xref: /openbmc/linux/fs/proc/base.c (revision 7abbaf94049914f074306d960b0f968ffe52e59f)
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49 
50 #include <asm/uaccess.h>
51 
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/printk.h>
77 #include <linux/cgroup.h>
78 #include <linux/cpuset.h>
79 #include <linux/audit.h>
80 #include <linux/poll.h>
81 #include <linux/nsproxy.h>
82 #include <linux/oom.h>
83 #include <linux/elf.h>
84 #include <linux/pid_namespace.h>
85 #include <linux/user_namespace.h>
86 #include <linux/fs_struct.h>
87 #include <linux/slab.h>
88 #include <linux/flex_array.h>
89 #include <linux/posix-timers.h>
90 #ifdef CONFIG_HARDWALL
91 #include <asm/hardwall.h>
92 #endif
93 #include <trace/events/oom.h>
94 #include "internal.h"
95 #include "fd.h"
96 
97 /* NOTE:
98  *	Implementing inode permission operations in /proc is almost
99  *	certainly an error.  Permission checks need to happen during
100  *	each system call not at open time.  The reason is that most of
101  *	what we wish to check for permissions in /proc varies at runtime.
102  *
103  *	The classic example of a problem is opening file descriptors
104  *	in /proc for a task before it execs a suid executable.
105  */
106 
107 struct pid_entry {
108 	const char *name;
109 	int len;
110 	umode_t mode;
111 	const struct inode_operations *iop;
112 	const struct file_operations *fop;
113 	union proc_op op;
114 };
115 
116 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
117 	.name = (NAME),					\
118 	.len  = sizeof(NAME) - 1,			\
119 	.mode = MODE,					\
120 	.iop  = IOP,					\
121 	.fop  = FOP,					\
122 	.op   = OP,					\
123 }
124 
125 #define DIR(NAME, MODE, iops, fops)	\
126 	NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
127 #define LNK(NAME, get_link)					\
128 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
129 		&proc_pid_link_inode_operations, NULL,		\
130 		{ .proc_get_link = get_link } )
131 #define REG(NAME, MODE, fops)				\
132 	NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
133 #define ONE(NAME, MODE, show)				\
134 	NOD(NAME, (S_IFREG|(MODE)), 			\
135 		NULL, &proc_single_file_operations,	\
136 		{ .proc_show = show } )
137 
138 /*
139  * Count the number of hardlinks for the pid_entry table, excluding the .
140  * and .. links.
141  */
142 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
143 	unsigned int n)
144 {
145 	unsigned int i;
146 	unsigned int count;
147 
148 	count = 0;
149 	for (i = 0; i < n; ++i) {
150 		if (S_ISDIR(entries[i].mode))
151 			++count;
152 	}
153 
154 	return count;
155 }
156 
157 static int get_task_root(struct task_struct *task, struct path *root)
158 {
159 	int result = -ENOENT;
160 
161 	task_lock(task);
162 	if (task->fs) {
163 		get_fs_root(task->fs, root);
164 		result = 0;
165 	}
166 	task_unlock(task);
167 	return result;
168 }
169 
170 static int proc_cwd_link(struct dentry *dentry, struct path *path)
171 {
172 	struct task_struct *task = get_proc_task(d_inode(dentry));
173 	int result = -ENOENT;
174 
175 	if (task) {
176 		task_lock(task);
177 		if (task->fs) {
178 			get_fs_pwd(task->fs, path);
179 			result = 0;
180 		}
181 		task_unlock(task);
182 		put_task_struct(task);
183 	}
184 	return result;
185 }
186 
187 static int proc_root_link(struct dentry *dentry, struct path *path)
188 {
189 	struct task_struct *task = get_proc_task(d_inode(dentry));
190 	int result = -ENOENT;
191 
192 	if (task) {
193 		result = get_task_root(task, path);
194 		put_task_struct(task);
195 	}
196 	return result;
197 }
198 
199 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
200 				     size_t _count, loff_t *pos)
201 {
202 	struct task_struct *tsk;
203 	struct mm_struct *mm;
204 	char *page;
205 	unsigned long count = _count;
206 	unsigned long arg_start, arg_end, env_start, env_end;
207 	unsigned long len1, len2, len;
208 	unsigned long p;
209 	char c;
210 	ssize_t rv;
211 
212 	BUG_ON(*pos < 0);
213 
214 	tsk = get_proc_task(file_inode(file));
215 	if (!tsk)
216 		return -ESRCH;
217 	mm = get_task_mm(tsk);
218 	put_task_struct(tsk);
219 	if (!mm)
220 		return 0;
221 	/* Check if process spawned far enough to have cmdline. */
222 	if (!mm->env_end) {
223 		rv = 0;
224 		goto out_mmput;
225 	}
226 
227 	page = (char *)__get_free_page(GFP_TEMPORARY);
228 	if (!page) {
229 		rv = -ENOMEM;
230 		goto out_mmput;
231 	}
232 
233 	down_read(&mm->mmap_sem);
234 	arg_start = mm->arg_start;
235 	arg_end = mm->arg_end;
236 	env_start = mm->env_start;
237 	env_end = mm->env_end;
238 	up_read(&mm->mmap_sem);
239 
240 	BUG_ON(arg_start > arg_end);
241 	BUG_ON(env_start > env_end);
242 
243 	len1 = arg_end - arg_start;
244 	len2 = env_end - env_start;
245 
246 	/* Empty ARGV. */
247 	if (len1 == 0) {
248 		rv = 0;
249 		goto out_free_page;
250 	}
251 	/*
252 	 * Inherently racy -- command line shares address space
253 	 * with code and data.
254 	 */
255 	rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
256 	if (rv <= 0)
257 		goto out_free_page;
258 
259 	rv = 0;
260 
261 	if (c == '\0') {
262 		/* Command line (set of strings) occupies whole ARGV. */
263 		if (len1 <= *pos)
264 			goto out_free_page;
265 
266 		p = arg_start + *pos;
267 		len = len1 - *pos;
268 		while (count > 0 && len > 0) {
269 			unsigned int _count;
270 			int nr_read;
271 
272 			_count = min3(count, len, PAGE_SIZE);
273 			nr_read = access_remote_vm(mm, p, page, _count, 0);
274 			if (nr_read < 0)
275 				rv = nr_read;
276 			if (nr_read <= 0)
277 				goto out_free_page;
278 
279 			if (copy_to_user(buf, page, nr_read)) {
280 				rv = -EFAULT;
281 				goto out_free_page;
282 			}
283 
284 			p	+= nr_read;
285 			len	-= nr_read;
286 			buf	+= nr_read;
287 			count	-= nr_read;
288 			rv	+= nr_read;
289 		}
290 	} else {
291 		/*
292 		 * Command line (1 string) occupies ARGV and maybe
293 		 * extends into ENVP.
294 		 */
295 		if (len1 + len2 <= *pos)
296 			goto skip_argv_envp;
297 		if (len1 <= *pos)
298 			goto skip_argv;
299 
300 		p = arg_start + *pos;
301 		len = len1 - *pos;
302 		while (count > 0 && len > 0) {
303 			unsigned int _count, l;
304 			int nr_read;
305 			bool final;
306 
307 			_count = min3(count, len, PAGE_SIZE);
308 			nr_read = access_remote_vm(mm, p, page, _count, 0);
309 			if (nr_read < 0)
310 				rv = nr_read;
311 			if (nr_read <= 0)
312 				goto out_free_page;
313 
314 			/*
315 			 * Command line can be shorter than whole ARGV
316 			 * even if last "marker" byte says it is not.
317 			 */
318 			final = false;
319 			l = strnlen(page, nr_read);
320 			if (l < nr_read) {
321 				nr_read = l;
322 				final = true;
323 			}
324 
325 			if (copy_to_user(buf, page, nr_read)) {
326 				rv = -EFAULT;
327 				goto out_free_page;
328 			}
329 
330 			p	+= nr_read;
331 			len	-= nr_read;
332 			buf	+= nr_read;
333 			count	-= nr_read;
334 			rv	+= nr_read;
335 
336 			if (final)
337 				goto out_free_page;
338 		}
339 skip_argv:
340 		/*
341 		 * Command line (1 string) occupies ARGV and
342 		 * extends into ENVP.
343 		 */
344 		if (len1 <= *pos) {
345 			p = env_start + *pos - len1;
346 			len = len1 + len2 - *pos;
347 		} else {
348 			p = env_start;
349 			len = len2;
350 		}
351 		while (count > 0 && len > 0) {
352 			unsigned int _count, l;
353 			int nr_read;
354 			bool final;
355 
356 			_count = min3(count, len, PAGE_SIZE);
357 			nr_read = access_remote_vm(mm, p, page, _count, 0);
358 			if (nr_read < 0)
359 				rv = nr_read;
360 			if (nr_read <= 0)
361 				goto out_free_page;
362 
363 			/* Find EOS. */
364 			final = false;
365 			l = strnlen(page, nr_read);
366 			if (l < nr_read) {
367 				nr_read = l;
368 				final = true;
369 			}
370 
371 			if (copy_to_user(buf, page, nr_read)) {
372 				rv = -EFAULT;
373 				goto out_free_page;
374 			}
375 
376 			p	+= nr_read;
377 			len	-= nr_read;
378 			buf	+= nr_read;
379 			count	-= nr_read;
380 			rv	+= nr_read;
381 
382 			if (final)
383 				goto out_free_page;
384 		}
385 skip_argv_envp:
386 		;
387 	}
388 
389 out_free_page:
390 	free_page((unsigned long)page);
391 out_mmput:
392 	mmput(mm);
393 	if (rv > 0)
394 		*pos += rv;
395 	return rv;
396 }
397 
398 static const struct file_operations proc_pid_cmdline_ops = {
399 	.read	= proc_pid_cmdline_read,
400 	.llseek	= generic_file_llseek,
401 };
402 
403 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
404 			 struct pid *pid, struct task_struct *task)
405 {
406 	struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
407 	if (mm && !IS_ERR(mm)) {
408 		unsigned int nwords = 0;
409 		do {
410 			nwords += 2;
411 		} while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
412 		seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
413 		mmput(mm);
414 		return 0;
415 	} else
416 		return PTR_ERR(mm);
417 }
418 
419 
420 #ifdef CONFIG_KALLSYMS
421 /*
422  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
423  * Returns the resolved symbol.  If that fails, simply return the address.
424  */
425 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
426 			  struct pid *pid, struct task_struct *task)
427 {
428 	unsigned long wchan;
429 	char symname[KSYM_NAME_LEN];
430 
431 	wchan = get_wchan(task);
432 
433 	if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
434 			&& !lookup_symbol_name(wchan, symname))
435 		seq_printf(m, "%s", symname);
436 	else
437 		seq_putc(m, '0');
438 
439 	return 0;
440 }
441 #endif /* CONFIG_KALLSYMS */
442 
443 static int lock_trace(struct task_struct *task)
444 {
445 	int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
446 	if (err)
447 		return err;
448 	if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
449 		mutex_unlock(&task->signal->cred_guard_mutex);
450 		return -EPERM;
451 	}
452 	return 0;
453 }
454 
455 static void unlock_trace(struct task_struct *task)
456 {
457 	mutex_unlock(&task->signal->cred_guard_mutex);
458 }
459 
460 #ifdef CONFIG_STACKTRACE
461 
462 #define MAX_STACK_TRACE_DEPTH	64
463 
464 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
465 			  struct pid *pid, struct task_struct *task)
466 {
467 	struct stack_trace trace;
468 	unsigned long *entries;
469 	int err;
470 	int i;
471 
472 	entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
473 	if (!entries)
474 		return -ENOMEM;
475 
476 	trace.nr_entries	= 0;
477 	trace.max_entries	= MAX_STACK_TRACE_DEPTH;
478 	trace.entries		= entries;
479 	trace.skip		= 0;
480 
481 	err = lock_trace(task);
482 	if (!err) {
483 		save_stack_trace_tsk(task, &trace);
484 
485 		for (i = 0; i < trace.nr_entries; i++) {
486 			seq_printf(m, "[<%pK>] %pB\n",
487 				   (void *)entries[i], (void *)entries[i]);
488 		}
489 		unlock_trace(task);
490 	}
491 	kfree(entries);
492 
493 	return err;
494 }
495 #endif
496 
497 #ifdef CONFIG_SCHED_INFO
498 /*
499  * Provides /proc/PID/schedstat
500  */
501 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
502 			      struct pid *pid, struct task_struct *task)
503 {
504 	if (unlikely(!sched_info_on()))
505 		seq_printf(m, "0 0 0\n");
506 	else
507 		seq_printf(m, "%llu %llu %lu\n",
508 		   (unsigned long long)task->se.sum_exec_runtime,
509 		   (unsigned long long)task->sched_info.run_delay,
510 		   task->sched_info.pcount);
511 
512 	return 0;
513 }
514 #endif
515 
516 #ifdef CONFIG_LATENCYTOP
517 static int lstats_show_proc(struct seq_file *m, void *v)
518 {
519 	int i;
520 	struct inode *inode = m->private;
521 	struct task_struct *task = get_proc_task(inode);
522 
523 	if (!task)
524 		return -ESRCH;
525 	seq_puts(m, "Latency Top version : v0.1\n");
526 	for (i = 0; i < 32; i++) {
527 		struct latency_record *lr = &task->latency_record[i];
528 		if (lr->backtrace[0]) {
529 			int q;
530 			seq_printf(m, "%i %li %li",
531 				   lr->count, lr->time, lr->max);
532 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
533 				unsigned long bt = lr->backtrace[q];
534 				if (!bt)
535 					break;
536 				if (bt == ULONG_MAX)
537 					break;
538 				seq_printf(m, " %ps", (void *)bt);
539 			}
540 			seq_putc(m, '\n');
541 		}
542 
543 	}
544 	put_task_struct(task);
545 	return 0;
546 }
547 
548 static int lstats_open(struct inode *inode, struct file *file)
549 {
550 	return single_open(file, lstats_show_proc, inode);
551 }
552 
553 static ssize_t lstats_write(struct file *file, const char __user *buf,
554 			    size_t count, loff_t *offs)
555 {
556 	struct task_struct *task = get_proc_task(file_inode(file));
557 
558 	if (!task)
559 		return -ESRCH;
560 	clear_all_latency_tracing(task);
561 	put_task_struct(task);
562 
563 	return count;
564 }
565 
566 static const struct file_operations proc_lstats_operations = {
567 	.open		= lstats_open,
568 	.read		= seq_read,
569 	.write		= lstats_write,
570 	.llseek		= seq_lseek,
571 	.release	= single_release,
572 };
573 
574 #endif
575 
576 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
577 			  struct pid *pid, struct task_struct *task)
578 {
579 	unsigned long totalpages = totalram_pages + total_swap_pages;
580 	unsigned long points = 0;
581 
582 	points = oom_badness(task, NULL, NULL, totalpages) *
583 					1000 / totalpages;
584 	seq_printf(m, "%lu\n", points);
585 
586 	return 0;
587 }
588 
589 struct limit_names {
590 	const char *name;
591 	const char *unit;
592 };
593 
594 static const struct limit_names lnames[RLIM_NLIMITS] = {
595 	[RLIMIT_CPU] = {"Max cpu time", "seconds"},
596 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
597 	[RLIMIT_DATA] = {"Max data size", "bytes"},
598 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
599 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
600 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
601 	[RLIMIT_NPROC] = {"Max processes", "processes"},
602 	[RLIMIT_NOFILE] = {"Max open files", "files"},
603 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
604 	[RLIMIT_AS] = {"Max address space", "bytes"},
605 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
606 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
607 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
608 	[RLIMIT_NICE] = {"Max nice priority", NULL},
609 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
610 	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
611 };
612 
613 /* Display limits for a process */
614 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
615 			   struct pid *pid, struct task_struct *task)
616 {
617 	unsigned int i;
618 	unsigned long flags;
619 
620 	struct rlimit rlim[RLIM_NLIMITS];
621 
622 	if (!lock_task_sighand(task, &flags))
623 		return 0;
624 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
625 	unlock_task_sighand(task, &flags);
626 
627 	/*
628 	 * print the file header
629 	 */
630        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
631 		  "Limit", "Soft Limit", "Hard Limit", "Units");
632 
633 	for (i = 0; i < RLIM_NLIMITS; i++) {
634 		if (rlim[i].rlim_cur == RLIM_INFINITY)
635 			seq_printf(m, "%-25s %-20s ",
636 				   lnames[i].name, "unlimited");
637 		else
638 			seq_printf(m, "%-25s %-20lu ",
639 				   lnames[i].name, rlim[i].rlim_cur);
640 
641 		if (rlim[i].rlim_max == RLIM_INFINITY)
642 			seq_printf(m, "%-20s ", "unlimited");
643 		else
644 			seq_printf(m, "%-20lu ", rlim[i].rlim_max);
645 
646 		if (lnames[i].unit)
647 			seq_printf(m, "%-10s\n", lnames[i].unit);
648 		else
649 			seq_putc(m, '\n');
650 	}
651 
652 	return 0;
653 }
654 
655 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
656 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
657 			    struct pid *pid, struct task_struct *task)
658 {
659 	long nr;
660 	unsigned long args[6], sp, pc;
661 	int res;
662 
663 	res = lock_trace(task);
664 	if (res)
665 		return res;
666 
667 	if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
668 		seq_puts(m, "running\n");
669 	else if (nr < 0)
670 		seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
671 	else
672 		seq_printf(m,
673 		       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
674 		       nr,
675 		       args[0], args[1], args[2], args[3], args[4], args[5],
676 		       sp, pc);
677 	unlock_trace(task);
678 
679 	return 0;
680 }
681 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
682 
683 /************************************************************************/
684 /*                       Here the fs part begins                        */
685 /************************************************************************/
686 
687 /* permission checks */
688 static int proc_fd_access_allowed(struct inode *inode)
689 {
690 	struct task_struct *task;
691 	int allowed = 0;
692 	/* Allow access to a task's file descriptors if it is us or we
693 	 * may use ptrace attach to the process and find out that
694 	 * information.
695 	 */
696 	task = get_proc_task(inode);
697 	if (task) {
698 		allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
699 		put_task_struct(task);
700 	}
701 	return allowed;
702 }
703 
704 int proc_setattr(struct dentry *dentry, struct iattr *attr)
705 {
706 	int error;
707 	struct inode *inode = d_inode(dentry);
708 
709 	if (attr->ia_valid & ATTR_MODE)
710 		return -EPERM;
711 
712 	error = inode_change_ok(inode, attr);
713 	if (error)
714 		return error;
715 
716 	setattr_copy(inode, attr);
717 	mark_inode_dirty(inode);
718 	return 0;
719 }
720 
721 /*
722  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
723  * or euid/egid (for hide_pid_min=2)?
724  */
725 static bool has_pid_permissions(struct pid_namespace *pid,
726 				 struct task_struct *task,
727 				 int hide_pid_min)
728 {
729 	if (pid->hide_pid < hide_pid_min)
730 		return true;
731 	if (in_group_p(pid->pid_gid))
732 		return true;
733 	return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
734 }
735 
736 
737 static int proc_pid_permission(struct inode *inode, int mask)
738 {
739 	struct pid_namespace *pid = inode->i_sb->s_fs_info;
740 	struct task_struct *task;
741 	bool has_perms;
742 
743 	task = get_proc_task(inode);
744 	if (!task)
745 		return -ESRCH;
746 	has_perms = has_pid_permissions(pid, task, 1);
747 	put_task_struct(task);
748 
749 	if (!has_perms) {
750 		if (pid->hide_pid == 2) {
751 			/*
752 			 * Let's make getdents(), stat(), and open()
753 			 * consistent with each other.  If a process
754 			 * may not stat() a file, it shouldn't be seen
755 			 * in procfs at all.
756 			 */
757 			return -ENOENT;
758 		}
759 
760 		return -EPERM;
761 	}
762 	return generic_permission(inode, mask);
763 }
764 
765 
766 
767 static const struct inode_operations proc_def_inode_operations = {
768 	.setattr	= proc_setattr,
769 };
770 
771 static int proc_single_show(struct seq_file *m, void *v)
772 {
773 	struct inode *inode = m->private;
774 	struct pid_namespace *ns;
775 	struct pid *pid;
776 	struct task_struct *task;
777 	int ret;
778 
779 	ns = inode->i_sb->s_fs_info;
780 	pid = proc_pid(inode);
781 	task = get_pid_task(pid, PIDTYPE_PID);
782 	if (!task)
783 		return -ESRCH;
784 
785 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
786 
787 	put_task_struct(task);
788 	return ret;
789 }
790 
791 static int proc_single_open(struct inode *inode, struct file *filp)
792 {
793 	return single_open(filp, proc_single_show, inode);
794 }
795 
796 static const struct file_operations proc_single_file_operations = {
797 	.open		= proc_single_open,
798 	.read		= seq_read,
799 	.llseek		= seq_lseek,
800 	.release	= single_release,
801 };
802 
803 
804 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
805 {
806 	struct task_struct *task = get_proc_task(inode);
807 	struct mm_struct *mm = ERR_PTR(-ESRCH);
808 
809 	if (task) {
810 		mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
811 		put_task_struct(task);
812 
813 		if (!IS_ERR_OR_NULL(mm)) {
814 			/* ensure this mm_struct can't be freed */
815 			atomic_inc(&mm->mm_count);
816 			/* but do not pin its memory */
817 			mmput(mm);
818 		}
819 	}
820 
821 	return mm;
822 }
823 
824 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
825 {
826 	struct mm_struct *mm = proc_mem_open(inode, mode);
827 
828 	if (IS_ERR(mm))
829 		return PTR_ERR(mm);
830 
831 	file->private_data = mm;
832 	return 0;
833 }
834 
835 static int mem_open(struct inode *inode, struct file *file)
836 {
837 	int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
838 
839 	/* OK to pass negative loff_t, we can catch out-of-range */
840 	file->f_mode |= FMODE_UNSIGNED_OFFSET;
841 
842 	return ret;
843 }
844 
845 static ssize_t mem_rw(struct file *file, char __user *buf,
846 			size_t count, loff_t *ppos, int write)
847 {
848 	struct mm_struct *mm = file->private_data;
849 	unsigned long addr = *ppos;
850 	ssize_t copied;
851 	char *page;
852 
853 	if (!mm)
854 		return 0;
855 
856 	page = (char *)__get_free_page(GFP_TEMPORARY);
857 	if (!page)
858 		return -ENOMEM;
859 
860 	copied = 0;
861 	if (!atomic_inc_not_zero(&mm->mm_users))
862 		goto free;
863 
864 	while (count > 0) {
865 		int this_len = min_t(int, count, PAGE_SIZE);
866 
867 		if (write && copy_from_user(page, buf, this_len)) {
868 			copied = -EFAULT;
869 			break;
870 		}
871 
872 		this_len = access_remote_vm(mm, addr, page, this_len, write);
873 		if (!this_len) {
874 			if (!copied)
875 				copied = -EIO;
876 			break;
877 		}
878 
879 		if (!write && copy_to_user(buf, page, this_len)) {
880 			copied = -EFAULT;
881 			break;
882 		}
883 
884 		buf += this_len;
885 		addr += this_len;
886 		copied += this_len;
887 		count -= this_len;
888 	}
889 	*ppos = addr;
890 
891 	mmput(mm);
892 free:
893 	free_page((unsigned long) page);
894 	return copied;
895 }
896 
897 static ssize_t mem_read(struct file *file, char __user *buf,
898 			size_t count, loff_t *ppos)
899 {
900 	return mem_rw(file, buf, count, ppos, 0);
901 }
902 
903 static ssize_t mem_write(struct file *file, const char __user *buf,
904 			 size_t count, loff_t *ppos)
905 {
906 	return mem_rw(file, (char __user*)buf, count, ppos, 1);
907 }
908 
909 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
910 {
911 	switch (orig) {
912 	case 0:
913 		file->f_pos = offset;
914 		break;
915 	case 1:
916 		file->f_pos += offset;
917 		break;
918 	default:
919 		return -EINVAL;
920 	}
921 	force_successful_syscall_return();
922 	return file->f_pos;
923 }
924 
925 static int mem_release(struct inode *inode, struct file *file)
926 {
927 	struct mm_struct *mm = file->private_data;
928 	if (mm)
929 		mmdrop(mm);
930 	return 0;
931 }
932 
933 static const struct file_operations proc_mem_operations = {
934 	.llseek		= mem_lseek,
935 	.read		= mem_read,
936 	.write		= mem_write,
937 	.open		= mem_open,
938 	.release	= mem_release,
939 };
940 
941 static int environ_open(struct inode *inode, struct file *file)
942 {
943 	return __mem_open(inode, file, PTRACE_MODE_READ);
944 }
945 
946 static ssize_t environ_read(struct file *file, char __user *buf,
947 			size_t count, loff_t *ppos)
948 {
949 	char *page;
950 	unsigned long src = *ppos;
951 	int ret = 0;
952 	struct mm_struct *mm = file->private_data;
953 	unsigned long env_start, env_end;
954 
955 	/* Ensure the process spawned far enough to have an environment. */
956 	if (!mm || !mm->env_end)
957 		return 0;
958 
959 	page = (char *)__get_free_page(GFP_TEMPORARY);
960 	if (!page)
961 		return -ENOMEM;
962 
963 	ret = 0;
964 	if (!atomic_inc_not_zero(&mm->mm_users))
965 		goto free;
966 
967 	down_read(&mm->mmap_sem);
968 	env_start = mm->env_start;
969 	env_end = mm->env_end;
970 	up_read(&mm->mmap_sem);
971 
972 	while (count > 0) {
973 		size_t this_len, max_len;
974 		int retval;
975 
976 		if (src >= (env_end - env_start))
977 			break;
978 
979 		this_len = env_end - (env_start + src);
980 
981 		max_len = min_t(size_t, PAGE_SIZE, count);
982 		this_len = min(max_len, this_len);
983 
984 		retval = access_remote_vm(mm, (env_start + src),
985 			page, this_len, 0);
986 
987 		if (retval <= 0) {
988 			ret = retval;
989 			break;
990 		}
991 
992 		if (copy_to_user(buf, page, retval)) {
993 			ret = -EFAULT;
994 			break;
995 		}
996 
997 		ret += retval;
998 		src += retval;
999 		buf += retval;
1000 		count -= retval;
1001 	}
1002 	*ppos = src;
1003 	mmput(mm);
1004 
1005 free:
1006 	free_page((unsigned long) page);
1007 	return ret;
1008 }
1009 
1010 static const struct file_operations proc_environ_operations = {
1011 	.open		= environ_open,
1012 	.read		= environ_read,
1013 	.llseek		= generic_file_llseek,
1014 	.release	= mem_release,
1015 };
1016 
1017 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1018 			    loff_t *ppos)
1019 {
1020 	struct task_struct *task = get_proc_task(file_inode(file));
1021 	char buffer[PROC_NUMBUF];
1022 	int oom_adj = OOM_ADJUST_MIN;
1023 	size_t len;
1024 
1025 	if (!task)
1026 		return -ESRCH;
1027 	if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1028 		oom_adj = OOM_ADJUST_MAX;
1029 	else
1030 		oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1031 			  OOM_SCORE_ADJ_MAX;
1032 	put_task_struct(task);
1033 	len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1034 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1035 }
1036 
1037 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1038 {
1039 	static DEFINE_MUTEX(oom_adj_mutex);
1040 	struct mm_struct *mm = NULL;
1041 	struct task_struct *task;
1042 	int err = 0;
1043 
1044 	task = get_proc_task(file_inode(file));
1045 	if (!task)
1046 		return -ESRCH;
1047 
1048 	mutex_lock(&oom_adj_mutex);
1049 	if (legacy) {
1050 		if (oom_adj < task->signal->oom_score_adj &&
1051 				!capable(CAP_SYS_RESOURCE)) {
1052 			err = -EACCES;
1053 			goto err_unlock;
1054 		}
1055 		/*
1056 		 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1057 		 * /proc/pid/oom_score_adj instead.
1058 		 */
1059 		pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1060 			  current->comm, task_pid_nr(current), task_pid_nr(task),
1061 			  task_pid_nr(task));
1062 	} else {
1063 		if ((short)oom_adj < task->signal->oom_score_adj_min &&
1064 				!capable(CAP_SYS_RESOURCE)) {
1065 			err = -EACCES;
1066 			goto err_unlock;
1067 		}
1068 	}
1069 
1070 	/*
1071 	 * Make sure we will check other processes sharing the mm if this is
1072 	 * not vfrok which wants its own oom_score_adj.
1073 	 * pin the mm so it doesn't go away and get reused after task_unlock
1074 	 */
1075 	if (!task->vfork_done) {
1076 		struct task_struct *p = find_lock_task_mm(task);
1077 
1078 		if (p) {
1079 			if (atomic_read(&p->mm->mm_users) > 1) {
1080 				mm = p->mm;
1081 				atomic_inc(&mm->mm_count);
1082 			}
1083 			task_unlock(p);
1084 		}
1085 	}
1086 
1087 	task->signal->oom_score_adj = oom_adj;
1088 	if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1089 		task->signal->oom_score_adj_min = (short)oom_adj;
1090 	trace_oom_score_adj_update(task);
1091 
1092 	if (mm) {
1093 		struct task_struct *p;
1094 
1095 		rcu_read_lock();
1096 		for_each_process(p) {
1097 			if (same_thread_group(task, p))
1098 				continue;
1099 
1100 			/* do not touch kernel threads or the global init */
1101 			if (p->flags & PF_KTHREAD || is_global_init(p))
1102 				continue;
1103 
1104 			task_lock(p);
1105 			if (!p->vfork_done && process_shares_mm(p, mm)) {
1106 				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",
1107 						task_pid_nr(p), p->comm,
1108 						p->signal->oom_score_adj, oom_adj,
1109 						task_pid_nr(task), task->comm);
1110 				p->signal->oom_score_adj = oom_adj;
1111 				if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1112 					p->signal->oom_score_adj_min = (short)oom_adj;
1113 			}
1114 			task_unlock(p);
1115 		}
1116 		rcu_read_unlock();
1117 		mmdrop(mm);
1118 	}
1119 err_unlock:
1120 	mutex_unlock(&oom_adj_mutex);
1121 	put_task_struct(task);
1122 	return err;
1123 }
1124 
1125 /*
1126  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1127  * kernels.  The effective policy is defined by oom_score_adj, which has a
1128  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1129  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1130  * Processes that become oom disabled via oom_adj will still be oom disabled
1131  * with this implementation.
1132  *
1133  * oom_adj cannot be removed since existing userspace binaries use it.
1134  */
1135 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1136 			     size_t count, loff_t *ppos)
1137 {
1138 	char buffer[PROC_NUMBUF];
1139 	int oom_adj;
1140 	int err;
1141 
1142 	memset(buffer, 0, sizeof(buffer));
1143 	if (count > sizeof(buffer) - 1)
1144 		count = sizeof(buffer) - 1;
1145 	if (copy_from_user(buffer, buf, count)) {
1146 		err = -EFAULT;
1147 		goto out;
1148 	}
1149 
1150 	err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1151 	if (err)
1152 		goto out;
1153 	if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1154 	     oom_adj != OOM_DISABLE) {
1155 		err = -EINVAL;
1156 		goto out;
1157 	}
1158 
1159 	/*
1160 	 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1161 	 * value is always attainable.
1162 	 */
1163 	if (oom_adj == OOM_ADJUST_MAX)
1164 		oom_adj = OOM_SCORE_ADJ_MAX;
1165 	else
1166 		oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1167 
1168 	err = __set_oom_adj(file, oom_adj, true);
1169 out:
1170 	return err < 0 ? err : count;
1171 }
1172 
1173 static const struct file_operations proc_oom_adj_operations = {
1174 	.read		= oom_adj_read,
1175 	.write		= oom_adj_write,
1176 	.llseek		= generic_file_llseek,
1177 };
1178 
1179 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1180 					size_t count, loff_t *ppos)
1181 {
1182 	struct task_struct *task = get_proc_task(file_inode(file));
1183 	char buffer[PROC_NUMBUF];
1184 	short oom_score_adj = OOM_SCORE_ADJ_MIN;
1185 	size_t len;
1186 
1187 	if (!task)
1188 		return -ESRCH;
1189 	oom_score_adj = task->signal->oom_score_adj;
1190 	put_task_struct(task);
1191 	len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1192 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1193 }
1194 
1195 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1196 					size_t count, loff_t *ppos)
1197 {
1198 	char buffer[PROC_NUMBUF];
1199 	int oom_score_adj;
1200 	int err;
1201 
1202 	memset(buffer, 0, sizeof(buffer));
1203 	if (count > sizeof(buffer) - 1)
1204 		count = sizeof(buffer) - 1;
1205 	if (copy_from_user(buffer, buf, count)) {
1206 		err = -EFAULT;
1207 		goto out;
1208 	}
1209 
1210 	err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1211 	if (err)
1212 		goto out;
1213 	if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1214 			oom_score_adj > OOM_SCORE_ADJ_MAX) {
1215 		err = -EINVAL;
1216 		goto out;
1217 	}
1218 
1219 	err = __set_oom_adj(file, oom_score_adj, false);
1220 out:
1221 	return err < 0 ? err : count;
1222 }
1223 
1224 static const struct file_operations proc_oom_score_adj_operations = {
1225 	.read		= oom_score_adj_read,
1226 	.write		= oom_score_adj_write,
1227 	.llseek		= default_llseek,
1228 };
1229 
1230 #ifdef CONFIG_AUDITSYSCALL
1231 #define TMPBUFLEN 21
1232 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1233 				  size_t count, loff_t *ppos)
1234 {
1235 	struct inode * inode = file_inode(file);
1236 	struct task_struct *task = get_proc_task(inode);
1237 	ssize_t length;
1238 	char tmpbuf[TMPBUFLEN];
1239 
1240 	if (!task)
1241 		return -ESRCH;
1242 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1243 			   from_kuid(file->f_cred->user_ns,
1244 				     audit_get_loginuid(task)));
1245 	put_task_struct(task);
1246 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1247 }
1248 
1249 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1250 				   size_t count, loff_t *ppos)
1251 {
1252 	struct inode * inode = file_inode(file);
1253 	uid_t loginuid;
1254 	kuid_t kloginuid;
1255 	int rv;
1256 
1257 	rcu_read_lock();
1258 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1259 		rcu_read_unlock();
1260 		return -EPERM;
1261 	}
1262 	rcu_read_unlock();
1263 
1264 	if (*ppos != 0) {
1265 		/* No partial writes. */
1266 		return -EINVAL;
1267 	}
1268 
1269 	rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1270 	if (rv < 0)
1271 		return rv;
1272 
1273 	/* is userspace tring to explicitly UNSET the loginuid? */
1274 	if (loginuid == AUDIT_UID_UNSET) {
1275 		kloginuid = INVALID_UID;
1276 	} else {
1277 		kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1278 		if (!uid_valid(kloginuid))
1279 			return -EINVAL;
1280 	}
1281 
1282 	rv = audit_set_loginuid(kloginuid);
1283 	if (rv < 0)
1284 		return rv;
1285 	return count;
1286 }
1287 
1288 static const struct file_operations proc_loginuid_operations = {
1289 	.read		= proc_loginuid_read,
1290 	.write		= proc_loginuid_write,
1291 	.llseek		= generic_file_llseek,
1292 };
1293 
1294 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1295 				  size_t count, loff_t *ppos)
1296 {
1297 	struct inode * inode = file_inode(file);
1298 	struct task_struct *task = get_proc_task(inode);
1299 	ssize_t length;
1300 	char tmpbuf[TMPBUFLEN];
1301 
1302 	if (!task)
1303 		return -ESRCH;
1304 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1305 				audit_get_sessionid(task));
1306 	put_task_struct(task);
1307 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1308 }
1309 
1310 static const struct file_operations proc_sessionid_operations = {
1311 	.read		= proc_sessionid_read,
1312 	.llseek		= generic_file_llseek,
1313 };
1314 #endif
1315 
1316 #ifdef CONFIG_FAULT_INJECTION
1317 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1318 				      size_t count, loff_t *ppos)
1319 {
1320 	struct task_struct *task = get_proc_task(file_inode(file));
1321 	char buffer[PROC_NUMBUF];
1322 	size_t len;
1323 	int make_it_fail;
1324 
1325 	if (!task)
1326 		return -ESRCH;
1327 	make_it_fail = task->make_it_fail;
1328 	put_task_struct(task);
1329 
1330 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1331 
1332 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1333 }
1334 
1335 static ssize_t proc_fault_inject_write(struct file * file,
1336 			const char __user * buf, size_t count, loff_t *ppos)
1337 {
1338 	struct task_struct *task;
1339 	char buffer[PROC_NUMBUF];
1340 	int make_it_fail;
1341 	int rv;
1342 
1343 	if (!capable(CAP_SYS_RESOURCE))
1344 		return -EPERM;
1345 	memset(buffer, 0, sizeof(buffer));
1346 	if (count > sizeof(buffer) - 1)
1347 		count = sizeof(buffer) - 1;
1348 	if (copy_from_user(buffer, buf, count))
1349 		return -EFAULT;
1350 	rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1351 	if (rv < 0)
1352 		return rv;
1353 	if (make_it_fail < 0 || make_it_fail > 1)
1354 		return -EINVAL;
1355 
1356 	task = get_proc_task(file_inode(file));
1357 	if (!task)
1358 		return -ESRCH;
1359 	task->make_it_fail = make_it_fail;
1360 	put_task_struct(task);
1361 
1362 	return count;
1363 }
1364 
1365 static const struct file_operations proc_fault_inject_operations = {
1366 	.read		= proc_fault_inject_read,
1367 	.write		= proc_fault_inject_write,
1368 	.llseek		= generic_file_llseek,
1369 };
1370 #endif
1371 
1372 
1373 #ifdef CONFIG_SCHED_DEBUG
1374 /*
1375  * Print out various scheduling related per-task fields:
1376  */
1377 static int sched_show(struct seq_file *m, void *v)
1378 {
1379 	struct inode *inode = m->private;
1380 	struct task_struct *p;
1381 
1382 	p = get_proc_task(inode);
1383 	if (!p)
1384 		return -ESRCH;
1385 	proc_sched_show_task(p, m);
1386 
1387 	put_task_struct(p);
1388 
1389 	return 0;
1390 }
1391 
1392 static ssize_t
1393 sched_write(struct file *file, const char __user *buf,
1394 	    size_t count, loff_t *offset)
1395 {
1396 	struct inode *inode = file_inode(file);
1397 	struct task_struct *p;
1398 
1399 	p = get_proc_task(inode);
1400 	if (!p)
1401 		return -ESRCH;
1402 	proc_sched_set_task(p);
1403 
1404 	put_task_struct(p);
1405 
1406 	return count;
1407 }
1408 
1409 static int sched_open(struct inode *inode, struct file *filp)
1410 {
1411 	return single_open(filp, sched_show, inode);
1412 }
1413 
1414 static const struct file_operations proc_pid_sched_operations = {
1415 	.open		= sched_open,
1416 	.read		= seq_read,
1417 	.write		= sched_write,
1418 	.llseek		= seq_lseek,
1419 	.release	= single_release,
1420 };
1421 
1422 #endif
1423 
1424 #ifdef CONFIG_SCHED_AUTOGROUP
1425 /*
1426  * Print out autogroup related information:
1427  */
1428 static int sched_autogroup_show(struct seq_file *m, void *v)
1429 {
1430 	struct inode *inode = m->private;
1431 	struct task_struct *p;
1432 
1433 	p = get_proc_task(inode);
1434 	if (!p)
1435 		return -ESRCH;
1436 	proc_sched_autogroup_show_task(p, m);
1437 
1438 	put_task_struct(p);
1439 
1440 	return 0;
1441 }
1442 
1443 static ssize_t
1444 sched_autogroup_write(struct file *file, const char __user *buf,
1445 	    size_t count, loff_t *offset)
1446 {
1447 	struct inode *inode = file_inode(file);
1448 	struct task_struct *p;
1449 	char buffer[PROC_NUMBUF];
1450 	int nice;
1451 	int err;
1452 
1453 	memset(buffer, 0, sizeof(buffer));
1454 	if (count > sizeof(buffer) - 1)
1455 		count = sizeof(buffer) - 1;
1456 	if (copy_from_user(buffer, buf, count))
1457 		return -EFAULT;
1458 
1459 	err = kstrtoint(strstrip(buffer), 0, &nice);
1460 	if (err < 0)
1461 		return err;
1462 
1463 	p = get_proc_task(inode);
1464 	if (!p)
1465 		return -ESRCH;
1466 
1467 	err = proc_sched_autogroup_set_nice(p, nice);
1468 	if (err)
1469 		count = err;
1470 
1471 	put_task_struct(p);
1472 
1473 	return count;
1474 }
1475 
1476 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1477 {
1478 	int ret;
1479 
1480 	ret = single_open(filp, sched_autogroup_show, NULL);
1481 	if (!ret) {
1482 		struct seq_file *m = filp->private_data;
1483 
1484 		m->private = inode;
1485 	}
1486 	return ret;
1487 }
1488 
1489 static const struct file_operations proc_pid_sched_autogroup_operations = {
1490 	.open		= sched_autogroup_open,
1491 	.read		= seq_read,
1492 	.write		= sched_autogroup_write,
1493 	.llseek		= seq_lseek,
1494 	.release	= single_release,
1495 };
1496 
1497 #endif /* CONFIG_SCHED_AUTOGROUP */
1498 
1499 static ssize_t comm_write(struct file *file, const char __user *buf,
1500 				size_t count, loff_t *offset)
1501 {
1502 	struct inode *inode = file_inode(file);
1503 	struct task_struct *p;
1504 	char buffer[TASK_COMM_LEN];
1505 	const size_t maxlen = sizeof(buffer) - 1;
1506 
1507 	memset(buffer, 0, sizeof(buffer));
1508 	if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1509 		return -EFAULT;
1510 
1511 	p = get_proc_task(inode);
1512 	if (!p)
1513 		return -ESRCH;
1514 
1515 	if (same_thread_group(current, p))
1516 		set_task_comm(p, buffer);
1517 	else
1518 		count = -EINVAL;
1519 
1520 	put_task_struct(p);
1521 
1522 	return count;
1523 }
1524 
1525 static int comm_show(struct seq_file *m, void *v)
1526 {
1527 	struct inode *inode = m->private;
1528 	struct task_struct *p;
1529 
1530 	p = get_proc_task(inode);
1531 	if (!p)
1532 		return -ESRCH;
1533 
1534 	task_lock(p);
1535 	seq_printf(m, "%s\n", p->comm);
1536 	task_unlock(p);
1537 
1538 	put_task_struct(p);
1539 
1540 	return 0;
1541 }
1542 
1543 static int comm_open(struct inode *inode, struct file *filp)
1544 {
1545 	return single_open(filp, comm_show, inode);
1546 }
1547 
1548 static const struct file_operations proc_pid_set_comm_operations = {
1549 	.open		= comm_open,
1550 	.read		= seq_read,
1551 	.write		= comm_write,
1552 	.llseek		= seq_lseek,
1553 	.release	= single_release,
1554 };
1555 
1556 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1557 {
1558 	struct task_struct *task;
1559 	struct file *exe_file;
1560 
1561 	task = get_proc_task(d_inode(dentry));
1562 	if (!task)
1563 		return -ENOENT;
1564 	exe_file = get_task_exe_file(task);
1565 	put_task_struct(task);
1566 	if (exe_file) {
1567 		*exe_path = exe_file->f_path;
1568 		path_get(&exe_file->f_path);
1569 		fput(exe_file);
1570 		return 0;
1571 	} else
1572 		return -ENOENT;
1573 }
1574 
1575 static const char *proc_pid_get_link(struct dentry *dentry,
1576 				     struct inode *inode,
1577 				     struct delayed_call *done)
1578 {
1579 	struct path path;
1580 	int error = -EACCES;
1581 
1582 	if (!dentry)
1583 		return ERR_PTR(-ECHILD);
1584 
1585 	/* Are we allowed to snoop on the tasks file descriptors? */
1586 	if (!proc_fd_access_allowed(inode))
1587 		goto out;
1588 
1589 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1590 	if (error)
1591 		goto out;
1592 
1593 	nd_jump_link(&path);
1594 	return NULL;
1595 out:
1596 	return ERR_PTR(error);
1597 }
1598 
1599 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1600 {
1601 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1602 	char *pathname;
1603 	int len;
1604 
1605 	if (!tmp)
1606 		return -ENOMEM;
1607 
1608 	pathname = d_path(path, tmp, PAGE_SIZE);
1609 	len = PTR_ERR(pathname);
1610 	if (IS_ERR(pathname))
1611 		goto out;
1612 	len = tmp + PAGE_SIZE - 1 - pathname;
1613 
1614 	if (len > buflen)
1615 		len = buflen;
1616 	if (copy_to_user(buffer, pathname, len))
1617 		len = -EFAULT;
1618  out:
1619 	free_page((unsigned long)tmp);
1620 	return len;
1621 }
1622 
1623 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1624 {
1625 	int error = -EACCES;
1626 	struct inode *inode = d_inode(dentry);
1627 	struct path path;
1628 
1629 	/* Are we allowed to snoop on the tasks file descriptors? */
1630 	if (!proc_fd_access_allowed(inode))
1631 		goto out;
1632 
1633 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1634 	if (error)
1635 		goto out;
1636 
1637 	error = do_proc_readlink(&path, buffer, buflen);
1638 	path_put(&path);
1639 out:
1640 	return error;
1641 }
1642 
1643 const struct inode_operations proc_pid_link_inode_operations = {
1644 	.readlink	= proc_pid_readlink,
1645 	.get_link	= proc_pid_get_link,
1646 	.setattr	= proc_setattr,
1647 };
1648 
1649 
1650 /* building an inode */
1651 
1652 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1653 {
1654 	struct inode * inode;
1655 	struct proc_inode *ei;
1656 	const struct cred *cred;
1657 
1658 	/* We need a new inode */
1659 
1660 	inode = new_inode(sb);
1661 	if (!inode)
1662 		goto out;
1663 
1664 	/* Common stuff */
1665 	ei = PROC_I(inode);
1666 	inode->i_ino = get_next_ino();
1667 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1668 	inode->i_op = &proc_def_inode_operations;
1669 
1670 	/*
1671 	 * grab the reference to task.
1672 	 */
1673 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1674 	if (!ei->pid)
1675 		goto out_unlock;
1676 
1677 	if (task_dumpable(task)) {
1678 		rcu_read_lock();
1679 		cred = __task_cred(task);
1680 		inode->i_uid = cred->euid;
1681 		inode->i_gid = cred->egid;
1682 		rcu_read_unlock();
1683 	}
1684 	security_task_to_inode(task, inode);
1685 
1686 out:
1687 	return inode;
1688 
1689 out_unlock:
1690 	iput(inode);
1691 	return NULL;
1692 }
1693 
1694 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1695 {
1696 	struct inode *inode = d_inode(dentry);
1697 	struct task_struct *task;
1698 	const struct cred *cred;
1699 	struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1700 
1701 	generic_fillattr(inode, stat);
1702 
1703 	rcu_read_lock();
1704 	stat->uid = GLOBAL_ROOT_UID;
1705 	stat->gid = GLOBAL_ROOT_GID;
1706 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1707 	if (task) {
1708 		if (!has_pid_permissions(pid, task, 2)) {
1709 			rcu_read_unlock();
1710 			/*
1711 			 * This doesn't prevent learning whether PID exists,
1712 			 * it only makes getattr() consistent with readdir().
1713 			 */
1714 			return -ENOENT;
1715 		}
1716 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1717 		    task_dumpable(task)) {
1718 			cred = __task_cred(task);
1719 			stat->uid = cred->euid;
1720 			stat->gid = cred->egid;
1721 		}
1722 	}
1723 	rcu_read_unlock();
1724 	return 0;
1725 }
1726 
1727 /* dentry stuff */
1728 
1729 /*
1730  *	Exceptional case: normally we are not allowed to unhash a busy
1731  * directory. In this case, however, we can do it - no aliasing problems
1732  * due to the way we treat inodes.
1733  *
1734  * Rewrite the inode's ownerships here because the owning task may have
1735  * performed a setuid(), etc.
1736  *
1737  * Before the /proc/pid/status file was created the only way to read
1738  * the effective uid of a /process was to stat /proc/pid.  Reading
1739  * /proc/pid/status is slow enough that procps and other packages
1740  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1741  * made this apply to all per process world readable and executable
1742  * directories.
1743  */
1744 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1745 {
1746 	struct inode *inode;
1747 	struct task_struct *task;
1748 	const struct cred *cred;
1749 
1750 	if (flags & LOOKUP_RCU)
1751 		return -ECHILD;
1752 
1753 	inode = d_inode(dentry);
1754 	task = get_proc_task(inode);
1755 
1756 	if (task) {
1757 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1758 		    task_dumpable(task)) {
1759 			rcu_read_lock();
1760 			cred = __task_cred(task);
1761 			inode->i_uid = cred->euid;
1762 			inode->i_gid = cred->egid;
1763 			rcu_read_unlock();
1764 		} else {
1765 			inode->i_uid = GLOBAL_ROOT_UID;
1766 			inode->i_gid = GLOBAL_ROOT_GID;
1767 		}
1768 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1769 		security_task_to_inode(task, inode);
1770 		put_task_struct(task);
1771 		return 1;
1772 	}
1773 	return 0;
1774 }
1775 
1776 static inline bool proc_inode_is_dead(struct inode *inode)
1777 {
1778 	return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1779 }
1780 
1781 int pid_delete_dentry(const struct dentry *dentry)
1782 {
1783 	/* Is the task we represent dead?
1784 	 * If so, then don't put the dentry on the lru list,
1785 	 * kill it immediately.
1786 	 */
1787 	return proc_inode_is_dead(d_inode(dentry));
1788 }
1789 
1790 const struct dentry_operations pid_dentry_operations =
1791 {
1792 	.d_revalidate	= pid_revalidate,
1793 	.d_delete	= pid_delete_dentry,
1794 };
1795 
1796 /* Lookups */
1797 
1798 /*
1799  * Fill a directory entry.
1800  *
1801  * If possible create the dcache entry and derive our inode number and
1802  * file type from dcache entry.
1803  *
1804  * Since all of the proc inode numbers are dynamically generated, the inode
1805  * numbers do not exist until the inode is cache.  This means creating the
1806  * the dcache entry in readdir is necessary to keep the inode numbers
1807  * reported by readdir in sync with the inode numbers reported
1808  * by stat.
1809  */
1810 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1811 	const char *name, int len,
1812 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1813 {
1814 	struct dentry *child, *dir = file->f_path.dentry;
1815 	struct qstr qname = QSTR_INIT(name, len);
1816 	struct inode *inode;
1817 	unsigned type;
1818 	ino_t ino;
1819 
1820 	child = d_hash_and_lookup(dir, &qname);
1821 	if (!child) {
1822 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1823 		child = d_alloc_parallel(dir, &qname, &wq);
1824 		if (IS_ERR(child))
1825 			goto end_instantiate;
1826 		if (d_in_lookup(child)) {
1827 			int err = instantiate(d_inode(dir), child, task, ptr);
1828 			d_lookup_done(child);
1829 			if (err < 0) {
1830 				dput(child);
1831 				goto end_instantiate;
1832 			}
1833 		}
1834 	}
1835 	inode = d_inode(child);
1836 	ino = inode->i_ino;
1837 	type = inode->i_mode >> 12;
1838 	dput(child);
1839 	return dir_emit(ctx, name, len, ino, type);
1840 
1841 end_instantiate:
1842 	return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1843 }
1844 
1845 /*
1846  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1847  * which represent vma start and end addresses.
1848  */
1849 static int dname_to_vma_addr(struct dentry *dentry,
1850 			     unsigned long *start, unsigned long *end)
1851 {
1852 	if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1853 		return -EINVAL;
1854 
1855 	return 0;
1856 }
1857 
1858 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1859 {
1860 	unsigned long vm_start, vm_end;
1861 	bool exact_vma_exists = false;
1862 	struct mm_struct *mm = NULL;
1863 	struct task_struct *task;
1864 	const struct cred *cred;
1865 	struct inode *inode;
1866 	int status = 0;
1867 
1868 	if (flags & LOOKUP_RCU)
1869 		return -ECHILD;
1870 
1871 	inode = d_inode(dentry);
1872 	task = get_proc_task(inode);
1873 	if (!task)
1874 		goto out_notask;
1875 
1876 	mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1877 	if (IS_ERR_OR_NULL(mm))
1878 		goto out;
1879 
1880 	if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1881 		down_read(&mm->mmap_sem);
1882 		exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1883 		up_read(&mm->mmap_sem);
1884 	}
1885 
1886 	mmput(mm);
1887 
1888 	if (exact_vma_exists) {
1889 		if (task_dumpable(task)) {
1890 			rcu_read_lock();
1891 			cred = __task_cred(task);
1892 			inode->i_uid = cred->euid;
1893 			inode->i_gid = cred->egid;
1894 			rcu_read_unlock();
1895 		} else {
1896 			inode->i_uid = GLOBAL_ROOT_UID;
1897 			inode->i_gid = GLOBAL_ROOT_GID;
1898 		}
1899 		security_task_to_inode(task, inode);
1900 		status = 1;
1901 	}
1902 
1903 out:
1904 	put_task_struct(task);
1905 
1906 out_notask:
1907 	return status;
1908 }
1909 
1910 static const struct dentry_operations tid_map_files_dentry_operations = {
1911 	.d_revalidate	= map_files_d_revalidate,
1912 	.d_delete	= pid_delete_dentry,
1913 };
1914 
1915 static int map_files_get_link(struct dentry *dentry, struct path *path)
1916 {
1917 	unsigned long vm_start, vm_end;
1918 	struct vm_area_struct *vma;
1919 	struct task_struct *task;
1920 	struct mm_struct *mm;
1921 	int rc;
1922 
1923 	rc = -ENOENT;
1924 	task = get_proc_task(d_inode(dentry));
1925 	if (!task)
1926 		goto out;
1927 
1928 	mm = get_task_mm(task);
1929 	put_task_struct(task);
1930 	if (!mm)
1931 		goto out;
1932 
1933 	rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1934 	if (rc)
1935 		goto out_mmput;
1936 
1937 	rc = -ENOENT;
1938 	down_read(&mm->mmap_sem);
1939 	vma = find_exact_vma(mm, vm_start, vm_end);
1940 	if (vma && vma->vm_file) {
1941 		*path = vma->vm_file->f_path;
1942 		path_get(path);
1943 		rc = 0;
1944 	}
1945 	up_read(&mm->mmap_sem);
1946 
1947 out_mmput:
1948 	mmput(mm);
1949 out:
1950 	return rc;
1951 }
1952 
1953 struct map_files_info {
1954 	fmode_t		mode;
1955 	unsigned long	len;
1956 	unsigned char	name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1957 };
1958 
1959 /*
1960  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1961  * symlinks may be used to bypass permissions on ancestor directories in the
1962  * path to the file in question.
1963  */
1964 static const char *
1965 proc_map_files_get_link(struct dentry *dentry,
1966 			struct inode *inode,
1967 		        struct delayed_call *done)
1968 {
1969 	if (!capable(CAP_SYS_ADMIN))
1970 		return ERR_PTR(-EPERM);
1971 
1972 	return proc_pid_get_link(dentry, inode, done);
1973 }
1974 
1975 /*
1976  * Identical to proc_pid_link_inode_operations except for get_link()
1977  */
1978 static const struct inode_operations proc_map_files_link_inode_operations = {
1979 	.readlink	= proc_pid_readlink,
1980 	.get_link	= proc_map_files_get_link,
1981 	.setattr	= proc_setattr,
1982 };
1983 
1984 static int
1985 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1986 			   struct task_struct *task, const void *ptr)
1987 {
1988 	fmode_t mode = (fmode_t)(unsigned long)ptr;
1989 	struct proc_inode *ei;
1990 	struct inode *inode;
1991 
1992 	inode = proc_pid_make_inode(dir->i_sb, task);
1993 	if (!inode)
1994 		return -ENOENT;
1995 
1996 	ei = PROC_I(inode);
1997 	ei->op.proc_get_link = map_files_get_link;
1998 
1999 	inode->i_op = &proc_map_files_link_inode_operations;
2000 	inode->i_size = 64;
2001 	inode->i_mode = S_IFLNK;
2002 
2003 	if (mode & FMODE_READ)
2004 		inode->i_mode |= S_IRUSR;
2005 	if (mode & FMODE_WRITE)
2006 		inode->i_mode |= S_IWUSR;
2007 
2008 	d_set_d_op(dentry, &tid_map_files_dentry_operations);
2009 	d_add(dentry, inode);
2010 
2011 	return 0;
2012 }
2013 
2014 static struct dentry *proc_map_files_lookup(struct inode *dir,
2015 		struct dentry *dentry, unsigned int flags)
2016 {
2017 	unsigned long vm_start, vm_end;
2018 	struct vm_area_struct *vma;
2019 	struct task_struct *task;
2020 	int result;
2021 	struct mm_struct *mm;
2022 
2023 	result = -ENOENT;
2024 	task = get_proc_task(dir);
2025 	if (!task)
2026 		goto out;
2027 
2028 	result = -EACCES;
2029 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2030 		goto out_put_task;
2031 
2032 	result = -ENOENT;
2033 	if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2034 		goto out_put_task;
2035 
2036 	mm = get_task_mm(task);
2037 	if (!mm)
2038 		goto out_put_task;
2039 
2040 	down_read(&mm->mmap_sem);
2041 	vma = find_exact_vma(mm, vm_start, vm_end);
2042 	if (!vma)
2043 		goto out_no_vma;
2044 
2045 	if (vma->vm_file)
2046 		result = proc_map_files_instantiate(dir, dentry, task,
2047 				(void *)(unsigned long)vma->vm_file->f_mode);
2048 
2049 out_no_vma:
2050 	up_read(&mm->mmap_sem);
2051 	mmput(mm);
2052 out_put_task:
2053 	put_task_struct(task);
2054 out:
2055 	return ERR_PTR(result);
2056 }
2057 
2058 static const struct inode_operations proc_map_files_inode_operations = {
2059 	.lookup		= proc_map_files_lookup,
2060 	.permission	= proc_fd_permission,
2061 	.setattr	= proc_setattr,
2062 };
2063 
2064 static int
2065 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2066 {
2067 	struct vm_area_struct *vma;
2068 	struct task_struct *task;
2069 	struct mm_struct *mm;
2070 	unsigned long nr_files, pos, i;
2071 	struct flex_array *fa = NULL;
2072 	struct map_files_info info;
2073 	struct map_files_info *p;
2074 	int ret;
2075 
2076 	ret = -ENOENT;
2077 	task = get_proc_task(file_inode(file));
2078 	if (!task)
2079 		goto out;
2080 
2081 	ret = -EACCES;
2082 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2083 		goto out_put_task;
2084 
2085 	ret = 0;
2086 	if (!dir_emit_dots(file, ctx))
2087 		goto out_put_task;
2088 
2089 	mm = get_task_mm(task);
2090 	if (!mm)
2091 		goto out_put_task;
2092 	down_read(&mm->mmap_sem);
2093 
2094 	nr_files = 0;
2095 
2096 	/*
2097 	 * We need two passes here:
2098 	 *
2099 	 *  1) Collect vmas of mapped files with mmap_sem taken
2100 	 *  2) Release mmap_sem and instantiate entries
2101 	 *
2102 	 * otherwise we get lockdep complained, since filldir()
2103 	 * routine might require mmap_sem taken in might_fault().
2104 	 */
2105 
2106 	for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2107 		if (vma->vm_file && ++pos > ctx->pos)
2108 			nr_files++;
2109 	}
2110 
2111 	if (nr_files) {
2112 		fa = flex_array_alloc(sizeof(info), nr_files,
2113 					GFP_KERNEL);
2114 		if (!fa || flex_array_prealloc(fa, 0, nr_files,
2115 						GFP_KERNEL)) {
2116 			ret = -ENOMEM;
2117 			if (fa)
2118 				flex_array_free(fa);
2119 			up_read(&mm->mmap_sem);
2120 			mmput(mm);
2121 			goto out_put_task;
2122 		}
2123 		for (i = 0, vma = mm->mmap, pos = 2; vma;
2124 				vma = vma->vm_next) {
2125 			if (!vma->vm_file)
2126 				continue;
2127 			if (++pos <= ctx->pos)
2128 				continue;
2129 
2130 			info.mode = vma->vm_file->f_mode;
2131 			info.len = snprintf(info.name,
2132 					sizeof(info.name), "%lx-%lx",
2133 					vma->vm_start, vma->vm_end);
2134 			if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2135 				BUG();
2136 		}
2137 	}
2138 	up_read(&mm->mmap_sem);
2139 
2140 	for (i = 0; i < nr_files; i++) {
2141 		p = flex_array_get(fa, i);
2142 		if (!proc_fill_cache(file, ctx,
2143 				      p->name, p->len,
2144 				      proc_map_files_instantiate,
2145 				      task,
2146 				      (void *)(unsigned long)p->mode))
2147 			break;
2148 		ctx->pos++;
2149 	}
2150 	if (fa)
2151 		flex_array_free(fa);
2152 	mmput(mm);
2153 
2154 out_put_task:
2155 	put_task_struct(task);
2156 out:
2157 	return ret;
2158 }
2159 
2160 static const struct file_operations proc_map_files_operations = {
2161 	.read		= generic_read_dir,
2162 	.iterate_shared	= proc_map_files_readdir,
2163 	.llseek		= generic_file_llseek,
2164 };
2165 
2166 #ifdef CONFIG_CHECKPOINT_RESTORE
2167 struct timers_private {
2168 	struct pid *pid;
2169 	struct task_struct *task;
2170 	struct sighand_struct *sighand;
2171 	struct pid_namespace *ns;
2172 	unsigned long flags;
2173 };
2174 
2175 static void *timers_start(struct seq_file *m, loff_t *pos)
2176 {
2177 	struct timers_private *tp = m->private;
2178 
2179 	tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2180 	if (!tp->task)
2181 		return ERR_PTR(-ESRCH);
2182 
2183 	tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2184 	if (!tp->sighand)
2185 		return ERR_PTR(-ESRCH);
2186 
2187 	return seq_list_start(&tp->task->signal->posix_timers, *pos);
2188 }
2189 
2190 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2191 {
2192 	struct timers_private *tp = m->private;
2193 	return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2194 }
2195 
2196 static void timers_stop(struct seq_file *m, void *v)
2197 {
2198 	struct timers_private *tp = m->private;
2199 
2200 	if (tp->sighand) {
2201 		unlock_task_sighand(tp->task, &tp->flags);
2202 		tp->sighand = NULL;
2203 	}
2204 
2205 	if (tp->task) {
2206 		put_task_struct(tp->task);
2207 		tp->task = NULL;
2208 	}
2209 }
2210 
2211 static int show_timer(struct seq_file *m, void *v)
2212 {
2213 	struct k_itimer *timer;
2214 	struct timers_private *tp = m->private;
2215 	int notify;
2216 	static const char * const nstr[] = {
2217 		[SIGEV_SIGNAL] = "signal",
2218 		[SIGEV_NONE] = "none",
2219 		[SIGEV_THREAD] = "thread",
2220 	};
2221 
2222 	timer = list_entry((struct list_head *)v, struct k_itimer, list);
2223 	notify = timer->it_sigev_notify;
2224 
2225 	seq_printf(m, "ID: %d\n", timer->it_id);
2226 	seq_printf(m, "signal: %d/%p\n",
2227 		   timer->sigq->info.si_signo,
2228 		   timer->sigq->info.si_value.sival_ptr);
2229 	seq_printf(m, "notify: %s/%s.%d\n",
2230 		   nstr[notify & ~SIGEV_THREAD_ID],
2231 		   (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2232 		   pid_nr_ns(timer->it_pid, tp->ns));
2233 	seq_printf(m, "ClockID: %d\n", timer->it_clock);
2234 
2235 	return 0;
2236 }
2237 
2238 static const struct seq_operations proc_timers_seq_ops = {
2239 	.start	= timers_start,
2240 	.next	= timers_next,
2241 	.stop	= timers_stop,
2242 	.show	= show_timer,
2243 };
2244 
2245 static int proc_timers_open(struct inode *inode, struct file *file)
2246 {
2247 	struct timers_private *tp;
2248 
2249 	tp = __seq_open_private(file, &proc_timers_seq_ops,
2250 			sizeof(struct timers_private));
2251 	if (!tp)
2252 		return -ENOMEM;
2253 
2254 	tp->pid = proc_pid(inode);
2255 	tp->ns = inode->i_sb->s_fs_info;
2256 	return 0;
2257 }
2258 
2259 static const struct file_operations proc_timers_operations = {
2260 	.open		= proc_timers_open,
2261 	.read		= seq_read,
2262 	.llseek		= seq_lseek,
2263 	.release	= seq_release_private,
2264 };
2265 #endif
2266 
2267 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2268 					size_t count, loff_t *offset)
2269 {
2270 	struct inode *inode = file_inode(file);
2271 	struct task_struct *p;
2272 	u64 slack_ns;
2273 	int err;
2274 
2275 	err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2276 	if (err < 0)
2277 		return err;
2278 
2279 	p = get_proc_task(inode);
2280 	if (!p)
2281 		return -ESRCH;
2282 
2283 	if (!capable(CAP_SYS_NICE)) {
2284 		count = -EPERM;
2285 		goto out;
2286 	}
2287 
2288 	task_lock(p);
2289 	if (slack_ns == 0)
2290 		p->timer_slack_ns = p->default_timer_slack_ns;
2291 	else
2292 		p->timer_slack_ns = slack_ns;
2293 	task_unlock(p);
2294 
2295 out:
2296 	put_task_struct(p);
2297 
2298 	return count;
2299 }
2300 
2301 static int timerslack_ns_show(struct seq_file *m, void *v)
2302 {
2303 	struct inode *inode = m->private;
2304 	struct task_struct *p;
2305 	int err = 0;
2306 
2307 	p = get_proc_task(inode);
2308 	if (!p)
2309 		return -ESRCH;
2310 
2311 	if (!capable(CAP_SYS_NICE)) {
2312 		err = -EPERM;
2313 		goto out;
2314 	}
2315 
2316 	task_lock(p);
2317 	seq_printf(m, "%llu\n", p->timer_slack_ns);
2318 	task_unlock(p);
2319 
2320 out:
2321 	put_task_struct(p);
2322 
2323 	return err;
2324 }
2325 
2326 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2327 {
2328 	return single_open(filp, timerslack_ns_show, inode);
2329 }
2330 
2331 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2332 	.open		= timerslack_ns_open,
2333 	.read		= seq_read,
2334 	.write		= timerslack_ns_write,
2335 	.llseek		= seq_lseek,
2336 	.release	= single_release,
2337 };
2338 
2339 static int proc_pident_instantiate(struct inode *dir,
2340 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2341 {
2342 	const struct pid_entry *p = ptr;
2343 	struct inode *inode;
2344 	struct proc_inode *ei;
2345 
2346 	inode = proc_pid_make_inode(dir->i_sb, task);
2347 	if (!inode)
2348 		goto out;
2349 
2350 	ei = PROC_I(inode);
2351 	inode->i_mode = p->mode;
2352 	if (S_ISDIR(inode->i_mode))
2353 		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
2354 	if (p->iop)
2355 		inode->i_op = p->iop;
2356 	if (p->fop)
2357 		inode->i_fop = p->fop;
2358 	ei->op = p->op;
2359 	d_set_d_op(dentry, &pid_dentry_operations);
2360 	d_add(dentry, inode);
2361 	/* Close the race of the process dying before we return the dentry */
2362 	if (pid_revalidate(dentry, 0))
2363 		return 0;
2364 out:
2365 	return -ENOENT;
2366 }
2367 
2368 static struct dentry *proc_pident_lookup(struct inode *dir,
2369 					 struct dentry *dentry,
2370 					 const struct pid_entry *ents,
2371 					 unsigned int nents)
2372 {
2373 	int error;
2374 	struct task_struct *task = get_proc_task(dir);
2375 	const struct pid_entry *p, *last;
2376 
2377 	error = -ENOENT;
2378 
2379 	if (!task)
2380 		goto out_no_task;
2381 
2382 	/*
2383 	 * Yes, it does not scale. And it should not. Don't add
2384 	 * new entries into /proc/<tgid>/ without very good reasons.
2385 	 */
2386 	last = &ents[nents - 1];
2387 	for (p = ents; p <= last; p++) {
2388 		if (p->len != dentry->d_name.len)
2389 			continue;
2390 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2391 			break;
2392 	}
2393 	if (p > last)
2394 		goto out;
2395 
2396 	error = proc_pident_instantiate(dir, dentry, task, p);
2397 out:
2398 	put_task_struct(task);
2399 out_no_task:
2400 	return ERR_PTR(error);
2401 }
2402 
2403 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2404 		const struct pid_entry *ents, unsigned int nents)
2405 {
2406 	struct task_struct *task = get_proc_task(file_inode(file));
2407 	const struct pid_entry *p;
2408 
2409 	if (!task)
2410 		return -ENOENT;
2411 
2412 	if (!dir_emit_dots(file, ctx))
2413 		goto out;
2414 
2415 	if (ctx->pos >= nents + 2)
2416 		goto out;
2417 
2418 	for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2419 		if (!proc_fill_cache(file, ctx, p->name, p->len,
2420 				proc_pident_instantiate, task, p))
2421 			break;
2422 		ctx->pos++;
2423 	}
2424 out:
2425 	put_task_struct(task);
2426 	return 0;
2427 }
2428 
2429 #ifdef CONFIG_SECURITY
2430 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2431 				  size_t count, loff_t *ppos)
2432 {
2433 	struct inode * inode = file_inode(file);
2434 	char *p = NULL;
2435 	ssize_t length;
2436 	struct task_struct *task = get_proc_task(inode);
2437 
2438 	if (!task)
2439 		return -ESRCH;
2440 
2441 	length = security_getprocattr(task,
2442 				      (char*)file->f_path.dentry->d_name.name,
2443 				      &p);
2444 	put_task_struct(task);
2445 	if (length > 0)
2446 		length = simple_read_from_buffer(buf, count, ppos, p, length);
2447 	kfree(p);
2448 	return length;
2449 }
2450 
2451 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2452 				   size_t count, loff_t *ppos)
2453 {
2454 	struct inode * inode = file_inode(file);
2455 	void *page;
2456 	ssize_t length;
2457 	struct task_struct *task = get_proc_task(inode);
2458 
2459 	length = -ESRCH;
2460 	if (!task)
2461 		goto out_no_task;
2462 	if (count > PAGE_SIZE)
2463 		count = PAGE_SIZE;
2464 
2465 	/* No partial writes. */
2466 	length = -EINVAL;
2467 	if (*ppos != 0)
2468 		goto out;
2469 
2470 	page = memdup_user(buf, count);
2471 	if (IS_ERR(page)) {
2472 		length = PTR_ERR(page);
2473 		goto out;
2474 	}
2475 
2476 	/* Guard against adverse ptrace interaction */
2477 	length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2478 	if (length < 0)
2479 		goto out_free;
2480 
2481 	length = security_setprocattr(task,
2482 				      (char*)file->f_path.dentry->d_name.name,
2483 				      page, count);
2484 	mutex_unlock(&task->signal->cred_guard_mutex);
2485 out_free:
2486 	kfree(page);
2487 out:
2488 	put_task_struct(task);
2489 out_no_task:
2490 	return length;
2491 }
2492 
2493 static const struct file_operations proc_pid_attr_operations = {
2494 	.read		= proc_pid_attr_read,
2495 	.write		= proc_pid_attr_write,
2496 	.llseek		= generic_file_llseek,
2497 };
2498 
2499 static const struct pid_entry attr_dir_stuff[] = {
2500 	REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2501 	REG("prev",       S_IRUGO,	   proc_pid_attr_operations),
2502 	REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2503 	REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2504 	REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2505 	REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2506 };
2507 
2508 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2509 {
2510 	return proc_pident_readdir(file, ctx,
2511 				   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2512 }
2513 
2514 static const struct file_operations proc_attr_dir_operations = {
2515 	.read		= generic_read_dir,
2516 	.iterate_shared	= proc_attr_dir_readdir,
2517 	.llseek		= generic_file_llseek,
2518 };
2519 
2520 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2521 				struct dentry *dentry, unsigned int flags)
2522 {
2523 	return proc_pident_lookup(dir, dentry,
2524 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2525 }
2526 
2527 static const struct inode_operations proc_attr_dir_inode_operations = {
2528 	.lookup		= proc_attr_dir_lookup,
2529 	.getattr	= pid_getattr,
2530 	.setattr	= proc_setattr,
2531 };
2532 
2533 #endif
2534 
2535 #ifdef CONFIG_ELF_CORE
2536 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2537 					 size_t count, loff_t *ppos)
2538 {
2539 	struct task_struct *task = get_proc_task(file_inode(file));
2540 	struct mm_struct *mm;
2541 	char buffer[PROC_NUMBUF];
2542 	size_t len;
2543 	int ret;
2544 
2545 	if (!task)
2546 		return -ESRCH;
2547 
2548 	ret = 0;
2549 	mm = get_task_mm(task);
2550 	if (mm) {
2551 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2552 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2553 				MMF_DUMP_FILTER_SHIFT));
2554 		mmput(mm);
2555 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2556 	}
2557 
2558 	put_task_struct(task);
2559 
2560 	return ret;
2561 }
2562 
2563 static ssize_t proc_coredump_filter_write(struct file *file,
2564 					  const char __user *buf,
2565 					  size_t count,
2566 					  loff_t *ppos)
2567 {
2568 	struct task_struct *task;
2569 	struct mm_struct *mm;
2570 	unsigned int val;
2571 	int ret;
2572 	int i;
2573 	unsigned long mask;
2574 
2575 	ret = kstrtouint_from_user(buf, count, 0, &val);
2576 	if (ret < 0)
2577 		return ret;
2578 
2579 	ret = -ESRCH;
2580 	task = get_proc_task(file_inode(file));
2581 	if (!task)
2582 		goto out_no_task;
2583 
2584 	mm = get_task_mm(task);
2585 	if (!mm)
2586 		goto out_no_mm;
2587 	ret = 0;
2588 
2589 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2590 		if (val & mask)
2591 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2592 		else
2593 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2594 	}
2595 
2596 	mmput(mm);
2597  out_no_mm:
2598 	put_task_struct(task);
2599  out_no_task:
2600 	if (ret < 0)
2601 		return ret;
2602 	return count;
2603 }
2604 
2605 static const struct file_operations proc_coredump_filter_operations = {
2606 	.read		= proc_coredump_filter_read,
2607 	.write		= proc_coredump_filter_write,
2608 	.llseek		= generic_file_llseek,
2609 };
2610 #endif
2611 
2612 #ifdef CONFIG_TASK_IO_ACCOUNTING
2613 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2614 {
2615 	struct task_io_accounting acct = task->ioac;
2616 	unsigned long flags;
2617 	int result;
2618 
2619 	result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2620 	if (result)
2621 		return result;
2622 
2623 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2624 		result = -EACCES;
2625 		goto out_unlock;
2626 	}
2627 
2628 	if (whole && lock_task_sighand(task, &flags)) {
2629 		struct task_struct *t = task;
2630 
2631 		task_io_accounting_add(&acct, &task->signal->ioac);
2632 		while_each_thread(task, t)
2633 			task_io_accounting_add(&acct, &t->ioac);
2634 
2635 		unlock_task_sighand(task, &flags);
2636 	}
2637 	seq_printf(m,
2638 		   "rchar: %llu\n"
2639 		   "wchar: %llu\n"
2640 		   "syscr: %llu\n"
2641 		   "syscw: %llu\n"
2642 		   "read_bytes: %llu\n"
2643 		   "write_bytes: %llu\n"
2644 		   "cancelled_write_bytes: %llu\n",
2645 		   (unsigned long long)acct.rchar,
2646 		   (unsigned long long)acct.wchar,
2647 		   (unsigned long long)acct.syscr,
2648 		   (unsigned long long)acct.syscw,
2649 		   (unsigned long long)acct.read_bytes,
2650 		   (unsigned long long)acct.write_bytes,
2651 		   (unsigned long long)acct.cancelled_write_bytes);
2652 	result = 0;
2653 
2654 out_unlock:
2655 	mutex_unlock(&task->signal->cred_guard_mutex);
2656 	return result;
2657 }
2658 
2659 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2660 				  struct pid *pid, struct task_struct *task)
2661 {
2662 	return do_io_accounting(task, m, 0);
2663 }
2664 
2665 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2666 				   struct pid *pid, struct task_struct *task)
2667 {
2668 	return do_io_accounting(task, m, 1);
2669 }
2670 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2671 
2672 #ifdef CONFIG_USER_NS
2673 static int proc_id_map_open(struct inode *inode, struct file *file,
2674 	const struct seq_operations *seq_ops)
2675 {
2676 	struct user_namespace *ns = NULL;
2677 	struct task_struct *task;
2678 	struct seq_file *seq;
2679 	int ret = -EINVAL;
2680 
2681 	task = get_proc_task(inode);
2682 	if (task) {
2683 		rcu_read_lock();
2684 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2685 		rcu_read_unlock();
2686 		put_task_struct(task);
2687 	}
2688 	if (!ns)
2689 		goto err;
2690 
2691 	ret = seq_open(file, seq_ops);
2692 	if (ret)
2693 		goto err_put_ns;
2694 
2695 	seq = file->private_data;
2696 	seq->private = ns;
2697 
2698 	return 0;
2699 err_put_ns:
2700 	put_user_ns(ns);
2701 err:
2702 	return ret;
2703 }
2704 
2705 static int proc_id_map_release(struct inode *inode, struct file *file)
2706 {
2707 	struct seq_file *seq = file->private_data;
2708 	struct user_namespace *ns = seq->private;
2709 	put_user_ns(ns);
2710 	return seq_release(inode, file);
2711 }
2712 
2713 static int proc_uid_map_open(struct inode *inode, struct file *file)
2714 {
2715 	return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2716 }
2717 
2718 static int proc_gid_map_open(struct inode *inode, struct file *file)
2719 {
2720 	return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2721 }
2722 
2723 static int proc_projid_map_open(struct inode *inode, struct file *file)
2724 {
2725 	return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2726 }
2727 
2728 static const struct file_operations proc_uid_map_operations = {
2729 	.open		= proc_uid_map_open,
2730 	.write		= proc_uid_map_write,
2731 	.read		= seq_read,
2732 	.llseek		= seq_lseek,
2733 	.release	= proc_id_map_release,
2734 };
2735 
2736 static const struct file_operations proc_gid_map_operations = {
2737 	.open		= proc_gid_map_open,
2738 	.write		= proc_gid_map_write,
2739 	.read		= seq_read,
2740 	.llseek		= seq_lseek,
2741 	.release	= proc_id_map_release,
2742 };
2743 
2744 static const struct file_operations proc_projid_map_operations = {
2745 	.open		= proc_projid_map_open,
2746 	.write		= proc_projid_map_write,
2747 	.read		= seq_read,
2748 	.llseek		= seq_lseek,
2749 	.release	= proc_id_map_release,
2750 };
2751 
2752 static int proc_setgroups_open(struct inode *inode, struct file *file)
2753 {
2754 	struct user_namespace *ns = NULL;
2755 	struct task_struct *task;
2756 	int ret;
2757 
2758 	ret = -ESRCH;
2759 	task = get_proc_task(inode);
2760 	if (task) {
2761 		rcu_read_lock();
2762 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2763 		rcu_read_unlock();
2764 		put_task_struct(task);
2765 	}
2766 	if (!ns)
2767 		goto err;
2768 
2769 	if (file->f_mode & FMODE_WRITE) {
2770 		ret = -EACCES;
2771 		if (!ns_capable(ns, CAP_SYS_ADMIN))
2772 			goto err_put_ns;
2773 	}
2774 
2775 	ret = single_open(file, &proc_setgroups_show, ns);
2776 	if (ret)
2777 		goto err_put_ns;
2778 
2779 	return 0;
2780 err_put_ns:
2781 	put_user_ns(ns);
2782 err:
2783 	return ret;
2784 }
2785 
2786 static int proc_setgroups_release(struct inode *inode, struct file *file)
2787 {
2788 	struct seq_file *seq = file->private_data;
2789 	struct user_namespace *ns = seq->private;
2790 	int ret = single_release(inode, file);
2791 	put_user_ns(ns);
2792 	return ret;
2793 }
2794 
2795 static const struct file_operations proc_setgroups_operations = {
2796 	.open		= proc_setgroups_open,
2797 	.write		= proc_setgroups_write,
2798 	.read		= seq_read,
2799 	.llseek		= seq_lseek,
2800 	.release	= proc_setgroups_release,
2801 };
2802 #endif /* CONFIG_USER_NS */
2803 
2804 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2805 				struct pid *pid, struct task_struct *task)
2806 {
2807 	int err = lock_trace(task);
2808 	if (!err) {
2809 		seq_printf(m, "%08x\n", task->personality);
2810 		unlock_trace(task);
2811 	}
2812 	return err;
2813 }
2814 
2815 /*
2816  * Thread groups
2817  */
2818 static const struct file_operations proc_task_operations;
2819 static const struct inode_operations proc_task_inode_operations;
2820 
2821 static const struct pid_entry tgid_base_stuff[] = {
2822 	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2823 	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2824 	DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2825 	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2826 	DIR("ns",	  S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2827 #ifdef CONFIG_NET
2828 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2829 #endif
2830 	REG("environ",    S_IRUSR, proc_environ_operations),
2831 	ONE("auxv",       S_IRUSR, proc_pid_auxv),
2832 	ONE("status",     S_IRUGO, proc_pid_status),
2833 	ONE("personality", S_IRUSR, proc_pid_personality),
2834 	ONE("limits",	  S_IRUGO, proc_pid_limits),
2835 #ifdef CONFIG_SCHED_DEBUG
2836 	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2837 #endif
2838 #ifdef CONFIG_SCHED_AUTOGROUP
2839 	REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2840 #endif
2841 	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2842 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2843 	ONE("syscall",    S_IRUSR, proc_pid_syscall),
2844 #endif
2845 	REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2846 	ONE("stat",       S_IRUGO, proc_tgid_stat),
2847 	ONE("statm",      S_IRUGO, proc_pid_statm),
2848 	REG("maps",       S_IRUGO, proc_pid_maps_operations),
2849 #ifdef CONFIG_NUMA
2850 	REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2851 #endif
2852 	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2853 	LNK("cwd",        proc_cwd_link),
2854 	LNK("root",       proc_root_link),
2855 	LNK("exe",        proc_exe_link),
2856 	REG("mounts",     S_IRUGO, proc_mounts_operations),
2857 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2858 	REG("mountstats", S_IRUSR, proc_mountstats_operations),
2859 #ifdef CONFIG_PROC_PAGE_MONITOR
2860 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2861 	REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2862 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2863 #endif
2864 #ifdef CONFIG_SECURITY
2865 	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2866 #endif
2867 #ifdef CONFIG_KALLSYMS
2868 	ONE("wchan",      S_IRUGO, proc_pid_wchan),
2869 #endif
2870 #ifdef CONFIG_STACKTRACE
2871 	ONE("stack",      S_IRUSR, proc_pid_stack),
2872 #endif
2873 #ifdef CONFIG_SCHED_INFO
2874 	ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2875 #endif
2876 #ifdef CONFIG_LATENCYTOP
2877 	REG("latency",  S_IRUGO, proc_lstats_operations),
2878 #endif
2879 #ifdef CONFIG_PROC_PID_CPUSET
2880 	ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2881 #endif
2882 #ifdef CONFIG_CGROUPS
2883 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2884 #endif
2885 	ONE("oom_score",  S_IRUGO, proc_oom_score),
2886 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2887 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2888 #ifdef CONFIG_AUDITSYSCALL
2889 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2890 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2891 #endif
2892 #ifdef CONFIG_FAULT_INJECTION
2893 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2894 #endif
2895 #ifdef CONFIG_ELF_CORE
2896 	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2897 #endif
2898 #ifdef CONFIG_TASK_IO_ACCOUNTING
2899 	ONE("io",	S_IRUSR, proc_tgid_io_accounting),
2900 #endif
2901 #ifdef CONFIG_HARDWALL
2902 	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2903 #endif
2904 #ifdef CONFIG_USER_NS
2905 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2906 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2907 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2908 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2909 #endif
2910 #ifdef CONFIG_CHECKPOINT_RESTORE
2911 	REG("timers",	  S_IRUGO, proc_timers_operations),
2912 #endif
2913 	REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2914 };
2915 
2916 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2917 {
2918 	return proc_pident_readdir(file, ctx,
2919 				   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2920 }
2921 
2922 static const struct file_operations proc_tgid_base_operations = {
2923 	.read		= generic_read_dir,
2924 	.iterate_shared	= proc_tgid_base_readdir,
2925 	.llseek		= generic_file_llseek,
2926 };
2927 
2928 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2929 {
2930 	return proc_pident_lookup(dir, dentry,
2931 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2932 }
2933 
2934 static const struct inode_operations proc_tgid_base_inode_operations = {
2935 	.lookup		= proc_tgid_base_lookup,
2936 	.getattr	= pid_getattr,
2937 	.setattr	= proc_setattr,
2938 	.permission	= proc_pid_permission,
2939 };
2940 
2941 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2942 {
2943 	struct dentry *dentry, *leader, *dir;
2944 	char buf[PROC_NUMBUF];
2945 	struct qstr name;
2946 
2947 	name.name = buf;
2948 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2949 	/* no ->d_hash() rejects on procfs */
2950 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2951 	if (dentry) {
2952 		d_invalidate(dentry);
2953 		dput(dentry);
2954 	}
2955 
2956 	if (pid == tgid)
2957 		return;
2958 
2959 	name.name = buf;
2960 	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2961 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2962 	if (!leader)
2963 		goto out;
2964 
2965 	name.name = "task";
2966 	name.len = strlen(name.name);
2967 	dir = d_hash_and_lookup(leader, &name);
2968 	if (!dir)
2969 		goto out_put_leader;
2970 
2971 	name.name = buf;
2972 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2973 	dentry = d_hash_and_lookup(dir, &name);
2974 	if (dentry) {
2975 		d_invalidate(dentry);
2976 		dput(dentry);
2977 	}
2978 
2979 	dput(dir);
2980 out_put_leader:
2981 	dput(leader);
2982 out:
2983 	return;
2984 }
2985 
2986 /**
2987  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2988  * @task: task that should be flushed.
2989  *
2990  * When flushing dentries from proc, one needs to flush them from global
2991  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2992  * in. This call is supposed to do all of this job.
2993  *
2994  * Looks in the dcache for
2995  * /proc/@pid
2996  * /proc/@tgid/task/@pid
2997  * if either directory is present flushes it and all of it'ts children
2998  * from the dcache.
2999  *
3000  * It is safe and reasonable to cache /proc entries for a task until
3001  * that task exits.  After that they just clog up the dcache with
3002  * useless entries, possibly causing useful dcache entries to be
3003  * flushed instead.  This routine is proved to flush those useless
3004  * dcache entries at process exit time.
3005  *
3006  * NOTE: This routine is just an optimization so it does not guarantee
3007  *       that no dcache entries will exist at process exit time it
3008  *       just makes it very unlikely that any will persist.
3009  */
3010 
3011 void proc_flush_task(struct task_struct *task)
3012 {
3013 	int i;
3014 	struct pid *pid, *tgid;
3015 	struct upid *upid;
3016 
3017 	pid = task_pid(task);
3018 	tgid = task_tgid(task);
3019 
3020 	for (i = 0; i <= pid->level; i++) {
3021 		upid = &pid->numbers[i];
3022 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3023 					tgid->numbers[i].nr);
3024 	}
3025 }
3026 
3027 static int proc_pid_instantiate(struct inode *dir,
3028 				   struct dentry * dentry,
3029 				   struct task_struct *task, const void *ptr)
3030 {
3031 	struct inode *inode;
3032 
3033 	inode = proc_pid_make_inode(dir->i_sb, task);
3034 	if (!inode)
3035 		goto out;
3036 
3037 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3038 	inode->i_op = &proc_tgid_base_inode_operations;
3039 	inode->i_fop = &proc_tgid_base_operations;
3040 	inode->i_flags|=S_IMMUTABLE;
3041 
3042 	set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3043 						  ARRAY_SIZE(tgid_base_stuff)));
3044 
3045 	d_set_d_op(dentry, &pid_dentry_operations);
3046 
3047 	d_add(dentry, inode);
3048 	/* Close the race of the process dying before we return the dentry */
3049 	if (pid_revalidate(dentry, 0))
3050 		return 0;
3051 out:
3052 	return -ENOENT;
3053 }
3054 
3055 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3056 {
3057 	int result = -ENOENT;
3058 	struct task_struct *task;
3059 	unsigned tgid;
3060 	struct pid_namespace *ns;
3061 
3062 	tgid = name_to_int(&dentry->d_name);
3063 	if (tgid == ~0U)
3064 		goto out;
3065 
3066 	ns = dentry->d_sb->s_fs_info;
3067 	rcu_read_lock();
3068 	task = find_task_by_pid_ns(tgid, ns);
3069 	if (task)
3070 		get_task_struct(task);
3071 	rcu_read_unlock();
3072 	if (!task)
3073 		goto out;
3074 
3075 	result = proc_pid_instantiate(dir, dentry, task, NULL);
3076 	put_task_struct(task);
3077 out:
3078 	return ERR_PTR(result);
3079 }
3080 
3081 /*
3082  * Find the first task with tgid >= tgid
3083  *
3084  */
3085 struct tgid_iter {
3086 	unsigned int tgid;
3087 	struct task_struct *task;
3088 };
3089 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3090 {
3091 	struct pid *pid;
3092 
3093 	if (iter.task)
3094 		put_task_struct(iter.task);
3095 	rcu_read_lock();
3096 retry:
3097 	iter.task = NULL;
3098 	pid = find_ge_pid(iter.tgid, ns);
3099 	if (pid) {
3100 		iter.tgid = pid_nr_ns(pid, ns);
3101 		iter.task = pid_task(pid, PIDTYPE_PID);
3102 		/* What we to know is if the pid we have find is the
3103 		 * pid of a thread_group_leader.  Testing for task
3104 		 * being a thread_group_leader is the obvious thing
3105 		 * todo but there is a window when it fails, due to
3106 		 * the pid transfer logic in de_thread.
3107 		 *
3108 		 * So we perform the straight forward test of seeing
3109 		 * if the pid we have found is the pid of a thread
3110 		 * group leader, and don't worry if the task we have
3111 		 * found doesn't happen to be a thread group leader.
3112 		 * As we don't care in the case of readdir.
3113 		 */
3114 		if (!iter.task || !has_group_leader_pid(iter.task)) {
3115 			iter.tgid += 1;
3116 			goto retry;
3117 		}
3118 		get_task_struct(iter.task);
3119 	}
3120 	rcu_read_unlock();
3121 	return iter;
3122 }
3123 
3124 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3125 
3126 /* for the /proc/ directory itself, after non-process stuff has been done */
3127 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3128 {
3129 	struct tgid_iter iter;
3130 	struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3131 	loff_t pos = ctx->pos;
3132 
3133 	if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3134 		return 0;
3135 
3136 	if (pos == TGID_OFFSET - 2) {
3137 		struct inode *inode = d_inode(ns->proc_self);
3138 		if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3139 			return 0;
3140 		ctx->pos = pos = pos + 1;
3141 	}
3142 	if (pos == TGID_OFFSET - 1) {
3143 		struct inode *inode = d_inode(ns->proc_thread_self);
3144 		if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3145 			return 0;
3146 		ctx->pos = pos = pos + 1;
3147 	}
3148 	iter.tgid = pos - TGID_OFFSET;
3149 	iter.task = NULL;
3150 	for (iter = next_tgid(ns, iter);
3151 	     iter.task;
3152 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
3153 		char name[PROC_NUMBUF];
3154 		int len;
3155 		if (!has_pid_permissions(ns, iter.task, 2))
3156 			continue;
3157 
3158 		len = snprintf(name, sizeof(name), "%d", iter.tgid);
3159 		ctx->pos = iter.tgid + TGID_OFFSET;
3160 		if (!proc_fill_cache(file, ctx, name, len,
3161 				     proc_pid_instantiate, iter.task, NULL)) {
3162 			put_task_struct(iter.task);
3163 			return 0;
3164 		}
3165 	}
3166 	ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3167 	return 0;
3168 }
3169 
3170 /*
3171  * proc_tid_comm_permission is a special permission function exclusively
3172  * used for the node /proc/<pid>/task/<tid>/comm.
3173  * It bypasses generic permission checks in the case where a task of the same
3174  * task group attempts to access the node.
3175  * The rationale behind this is that glibc and bionic access this node for
3176  * cross thread naming (pthread_set/getname_np(!self)). However, if
3177  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3178  * which locks out the cross thread naming implementation.
3179  * This function makes sure that the node is always accessible for members of
3180  * same thread group.
3181  */
3182 static int proc_tid_comm_permission(struct inode *inode, int mask)
3183 {
3184 	bool is_same_tgroup;
3185 	struct task_struct *task;
3186 
3187 	task = get_proc_task(inode);
3188 	if (!task)
3189 		return -ESRCH;
3190 	is_same_tgroup = same_thread_group(current, task);
3191 	put_task_struct(task);
3192 
3193 	if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3194 		/* This file (/proc/<pid>/task/<tid>/comm) can always be
3195 		 * read or written by the members of the corresponding
3196 		 * thread group.
3197 		 */
3198 		return 0;
3199 	}
3200 
3201 	return generic_permission(inode, mask);
3202 }
3203 
3204 static const struct inode_operations proc_tid_comm_inode_operations = {
3205 		.permission = proc_tid_comm_permission,
3206 };
3207 
3208 /*
3209  * Tasks
3210  */
3211 static const struct pid_entry tid_base_stuff[] = {
3212 	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3213 	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3214 	DIR("ns",	 S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3215 #ifdef CONFIG_NET
3216 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3217 #endif
3218 	REG("environ",   S_IRUSR, proc_environ_operations),
3219 	ONE("auxv",      S_IRUSR, proc_pid_auxv),
3220 	ONE("status",    S_IRUGO, proc_pid_status),
3221 	ONE("personality", S_IRUSR, proc_pid_personality),
3222 	ONE("limits",	 S_IRUGO, proc_pid_limits),
3223 #ifdef CONFIG_SCHED_DEBUG
3224 	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3225 #endif
3226 	NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3227 			 &proc_tid_comm_inode_operations,
3228 			 &proc_pid_set_comm_operations, {}),
3229 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3230 	ONE("syscall",   S_IRUSR, proc_pid_syscall),
3231 #endif
3232 	REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3233 	ONE("stat",      S_IRUGO, proc_tid_stat),
3234 	ONE("statm",     S_IRUGO, proc_pid_statm),
3235 	REG("maps",      S_IRUGO, proc_tid_maps_operations),
3236 #ifdef CONFIG_PROC_CHILDREN
3237 	REG("children",  S_IRUGO, proc_tid_children_operations),
3238 #endif
3239 #ifdef CONFIG_NUMA
3240 	REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3241 #endif
3242 	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3243 	LNK("cwd",       proc_cwd_link),
3244 	LNK("root",      proc_root_link),
3245 	LNK("exe",       proc_exe_link),
3246 	REG("mounts",    S_IRUGO, proc_mounts_operations),
3247 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3248 #ifdef CONFIG_PROC_PAGE_MONITOR
3249 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3250 	REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3251 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3252 #endif
3253 #ifdef CONFIG_SECURITY
3254 	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3255 #endif
3256 #ifdef CONFIG_KALLSYMS
3257 	ONE("wchan",     S_IRUGO, proc_pid_wchan),
3258 #endif
3259 #ifdef CONFIG_STACKTRACE
3260 	ONE("stack",      S_IRUSR, proc_pid_stack),
3261 #endif
3262 #ifdef CONFIG_SCHED_INFO
3263 	ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3264 #endif
3265 #ifdef CONFIG_LATENCYTOP
3266 	REG("latency",  S_IRUGO, proc_lstats_operations),
3267 #endif
3268 #ifdef CONFIG_PROC_PID_CPUSET
3269 	ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3270 #endif
3271 #ifdef CONFIG_CGROUPS
3272 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3273 #endif
3274 	ONE("oom_score", S_IRUGO, proc_oom_score),
3275 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3276 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3277 #ifdef CONFIG_AUDITSYSCALL
3278 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3279 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3280 #endif
3281 #ifdef CONFIG_FAULT_INJECTION
3282 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3283 #endif
3284 #ifdef CONFIG_TASK_IO_ACCOUNTING
3285 	ONE("io",	S_IRUSR, proc_tid_io_accounting),
3286 #endif
3287 #ifdef CONFIG_HARDWALL
3288 	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3289 #endif
3290 #ifdef CONFIG_USER_NS
3291 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3292 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3293 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3294 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3295 #endif
3296 };
3297 
3298 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3299 {
3300 	return proc_pident_readdir(file, ctx,
3301 				   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3302 }
3303 
3304 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3305 {
3306 	return proc_pident_lookup(dir, dentry,
3307 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3308 }
3309 
3310 static const struct file_operations proc_tid_base_operations = {
3311 	.read		= generic_read_dir,
3312 	.iterate_shared	= proc_tid_base_readdir,
3313 	.llseek		= generic_file_llseek,
3314 };
3315 
3316 static const struct inode_operations proc_tid_base_inode_operations = {
3317 	.lookup		= proc_tid_base_lookup,
3318 	.getattr	= pid_getattr,
3319 	.setattr	= proc_setattr,
3320 };
3321 
3322 static int proc_task_instantiate(struct inode *dir,
3323 	struct dentry *dentry, struct task_struct *task, const void *ptr)
3324 {
3325 	struct inode *inode;
3326 	inode = proc_pid_make_inode(dir->i_sb, task);
3327 
3328 	if (!inode)
3329 		goto out;
3330 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3331 	inode->i_op = &proc_tid_base_inode_operations;
3332 	inode->i_fop = &proc_tid_base_operations;
3333 	inode->i_flags|=S_IMMUTABLE;
3334 
3335 	set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3336 						  ARRAY_SIZE(tid_base_stuff)));
3337 
3338 	d_set_d_op(dentry, &pid_dentry_operations);
3339 
3340 	d_add(dentry, inode);
3341 	/* Close the race of the process dying before we return the dentry */
3342 	if (pid_revalidate(dentry, 0))
3343 		return 0;
3344 out:
3345 	return -ENOENT;
3346 }
3347 
3348 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3349 {
3350 	int result = -ENOENT;
3351 	struct task_struct *task;
3352 	struct task_struct *leader = get_proc_task(dir);
3353 	unsigned tid;
3354 	struct pid_namespace *ns;
3355 
3356 	if (!leader)
3357 		goto out_no_task;
3358 
3359 	tid = name_to_int(&dentry->d_name);
3360 	if (tid == ~0U)
3361 		goto out;
3362 
3363 	ns = dentry->d_sb->s_fs_info;
3364 	rcu_read_lock();
3365 	task = find_task_by_pid_ns(tid, ns);
3366 	if (task)
3367 		get_task_struct(task);
3368 	rcu_read_unlock();
3369 	if (!task)
3370 		goto out;
3371 	if (!same_thread_group(leader, task))
3372 		goto out_drop_task;
3373 
3374 	result = proc_task_instantiate(dir, dentry, task, NULL);
3375 out_drop_task:
3376 	put_task_struct(task);
3377 out:
3378 	put_task_struct(leader);
3379 out_no_task:
3380 	return ERR_PTR(result);
3381 }
3382 
3383 /*
3384  * Find the first tid of a thread group to return to user space.
3385  *
3386  * Usually this is just the thread group leader, but if the users
3387  * buffer was too small or there was a seek into the middle of the
3388  * directory we have more work todo.
3389  *
3390  * In the case of a short read we start with find_task_by_pid.
3391  *
3392  * In the case of a seek we start with the leader and walk nr
3393  * threads past it.
3394  */
3395 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3396 					struct pid_namespace *ns)
3397 {
3398 	struct task_struct *pos, *task;
3399 	unsigned long nr = f_pos;
3400 
3401 	if (nr != f_pos)	/* 32bit overflow? */
3402 		return NULL;
3403 
3404 	rcu_read_lock();
3405 	task = pid_task(pid, PIDTYPE_PID);
3406 	if (!task)
3407 		goto fail;
3408 
3409 	/* Attempt to start with the tid of a thread */
3410 	if (tid && nr) {
3411 		pos = find_task_by_pid_ns(tid, ns);
3412 		if (pos && same_thread_group(pos, task))
3413 			goto found;
3414 	}
3415 
3416 	/* If nr exceeds the number of threads there is nothing todo */
3417 	if (nr >= get_nr_threads(task))
3418 		goto fail;
3419 
3420 	/* If we haven't found our starting place yet start
3421 	 * with the leader and walk nr threads forward.
3422 	 */
3423 	pos = task = task->group_leader;
3424 	do {
3425 		if (!nr--)
3426 			goto found;
3427 	} while_each_thread(task, pos);
3428 fail:
3429 	pos = NULL;
3430 	goto out;
3431 found:
3432 	get_task_struct(pos);
3433 out:
3434 	rcu_read_unlock();
3435 	return pos;
3436 }
3437 
3438 /*
3439  * Find the next thread in the thread list.
3440  * Return NULL if there is an error or no next thread.
3441  *
3442  * The reference to the input task_struct is released.
3443  */
3444 static struct task_struct *next_tid(struct task_struct *start)
3445 {
3446 	struct task_struct *pos = NULL;
3447 	rcu_read_lock();
3448 	if (pid_alive(start)) {
3449 		pos = next_thread(start);
3450 		if (thread_group_leader(pos))
3451 			pos = NULL;
3452 		else
3453 			get_task_struct(pos);
3454 	}
3455 	rcu_read_unlock();
3456 	put_task_struct(start);
3457 	return pos;
3458 }
3459 
3460 /* for the /proc/TGID/task/ directories */
3461 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3462 {
3463 	struct inode *inode = file_inode(file);
3464 	struct task_struct *task;
3465 	struct pid_namespace *ns;
3466 	int tid;
3467 
3468 	if (proc_inode_is_dead(inode))
3469 		return -ENOENT;
3470 
3471 	if (!dir_emit_dots(file, ctx))
3472 		return 0;
3473 
3474 	/* f_version caches the tgid value that the last readdir call couldn't
3475 	 * return. lseek aka telldir automagically resets f_version to 0.
3476 	 */
3477 	ns = inode->i_sb->s_fs_info;
3478 	tid = (int)file->f_version;
3479 	file->f_version = 0;
3480 	for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3481 	     task;
3482 	     task = next_tid(task), ctx->pos++) {
3483 		char name[PROC_NUMBUF];
3484 		int len;
3485 		tid = task_pid_nr_ns(task, ns);
3486 		len = snprintf(name, sizeof(name), "%d", tid);
3487 		if (!proc_fill_cache(file, ctx, name, len,
3488 				proc_task_instantiate, task, NULL)) {
3489 			/* returning this tgid failed, save it as the first
3490 			 * pid for the next readir call */
3491 			file->f_version = (u64)tid;
3492 			put_task_struct(task);
3493 			break;
3494 		}
3495 	}
3496 
3497 	return 0;
3498 }
3499 
3500 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3501 {
3502 	struct inode *inode = d_inode(dentry);
3503 	struct task_struct *p = get_proc_task(inode);
3504 	generic_fillattr(inode, stat);
3505 
3506 	if (p) {
3507 		stat->nlink += get_nr_threads(p);
3508 		put_task_struct(p);
3509 	}
3510 
3511 	return 0;
3512 }
3513 
3514 static const struct inode_operations proc_task_inode_operations = {
3515 	.lookup		= proc_task_lookup,
3516 	.getattr	= proc_task_getattr,
3517 	.setattr	= proc_setattr,
3518 	.permission	= proc_pid_permission,
3519 };
3520 
3521 static const struct file_operations proc_task_operations = {
3522 	.read		= generic_read_dir,
3523 	.iterate_shared	= proc_task_readdir,
3524 	.llseek		= generic_file_llseek,
3525 };
3526