONE("wchan",      S_IRUGO, proc_pid_wchan),
2897 #endif
2898 #ifdef CONFIG_STACKTRACE
2899 	ONE("stack",      S_IRUSR, proc_pid_stack),
2900 #endif
2901 #ifdef CONFIG_SCHED_INFO
2902 	ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2903 #endif
2904 #ifdef CONFIG_LATENCYTOP
2905 	REG("latency",  S_IRUGO, proc_lstats_operations),
2906 #endif
2907 #ifdef CONFIG_PROC_PID_CPUSET
2908 	ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2909 #endif
2910 #ifdef CONFIG_CGROUPS
2911 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2912 #endif
2913 	ONE("oom_score",  S_IRUGO, proc_oom_score),
2914 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2915 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2916 #ifdef CONFIG_AUDITSYSCALL
2917 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2918 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2919 #endif
2920 #ifdef CONFIG_FAULT_INJECTION
2921 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2922 #endif
2923 #ifdef CONFIG_ELF_CORE
2924 	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2925 #endif
2926 #ifdef CONFIG_TASK_IO_ACCOUNTING
2927 	ONE("io",	S_IRUSR, proc_tgid_io_accounting),
2928 #endif
2929 #ifdef CONFIG_HARDWALL
2930 	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2931 #endif
2932 #ifdef CONFIG_USER_NS
2933 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2934 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2935 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2936 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2937 #endif
2938 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2939 	REG("timers",	  S_IRUGO, proc_timers_operations),
2940 #endif
2941 	REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2942 #ifdef CONFIG_LIVEPATCH
2943 	ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
2944 #endif
2945 };
2946 
2947 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2948 {
2949 	return proc_pident_readdir(file, ctx,
2950 				   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2951 }
2952 
2953 static const struct file_operations proc_tgid_base_operations = {
2954 	.read		= generic_read_dir,
2955 	.iterate_shared	= proc_tgid_base_readdir,
2956 	.llseek		= generic_file_llseek,
2957 };
2958 
2959 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2960 {
2961 	return proc_pident_lookup(dir, dentry,
2962 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2963 }
2964 
2965 static const struct inode_operations proc_tgid_base_inode_operations = {
2966 	.lookup		= proc_tgid_base_lookup,
2967 	.getattr	= pid_getattr,
2968 	.setattr	= proc_setattr,
2969 	.permission	= proc_pid_permission,
2970 };
2971 
2972 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2973 {
2974 	struct dentry *dentry, *leader, *dir;
2975 	char buf[PROC_NUMBUF];
2976 	struct qstr name;
2977 
2978 	name.name = buf;
2979 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2980 	/* no ->d_hash() rejects on procfs */
2981 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2982 	if (dentry) {
2983 		d_invalidate(dentry);
2984 		dput(dentry);
2985 	}
2986 
2987 	if (pid == tgid)
2988 		return;
2989 
2990 	name.name = buf;
2991 	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2992 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2993 	if (!leader)
2994 		goto out;
2995 
2996 	name.name = "task";
2997 	name.len = strlen(name.name);
2998 	dir = d_hash_and_lookup(leader, &name);
2999 	if (!dir)
3000 		goto out_put_leader;
3001 
3002 	name.name = buf;
3003 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
3004 	dentry = d_hash_and_lookup(dir, &name);
3005 	if (dentry) {
3006 		d_invalidate(dentry);
3007 		dput(dentry);
3008 	}
3009 
3010 	dput(dir);
3011 out_put_leader:
3012 	dput(leader);
3013 out:
3014 	return;
3015 }
3016 
3017 /**
3018  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3019  * @task: task that should be flushed.
3020  *
3021  * When flushing dentries from proc, one needs to flush them from global
3022  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3023  * in. This call is supposed to do all of this job.
3024  *
3025  * Looks in the dcache for
3026  * /proc/@pid
3027  * /proc/@tgid/task/@pid
3028  * if either directory is present flushes it and all of it'ts children
3029  * from the dcache.
3030  *
3031  * It is safe and reasonable to cache /proc entries for a task until
3032  * that task exits.  After that they just clog up the dcache with
3033  * useless entries, possibly causing useful dcache entries to be
3034  * flushed instead.  This routine is proved to flush those useless
3035  * dcache entries at process exit time.
3036  *
3037  * NOTE: This routine is just an optimization so it does not guarantee
3038  *       that no dcache entries will exist at process exit time it
3039  *       just makes it very unlikely that any will persist.
3040  */
3041 
3042 void proc_flush_task(struct task_struct *task)
3043 {
3044 	int i;
3045 	struct pid *pid, *tgid;
3046 	struct upid *upid;
3047 
3048 	pid = task_pid(task);
3049 	tgid = task_tgid(task);
3050 
3051 	for (i = 0; i <= pid->level; i++) {
3052 		upid = &pid->numbers[i];
3053 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3054 					tgid->numbers[i].nr);
3055 	}
3056 }
3057 
3058 static int proc_pid_instantiate(struct inode *dir,
3059 				   struct dentry * dentry,
3060 				   struct task_struct *task, const void *ptr)
3061 {
3062 	struct inode *inode;
3063 
3064 	inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3065 	if (!inode)
3066 		goto out;
3067 
3068 	inode->i_op = &proc_tgid_base_inode_operations;
3069 	inode->i_fop = &proc_tgid_base_operations;
3070 	inode->i_flags|=S_IMMUTABLE;
3071 
3072 	set_nlink(inode, nlink_tgid);
3073 
3074 	d_set_d_op(dentry, &pid_dentry_operations);
3075 
3076 	d_add(dentry, inode);
3077 	/* Close the race of the process dying before we return the dentry */
3078 	if (pid_revalidate(dentry, 0))
3079 		return 0;
3080 out:
3081 	return -ENOENT;
3082 }
3083 
3084 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3085 {
3086 	int result = -ENOENT;
3087 	struct task_struct *task;
3088 	unsigned tgid;
3089 	struct pid_namespace *ns;
3090 
3091 	tgid = name_to_int(&dentry->d_name);
3092 	if (tgid == ~0U)
3093 		goto out;
3094 
3095 	ns = dentry->d_sb->s_fs_info;
3096 	rcu_read_lock();
3097 	task = find_task_by_pid_ns(tgid, ns);
3098 	if (task)
3099 		get_task_struct(task);
3100 	rcu_read_unlock();
3101 	if (!task)
3102 		goto out;
3103 
3104 	result = proc_pid_instantiate(dir, dentry, task, NULL);
3105 	put_task_struct(task);
3106 out:
3107 	return ERR_PTR(result);
3108 }
3109 
3110 /*
3111  * Find the first task with tgid >= tgid
3112  *
3113  */
3114 struct tgid_iter {
3115 	unsigned int tgid;
3116 	struct task_struct *task;
3117 };
3118 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3119 {
3120 	struct pid *pid;
3121 
3122 	if (iter.task)
3123 		put_task_struct(iter.task);
3124 	rcu_read_lock();
3125 retry:
3126 	iter.task = NULL;
3127 	pid = find_ge_pid(iter.tgid, ns);
3128 	if (pid) {
3129 		iter.tgid = pid_nr_ns(pid, ns);
3130 		iter.task = pid_task(pid, PIDTYPE_PID);
3131 		/* What we to know is if the pid we have find is the
3132 		 * pid of a thread_group_leader.  Testing for task
3133 		 * being a thread_group_leader is the obvious thing
3134 		 * todo but there is a window when it fails, due to
3135 		 * the pid transfer logic in de_thread.
3136 		 *
3137 		 * So we perform the straight forward test of seeing
3138 		 * if the pid we have found is the pid of a thread
3139 		 * group leader, and don't worry if the task we have
3140 		 * found doesn't happen to be a thread group leader.
3141 		 * As we don't care in the case of readdir.
3142 		 */
3143 		if (!iter.task || !has_group_leader_pid(iter.task)) {
3144 			iter.tgid += 1;
3145 			goto retry;
3146 		}
3147 		get_task_struct(iter.task);
3148 	}
3149 	rcu_read_unlock();
3150 	return iter;
3151 }
3152 
3153 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3154 
3155 /* for the /proc/ directory itself, after non-process stuff has been done */
3156 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3157 {
3158 	struct tgid_iter iter;
3159 	struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3160 	loff_t pos = ctx->pos;
3161 
3162 	if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3163 		return 0;
3164 
3165 	if (pos == TGID_OFFSET - 2) {
3166 		struct inode *inode = d_inode(ns->proc_self);
3167 		if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3168 			return 0;
3169 		ctx->pos = pos = pos + 1;
3170 	}
3171 	if (pos == TGID_OFFSET - 1) {
3172 		struct inode *inode = d_inode(ns->proc_thread_self);
3173 		if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3174 			return 0;
3175 		ctx->pos = pos = pos + 1;
3176 	}
3177 	iter.tgid = pos - TGID_OFFSET;
3178 	iter.task = NULL;
3179 	for (iter = next_tgid(ns, iter);
3180 	     iter.task;
3181 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
3182 		char name[PROC_NUMBUF];
3183 		int len;
3184 
3185 		cond_resched();
3186 		if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3187 			continue;
3188 
3189 		len = snprintf(name, sizeof(name), "%d", iter.tgid);
3190 		ctx->pos = iter.tgid + TGID_OFFSET;
3191 		if (!proc_fill_cache(file, ctx, name, len,
3192 				     proc_pid_instantiate, iter.task, NULL)) {
3193 			put_task_struct(iter.task);
3194 			return 0;
3195 		}
3196 	}
3197 	ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3198 	return 0;
3199 }
3200 
3201 /*
3202  * proc_tid_comm_permission is a special permission function exclusively
3203  * used for the node /proc/<pid>/task/<tid>/comm.
3204  * It bypasses generic permission checks in the case where a task of the same
3205  * task group attempts to access the node.
3206  * The rationale behind this is that glibc and bionic access this node for
3207  * cross thread naming (pthread_set/getname_np(!self)). However, if
3208  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3209  * which locks out the cross thread naming implementation.
3210  * This function makes sure that the node is always accessible for members of
3211  * same thread group.
3212  */
3213 static int proc_tid_comm_permission(struct inode *inode, int mask)
3214 {
3215 	bool is_same_tgroup;
3216 	struct task_struct *task;
3217 
3218 	task = get_proc_task(inode);
3219 	if (!task)
3220 		return -ESRCH;
3221 	is_same_tgroup = same_thread_group(current, task);
3222 	put_task_struct(task);
3223 
3224 	if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3225 		/* This file (/proc/<pid>/task/<tid>/comm) can always be
3226 		 * read or written by the members of the corresponding
3227 		 * thread group.
3228 		 */
3229 		return 0;
3230 	}
3231 
3232 	return generic_permission(inode, mask);
3233 }
3234 
3235 static const struct inode_operations proc_tid_comm_inode_operations = {
3236 		.permission = proc_tid_comm_permission,
3237 };
3238 
3239 /*
3240  * Tasks
3241  */
3242 static const struct pid_entry tid_base_stuff[] = {
3243 	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3244 	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3245 	DIR("ns",	 S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3246 #ifdef CONFIG_NET
3247 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3248 #endif
3249 	REG("environ",   S_IRUSR, proc_environ_operations),
3250 	REG("auxv",      S_IRUSR, proc_auxv_operations),
3251 	ONE("status",    S_IRUGO, proc_pid_status),
3252 	ONE("personality", S_IRUSR, proc_pid_personality),
3253 	ONE("limits",	 S_IRUGO, proc_pid_limits),
3254 #ifdef CONFIG_SCHED_DEBUG
3255 	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3256 #endif
3257 	NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3258 			 &proc_tid_comm_inode_operations,
3259 			 &proc_pid_set_comm_operations, {}),
3260 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3261 	ONE("syscall",   S_IRUSR, proc_pid_syscall),
3262 #endif
3263 	REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3264 	ONE("stat",      S_IRUGO, proc_tid_stat),
3265 	ONE("statm",     S_IRUGO, proc_pid_statm),
3266 	REG("maps",      S_IRUGO, proc_tid_maps_operations),
3267 #ifdef CONFIG_PROC_CHILDREN
3268 	REG("children",  S_IRUGO, proc_tid_children_operations),
3269 #endif
3270 #ifdef CONFIG_NUMA
3271 	REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3272 #endif
3273 	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3274 	LNK("cwd",       proc_cwd_link),
3275 	LNK("root",      proc_root_link),
3276 	LNK("exe",       proc_exe_link),
3277 	REG("mounts",    S_IRUGO, proc_mounts_operations),
3278 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3279 #ifdef CONFIG_PROC_PAGE_MONITOR
3280 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3281 	REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3282 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3283 #endif
3284 #ifdef CONFIG_SECURITY
3285 	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3286 #endif
3287 #ifdef CONFIG_KALLSYMS
3288 	ONE("wchan",     S_IRUGO, proc_pid_wchan),
3289 #endif
3290 #ifdef CONFIG_STACKTRACE
3291 	ONE("stack",      S_IRUSR, proc_pid_stack),
3292 #endif
3293 #ifdef CONFIG_SCHED_INFO
3294 	ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3295 #endif
3296 #ifdef CONFIG_LATENCYTOP
3297 	REG("latency",  S_IRUGO, proc_lstats_operations),
3298 #endif
3299 #ifdef CONFIG_PROC_PID_CPUSET
3300 	ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3301 #endif
3302 #ifdef CONFIG_CGROUPS
3303 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3304 #endif
3305 	ONE("oom_score", S_IRUGO, proc_oom_score),
3306 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3307 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3308 #ifdef CONFIG_AUDITSYSCALL
3309 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3310 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3311 #endif
3312 #ifdef CONFIG_FAULT_INJECTION
3313 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3314 #endif
3315 #ifdef CONFIG_TASK_IO_ACCOUNTING
3316 	ONE("io",	S_IRUSR, proc_tid_io_accounting),
3317 #endif
3318 #ifdef CONFIG_HARDWALL
3319 	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3320 #endif
3321 #ifdef CONFIG_USER_NS
3322 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3323 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3324 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3325 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3326 #endif
3327 #ifdef CONFIG_LIVEPATCH
3328 	ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3329 #endif
3330 };
3331 
3332 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3333 {
3334 	return proc_pident_readdir(file, ctx,
3335 				   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3336 }
3337 
3338 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3339 {
3340 	return proc_pident_lookup(dir, dentry,
3341 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3342 }
3343 
3344 static const struct file_operations proc_tid_base_operations = {
3345 	.read		= generic_read_dir,
3346 	.iterate_shared	= proc_tid_base_readdir,
3347 	.llseek		= generic_file_llseek,
3348 };
3349 
3350 static const struct inode_operations proc_tid_base_inode_operations = {
3351 	.lookup		= proc_tid_base_lookup,
3352 	.getattr	= pid_getattr,
3353 	.setattr	= proc_setattr,
3354 };
3355 
3356 static int proc_task_instantiate(struct inode *dir,
3357 	struct dentry *dentry, struct task_struct *task, const void *ptr)
3358 {
3359 	struct inode *inode;
3360 	inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3361 
3362 	if (!inode)
3363 		goto out;
3364 	inode->i_op = &proc_tid_base_inode_operations;
3365 	inode->i_fop = &proc_tid_base_operations;
3366 	inode->i_flags|=S_IMMUTABLE;
3367 
3368 	set_nlink(inode, nlink_tid);
3369 
3370 	d_set_d_op(dentry, &pid_dentry_operations);
3371 
3372 	d_add(dentry, inode);
3373 	/* Close the race of the process dying before we return the dentry */
3374 	if (pid_revalidate(dentry, 0))
3375 		return 0;
3376 out:
3377 	return -ENOENT;
3378 }
3379 
3380 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3381 {
3382 	int result = -ENOENT;
3383 	struct task_struct *task;
3384 	struct task_struct *leader = get_proc_task(dir);
3385 	unsigned tid;
3386 	struct pid_namespace *ns;
3387 
3388 	if (!leader)
3389 		goto out_no_task;
3390 
3391 	tid = name_to_int(&dentry->d_name);
3392 	if (tid == ~0U)
3393 		goto out;
3394 
3395 	ns = dentry->d_sb->s_fs_info;
3396 	rcu_read_lock();
3397 	task = find_task_by_pid_ns(tid, ns);
3398 	if (task)
3399 		get_task_struct(task);
3400 	rcu_read_unlock();
3401 	if (!task)
3402 		goto out;
3403 	if (!same_thread_group(leader, task))
3404 		goto out_drop_task;
3405 
3406 	result = proc_task_instantiate(dir, dentry, task, NULL);
3407 out_drop_task:
3408 	put_task_struct(task);
3409 out:
3410 	put_task_struct(leader);
3411 out_no_task:
3412 	return ERR_PTR(result);
3413 }
3414 
3415 /*
3416  * Find the first tid of a thread group to return to user space.
3417  *
3418  * Usually this is just the thread group leader, but if the users
3419  * buffer was too small or there was a seek into the middle of the
3420  * directory we have more work todo.
3421  *
3422  * In the case of a short read we start with find_task_by_pid.
3423  *
3424  * In the case of a seek we start with the leader and walk nr
3425  * threads past it.
3426  */
3427 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3428 					struct pid_namespace *ns)
3429 {
3430 	struct task_struct *pos, *task;
3431 	unsigned long nr = f_pos;
3432 
3433 	if (nr != f_pos)	/* 32bit overflow? */
3434 		return NULL;
3435 
3436 	rcu_read_lock();
3437 	task = pid_task(pid, PIDTYPE_PID);
3438 	if (!task)
3439 		goto fail;
3440 
3441 	/* Attempt to start with the tid of a thread */
3442 	if (tid && nr) {
3443 		pos = find_task_by_pid_ns(tid, ns);
3444 		if (pos && same_thread_group(pos, task))
3445 			goto found;
3446 	}
3447 
3448 	/* If nr exceeds the number of threads there is nothing todo */
3449 	if (nr >= get_nr_threads(task))
3450 		goto fail;
3451 
3452 	/* If we haven't found our starting place yet start
3453 	 * with the leader and walk nr threads forward.
3454 	 */
3455 	pos = task = task->group_leader;
3456 	do {
3457 		if (!nr--)
3458 			goto found;
3459 	} while_each_thread(task, pos);
3460 fail:
3461 	pos = NULL;
3462 	goto out;
3463 found:
3464 	get_task_struct(pos);
3465 out:
3466 	rcu_read_unlock();
3467 	return pos;
3468 }
3469 
3470 /*
3471  * Find the next thread in the thread list.
3472  * Return NULL if there is an error or no next thread.
3473  *
3474  * The reference to the input task_struct is released.
3475  */
3476 static struct task_struct *next_tid(struct task_struct *start)
3477 {
3478 	struct task_struct *pos = NULL;
3479 	rcu_read_lock();
3480 	if (pid_alive(start)) {
3481 		pos = next_thread(start);
3482 		if (thread_group_leader(pos))
3483 			pos = NULL;
3484 		else
3485 			get_task_struct(pos);
3486 	}
3487 	rcu_read_unlock();
3488 	put_task_struct(start);
3489 	return pos;
3490 }
3491 
3492 /* for the /proc/TGID/task/ directories */
3493 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3494 {
3495 	struct inode *inode = file_inode(file);
3496 	struct task_struct *task;
3497 	struct pid_namespace *ns;
3498 	int tid;
3499 
3500 	if (proc_inode_is_dead(inode))
3501 		return -ENOENT;
3502 
3503 	if (!dir_emit_dots(file, ctx))
3504 		return 0;
3505 
3506 	/* f_version caches the tgid value that the last readdir call couldn't
3507 	 * return. lseek aka telldir automagically resets f_version to 0.
3508 	 */
3509 	ns = inode->i_sb->s_fs_info;
3510 	tid = (int)file->f_version;
3511 	file->f_version = 0;
3512 	for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3513 	     task;
3514 	     task = next_tid(task), ctx->pos++) {
3515 		char name[PROC_NUMBUF];
3516 		int len;
3517 		tid = task_pid_nr_ns(task, ns);
3518 		len = snprintf(name, sizeof(name), "%d", tid);
3519 		if (!proc_fill_cache(file, ctx, name, len,
3520 				proc_task_instantiate, task, NULL)) {
3521 			/* returning this tgid failed, save it as the first
3522 			 * pid for the next readir call */
3523 			file->f_version = (u64)tid;
3524 			put_task_struct(task);
3525 			break;
3526 		}
3527 	}
3528 
3529 	return 0;
3530 }
3531 
3532 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3533 			     u32 request_mask, unsigned int query_flags)
3534 {
3535 	struct inode *inode = d_inode(path->dentry);
3536 	struct task_struct *p = get_proc_task(inode);
3537 	generic_fillattr(inode, stat);
3538 
3539 	if (p) {
3540 		stat->nlink += get_nr_threads(p);
3541 		put_task_struct(p);
3542 	}
3543 
3544 	return 0;
3545 }
3546 
3547 static const struct inode_operations proc_task_inode_operations = {
3548 	.lookup		= proc_task_lookup,
3549 	.getattr	= proc_task_getattr,
3550 	.setattr	= proc_setattr,
3551 	.permission	= proc_pid_permission,
3552 };
3553 
3554 static const struct file_operations proc_task_operations = {
3555 	.read		= generic_read_dir,
3556 	.iterate_shared	= proc_task_readdir,
3557 	.llseek		= generic_file_llseek,
3558 };
3559 
3560 void __init set_proc_pid_nlink(void)
3561 {
3562 	nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3563 	nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3564 }
3565