xref: /openbmc/linux/fs/proc/base.c (revision 79f08d9e)
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 	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 INF(NAME, MODE, read)				\
134 	NOD(NAME, (S_IFREG|(MODE)), 			\
135 		NULL, &proc_info_file_operations,	\
136 		{ .proc_read = read } )
137 #define ONE(NAME, MODE, show)				\
138 	NOD(NAME, (S_IFREG|(MODE)), 			\
139 		NULL, &proc_single_file_operations,	\
140 		{ .proc_show = show } )
141 
142 /*
143  * Count the number of hardlinks for the pid_entry table, excluding the .
144  * and .. links.
145  */
146 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
147 	unsigned int n)
148 {
149 	unsigned int i;
150 	unsigned int count;
151 
152 	count = 0;
153 	for (i = 0; i < n; ++i) {
154 		if (S_ISDIR(entries[i].mode))
155 			++count;
156 	}
157 
158 	return count;
159 }
160 
161 static int get_task_root(struct task_struct *task, struct path *root)
162 {
163 	int result = -ENOENT;
164 
165 	task_lock(task);
166 	if (task->fs) {
167 		get_fs_root(task->fs, root);
168 		result = 0;
169 	}
170 	task_unlock(task);
171 	return result;
172 }
173 
174 static int proc_cwd_link(struct dentry *dentry, struct path *path)
175 {
176 	struct task_struct *task = get_proc_task(dentry->d_inode);
177 	int result = -ENOENT;
178 
179 	if (task) {
180 		task_lock(task);
181 		if (task->fs) {
182 			get_fs_pwd(task->fs, path);
183 			result = 0;
184 		}
185 		task_unlock(task);
186 		put_task_struct(task);
187 	}
188 	return result;
189 }
190 
191 static int proc_root_link(struct dentry *dentry, struct path *path)
192 {
193 	struct task_struct *task = get_proc_task(dentry->d_inode);
194 	int result = -ENOENT;
195 
196 	if (task) {
197 		result = get_task_root(task, path);
198 		put_task_struct(task);
199 	}
200 	return result;
201 }
202 
203 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
204 {
205 	int res = 0;
206 	unsigned int len;
207 	struct mm_struct *mm = get_task_mm(task);
208 	if (!mm)
209 		goto out;
210 	if (!mm->arg_end)
211 		goto out_mm;	/* Shh! No looking before we're done */
212 
213  	len = mm->arg_end - mm->arg_start;
214 
215 	if (len > PAGE_SIZE)
216 		len = PAGE_SIZE;
217 
218 	res = access_process_vm(task, mm->arg_start, buffer, len, 0);
219 
220 	// If the nul at the end of args has been overwritten, then
221 	// assume application is using setproctitle(3).
222 	if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
223 		len = strnlen(buffer, res);
224 		if (len < res) {
225 		    res = len;
226 		} else {
227 			len = mm->env_end - mm->env_start;
228 			if (len > PAGE_SIZE - res)
229 				len = PAGE_SIZE - res;
230 			res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
231 			res = strnlen(buffer, res);
232 		}
233 	}
234 out_mm:
235 	mmput(mm);
236 out:
237 	return res;
238 }
239 
240 static int proc_pid_auxv(struct task_struct *task, char *buffer)
241 {
242 	struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
243 	int res = PTR_ERR(mm);
244 	if (mm && !IS_ERR(mm)) {
245 		unsigned int nwords = 0;
246 		do {
247 			nwords += 2;
248 		} while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
249 		res = nwords * sizeof(mm->saved_auxv[0]);
250 		if (res > PAGE_SIZE)
251 			res = PAGE_SIZE;
252 		memcpy(buffer, mm->saved_auxv, res);
253 		mmput(mm);
254 	}
255 	return res;
256 }
257 
258 
259 #ifdef CONFIG_KALLSYMS
260 /*
261  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
262  * Returns the resolved symbol.  If that fails, simply return the address.
263  */
264 static int proc_pid_wchan(struct task_struct *task, char *buffer)
265 {
266 	unsigned long wchan;
267 	char symname[KSYM_NAME_LEN];
268 
269 	wchan = get_wchan(task);
270 
271 	if (lookup_symbol_name(wchan, symname) < 0)
272 		if (!ptrace_may_access(task, PTRACE_MODE_READ))
273 			return 0;
274 		else
275 			return sprintf(buffer, "%lu", wchan);
276 	else
277 		return sprintf(buffer, "%s", symname);
278 }
279 #endif /* CONFIG_KALLSYMS */
280 
281 static int lock_trace(struct task_struct *task)
282 {
283 	int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
284 	if (err)
285 		return err;
286 	if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
287 		mutex_unlock(&task->signal->cred_guard_mutex);
288 		return -EPERM;
289 	}
290 	return 0;
291 }
292 
293 static void unlock_trace(struct task_struct *task)
294 {
295 	mutex_unlock(&task->signal->cred_guard_mutex);
296 }
297 
298 #ifdef CONFIG_STACKTRACE
299 
300 #define MAX_STACK_TRACE_DEPTH	64
301 
302 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
303 			  struct pid *pid, struct task_struct *task)
304 {
305 	struct stack_trace trace;
306 	unsigned long *entries;
307 	int err;
308 	int i;
309 
310 	entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
311 	if (!entries)
312 		return -ENOMEM;
313 
314 	trace.nr_entries	= 0;
315 	trace.max_entries	= MAX_STACK_TRACE_DEPTH;
316 	trace.entries		= entries;
317 	trace.skip		= 0;
318 
319 	err = lock_trace(task);
320 	if (!err) {
321 		save_stack_trace_tsk(task, &trace);
322 
323 		for (i = 0; i < trace.nr_entries; i++) {
324 			seq_printf(m, "[<%pK>] %pS\n",
325 				   (void *)entries[i], (void *)entries[i]);
326 		}
327 		unlock_trace(task);
328 	}
329 	kfree(entries);
330 
331 	return err;
332 }
333 #endif
334 
335 #ifdef CONFIG_SCHEDSTATS
336 /*
337  * Provides /proc/PID/schedstat
338  */
339 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
340 {
341 	return sprintf(buffer, "%llu %llu %lu\n",
342 			(unsigned long long)task->se.sum_exec_runtime,
343 			(unsigned long long)task->sched_info.run_delay,
344 			task->sched_info.pcount);
345 }
346 #endif
347 
348 #ifdef CONFIG_LATENCYTOP
349 static int lstats_show_proc(struct seq_file *m, void *v)
350 {
351 	int i;
352 	struct inode *inode = m->private;
353 	struct task_struct *task = get_proc_task(inode);
354 
355 	if (!task)
356 		return -ESRCH;
357 	seq_puts(m, "Latency Top version : v0.1\n");
358 	for (i = 0; i < 32; i++) {
359 		struct latency_record *lr = &task->latency_record[i];
360 		if (lr->backtrace[0]) {
361 			int q;
362 			seq_printf(m, "%i %li %li",
363 				   lr->count, lr->time, lr->max);
364 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
365 				unsigned long bt = lr->backtrace[q];
366 				if (!bt)
367 					break;
368 				if (bt == ULONG_MAX)
369 					break;
370 				seq_printf(m, " %ps", (void *)bt);
371 			}
372 			seq_putc(m, '\n');
373 		}
374 
375 	}
376 	put_task_struct(task);
377 	return 0;
378 }
379 
380 static int lstats_open(struct inode *inode, struct file *file)
381 {
382 	return single_open(file, lstats_show_proc, inode);
383 }
384 
385 static ssize_t lstats_write(struct file *file, const char __user *buf,
386 			    size_t count, loff_t *offs)
387 {
388 	struct task_struct *task = get_proc_task(file_inode(file));
389 
390 	if (!task)
391 		return -ESRCH;
392 	clear_all_latency_tracing(task);
393 	put_task_struct(task);
394 
395 	return count;
396 }
397 
398 static const struct file_operations proc_lstats_operations = {
399 	.open		= lstats_open,
400 	.read		= seq_read,
401 	.write		= lstats_write,
402 	.llseek		= seq_lseek,
403 	.release	= single_release,
404 };
405 
406 #endif
407 
408 #ifdef CONFIG_CGROUPS
409 static int cgroup_open(struct inode *inode, struct file *file)
410 {
411 	struct pid *pid = PROC_I(inode)->pid;
412 	return single_open(file, proc_cgroup_show, pid);
413 }
414 
415 static const struct file_operations proc_cgroup_operations = {
416 	.open		= cgroup_open,
417 	.read		= seq_read,
418 	.llseek		= seq_lseek,
419 	.release	= single_release,
420 };
421 #endif
422 
423 #ifdef CONFIG_PROC_PID_CPUSET
424 
425 static int cpuset_open(struct inode *inode, struct file *file)
426 {
427 	struct pid *pid = PROC_I(inode)->pid;
428 	return single_open(file, proc_cpuset_show, pid);
429 }
430 
431 static const struct file_operations proc_cpuset_operations = {
432 	.open		= cpuset_open,
433 	.read		= seq_read,
434 	.llseek		= seq_lseek,
435 	.release	= single_release,
436 };
437 #endif
438 
439 static int proc_oom_score(struct task_struct *task, char *buffer)
440 {
441 	unsigned long totalpages = totalram_pages + total_swap_pages;
442 	unsigned long points = 0;
443 
444 	read_lock(&tasklist_lock);
445 	if (pid_alive(task))
446 		points = oom_badness(task, NULL, NULL, totalpages) *
447 						1000 / totalpages;
448 	read_unlock(&tasklist_lock);
449 	return sprintf(buffer, "%lu\n", points);
450 }
451 
452 struct limit_names {
453 	char *name;
454 	char *unit;
455 };
456 
457 static const struct limit_names lnames[RLIM_NLIMITS] = {
458 	[RLIMIT_CPU] = {"Max cpu time", "seconds"},
459 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
460 	[RLIMIT_DATA] = {"Max data size", "bytes"},
461 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
462 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
463 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
464 	[RLIMIT_NPROC] = {"Max processes", "processes"},
465 	[RLIMIT_NOFILE] = {"Max open files", "files"},
466 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
467 	[RLIMIT_AS] = {"Max address space", "bytes"},
468 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
469 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
470 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
471 	[RLIMIT_NICE] = {"Max nice priority", NULL},
472 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
473 	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
474 };
475 
476 /* Display limits for a process */
477 static int proc_pid_limits(struct task_struct *task, char *buffer)
478 {
479 	unsigned int i;
480 	int count = 0;
481 	unsigned long flags;
482 	char *bufptr = buffer;
483 
484 	struct rlimit rlim[RLIM_NLIMITS];
485 
486 	if (!lock_task_sighand(task, &flags))
487 		return 0;
488 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
489 	unlock_task_sighand(task, &flags);
490 
491 	/*
492 	 * print the file header
493 	 */
494 	count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
495 			"Limit", "Soft Limit", "Hard Limit", "Units");
496 
497 	for (i = 0; i < RLIM_NLIMITS; i++) {
498 		if (rlim[i].rlim_cur == RLIM_INFINITY)
499 			count += sprintf(&bufptr[count], "%-25s %-20s ",
500 					 lnames[i].name, "unlimited");
501 		else
502 			count += sprintf(&bufptr[count], "%-25s %-20lu ",
503 					 lnames[i].name, rlim[i].rlim_cur);
504 
505 		if (rlim[i].rlim_max == RLIM_INFINITY)
506 			count += sprintf(&bufptr[count], "%-20s ", "unlimited");
507 		else
508 			count += sprintf(&bufptr[count], "%-20lu ",
509 					 rlim[i].rlim_max);
510 
511 		if (lnames[i].unit)
512 			count += sprintf(&bufptr[count], "%-10s\n",
513 					 lnames[i].unit);
514 		else
515 			count += sprintf(&bufptr[count], "\n");
516 	}
517 
518 	return count;
519 }
520 
521 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
522 static int proc_pid_syscall(struct task_struct *task, char *buffer)
523 {
524 	long nr;
525 	unsigned long args[6], sp, pc;
526 	int res = lock_trace(task);
527 	if (res)
528 		return res;
529 
530 	if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
531 		res = sprintf(buffer, "running\n");
532 	else if (nr < 0)
533 		res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
534 	else
535 		res = sprintf(buffer,
536 		       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
537 		       nr,
538 		       args[0], args[1], args[2], args[3], args[4], args[5],
539 		       sp, pc);
540 	unlock_trace(task);
541 	return res;
542 }
543 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
544 
545 /************************************************************************/
546 /*                       Here the fs part begins                        */
547 /************************************************************************/
548 
549 /* permission checks */
550 static int proc_fd_access_allowed(struct inode *inode)
551 {
552 	struct task_struct *task;
553 	int allowed = 0;
554 	/* Allow access to a task's file descriptors if it is us or we
555 	 * may use ptrace attach to the process and find out that
556 	 * information.
557 	 */
558 	task = get_proc_task(inode);
559 	if (task) {
560 		allowed = ptrace_may_access(task, PTRACE_MODE_READ);
561 		put_task_struct(task);
562 	}
563 	return allowed;
564 }
565 
566 int proc_setattr(struct dentry *dentry, struct iattr *attr)
567 {
568 	int error;
569 	struct inode *inode = dentry->d_inode;
570 
571 	if (attr->ia_valid & ATTR_MODE)
572 		return -EPERM;
573 
574 	error = inode_change_ok(inode, attr);
575 	if (error)
576 		return error;
577 
578 	setattr_copy(inode, attr);
579 	mark_inode_dirty(inode);
580 	return 0;
581 }
582 
583 /*
584  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
585  * or euid/egid (for hide_pid_min=2)?
586  */
587 static bool has_pid_permissions(struct pid_namespace *pid,
588 				 struct task_struct *task,
589 				 int hide_pid_min)
590 {
591 	if (pid->hide_pid < hide_pid_min)
592 		return true;
593 	if (in_group_p(pid->pid_gid))
594 		return true;
595 	return ptrace_may_access(task, PTRACE_MODE_READ);
596 }
597 
598 
599 static int proc_pid_permission(struct inode *inode, int mask)
600 {
601 	struct pid_namespace *pid = inode->i_sb->s_fs_info;
602 	struct task_struct *task;
603 	bool has_perms;
604 
605 	task = get_proc_task(inode);
606 	if (!task)
607 		return -ESRCH;
608 	has_perms = has_pid_permissions(pid, task, 1);
609 	put_task_struct(task);
610 
611 	if (!has_perms) {
612 		if (pid->hide_pid == 2) {
613 			/*
614 			 * Let's make getdents(), stat(), and open()
615 			 * consistent with each other.  If a process
616 			 * may not stat() a file, it shouldn't be seen
617 			 * in procfs at all.
618 			 */
619 			return -ENOENT;
620 		}
621 
622 		return -EPERM;
623 	}
624 	return generic_permission(inode, mask);
625 }
626 
627 
628 
629 static const struct inode_operations proc_def_inode_operations = {
630 	.setattr	= proc_setattr,
631 };
632 
633 #define PROC_BLOCK_SIZE	(3*1024)		/* 4K page size but our output routines use some slack for overruns */
634 
635 static ssize_t proc_info_read(struct file * file, char __user * buf,
636 			  size_t count, loff_t *ppos)
637 {
638 	struct inode * inode = file_inode(file);
639 	unsigned long page;
640 	ssize_t length;
641 	struct task_struct *task = get_proc_task(inode);
642 
643 	length = -ESRCH;
644 	if (!task)
645 		goto out_no_task;
646 
647 	if (count > PROC_BLOCK_SIZE)
648 		count = PROC_BLOCK_SIZE;
649 
650 	length = -ENOMEM;
651 	if (!(page = __get_free_page(GFP_TEMPORARY)))
652 		goto out;
653 
654 	length = PROC_I(inode)->op.proc_read(task, (char*)page);
655 
656 	if (length >= 0)
657 		length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
658 	free_page(page);
659 out:
660 	put_task_struct(task);
661 out_no_task:
662 	return length;
663 }
664 
665 static const struct file_operations proc_info_file_operations = {
666 	.read		= proc_info_read,
667 	.llseek		= generic_file_llseek,
668 };
669 
670 static int proc_single_show(struct seq_file *m, void *v)
671 {
672 	struct inode *inode = m->private;
673 	struct pid_namespace *ns;
674 	struct pid *pid;
675 	struct task_struct *task;
676 	int ret;
677 
678 	ns = inode->i_sb->s_fs_info;
679 	pid = proc_pid(inode);
680 	task = get_pid_task(pid, PIDTYPE_PID);
681 	if (!task)
682 		return -ESRCH;
683 
684 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
685 
686 	put_task_struct(task);
687 	return ret;
688 }
689 
690 static int proc_single_open(struct inode *inode, struct file *filp)
691 {
692 	return single_open(filp, proc_single_show, inode);
693 }
694 
695 static const struct file_operations proc_single_file_operations = {
696 	.open		= proc_single_open,
697 	.read		= seq_read,
698 	.llseek		= seq_lseek,
699 	.release	= single_release,
700 };
701 
702 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
703 {
704 	struct task_struct *task = get_proc_task(file_inode(file));
705 	struct mm_struct *mm;
706 
707 	if (!task)
708 		return -ESRCH;
709 
710 	mm = mm_access(task, mode);
711 	put_task_struct(task);
712 
713 	if (IS_ERR(mm))
714 		return PTR_ERR(mm);
715 
716 	if (mm) {
717 		/* ensure this mm_struct can't be freed */
718 		atomic_inc(&mm->mm_count);
719 		/* but do not pin its memory */
720 		mmput(mm);
721 	}
722 
723 	file->private_data = mm;
724 
725 	return 0;
726 }
727 
728 static int mem_open(struct inode *inode, struct file *file)
729 {
730 	int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
731 
732 	/* OK to pass negative loff_t, we can catch out-of-range */
733 	file->f_mode |= FMODE_UNSIGNED_OFFSET;
734 
735 	return ret;
736 }
737 
738 static ssize_t mem_rw(struct file *file, char __user *buf,
739 			size_t count, loff_t *ppos, int write)
740 {
741 	struct mm_struct *mm = file->private_data;
742 	unsigned long addr = *ppos;
743 	ssize_t copied;
744 	char *page;
745 
746 	if (!mm)
747 		return 0;
748 
749 	page = (char *)__get_free_page(GFP_TEMPORARY);
750 	if (!page)
751 		return -ENOMEM;
752 
753 	copied = 0;
754 	if (!atomic_inc_not_zero(&mm->mm_users))
755 		goto free;
756 
757 	while (count > 0) {
758 		int this_len = min_t(int, count, PAGE_SIZE);
759 
760 		if (write && copy_from_user(page, buf, this_len)) {
761 			copied = -EFAULT;
762 			break;
763 		}
764 
765 		this_len = access_remote_vm(mm, addr, page, this_len, write);
766 		if (!this_len) {
767 			if (!copied)
768 				copied = -EIO;
769 			break;
770 		}
771 
772 		if (!write && copy_to_user(buf, page, this_len)) {
773 			copied = -EFAULT;
774 			break;
775 		}
776 
777 		buf += this_len;
778 		addr += this_len;
779 		copied += this_len;
780 		count -= this_len;
781 	}
782 	*ppos = addr;
783 
784 	mmput(mm);
785 free:
786 	free_page((unsigned long) page);
787 	return copied;
788 }
789 
790 static ssize_t mem_read(struct file *file, char __user *buf,
791 			size_t count, loff_t *ppos)
792 {
793 	return mem_rw(file, buf, count, ppos, 0);
794 }
795 
796 static ssize_t mem_write(struct file *file, const char __user *buf,
797 			 size_t count, loff_t *ppos)
798 {
799 	return mem_rw(file, (char __user*)buf, count, ppos, 1);
800 }
801 
802 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
803 {
804 	switch (orig) {
805 	case 0:
806 		file->f_pos = offset;
807 		break;
808 	case 1:
809 		file->f_pos += offset;
810 		break;
811 	default:
812 		return -EINVAL;
813 	}
814 	force_successful_syscall_return();
815 	return file->f_pos;
816 }
817 
818 static int mem_release(struct inode *inode, struct file *file)
819 {
820 	struct mm_struct *mm = file->private_data;
821 	if (mm)
822 		mmdrop(mm);
823 	return 0;
824 }
825 
826 static const struct file_operations proc_mem_operations = {
827 	.llseek		= mem_lseek,
828 	.read		= mem_read,
829 	.write		= mem_write,
830 	.open		= mem_open,
831 	.release	= mem_release,
832 };
833 
834 static int environ_open(struct inode *inode, struct file *file)
835 {
836 	return __mem_open(inode, file, PTRACE_MODE_READ);
837 }
838 
839 static ssize_t environ_read(struct file *file, char __user *buf,
840 			size_t count, loff_t *ppos)
841 {
842 	char *page;
843 	unsigned long src = *ppos;
844 	int ret = 0;
845 	struct mm_struct *mm = file->private_data;
846 
847 	if (!mm)
848 		return 0;
849 
850 	page = (char *)__get_free_page(GFP_TEMPORARY);
851 	if (!page)
852 		return -ENOMEM;
853 
854 	ret = 0;
855 	if (!atomic_inc_not_zero(&mm->mm_users))
856 		goto free;
857 	while (count > 0) {
858 		size_t this_len, max_len;
859 		int retval;
860 
861 		if (src >= (mm->env_end - mm->env_start))
862 			break;
863 
864 		this_len = mm->env_end - (mm->env_start + src);
865 
866 		max_len = min_t(size_t, PAGE_SIZE, count);
867 		this_len = min(max_len, this_len);
868 
869 		retval = access_remote_vm(mm, (mm->env_start + src),
870 			page, this_len, 0);
871 
872 		if (retval <= 0) {
873 			ret = retval;
874 			break;
875 		}
876 
877 		if (copy_to_user(buf, page, retval)) {
878 			ret = -EFAULT;
879 			break;
880 		}
881 
882 		ret += retval;
883 		src += retval;
884 		buf += retval;
885 		count -= retval;
886 	}
887 	*ppos = src;
888 	mmput(mm);
889 
890 free:
891 	free_page((unsigned long) page);
892 	return ret;
893 }
894 
895 static const struct file_operations proc_environ_operations = {
896 	.open		= environ_open,
897 	.read		= environ_read,
898 	.llseek		= generic_file_llseek,
899 	.release	= mem_release,
900 };
901 
902 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
903 			    loff_t *ppos)
904 {
905 	struct task_struct *task = get_proc_task(file_inode(file));
906 	char buffer[PROC_NUMBUF];
907 	int oom_adj = OOM_ADJUST_MIN;
908 	size_t len;
909 	unsigned long flags;
910 
911 	if (!task)
912 		return -ESRCH;
913 	if (lock_task_sighand(task, &flags)) {
914 		if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
915 			oom_adj = OOM_ADJUST_MAX;
916 		else
917 			oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
918 				  OOM_SCORE_ADJ_MAX;
919 		unlock_task_sighand(task, &flags);
920 	}
921 	put_task_struct(task);
922 	len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
923 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
924 }
925 
926 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
927 			     size_t count, loff_t *ppos)
928 {
929 	struct task_struct *task;
930 	char buffer[PROC_NUMBUF];
931 	int oom_adj;
932 	unsigned long flags;
933 	int err;
934 
935 	memset(buffer, 0, sizeof(buffer));
936 	if (count > sizeof(buffer) - 1)
937 		count = sizeof(buffer) - 1;
938 	if (copy_from_user(buffer, buf, count)) {
939 		err = -EFAULT;
940 		goto out;
941 	}
942 
943 	err = kstrtoint(strstrip(buffer), 0, &oom_adj);
944 	if (err)
945 		goto out;
946 	if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
947 	     oom_adj != OOM_DISABLE) {
948 		err = -EINVAL;
949 		goto out;
950 	}
951 
952 	task = get_proc_task(file_inode(file));
953 	if (!task) {
954 		err = -ESRCH;
955 		goto out;
956 	}
957 
958 	task_lock(task);
959 	if (!task->mm) {
960 		err = -EINVAL;
961 		goto err_task_lock;
962 	}
963 
964 	if (!lock_task_sighand(task, &flags)) {
965 		err = -ESRCH;
966 		goto err_task_lock;
967 	}
968 
969 	/*
970 	 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
971 	 * value is always attainable.
972 	 */
973 	if (oom_adj == OOM_ADJUST_MAX)
974 		oom_adj = OOM_SCORE_ADJ_MAX;
975 	else
976 		oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
977 
978 	if (oom_adj < task->signal->oom_score_adj &&
979 	    !capable(CAP_SYS_RESOURCE)) {
980 		err = -EACCES;
981 		goto err_sighand;
982 	}
983 
984 	/*
985 	 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
986 	 * /proc/pid/oom_score_adj instead.
987 	 */
988 	pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
989 		  current->comm, task_pid_nr(current), task_pid_nr(task),
990 		  task_pid_nr(task));
991 
992 	task->signal->oom_score_adj = oom_adj;
993 	trace_oom_score_adj_update(task);
994 err_sighand:
995 	unlock_task_sighand(task, &flags);
996 err_task_lock:
997 	task_unlock(task);
998 	put_task_struct(task);
999 out:
1000 	return err < 0 ? err : count;
1001 }
1002 
1003 static const struct file_operations proc_oom_adj_operations = {
1004 	.read		= oom_adj_read,
1005 	.write		= oom_adj_write,
1006 	.llseek		= generic_file_llseek,
1007 };
1008 
1009 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1010 					size_t count, loff_t *ppos)
1011 {
1012 	struct task_struct *task = get_proc_task(file_inode(file));
1013 	char buffer[PROC_NUMBUF];
1014 	short oom_score_adj = OOM_SCORE_ADJ_MIN;
1015 	unsigned long flags;
1016 	size_t len;
1017 
1018 	if (!task)
1019 		return -ESRCH;
1020 	if (lock_task_sighand(task, &flags)) {
1021 		oom_score_adj = task->signal->oom_score_adj;
1022 		unlock_task_sighand(task, &flags);
1023 	}
1024 	put_task_struct(task);
1025 	len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1026 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1027 }
1028 
1029 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1030 					size_t count, loff_t *ppos)
1031 {
1032 	struct task_struct *task;
1033 	char buffer[PROC_NUMBUF];
1034 	unsigned long flags;
1035 	int oom_score_adj;
1036 	int err;
1037 
1038 	memset(buffer, 0, sizeof(buffer));
1039 	if (count > sizeof(buffer) - 1)
1040 		count = sizeof(buffer) - 1;
1041 	if (copy_from_user(buffer, buf, count)) {
1042 		err = -EFAULT;
1043 		goto out;
1044 	}
1045 
1046 	err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1047 	if (err)
1048 		goto out;
1049 	if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1050 			oom_score_adj > OOM_SCORE_ADJ_MAX) {
1051 		err = -EINVAL;
1052 		goto out;
1053 	}
1054 
1055 	task = get_proc_task(file_inode(file));
1056 	if (!task) {
1057 		err = -ESRCH;
1058 		goto out;
1059 	}
1060 
1061 	task_lock(task);
1062 	if (!task->mm) {
1063 		err = -EINVAL;
1064 		goto err_task_lock;
1065 	}
1066 
1067 	if (!lock_task_sighand(task, &flags)) {
1068 		err = -ESRCH;
1069 		goto err_task_lock;
1070 	}
1071 
1072 	if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1073 			!capable(CAP_SYS_RESOURCE)) {
1074 		err = -EACCES;
1075 		goto err_sighand;
1076 	}
1077 
1078 	task->signal->oom_score_adj = (short)oom_score_adj;
1079 	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1080 		task->signal->oom_score_adj_min = (short)oom_score_adj;
1081 	trace_oom_score_adj_update(task);
1082 
1083 err_sighand:
1084 	unlock_task_sighand(task, &flags);
1085 err_task_lock:
1086 	task_unlock(task);
1087 	put_task_struct(task);
1088 out:
1089 	return err < 0 ? err : count;
1090 }
1091 
1092 static const struct file_operations proc_oom_score_adj_operations = {
1093 	.read		= oom_score_adj_read,
1094 	.write		= oom_score_adj_write,
1095 	.llseek		= default_llseek,
1096 };
1097 
1098 #ifdef CONFIG_AUDITSYSCALL
1099 #define TMPBUFLEN 21
1100 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1101 				  size_t count, loff_t *ppos)
1102 {
1103 	struct inode * inode = file_inode(file);
1104 	struct task_struct *task = get_proc_task(inode);
1105 	ssize_t length;
1106 	char tmpbuf[TMPBUFLEN];
1107 
1108 	if (!task)
1109 		return -ESRCH;
1110 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1111 			   from_kuid(file->f_cred->user_ns,
1112 				     audit_get_loginuid(task)));
1113 	put_task_struct(task);
1114 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1115 }
1116 
1117 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1118 				   size_t count, loff_t *ppos)
1119 {
1120 	struct inode * inode = file_inode(file);
1121 	char *page, *tmp;
1122 	ssize_t length;
1123 	uid_t loginuid;
1124 	kuid_t kloginuid;
1125 
1126 	rcu_read_lock();
1127 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1128 		rcu_read_unlock();
1129 		return -EPERM;
1130 	}
1131 	rcu_read_unlock();
1132 
1133 	if (count >= PAGE_SIZE)
1134 		count = PAGE_SIZE - 1;
1135 
1136 	if (*ppos != 0) {
1137 		/* No partial writes. */
1138 		return -EINVAL;
1139 	}
1140 	page = (char*)__get_free_page(GFP_TEMPORARY);
1141 	if (!page)
1142 		return -ENOMEM;
1143 	length = -EFAULT;
1144 	if (copy_from_user(page, buf, count))
1145 		goto out_free_page;
1146 
1147 	page[count] = '\0';
1148 	loginuid = simple_strtoul(page, &tmp, 10);
1149 	if (tmp == page) {
1150 		length = -EINVAL;
1151 		goto out_free_page;
1152 
1153 	}
1154 
1155 	/* is userspace tring to explicitly UNSET the loginuid? */
1156 	if (loginuid == AUDIT_UID_UNSET) {
1157 		kloginuid = INVALID_UID;
1158 	} else {
1159 		kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1160 		if (!uid_valid(kloginuid)) {
1161 			length = -EINVAL;
1162 			goto out_free_page;
1163 		}
1164 	}
1165 
1166 	length = audit_set_loginuid(kloginuid);
1167 	if (likely(length == 0))
1168 		length = count;
1169 
1170 out_free_page:
1171 	free_page((unsigned long) page);
1172 	return length;
1173 }
1174 
1175 static const struct file_operations proc_loginuid_operations = {
1176 	.read		= proc_loginuid_read,
1177 	.write		= proc_loginuid_write,
1178 	.llseek		= generic_file_llseek,
1179 };
1180 
1181 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1182 				  size_t count, loff_t *ppos)
1183 {
1184 	struct inode * inode = file_inode(file);
1185 	struct task_struct *task = get_proc_task(inode);
1186 	ssize_t length;
1187 	char tmpbuf[TMPBUFLEN];
1188 
1189 	if (!task)
1190 		return -ESRCH;
1191 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1192 				audit_get_sessionid(task));
1193 	put_task_struct(task);
1194 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1195 }
1196 
1197 static const struct file_operations proc_sessionid_operations = {
1198 	.read		= proc_sessionid_read,
1199 	.llseek		= generic_file_llseek,
1200 };
1201 #endif
1202 
1203 #ifdef CONFIG_FAULT_INJECTION
1204 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1205 				      size_t count, loff_t *ppos)
1206 {
1207 	struct task_struct *task = get_proc_task(file_inode(file));
1208 	char buffer[PROC_NUMBUF];
1209 	size_t len;
1210 	int make_it_fail;
1211 
1212 	if (!task)
1213 		return -ESRCH;
1214 	make_it_fail = task->make_it_fail;
1215 	put_task_struct(task);
1216 
1217 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1218 
1219 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1220 }
1221 
1222 static ssize_t proc_fault_inject_write(struct file * file,
1223 			const char __user * buf, size_t count, loff_t *ppos)
1224 {
1225 	struct task_struct *task;
1226 	char buffer[PROC_NUMBUF], *end;
1227 	int make_it_fail;
1228 
1229 	if (!capable(CAP_SYS_RESOURCE))
1230 		return -EPERM;
1231 	memset(buffer, 0, sizeof(buffer));
1232 	if (count > sizeof(buffer) - 1)
1233 		count = sizeof(buffer) - 1;
1234 	if (copy_from_user(buffer, buf, count))
1235 		return -EFAULT;
1236 	make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1237 	if (*end)
1238 		return -EINVAL;
1239 	task = get_proc_task(file_inode(file));
1240 	if (!task)
1241 		return -ESRCH;
1242 	task->make_it_fail = make_it_fail;
1243 	put_task_struct(task);
1244 
1245 	return count;
1246 }
1247 
1248 static const struct file_operations proc_fault_inject_operations = {
1249 	.read		= proc_fault_inject_read,
1250 	.write		= proc_fault_inject_write,
1251 	.llseek		= generic_file_llseek,
1252 };
1253 #endif
1254 
1255 
1256 #ifdef CONFIG_SCHED_DEBUG
1257 /*
1258  * Print out various scheduling related per-task fields:
1259  */
1260 static int sched_show(struct seq_file *m, void *v)
1261 {
1262 	struct inode *inode = m->private;
1263 	struct task_struct *p;
1264 
1265 	p = get_proc_task(inode);
1266 	if (!p)
1267 		return -ESRCH;
1268 	proc_sched_show_task(p, m);
1269 
1270 	put_task_struct(p);
1271 
1272 	return 0;
1273 }
1274 
1275 static ssize_t
1276 sched_write(struct file *file, const char __user *buf,
1277 	    size_t count, loff_t *offset)
1278 {
1279 	struct inode *inode = file_inode(file);
1280 	struct task_struct *p;
1281 
1282 	p = get_proc_task(inode);
1283 	if (!p)
1284 		return -ESRCH;
1285 	proc_sched_set_task(p);
1286 
1287 	put_task_struct(p);
1288 
1289 	return count;
1290 }
1291 
1292 static int sched_open(struct inode *inode, struct file *filp)
1293 {
1294 	return single_open(filp, sched_show, inode);
1295 }
1296 
1297 static const struct file_operations proc_pid_sched_operations = {
1298 	.open		= sched_open,
1299 	.read		= seq_read,
1300 	.write		= sched_write,
1301 	.llseek		= seq_lseek,
1302 	.release	= single_release,
1303 };
1304 
1305 #endif
1306 
1307 #ifdef CONFIG_SCHED_AUTOGROUP
1308 /*
1309  * Print out autogroup related information:
1310  */
1311 static int sched_autogroup_show(struct seq_file *m, void *v)
1312 {
1313 	struct inode *inode = m->private;
1314 	struct task_struct *p;
1315 
1316 	p = get_proc_task(inode);
1317 	if (!p)
1318 		return -ESRCH;
1319 	proc_sched_autogroup_show_task(p, m);
1320 
1321 	put_task_struct(p);
1322 
1323 	return 0;
1324 }
1325 
1326 static ssize_t
1327 sched_autogroup_write(struct file *file, const char __user *buf,
1328 	    size_t count, loff_t *offset)
1329 {
1330 	struct inode *inode = file_inode(file);
1331 	struct task_struct *p;
1332 	char buffer[PROC_NUMBUF];
1333 	int nice;
1334 	int err;
1335 
1336 	memset(buffer, 0, sizeof(buffer));
1337 	if (count > sizeof(buffer) - 1)
1338 		count = sizeof(buffer) - 1;
1339 	if (copy_from_user(buffer, buf, count))
1340 		return -EFAULT;
1341 
1342 	err = kstrtoint(strstrip(buffer), 0, &nice);
1343 	if (err < 0)
1344 		return err;
1345 
1346 	p = get_proc_task(inode);
1347 	if (!p)
1348 		return -ESRCH;
1349 
1350 	err = proc_sched_autogroup_set_nice(p, nice);
1351 	if (err)
1352 		count = err;
1353 
1354 	put_task_struct(p);
1355 
1356 	return count;
1357 }
1358 
1359 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1360 {
1361 	int ret;
1362 
1363 	ret = single_open(filp, sched_autogroup_show, NULL);
1364 	if (!ret) {
1365 		struct seq_file *m = filp->private_data;
1366 
1367 		m->private = inode;
1368 	}
1369 	return ret;
1370 }
1371 
1372 static const struct file_operations proc_pid_sched_autogroup_operations = {
1373 	.open		= sched_autogroup_open,
1374 	.read		= seq_read,
1375 	.write		= sched_autogroup_write,
1376 	.llseek		= seq_lseek,
1377 	.release	= single_release,
1378 };
1379 
1380 #endif /* CONFIG_SCHED_AUTOGROUP */
1381 
1382 static ssize_t comm_write(struct file *file, const char __user *buf,
1383 				size_t count, loff_t *offset)
1384 {
1385 	struct inode *inode = file_inode(file);
1386 	struct task_struct *p;
1387 	char buffer[TASK_COMM_LEN];
1388 	const size_t maxlen = sizeof(buffer) - 1;
1389 
1390 	memset(buffer, 0, sizeof(buffer));
1391 	if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1392 		return -EFAULT;
1393 
1394 	p = get_proc_task(inode);
1395 	if (!p)
1396 		return -ESRCH;
1397 
1398 	if (same_thread_group(current, p))
1399 		set_task_comm(p, buffer);
1400 	else
1401 		count = -EINVAL;
1402 
1403 	put_task_struct(p);
1404 
1405 	return count;
1406 }
1407 
1408 static int comm_show(struct seq_file *m, void *v)
1409 {
1410 	struct inode *inode = m->private;
1411 	struct task_struct *p;
1412 
1413 	p = get_proc_task(inode);
1414 	if (!p)
1415 		return -ESRCH;
1416 
1417 	task_lock(p);
1418 	seq_printf(m, "%s\n", p->comm);
1419 	task_unlock(p);
1420 
1421 	put_task_struct(p);
1422 
1423 	return 0;
1424 }
1425 
1426 static int comm_open(struct inode *inode, struct file *filp)
1427 {
1428 	return single_open(filp, comm_show, inode);
1429 }
1430 
1431 static const struct file_operations proc_pid_set_comm_operations = {
1432 	.open		= comm_open,
1433 	.read		= seq_read,
1434 	.write		= comm_write,
1435 	.llseek		= seq_lseek,
1436 	.release	= single_release,
1437 };
1438 
1439 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1440 {
1441 	struct task_struct *task;
1442 	struct mm_struct *mm;
1443 	struct file *exe_file;
1444 
1445 	task = get_proc_task(dentry->d_inode);
1446 	if (!task)
1447 		return -ENOENT;
1448 	mm = get_task_mm(task);
1449 	put_task_struct(task);
1450 	if (!mm)
1451 		return -ENOENT;
1452 	exe_file = get_mm_exe_file(mm);
1453 	mmput(mm);
1454 	if (exe_file) {
1455 		*exe_path = exe_file->f_path;
1456 		path_get(&exe_file->f_path);
1457 		fput(exe_file);
1458 		return 0;
1459 	} else
1460 		return -ENOENT;
1461 }
1462 
1463 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1464 {
1465 	struct inode *inode = dentry->d_inode;
1466 	struct path path;
1467 	int error = -EACCES;
1468 
1469 	/* Are we allowed to snoop on the tasks file descriptors? */
1470 	if (!proc_fd_access_allowed(inode))
1471 		goto out;
1472 
1473 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1474 	if (error)
1475 		goto out;
1476 
1477 	nd_jump_link(nd, &path);
1478 	return NULL;
1479 out:
1480 	return ERR_PTR(error);
1481 }
1482 
1483 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1484 {
1485 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1486 	char *pathname;
1487 	int len;
1488 
1489 	if (!tmp)
1490 		return -ENOMEM;
1491 
1492 	pathname = d_path(path, tmp, PAGE_SIZE);
1493 	len = PTR_ERR(pathname);
1494 	if (IS_ERR(pathname))
1495 		goto out;
1496 	len = tmp + PAGE_SIZE - 1 - pathname;
1497 
1498 	if (len > buflen)
1499 		len = buflen;
1500 	if (copy_to_user(buffer, pathname, len))
1501 		len = -EFAULT;
1502  out:
1503 	free_page((unsigned long)tmp);
1504 	return len;
1505 }
1506 
1507 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1508 {
1509 	int error = -EACCES;
1510 	struct inode *inode = dentry->d_inode;
1511 	struct path path;
1512 
1513 	/* Are we allowed to snoop on the tasks file descriptors? */
1514 	if (!proc_fd_access_allowed(inode))
1515 		goto out;
1516 
1517 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1518 	if (error)
1519 		goto out;
1520 
1521 	error = do_proc_readlink(&path, buffer, buflen);
1522 	path_put(&path);
1523 out:
1524 	return error;
1525 }
1526 
1527 const struct inode_operations proc_pid_link_inode_operations = {
1528 	.readlink	= proc_pid_readlink,
1529 	.follow_link	= proc_pid_follow_link,
1530 	.setattr	= proc_setattr,
1531 };
1532 
1533 
1534 /* building an inode */
1535 
1536 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1537 {
1538 	struct inode * inode;
1539 	struct proc_inode *ei;
1540 	const struct cred *cred;
1541 
1542 	/* We need a new inode */
1543 
1544 	inode = new_inode(sb);
1545 	if (!inode)
1546 		goto out;
1547 
1548 	/* Common stuff */
1549 	ei = PROC_I(inode);
1550 	inode->i_ino = get_next_ino();
1551 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1552 	inode->i_op = &proc_def_inode_operations;
1553 
1554 	/*
1555 	 * grab the reference to task.
1556 	 */
1557 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1558 	if (!ei->pid)
1559 		goto out_unlock;
1560 
1561 	if (task_dumpable(task)) {
1562 		rcu_read_lock();
1563 		cred = __task_cred(task);
1564 		inode->i_uid = cred->euid;
1565 		inode->i_gid = cred->egid;
1566 		rcu_read_unlock();
1567 	}
1568 	security_task_to_inode(task, inode);
1569 
1570 out:
1571 	return inode;
1572 
1573 out_unlock:
1574 	iput(inode);
1575 	return NULL;
1576 }
1577 
1578 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1579 {
1580 	struct inode *inode = dentry->d_inode;
1581 	struct task_struct *task;
1582 	const struct cred *cred;
1583 	struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1584 
1585 	generic_fillattr(inode, stat);
1586 
1587 	rcu_read_lock();
1588 	stat->uid = GLOBAL_ROOT_UID;
1589 	stat->gid = GLOBAL_ROOT_GID;
1590 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1591 	if (task) {
1592 		if (!has_pid_permissions(pid, task, 2)) {
1593 			rcu_read_unlock();
1594 			/*
1595 			 * This doesn't prevent learning whether PID exists,
1596 			 * it only makes getattr() consistent with readdir().
1597 			 */
1598 			return -ENOENT;
1599 		}
1600 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1601 		    task_dumpable(task)) {
1602 			cred = __task_cred(task);
1603 			stat->uid = cred->euid;
1604 			stat->gid = cred->egid;
1605 		}
1606 	}
1607 	rcu_read_unlock();
1608 	return 0;
1609 }
1610 
1611 /* dentry stuff */
1612 
1613 /*
1614  *	Exceptional case: normally we are not allowed to unhash a busy
1615  * directory. In this case, however, we can do it - no aliasing problems
1616  * due to the way we treat inodes.
1617  *
1618  * Rewrite the inode's ownerships here because the owning task may have
1619  * performed a setuid(), etc.
1620  *
1621  * Before the /proc/pid/status file was created the only way to read
1622  * the effective uid of a /process was to stat /proc/pid.  Reading
1623  * /proc/pid/status is slow enough that procps and other packages
1624  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1625  * made this apply to all per process world readable and executable
1626  * directories.
1627  */
1628 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1629 {
1630 	struct inode *inode;
1631 	struct task_struct *task;
1632 	const struct cred *cred;
1633 
1634 	if (flags & LOOKUP_RCU)
1635 		return -ECHILD;
1636 
1637 	inode = dentry->d_inode;
1638 	task = get_proc_task(inode);
1639 
1640 	if (task) {
1641 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1642 		    task_dumpable(task)) {
1643 			rcu_read_lock();
1644 			cred = __task_cred(task);
1645 			inode->i_uid = cred->euid;
1646 			inode->i_gid = cred->egid;
1647 			rcu_read_unlock();
1648 		} else {
1649 			inode->i_uid = GLOBAL_ROOT_UID;
1650 			inode->i_gid = GLOBAL_ROOT_GID;
1651 		}
1652 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1653 		security_task_to_inode(task, inode);
1654 		put_task_struct(task);
1655 		return 1;
1656 	}
1657 	d_drop(dentry);
1658 	return 0;
1659 }
1660 
1661 int pid_delete_dentry(const struct dentry *dentry)
1662 {
1663 	/* Is the task we represent dead?
1664 	 * If so, then don't put the dentry on the lru list,
1665 	 * kill it immediately.
1666 	 */
1667 	return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1668 }
1669 
1670 const struct dentry_operations pid_dentry_operations =
1671 {
1672 	.d_revalidate	= pid_revalidate,
1673 	.d_delete	= pid_delete_dentry,
1674 };
1675 
1676 /* Lookups */
1677 
1678 /*
1679  * Fill a directory entry.
1680  *
1681  * If possible create the dcache entry and derive our inode number and
1682  * file type from dcache entry.
1683  *
1684  * Since all of the proc inode numbers are dynamically generated, the inode
1685  * numbers do not exist until the inode is cache.  This means creating the
1686  * the dcache entry in readdir is necessary to keep the inode numbers
1687  * reported by readdir in sync with the inode numbers reported
1688  * by stat.
1689  */
1690 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1691 	const char *name, int len,
1692 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1693 {
1694 	struct dentry *child, *dir = file->f_path.dentry;
1695 	struct qstr qname = QSTR_INIT(name, len);
1696 	struct inode *inode;
1697 	unsigned type;
1698 	ino_t ino;
1699 
1700 	child = d_hash_and_lookup(dir, &qname);
1701 	if (!child) {
1702 		child = d_alloc(dir, &qname);
1703 		if (!child)
1704 			goto end_instantiate;
1705 		if (instantiate(dir->d_inode, child, task, ptr) < 0) {
1706 			dput(child);
1707 			goto end_instantiate;
1708 		}
1709 	}
1710 	inode = child->d_inode;
1711 	ino = inode->i_ino;
1712 	type = inode->i_mode >> 12;
1713 	dput(child);
1714 	return dir_emit(ctx, name, len, ino, type);
1715 
1716 end_instantiate:
1717 	return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1718 }
1719 
1720 #ifdef CONFIG_CHECKPOINT_RESTORE
1721 
1722 /*
1723  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1724  * which represent vma start and end addresses.
1725  */
1726 static int dname_to_vma_addr(struct dentry *dentry,
1727 			     unsigned long *start, unsigned long *end)
1728 {
1729 	if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1730 		return -EINVAL;
1731 
1732 	return 0;
1733 }
1734 
1735 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1736 {
1737 	unsigned long vm_start, vm_end;
1738 	bool exact_vma_exists = false;
1739 	struct mm_struct *mm = NULL;
1740 	struct task_struct *task;
1741 	const struct cred *cred;
1742 	struct inode *inode;
1743 	int status = 0;
1744 
1745 	if (flags & LOOKUP_RCU)
1746 		return -ECHILD;
1747 
1748 	if (!capable(CAP_SYS_ADMIN)) {
1749 		status = -EPERM;
1750 		goto out_notask;
1751 	}
1752 
1753 	inode = dentry->d_inode;
1754 	task = get_proc_task(inode);
1755 	if (!task)
1756 		goto out_notask;
1757 
1758 	mm = mm_access(task, PTRACE_MODE_READ);
1759 	if (IS_ERR_OR_NULL(mm))
1760 		goto out;
1761 
1762 	if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1763 		down_read(&mm->mmap_sem);
1764 		exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1765 		up_read(&mm->mmap_sem);
1766 	}
1767 
1768 	mmput(mm);
1769 
1770 	if (exact_vma_exists) {
1771 		if (task_dumpable(task)) {
1772 			rcu_read_lock();
1773 			cred = __task_cred(task);
1774 			inode->i_uid = cred->euid;
1775 			inode->i_gid = cred->egid;
1776 			rcu_read_unlock();
1777 		} else {
1778 			inode->i_uid = GLOBAL_ROOT_UID;
1779 			inode->i_gid = GLOBAL_ROOT_GID;
1780 		}
1781 		security_task_to_inode(task, inode);
1782 		status = 1;
1783 	}
1784 
1785 out:
1786 	put_task_struct(task);
1787 
1788 out_notask:
1789 	if (status <= 0)
1790 		d_drop(dentry);
1791 
1792 	return status;
1793 }
1794 
1795 static const struct dentry_operations tid_map_files_dentry_operations = {
1796 	.d_revalidate	= map_files_d_revalidate,
1797 	.d_delete	= pid_delete_dentry,
1798 };
1799 
1800 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1801 {
1802 	unsigned long vm_start, vm_end;
1803 	struct vm_area_struct *vma;
1804 	struct task_struct *task;
1805 	struct mm_struct *mm;
1806 	int rc;
1807 
1808 	rc = -ENOENT;
1809 	task = get_proc_task(dentry->d_inode);
1810 	if (!task)
1811 		goto out;
1812 
1813 	mm = get_task_mm(task);
1814 	put_task_struct(task);
1815 	if (!mm)
1816 		goto out;
1817 
1818 	rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1819 	if (rc)
1820 		goto out_mmput;
1821 
1822 	down_read(&mm->mmap_sem);
1823 	vma = find_exact_vma(mm, vm_start, vm_end);
1824 	if (vma && vma->vm_file) {
1825 		*path = vma->vm_file->f_path;
1826 		path_get(path);
1827 		rc = 0;
1828 	}
1829 	up_read(&mm->mmap_sem);
1830 
1831 out_mmput:
1832 	mmput(mm);
1833 out:
1834 	return rc;
1835 }
1836 
1837 struct map_files_info {
1838 	fmode_t		mode;
1839 	unsigned long	len;
1840 	unsigned char	name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1841 };
1842 
1843 static int
1844 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1845 			   struct task_struct *task, const void *ptr)
1846 {
1847 	fmode_t mode = (fmode_t)(unsigned long)ptr;
1848 	struct proc_inode *ei;
1849 	struct inode *inode;
1850 
1851 	inode = proc_pid_make_inode(dir->i_sb, task);
1852 	if (!inode)
1853 		return -ENOENT;
1854 
1855 	ei = PROC_I(inode);
1856 	ei->op.proc_get_link = proc_map_files_get_link;
1857 
1858 	inode->i_op = &proc_pid_link_inode_operations;
1859 	inode->i_size = 64;
1860 	inode->i_mode = S_IFLNK;
1861 
1862 	if (mode & FMODE_READ)
1863 		inode->i_mode |= S_IRUSR;
1864 	if (mode & FMODE_WRITE)
1865 		inode->i_mode |= S_IWUSR;
1866 
1867 	d_set_d_op(dentry, &tid_map_files_dentry_operations);
1868 	d_add(dentry, inode);
1869 
1870 	return 0;
1871 }
1872 
1873 static struct dentry *proc_map_files_lookup(struct inode *dir,
1874 		struct dentry *dentry, unsigned int flags)
1875 {
1876 	unsigned long vm_start, vm_end;
1877 	struct vm_area_struct *vma;
1878 	struct task_struct *task;
1879 	int result;
1880 	struct mm_struct *mm;
1881 
1882 	result = -EPERM;
1883 	if (!capable(CAP_SYS_ADMIN))
1884 		goto out;
1885 
1886 	result = -ENOENT;
1887 	task = get_proc_task(dir);
1888 	if (!task)
1889 		goto out;
1890 
1891 	result = -EACCES;
1892 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
1893 		goto out_put_task;
1894 
1895 	result = -ENOENT;
1896 	if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1897 		goto out_put_task;
1898 
1899 	mm = get_task_mm(task);
1900 	if (!mm)
1901 		goto out_put_task;
1902 
1903 	down_read(&mm->mmap_sem);
1904 	vma = find_exact_vma(mm, vm_start, vm_end);
1905 	if (!vma)
1906 		goto out_no_vma;
1907 
1908 	if (vma->vm_file)
1909 		result = proc_map_files_instantiate(dir, dentry, task,
1910 				(void *)(unsigned long)vma->vm_file->f_mode);
1911 
1912 out_no_vma:
1913 	up_read(&mm->mmap_sem);
1914 	mmput(mm);
1915 out_put_task:
1916 	put_task_struct(task);
1917 out:
1918 	return ERR_PTR(result);
1919 }
1920 
1921 static const struct inode_operations proc_map_files_inode_operations = {
1922 	.lookup		= proc_map_files_lookup,
1923 	.permission	= proc_fd_permission,
1924 	.setattr	= proc_setattr,
1925 };
1926 
1927 static int
1928 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
1929 {
1930 	struct vm_area_struct *vma;
1931 	struct task_struct *task;
1932 	struct mm_struct *mm;
1933 	unsigned long nr_files, pos, i;
1934 	struct flex_array *fa = NULL;
1935 	struct map_files_info info;
1936 	struct map_files_info *p;
1937 	int ret;
1938 
1939 	ret = -EPERM;
1940 	if (!capable(CAP_SYS_ADMIN))
1941 		goto out;
1942 
1943 	ret = -ENOENT;
1944 	task = get_proc_task(file_inode(file));
1945 	if (!task)
1946 		goto out;
1947 
1948 	ret = -EACCES;
1949 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
1950 		goto out_put_task;
1951 
1952 	ret = 0;
1953 	if (!dir_emit_dots(file, ctx))
1954 		goto out_put_task;
1955 
1956 	mm = get_task_mm(task);
1957 	if (!mm)
1958 		goto out_put_task;
1959 	down_read(&mm->mmap_sem);
1960 
1961 	nr_files = 0;
1962 
1963 	/*
1964 	 * We need two passes here:
1965 	 *
1966 	 *  1) Collect vmas of mapped files with mmap_sem taken
1967 	 *  2) Release mmap_sem and instantiate entries
1968 	 *
1969 	 * otherwise we get lockdep complained, since filldir()
1970 	 * routine might require mmap_sem taken in might_fault().
1971 	 */
1972 
1973 	for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1974 		if (vma->vm_file && ++pos > ctx->pos)
1975 			nr_files++;
1976 	}
1977 
1978 	if (nr_files) {
1979 		fa = flex_array_alloc(sizeof(info), nr_files,
1980 					GFP_KERNEL);
1981 		if (!fa || flex_array_prealloc(fa, 0, nr_files,
1982 						GFP_KERNEL)) {
1983 			ret = -ENOMEM;
1984 			if (fa)
1985 				flex_array_free(fa);
1986 			up_read(&mm->mmap_sem);
1987 			mmput(mm);
1988 			goto out_put_task;
1989 		}
1990 		for (i = 0, vma = mm->mmap, pos = 2; vma;
1991 				vma = vma->vm_next) {
1992 			if (!vma->vm_file)
1993 				continue;
1994 			if (++pos <= ctx->pos)
1995 				continue;
1996 
1997 			info.mode = vma->vm_file->f_mode;
1998 			info.len = snprintf(info.name,
1999 					sizeof(info.name), "%lx-%lx",
2000 					vma->vm_start, vma->vm_end);
2001 			if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2002 				BUG();
2003 		}
2004 	}
2005 	up_read(&mm->mmap_sem);
2006 
2007 	for (i = 0; i < nr_files; i++) {
2008 		p = flex_array_get(fa, i);
2009 		if (!proc_fill_cache(file, ctx,
2010 				      p->name, p->len,
2011 				      proc_map_files_instantiate,
2012 				      task,
2013 				      (void *)(unsigned long)p->mode))
2014 			break;
2015 		ctx->pos++;
2016 	}
2017 	if (fa)
2018 		flex_array_free(fa);
2019 	mmput(mm);
2020 
2021 out_put_task:
2022 	put_task_struct(task);
2023 out:
2024 	return ret;
2025 }
2026 
2027 static const struct file_operations proc_map_files_operations = {
2028 	.read		= generic_read_dir,
2029 	.iterate	= proc_map_files_readdir,
2030 	.llseek		= default_llseek,
2031 };
2032 
2033 struct timers_private {
2034 	struct pid *pid;
2035 	struct task_struct *task;
2036 	struct sighand_struct *sighand;
2037 	struct pid_namespace *ns;
2038 	unsigned long flags;
2039 };
2040 
2041 static void *timers_start(struct seq_file *m, loff_t *pos)
2042 {
2043 	struct timers_private *tp = m->private;
2044 
2045 	tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2046 	if (!tp->task)
2047 		return ERR_PTR(-ESRCH);
2048 
2049 	tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2050 	if (!tp->sighand)
2051 		return ERR_PTR(-ESRCH);
2052 
2053 	return seq_list_start(&tp->task->signal->posix_timers, *pos);
2054 }
2055 
2056 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2057 {
2058 	struct timers_private *tp = m->private;
2059 	return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2060 }
2061 
2062 static void timers_stop(struct seq_file *m, void *v)
2063 {
2064 	struct timers_private *tp = m->private;
2065 
2066 	if (tp->sighand) {
2067 		unlock_task_sighand(tp->task, &tp->flags);
2068 		tp->sighand = NULL;
2069 	}
2070 
2071 	if (tp->task) {
2072 		put_task_struct(tp->task);
2073 		tp->task = NULL;
2074 	}
2075 }
2076 
2077 static int show_timer(struct seq_file *m, void *v)
2078 {
2079 	struct k_itimer *timer;
2080 	struct timers_private *tp = m->private;
2081 	int notify;
2082 	static char *nstr[] = {
2083 		[SIGEV_SIGNAL] = "signal",
2084 		[SIGEV_NONE] = "none",
2085 		[SIGEV_THREAD] = "thread",
2086 	};
2087 
2088 	timer = list_entry((struct list_head *)v, struct k_itimer, list);
2089 	notify = timer->it_sigev_notify;
2090 
2091 	seq_printf(m, "ID: %d\n", timer->it_id);
2092 	seq_printf(m, "signal: %d/%p\n", timer->sigq->info.si_signo,
2093 			timer->sigq->info.si_value.sival_ptr);
2094 	seq_printf(m, "notify: %s/%s.%d\n",
2095 		nstr[notify & ~SIGEV_THREAD_ID],
2096 		(notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2097 		pid_nr_ns(timer->it_pid, tp->ns));
2098 	seq_printf(m, "ClockID: %d\n", timer->it_clock);
2099 
2100 	return 0;
2101 }
2102 
2103 static const struct seq_operations proc_timers_seq_ops = {
2104 	.start	= timers_start,
2105 	.next	= timers_next,
2106 	.stop	= timers_stop,
2107 	.show	= show_timer,
2108 };
2109 
2110 static int proc_timers_open(struct inode *inode, struct file *file)
2111 {
2112 	struct timers_private *tp;
2113 
2114 	tp = __seq_open_private(file, &proc_timers_seq_ops,
2115 			sizeof(struct timers_private));
2116 	if (!tp)
2117 		return -ENOMEM;
2118 
2119 	tp->pid = proc_pid(inode);
2120 	tp->ns = inode->i_sb->s_fs_info;
2121 	return 0;
2122 }
2123 
2124 static const struct file_operations proc_timers_operations = {
2125 	.open		= proc_timers_open,
2126 	.read		= seq_read,
2127 	.llseek		= seq_lseek,
2128 	.release	= seq_release_private,
2129 };
2130 #endif /* CONFIG_CHECKPOINT_RESTORE */
2131 
2132 static int proc_pident_instantiate(struct inode *dir,
2133 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2134 {
2135 	const struct pid_entry *p = ptr;
2136 	struct inode *inode;
2137 	struct proc_inode *ei;
2138 
2139 	inode = proc_pid_make_inode(dir->i_sb, task);
2140 	if (!inode)
2141 		goto out;
2142 
2143 	ei = PROC_I(inode);
2144 	inode->i_mode = p->mode;
2145 	if (S_ISDIR(inode->i_mode))
2146 		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
2147 	if (p->iop)
2148 		inode->i_op = p->iop;
2149 	if (p->fop)
2150 		inode->i_fop = p->fop;
2151 	ei->op = p->op;
2152 	d_set_d_op(dentry, &pid_dentry_operations);
2153 	d_add(dentry, inode);
2154 	/* Close the race of the process dying before we return the dentry */
2155 	if (pid_revalidate(dentry, 0))
2156 		return 0;
2157 out:
2158 	return -ENOENT;
2159 }
2160 
2161 static struct dentry *proc_pident_lookup(struct inode *dir,
2162 					 struct dentry *dentry,
2163 					 const struct pid_entry *ents,
2164 					 unsigned int nents)
2165 {
2166 	int error;
2167 	struct task_struct *task = get_proc_task(dir);
2168 	const struct pid_entry *p, *last;
2169 
2170 	error = -ENOENT;
2171 
2172 	if (!task)
2173 		goto out_no_task;
2174 
2175 	/*
2176 	 * Yes, it does not scale. And it should not. Don't add
2177 	 * new entries into /proc/<tgid>/ without very good reasons.
2178 	 */
2179 	last = &ents[nents - 1];
2180 	for (p = ents; p <= last; p++) {
2181 		if (p->len != dentry->d_name.len)
2182 			continue;
2183 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2184 			break;
2185 	}
2186 	if (p > last)
2187 		goto out;
2188 
2189 	error = proc_pident_instantiate(dir, dentry, task, p);
2190 out:
2191 	put_task_struct(task);
2192 out_no_task:
2193 	return ERR_PTR(error);
2194 }
2195 
2196 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2197 		const struct pid_entry *ents, unsigned int nents)
2198 {
2199 	struct task_struct *task = get_proc_task(file_inode(file));
2200 	const struct pid_entry *p;
2201 
2202 	if (!task)
2203 		return -ENOENT;
2204 
2205 	if (!dir_emit_dots(file, ctx))
2206 		goto out;
2207 
2208 	if (ctx->pos >= nents + 2)
2209 		goto out;
2210 
2211 	for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2212 		if (!proc_fill_cache(file, ctx, p->name, p->len,
2213 				proc_pident_instantiate, task, p))
2214 			break;
2215 		ctx->pos++;
2216 	}
2217 out:
2218 	put_task_struct(task);
2219 	return 0;
2220 }
2221 
2222 #ifdef CONFIG_SECURITY
2223 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2224 				  size_t count, loff_t *ppos)
2225 {
2226 	struct inode * inode = file_inode(file);
2227 	char *p = NULL;
2228 	ssize_t length;
2229 	struct task_struct *task = get_proc_task(inode);
2230 
2231 	if (!task)
2232 		return -ESRCH;
2233 
2234 	length = security_getprocattr(task,
2235 				      (char*)file->f_path.dentry->d_name.name,
2236 				      &p);
2237 	put_task_struct(task);
2238 	if (length > 0)
2239 		length = simple_read_from_buffer(buf, count, ppos, p, length);
2240 	kfree(p);
2241 	return length;
2242 }
2243 
2244 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2245 				   size_t count, loff_t *ppos)
2246 {
2247 	struct inode * inode = file_inode(file);
2248 	char *page;
2249 	ssize_t length;
2250 	struct task_struct *task = get_proc_task(inode);
2251 
2252 	length = -ESRCH;
2253 	if (!task)
2254 		goto out_no_task;
2255 	if (count > PAGE_SIZE)
2256 		count = PAGE_SIZE;
2257 
2258 	/* No partial writes. */
2259 	length = -EINVAL;
2260 	if (*ppos != 0)
2261 		goto out;
2262 
2263 	length = -ENOMEM;
2264 	page = (char*)__get_free_page(GFP_TEMPORARY);
2265 	if (!page)
2266 		goto out;
2267 
2268 	length = -EFAULT;
2269 	if (copy_from_user(page, buf, count))
2270 		goto out_free;
2271 
2272 	/* Guard against adverse ptrace interaction */
2273 	length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2274 	if (length < 0)
2275 		goto out_free;
2276 
2277 	length = security_setprocattr(task,
2278 				      (char*)file->f_path.dentry->d_name.name,
2279 				      (void*)page, count);
2280 	mutex_unlock(&task->signal->cred_guard_mutex);
2281 out_free:
2282 	free_page((unsigned long) page);
2283 out:
2284 	put_task_struct(task);
2285 out_no_task:
2286 	return length;
2287 }
2288 
2289 static const struct file_operations proc_pid_attr_operations = {
2290 	.read		= proc_pid_attr_read,
2291 	.write		= proc_pid_attr_write,
2292 	.llseek		= generic_file_llseek,
2293 };
2294 
2295 static const struct pid_entry attr_dir_stuff[] = {
2296 	REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2297 	REG("prev",       S_IRUGO,	   proc_pid_attr_operations),
2298 	REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2299 	REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2300 	REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2301 	REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2302 };
2303 
2304 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2305 {
2306 	return proc_pident_readdir(file, ctx,
2307 				   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2308 }
2309 
2310 static const struct file_operations proc_attr_dir_operations = {
2311 	.read		= generic_read_dir,
2312 	.iterate	= proc_attr_dir_readdir,
2313 	.llseek		= default_llseek,
2314 };
2315 
2316 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2317 				struct dentry *dentry, unsigned int flags)
2318 {
2319 	return proc_pident_lookup(dir, dentry,
2320 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2321 }
2322 
2323 static const struct inode_operations proc_attr_dir_inode_operations = {
2324 	.lookup		= proc_attr_dir_lookup,
2325 	.getattr	= pid_getattr,
2326 	.setattr	= proc_setattr,
2327 };
2328 
2329 #endif
2330 
2331 #ifdef CONFIG_ELF_CORE
2332 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2333 					 size_t count, loff_t *ppos)
2334 {
2335 	struct task_struct *task = get_proc_task(file_inode(file));
2336 	struct mm_struct *mm;
2337 	char buffer[PROC_NUMBUF];
2338 	size_t len;
2339 	int ret;
2340 
2341 	if (!task)
2342 		return -ESRCH;
2343 
2344 	ret = 0;
2345 	mm = get_task_mm(task);
2346 	if (mm) {
2347 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2348 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2349 				MMF_DUMP_FILTER_SHIFT));
2350 		mmput(mm);
2351 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2352 	}
2353 
2354 	put_task_struct(task);
2355 
2356 	return ret;
2357 }
2358 
2359 static ssize_t proc_coredump_filter_write(struct file *file,
2360 					  const char __user *buf,
2361 					  size_t count,
2362 					  loff_t *ppos)
2363 {
2364 	struct task_struct *task;
2365 	struct mm_struct *mm;
2366 	char buffer[PROC_NUMBUF], *end;
2367 	unsigned int val;
2368 	int ret;
2369 	int i;
2370 	unsigned long mask;
2371 
2372 	ret = -EFAULT;
2373 	memset(buffer, 0, sizeof(buffer));
2374 	if (count > sizeof(buffer) - 1)
2375 		count = sizeof(buffer) - 1;
2376 	if (copy_from_user(buffer, buf, count))
2377 		goto out_no_task;
2378 
2379 	ret = -EINVAL;
2380 	val = (unsigned int)simple_strtoul(buffer, &end, 0);
2381 	if (*end == '\n')
2382 		end++;
2383 	if (end - buffer == 0)
2384 		goto out_no_task;
2385 
2386 	ret = -ESRCH;
2387 	task = get_proc_task(file_inode(file));
2388 	if (!task)
2389 		goto out_no_task;
2390 
2391 	ret = end - buffer;
2392 	mm = get_task_mm(task);
2393 	if (!mm)
2394 		goto out_no_mm;
2395 
2396 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2397 		if (val & mask)
2398 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2399 		else
2400 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2401 	}
2402 
2403 	mmput(mm);
2404  out_no_mm:
2405 	put_task_struct(task);
2406  out_no_task:
2407 	return ret;
2408 }
2409 
2410 static const struct file_operations proc_coredump_filter_operations = {
2411 	.read		= proc_coredump_filter_read,
2412 	.write		= proc_coredump_filter_write,
2413 	.llseek		= generic_file_llseek,
2414 };
2415 #endif
2416 
2417 #ifdef CONFIG_TASK_IO_ACCOUNTING
2418 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2419 {
2420 	struct task_io_accounting acct = task->ioac;
2421 	unsigned long flags;
2422 	int result;
2423 
2424 	result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2425 	if (result)
2426 		return result;
2427 
2428 	if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2429 		result = -EACCES;
2430 		goto out_unlock;
2431 	}
2432 
2433 	if (whole && lock_task_sighand(task, &flags)) {
2434 		struct task_struct *t = task;
2435 
2436 		task_io_accounting_add(&acct, &task->signal->ioac);
2437 		while_each_thread(task, t)
2438 			task_io_accounting_add(&acct, &t->ioac);
2439 
2440 		unlock_task_sighand(task, &flags);
2441 	}
2442 	result = sprintf(buffer,
2443 			"rchar: %llu\n"
2444 			"wchar: %llu\n"
2445 			"syscr: %llu\n"
2446 			"syscw: %llu\n"
2447 			"read_bytes: %llu\n"
2448 			"write_bytes: %llu\n"
2449 			"cancelled_write_bytes: %llu\n",
2450 			(unsigned long long)acct.rchar,
2451 			(unsigned long long)acct.wchar,
2452 			(unsigned long long)acct.syscr,
2453 			(unsigned long long)acct.syscw,
2454 			(unsigned long long)acct.read_bytes,
2455 			(unsigned long long)acct.write_bytes,
2456 			(unsigned long long)acct.cancelled_write_bytes);
2457 out_unlock:
2458 	mutex_unlock(&task->signal->cred_guard_mutex);
2459 	return result;
2460 }
2461 
2462 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2463 {
2464 	return do_io_accounting(task, buffer, 0);
2465 }
2466 
2467 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2468 {
2469 	return do_io_accounting(task, buffer, 1);
2470 }
2471 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2472 
2473 #ifdef CONFIG_USER_NS
2474 static int proc_id_map_open(struct inode *inode, struct file *file,
2475 	struct seq_operations *seq_ops)
2476 {
2477 	struct user_namespace *ns = NULL;
2478 	struct task_struct *task;
2479 	struct seq_file *seq;
2480 	int ret = -EINVAL;
2481 
2482 	task = get_proc_task(inode);
2483 	if (task) {
2484 		rcu_read_lock();
2485 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2486 		rcu_read_unlock();
2487 		put_task_struct(task);
2488 	}
2489 	if (!ns)
2490 		goto err;
2491 
2492 	ret = seq_open(file, seq_ops);
2493 	if (ret)
2494 		goto err_put_ns;
2495 
2496 	seq = file->private_data;
2497 	seq->private = ns;
2498 
2499 	return 0;
2500 err_put_ns:
2501 	put_user_ns(ns);
2502 err:
2503 	return ret;
2504 }
2505 
2506 static int proc_id_map_release(struct inode *inode, struct file *file)
2507 {
2508 	struct seq_file *seq = file->private_data;
2509 	struct user_namespace *ns = seq->private;
2510 	put_user_ns(ns);
2511 	return seq_release(inode, file);
2512 }
2513 
2514 static int proc_uid_map_open(struct inode *inode, struct file *file)
2515 {
2516 	return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2517 }
2518 
2519 static int proc_gid_map_open(struct inode *inode, struct file *file)
2520 {
2521 	return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2522 }
2523 
2524 static int proc_projid_map_open(struct inode *inode, struct file *file)
2525 {
2526 	return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2527 }
2528 
2529 static const struct file_operations proc_uid_map_operations = {
2530 	.open		= proc_uid_map_open,
2531 	.write		= proc_uid_map_write,
2532 	.read		= seq_read,
2533 	.llseek		= seq_lseek,
2534 	.release	= proc_id_map_release,
2535 };
2536 
2537 static const struct file_operations proc_gid_map_operations = {
2538 	.open		= proc_gid_map_open,
2539 	.write		= proc_gid_map_write,
2540 	.read		= seq_read,
2541 	.llseek		= seq_lseek,
2542 	.release	= proc_id_map_release,
2543 };
2544 
2545 static const struct file_operations proc_projid_map_operations = {
2546 	.open		= proc_projid_map_open,
2547 	.write		= proc_projid_map_write,
2548 	.read		= seq_read,
2549 	.llseek		= seq_lseek,
2550 	.release	= proc_id_map_release,
2551 };
2552 #endif /* CONFIG_USER_NS */
2553 
2554 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2555 				struct pid *pid, struct task_struct *task)
2556 {
2557 	int err = lock_trace(task);
2558 	if (!err) {
2559 		seq_printf(m, "%08x\n", task->personality);
2560 		unlock_trace(task);
2561 	}
2562 	return err;
2563 }
2564 
2565 /*
2566  * Thread groups
2567  */
2568 static const struct file_operations proc_task_operations;
2569 static const struct inode_operations proc_task_inode_operations;
2570 
2571 static const struct pid_entry tgid_base_stuff[] = {
2572 	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2573 	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2574 #ifdef CONFIG_CHECKPOINT_RESTORE
2575 	DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2576 #endif
2577 	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2578 	DIR("ns",	  S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2579 #ifdef CONFIG_NET
2580 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2581 #endif
2582 	REG("environ",    S_IRUSR, proc_environ_operations),
2583 	INF("auxv",       S_IRUSR, proc_pid_auxv),
2584 	ONE("status",     S_IRUGO, proc_pid_status),
2585 	ONE("personality", S_IRUGO, proc_pid_personality),
2586 	INF("limits",	  S_IRUGO, proc_pid_limits),
2587 #ifdef CONFIG_SCHED_DEBUG
2588 	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2589 #endif
2590 #ifdef CONFIG_SCHED_AUTOGROUP
2591 	REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2592 #endif
2593 	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2594 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2595 	INF("syscall",    S_IRUGO, proc_pid_syscall),
2596 #endif
2597 	INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2598 	ONE("stat",       S_IRUGO, proc_tgid_stat),
2599 	ONE("statm",      S_IRUGO, proc_pid_statm),
2600 	REG("maps",       S_IRUGO, proc_pid_maps_operations),
2601 #ifdef CONFIG_NUMA
2602 	REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2603 #endif
2604 	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2605 	LNK("cwd",        proc_cwd_link),
2606 	LNK("root",       proc_root_link),
2607 	LNK("exe",        proc_exe_link),
2608 	REG("mounts",     S_IRUGO, proc_mounts_operations),
2609 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2610 	REG("mountstats", S_IRUSR, proc_mountstats_operations),
2611 #ifdef CONFIG_PROC_PAGE_MONITOR
2612 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2613 	REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2614 	REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2615 #endif
2616 #ifdef CONFIG_SECURITY
2617 	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2618 #endif
2619 #ifdef CONFIG_KALLSYMS
2620 	INF("wchan",      S_IRUGO, proc_pid_wchan),
2621 #endif
2622 #ifdef CONFIG_STACKTRACE
2623 	ONE("stack",      S_IRUGO, proc_pid_stack),
2624 #endif
2625 #ifdef CONFIG_SCHEDSTATS
2626 	INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2627 #endif
2628 #ifdef CONFIG_LATENCYTOP
2629 	REG("latency",  S_IRUGO, proc_lstats_operations),
2630 #endif
2631 #ifdef CONFIG_PROC_PID_CPUSET
2632 	REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2633 #endif
2634 #ifdef CONFIG_CGROUPS
2635 	REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2636 #endif
2637 	INF("oom_score",  S_IRUGO, proc_oom_score),
2638 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2639 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2640 #ifdef CONFIG_AUDITSYSCALL
2641 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2642 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2643 #endif
2644 #ifdef CONFIG_FAULT_INJECTION
2645 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2646 #endif
2647 #ifdef CONFIG_ELF_CORE
2648 	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2649 #endif
2650 #ifdef CONFIG_TASK_IO_ACCOUNTING
2651 	INF("io",	S_IRUSR, proc_tgid_io_accounting),
2652 #endif
2653 #ifdef CONFIG_HARDWALL
2654 	INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2655 #endif
2656 #ifdef CONFIG_USER_NS
2657 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2658 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2659 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2660 #endif
2661 #ifdef CONFIG_CHECKPOINT_RESTORE
2662 	REG("timers",	  S_IRUGO, proc_timers_operations),
2663 #endif
2664 };
2665 
2666 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2667 {
2668 	return proc_pident_readdir(file, ctx,
2669 				   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2670 }
2671 
2672 static const struct file_operations proc_tgid_base_operations = {
2673 	.read		= generic_read_dir,
2674 	.iterate	= proc_tgid_base_readdir,
2675 	.llseek		= default_llseek,
2676 };
2677 
2678 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2679 {
2680 	return proc_pident_lookup(dir, dentry,
2681 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2682 }
2683 
2684 static const struct inode_operations proc_tgid_base_inode_operations = {
2685 	.lookup		= proc_tgid_base_lookup,
2686 	.getattr	= pid_getattr,
2687 	.setattr	= proc_setattr,
2688 	.permission	= proc_pid_permission,
2689 };
2690 
2691 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2692 {
2693 	struct dentry *dentry, *leader, *dir;
2694 	char buf[PROC_NUMBUF];
2695 	struct qstr name;
2696 
2697 	name.name = buf;
2698 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2699 	/* no ->d_hash() rejects on procfs */
2700 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2701 	if (dentry) {
2702 		shrink_dcache_parent(dentry);
2703 		d_drop(dentry);
2704 		dput(dentry);
2705 	}
2706 
2707 	name.name = buf;
2708 	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2709 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2710 	if (!leader)
2711 		goto out;
2712 
2713 	name.name = "task";
2714 	name.len = strlen(name.name);
2715 	dir = d_hash_and_lookup(leader, &name);
2716 	if (!dir)
2717 		goto out_put_leader;
2718 
2719 	name.name = buf;
2720 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2721 	dentry = d_hash_and_lookup(dir, &name);
2722 	if (dentry) {
2723 		shrink_dcache_parent(dentry);
2724 		d_drop(dentry);
2725 		dput(dentry);
2726 	}
2727 
2728 	dput(dir);
2729 out_put_leader:
2730 	dput(leader);
2731 out:
2732 	return;
2733 }
2734 
2735 /**
2736  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2737  * @task: task that should be flushed.
2738  *
2739  * When flushing dentries from proc, one needs to flush them from global
2740  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2741  * in. This call is supposed to do all of this job.
2742  *
2743  * Looks in the dcache for
2744  * /proc/@pid
2745  * /proc/@tgid/task/@pid
2746  * if either directory is present flushes it and all of it'ts children
2747  * from the dcache.
2748  *
2749  * It is safe and reasonable to cache /proc entries for a task until
2750  * that task exits.  After that they just clog up the dcache with
2751  * useless entries, possibly causing useful dcache entries to be
2752  * flushed instead.  This routine is proved to flush those useless
2753  * dcache entries at process exit time.
2754  *
2755  * NOTE: This routine is just an optimization so it does not guarantee
2756  *       that no dcache entries will exist at process exit time it
2757  *       just makes it very unlikely that any will persist.
2758  */
2759 
2760 void proc_flush_task(struct task_struct *task)
2761 {
2762 	int i;
2763 	struct pid *pid, *tgid;
2764 	struct upid *upid;
2765 
2766 	pid = task_pid(task);
2767 	tgid = task_tgid(task);
2768 
2769 	for (i = 0; i <= pid->level; i++) {
2770 		upid = &pid->numbers[i];
2771 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2772 					tgid->numbers[i].nr);
2773 	}
2774 }
2775 
2776 static int proc_pid_instantiate(struct inode *dir,
2777 				   struct dentry * dentry,
2778 				   struct task_struct *task, const void *ptr)
2779 {
2780 	struct inode *inode;
2781 
2782 	inode = proc_pid_make_inode(dir->i_sb, task);
2783 	if (!inode)
2784 		goto out;
2785 
2786 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2787 	inode->i_op = &proc_tgid_base_inode_operations;
2788 	inode->i_fop = &proc_tgid_base_operations;
2789 	inode->i_flags|=S_IMMUTABLE;
2790 
2791 	set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2792 						  ARRAY_SIZE(tgid_base_stuff)));
2793 
2794 	d_set_d_op(dentry, &pid_dentry_operations);
2795 
2796 	d_add(dentry, inode);
2797 	/* Close the race of the process dying before we return the dentry */
2798 	if (pid_revalidate(dentry, 0))
2799 		return 0;
2800 out:
2801 	return -ENOENT;
2802 }
2803 
2804 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2805 {
2806 	int result = 0;
2807 	struct task_struct *task;
2808 	unsigned tgid;
2809 	struct pid_namespace *ns;
2810 
2811 	tgid = name_to_int(dentry);
2812 	if (tgid == ~0U)
2813 		goto out;
2814 
2815 	ns = dentry->d_sb->s_fs_info;
2816 	rcu_read_lock();
2817 	task = find_task_by_pid_ns(tgid, ns);
2818 	if (task)
2819 		get_task_struct(task);
2820 	rcu_read_unlock();
2821 	if (!task)
2822 		goto out;
2823 
2824 	result = proc_pid_instantiate(dir, dentry, task, NULL);
2825 	put_task_struct(task);
2826 out:
2827 	return ERR_PTR(result);
2828 }
2829 
2830 /*
2831  * Find the first task with tgid >= tgid
2832  *
2833  */
2834 struct tgid_iter {
2835 	unsigned int tgid;
2836 	struct task_struct *task;
2837 };
2838 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2839 {
2840 	struct pid *pid;
2841 
2842 	if (iter.task)
2843 		put_task_struct(iter.task);
2844 	rcu_read_lock();
2845 retry:
2846 	iter.task = NULL;
2847 	pid = find_ge_pid(iter.tgid, ns);
2848 	if (pid) {
2849 		iter.tgid = pid_nr_ns(pid, ns);
2850 		iter.task = pid_task(pid, PIDTYPE_PID);
2851 		/* What we to know is if the pid we have find is the
2852 		 * pid of a thread_group_leader.  Testing for task
2853 		 * being a thread_group_leader is the obvious thing
2854 		 * todo but there is a window when it fails, due to
2855 		 * the pid transfer logic in de_thread.
2856 		 *
2857 		 * So we perform the straight forward test of seeing
2858 		 * if the pid we have found is the pid of a thread
2859 		 * group leader, and don't worry if the task we have
2860 		 * found doesn't happen to be a thread group leader.
2861 		 * As we don't care in the case of readdir.
2862 		 */
2863 		if (!iter.task || !has_group_leader_pid(iter.task)) {
2864 			iter.tgid += 1;
2865 			goto retry;
2866 		}
2867 		get_task_struct(iter.task);
2868 	}
2869 	rcu_read_unlock();
2870 	return iter;
2871 }
2872 
2873 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 1)
2874 
2875 /* for the /proc/ directory itself, after non-process stuff has been done */
2876 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
2877 {
2878 	struct tgid_iter iter;
2879 	struct pid_namespace *ns = file->f_dentry->d_sb->s_fs_info;
2880 	loff_t pos = ctx->pos;
2881 
2882 	if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
2883 		return 0;
2884 
2885 	if (pos == TGID_OFFSET - 1) {
2886 		struct inode *inode = ns->proc_self->d_inode;
2887 		if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
2888 			return 0;
2889 		iter.tgid = 0;
2890 	} else {
2891 		iter.tgid = pos - TGID_OFFSET;
2892 	}
2893 	iter.task = NULL;
2894 	for (iter = next_tgid(ns, iter);
2895 	     iter.task;
2896 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
2897 		char name[PROC_NUMBUF];
2898 		int len;
2899 		if (!has_pid_permissions(ns, iter.task, 2))
2900 			continue;
2901 
2902 		len = snprintf(name, sizeof(name), "%d", iter.tgid);
2903 		ctx->pos = iter.tgid + TGID_OFFSET;
2904 		if (!proc_fill_cache(file, ctx, name, len,
2905 				     proc_pid_instantiate, iter.task, NULL)) {
2906 			put_task_struct(iter.task);
2907 			return 0;
2908 		}
2909 	}
2910 	ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
2911 	return 0;
2912 }
2913 
2914 /*
2915  * Tasks
2916  */
2917 static const struct pid_entry tid_base_stuff[] = {
2918 	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2919 	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2920 	DIR("ns",	 S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2921 	REG("environ",   S_IRUSR, proc_environ_operations),
2922 	INF("auxv",      S_IRUSR, proc_pid_auxv),
2923 	ONE("status",    S_IRUGO, proc_pid_status),
2924 	ONE("personality", S_IRUGO, proc_pid_personality),
2925 	INF("limits",	 S_IRUGO, proc_pid_limits),
2926 #ifdef CONFIG_SCHED_DEBUG
2927 	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2928 #endif
2929 	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2930 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2931 	INF("syscall",   S_IRUGO, proc_pid_syscall),
2932 #endif
2933 	INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2934 	ONE("stat",      S_IRUGO, proc_tid_stat),
2935 	ONE("statm",     S_IRUGO, proc_pid_statm),
2936 	REG("maps",      S_IRUGO, proc_tid_maps_operations),
2937 #ifdef CONFIG_CHECKPOINT_RESTORE
2938 	REG("children",  S_IRUGO, proc_tid_children_operations),
2939 #endif
2940 #ifdef CONFIG_NUMA
2941 	REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2942 #endif
2943 	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2944 	LNK("cwd",       proc_cwd_link),
2945 	LNK("root",      proc_root_link),
2946 	LNK("exe",       proc_exe_link),
2947 	REG("mounts",    S_IRUGO, proc_mounts_operations),
2948 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2949 #ifdef CONFIG_PROC_PAGE_MONITOR
2950 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2951 	REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2952 	REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2953 #endif
2954 #ifdef CONFIG_SECURITY
2955 	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2956 #endif
2957 #ifdef CONFIG_KALLSYMS
2958 	INF("wchan",     S_IRUGO, proc_pid_wchan),
2959 #endif
2960 #ifdef CONFIG_STACKTRACE
2961 	ONE("stack",      S_IRUGO, proc_pid_stack),
2962 #endif
2963 #ifdef CONFIG_SCHEDSTATS
2964 	INF("schedstat", S_IRUGO, proc_pid_schedstat),
2965 #endif
2966 #ifdef CONFIG_LATENCYTOP
2967 	REG("latency",  S_IRUGO, proc_lstats_operations),
2968 #endif
2969 #ifdef CONFIG_PROC_PID_CPUSET
2970 	REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2971 #endif
2972 #ifdef CONFIG_CGROUPS
2973 	REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2974 #endif
2975 	INF("oom_score", S_IRUGO, proc_oom_score),
2976 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2977 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2978 #ifdef CONFIG_AUDITSYSCALL
2979 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2980 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2981 #endif
2982 #ifdef CONFIG_FAULT_INJECTION
2983 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2984 #endif
2985 #ifdef CONFIG_TASK_IO_ACCOUNTING
2986 	INF("io",	S_IRUSR, proc_tid_io_accounting),
2987 #endif
2988 #ifdef CONFIG_HARDWALL
2989 	INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2990 #endif
2991 #ifdef CONFIG_USER_NS
2992 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2993 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2994 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2995 #endif
2996 };
2997 
2998 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
2999 {
3000 	return proc_pident_readdir(file, ctx,
3001 				   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3002 }
3003 
3004 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3005 {
3006 	return proc_pident_lookup(dir, dentry,
3007 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3008 }
3009 
3010 static const struct file_operations proc_tid_base_operations = {
3011 	.read		= generic_read_dir,
3012 	.iterate	= proc_tid_base_readdir,
3013 	.llseek		= default_llseek,
3014 };
3015 
3016 static const struct inode_operations proc_tid_base_inode_operations = {
3017 	.lookup		= proc_tid_base_lookup,
3018 	.getattr	= pid_getattr,
3019 	.setattr	= proc_setattr,
3020 };
3021 
3022 static int proc_task_instantiate(struct inode *dir,
3023 	struct dentry *dentry, struct task_struct *task, const void *ptr)
3024 {
3025 	struct inode *inode;
3026 	inode = proc_pid_make_inode(dir->i_sb, task);
3027 
3028 	if (!inode)
3029 		goto out;
3030 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3031 	inode->i_op = &proc_tid_base_inode_operations;
3032 	inode->i_fop = &proc_tid_base_operations;
3033 	inode->i_flags|=S_IMMUTABLE;
3034 
3035 	set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3036 						  ARRAY_SIZE(tid_base_stuff)));
3037 
3038 	d_set_d_op(dentry, &pid_dentry_operations);
3039 
3040 	d_add(dentry, inode);
3041 	/* Close the race of the process dying before we return the dentry */
3042 	if (pid_revalidate(dentry, 0))
3043 		return 0;
3044 out:
3045 	return -ENOENT;
3046 }
3047 
3048 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3049 {
3050 	int result = -ENOENT;
3051 	struct task_struct *task;
3052 	struct task_struct *leader = get_proc_task(dir);
3053 	unsigned tid;
3054 	struct pid_namespace *ns;
3055 
3056 	if (!leader)
3057 		goto out_no_task;
3058 
3059 	tid = name_to_int(dentry);
3060 	if (tid == ~0U)
3061 		goto out;
3062 
3063 	ns = dentry->d_sb->s_fs_info;
3064 	rcu_read_lock();
3065 	task = find_task_by_pid_ns(tid, ns);
3066 	if (task)
3067 		get_task_struct(task);
3068 	rcu_read_unlock();
3069 	if (!task)
3070 		goto out;
3071 	if (!same_thread_group(leader, task))
3072 		goto out_drop_task;
3073 
3074 	result = proc_task_instantiate(dir, dentry, task, NULL);
3075 out_drop_task:
3076 	put_task_struct(task);
3077 out:
3078 	put_task_struct(leader);
3079 out_no_task:
3080 	return ERR_PTR(result);
3081 }
3082 
3083 /*
3084  * Find the first tid of a thread group to return to user space.
3085  *
3086  * Usually this is just the thread group leader, but if the users
3087  * buffer was too small or there was a seek into the middle of the
3088  * directory we have more work todo.
3089  *
3090  * In the case of a short read we start with find_task_by_pid.
3091  *
3092  * In the case of a seek we start with the leader and walk nr
3093  * threads past it.
3094  */
3095 static struct task_struct *first_tid(struct task_struct *leader,
3096 		int tid, int nr, struct pid_namespace *ns)
3097 {
3098 	struct task_struct *pos;
3099 
3100 	rcu_read_lock();
3101 	/* Attempt to start with the pid of a thread */
3102 	if (tid && (nr > 0)) {
3103 		pos = find_task_by_pid_ns(tid, ns);
3104 		if (pos && (pos->group_leader == leader))
3105 			goto found;
3106 	}
3107 
3108 	/* If nr exceeds the number of threads there is nothing todo */
3109 	pos = NULL;
3110 	if (nr && nr >= get_nr_threads(leader))
3111 		goto out;
3112 
3113 	/* If we haven't found our starting place yet start
3114 	 * with the leader and walk nr threads forward.
3115 	 */
3116 	for (pos = leader; nr > 0; --nr) {
3117 		pos = next_thread(pos);
3118 		if (pos == leader) {
3119 			pos = NULL;
3120 			goto out;
3121 		}
3122 	}
3123 found:
3124 	get_task_struct(pos);
3125 out:
3126 	rcu_read_unlock();
3127 	return pos;
3128 }
3129 
3130 /*
3131  * Find the next thread in the thread list.
3132  * Return NULL if there is an error or no next thread.
3133  *
3134  * The reference to the input task_struct is released.
3135  */
3136 static struct task_struct *next_tid(struct task_struct *start)
3137 {
3138 	struct task_struct *pos = NULL;
3139 	rcu_read_lock();
3140 	if (pid_alive(start)) {
3141 		pos = next_thread(start);
3142 		if (thread_group_leader(pos))
3143 			pos = NULL;
3144 		else
3145 			get_task_struct(pos);
3146 	}
3147 	rcu_read_unlock();
3148 	put_task_struct(start);
3149 	return pos;
3150 }
3151 
3152 /* for the /proc/TGID/task/ directories */
3153 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3154 {
3155 	struct task_struct *leader = NULL;
3156 	struct task_struct *task = get_proc_task(file_inode(file));
3157 	struct pid_namespace *ns;
3158 	int tid;
3159 
3160 	if (!task)
3161 		return -ENOENT;
3162 	rcu_read_lock();
3163 	if (pid_alive(task)) {
3164 		leader = task->group_leader;
3165 		get_task_struct(leader);
3166 	}
3167 	rcu_read_unlock();
3168 	put_task_struct(task);
3169 	if (!leader)
3170 		return -ENOENT;
3171 
3172 	if (!dir_emit_dots(file, ctx))
3173 		goto out;
3174 
3175 	/* f_version caches the tgid value that the last readdir call couldn't
3176 	 * return. lseek aka telldir automagically resets f_version to 0.
3177 	 */
3178 	ns = file->f_dentry->d_sb->s_fs_info;
3179 	tid = (int)file->f_version;
3180 	file->f_version = 0;
3181 	for (task = first_tid(leader, tid, ctx->pos - 2, ns);
3182 	     task;
3183 	     task = next_tid(task), ctx->pos++) {
3184 		char name[PROC_NUMBUF];
3185 		int len;
3186 		tid = task_pid_nr_ns(task, ns);
3187 		len = snprintf(name, sizeof(name), "%d", tid);
3188 		if (!proc_fill_cache(file, ctx, name, len,
3189 				proc_task_instantiate, task, NULL)) {
3190 			/* returning this tgid failed, save it as the first
3191 			 * pid for the next readir call */
3192 			file->f_version = (u64)tid;
3193 			put_task_struct(task);
3194 			break;
3195 		}
3196 	}
3197 out:
3198 	put_task_struct(leader);
3199 	return 0;
3200 }
3201 
3202 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3203 {
3204 	struct inode *inode = dentry->d_inode;
3205 	struct task_struct *p = get_proc_task(inode);
3206 	generic_fillattr(inode, stat);
3207 
3208 	if (p) {
3209 		stat->nlink += get_nr_threads(p);
3210 		put_task_struct(p);
3211 	}
3212 
3213 	return 0;
3214 }
3215 
3216 static const struct inode_operations proc_task_inode_operations = {
3217 	.lookup		= proc_task_lookup,
3218 	.getattr	= proc_task_getattr,
3219 	.setattr	= proc_setattr,
3220 	.permission	= proc_pid_permission,
3221 };
3222 
3223 static const struct file_operations proc_task_operations = {
3224 	.read		= generic_read_dir,
3225 	.iterate	= proc_task_readdir,
3226 	.llseek		= default_llseek,
3227 };
3228