xref: /linux/fs/proc/base.c (revision ed3174d93c342b8b2eeba6bbd124707d55304a7b)
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/init.h>
57 #include <linux/capability.h>
58 #include <linux/file.h>
59 #include <linux/string.h>
60 #include <linux/seq_file.h>
61 #include <linux/namei.h>
62 #include <linux/mnt_namespace.h>
63 #include <linux/mm.h>
64 #include <linux/rcupdate.h>
65 #include <linux/kallsyms.h>
66 #include <linux/resource.h>
67 #include <linux/module.h>
68 #include <linux/mount.h>
69 #include <linux/security.h>
70 #include <linux/ptrace.h>
71 #include <linux/cgroup.h>
72 #include <linux/cpuset.h>
73 #include <linux/audit.h>
74 #include <linux/poll.h>
75 #include <linux/nsproxy.h>
76 #include <linux/oom.h>
77 #include <linux/elf.h>
78 #include <linux/pid_namespace.h>
79 #include "internal.h"
80 
81 /* NOTE:
82  *	Implementing inode permission operations in /proc is almost
83  *	certainly an error.  Permission checks need to happen during
84  *	each system call not at open time.  The reason is that most of
85  *	what we wish to check for permissions in /proc varies at runtime.
86  *
87  *	The classic example of a problem is opening file descriptors
88  *	in /proc for a task before it execs a suid executable.
89  */
90 
91 struct pid_entry {
92 	char *name;
93 	int len;
94 	mode_t mode;
95 	const struct inode_operations *iop;
96 	const struct file_operations *fop;
97 	union proc_op op;
98 };
99 
100 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
101 	.name = (NAME),					\
102 	.len  = sizeof(NAME) - 1,			\
103 	.mode = MODE,					\
104 	.iop  = IOP,					\
105 	.fop  = FOP,					\
106 	.op   = OP,					\
107 }
108 
109 #define DIR(NAME, MODE, OTYPE)							\
110 	NOD(NAME, (S_IFDIR|(MODE)),						\
111 		&proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,	\
112 		{} )
113 #define LNK(NAME, OTYPE)					\
114 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
115 		&proc_pid_link_inode_operations, NULL,		\
116 		{ .proc_get_link = &proc_##OTYPE##_link } )
117 #define REG(NAME, MODE, OTYPE)				\
118 	NOD(NAME, (S_IFREG|(MODE)), NULL,		\
119 		&proc_##OTYPE##_operations, {})
120 #define INF(NAME, MODE, OTYPE)				\
121 	NOD(NAME, (S_IFREG|(MODE)), 			\
122 		NULL, &proc_info_file_operations,	\
123 		{ .proc_read = &proc_##OTYPE } )
124 #define ONE(NAME, MODE, OTYPE)				\
125 	NOD(NAME, (S_IFREG|(MODE)), 			\
126 		NULL, &proc_single_file_operations,	\
127 		{ .proc_show = &proc_##OTYPE } )
128 
129 int maps_protect;
130 EXPORT_SYMBOL(maps_protect);
131 
132 static struct fs_struct *get_fs_struct(struct task_struct *task)
133 {
134 	struct fs_struct *fs;
135 	task_lock(task);
136 	fs = task->fs;
137 	if(fs)
138 		atomic_inc(&fs->count);
139 	task_unlock(task);
140 	return fs;
141 }
142 
143 static int get_nr_threads(struct task_struct *tsk)
144 {
145 	/* Must be called with the rcu_read_lock held */
146 	unsigned long flags;
147 	int count = 0;
148 
149 	if (lock_task_sighand(tsk, &flags)) {
150 		count = atomic_read(&tsk->signal->count);
151 		unlock_task_sighand(tsk, &flags);
152 	}
153 	return count;
154 }
155 
156 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
157 {
158 	struct task_struct *task = get_proc_task(inode);
159 	struct fs_struct *fs = NULL;
160 	int result = -ENOENT;
161 
162 	if (task) {
163 		fs = get_fs_struct(task);
164 		put_task_struct(task);
165 	}
166 	if (fs) {
167 		read_lock(&fs->lock);
168 		*mnt = mntget(fs->pwdmnt);
169 		*dentry = dget(fs->pwd);
170 		read_unlock(&fs->lock);
171 		result = 0;
172 		put_fs_struct(fs);
173 	}
174 	return result;
175 }
176 
177 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
178 {
179 	struct task_struct *task = get_proc_task(inode);
180 	struct fs_struct *fs = NULL;
181 	int result = -ENOENT;
182 
183 	if (task) {
184 		fs = get_fs_struct(task);
185 		put_task_struct(task);
186 	}
187 	if (fs) {
188 		read_lock(&fs->lock);
189 		*mnt = mntget(fs->rootmnt);
190 		*dentry = dget(fs->root);
191 		read_unlock(&fs->lock);
192 		result = 0;
193 		put_fs_struct(fs);
194 	}
195 	return result;
196 }
197 
198 #define MAY_PTRACE(task) \
199 	(task == current || \
200 	(task->parent == current && \
201 	(task->ptrace & PT_PTRACED) && \
202 	 (task_is_stopped_or_traced(task)) && \
203 	 security_ptrace(current,task) == 0))
204 
205 struct mm_struct *mm_for_maps(struct task_struct *task)
206 {
207 	struct mm_struct *mm = get_task_mm(task);
208 	if (!mm)
209 		return NULL;
210 	down_read(&mm->mmap_sem);
211 	task_lock(task);
212 	if (task->mm != mm)
213 		goto out;
214 	if (task->mm != current->mm && __ptrace_may_attach(task) < 0)
215 		goto out;
216 	task_unlock(task);
217 	return mm;
218 out:
219 	task_unlock(task);
220 	up_read(&mm->mmap_sem);
221 	mmput(mm);
222 	return NULL;
223 }
224 
225 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
226 {
227 	int res = 0;
228 	unsigned int len;
229 	struct mm_struct *mm = get_task_mm(task);
230 	if (!mm)
231 		goto out;
232 	if (!mm->arg_end)
233 		goto out_mm;	/* Shh! No looking before we're done */
234 
235  	len = mm->arg_end - mm->arg_start;
236 
237 	if (len > PAGE_SIZE)
238 		len = PAGE_SIZE;
239 
240 	res = access_process_vm(task, mm->arg_start, buffer, len, 0);
241 
242 	// If the nul at the end of args has been overwritten, then
243 	// assume application is using setproctitle(3).
244 	if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
245 		len = strnlen(buffer, res);
246 		if (len < res) {
247 		    res = len;
248 		} else {
249 			len = mm->env_end - mm->env_start;
250 			if (len > PAGE_SIZE - res)
251 				len = PAGE_SIZE - res;
252 			res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
253 			res = strnlen(buffer, res);
254 		}
255 	}
256 out_mm:
257 	mmput(mm);
258 out:
259 	return res;
260 }
261 
262 static int proc_pid_auxv(struct task_struct *task, char *buffer)
263 {
264 	int res = 0;
265 	struct mm_struct *mm = get_task_mm(task);
266 	if (mm) {
267 		unsigned int nwords = 0;
268 		do
269 			nwords += 2;
270 		while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
271 		res = nwords * sizeof(mm->saved_auxv[0]);
272 		if (res > PAGE_SIZE)
273 			res = PAGE_SIZE;
274 		memcpy(buffer, mm->saved_auxv, res);
275 		mmput(mm);
276 	}
277 	return res;
278 }
279 
280 
281 #ifdef CONFIG_KALLSYMS
282 /*
283  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
284  * Returns the resolved symbol.  If that fails, simply return the address.
285  */
286 static int proc_pid_wchan(struct task_struct *task, char *buffer)
287 {
288 	unsigned long wchan;
289 	char symname[KSYM_NAME_LEN];
290 
291 	wchan = get_wchan(task);
292 
293 	if (lookup_symbol_name(wchan, symname) < 0)
294 		return sprintf(buffer, "%lu", wchan);
295 	else
296 		return sprintf(buffer, "%s", symname);
297 }
298 #endif /* CONFIG_KALLSYMS */
299 
300 #ifdef CONFIG_SCHEDSTATS
301 /*
302  * Provides /proc/PID/schedstat
303  */
304 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
305 {
306 	return sprintf(buffer, "%llu %llu %lu\n",
307 			task->sched_info.cpu_time,
308 			task->sched_info.run_delay,
309 			task->sched_info.pcount);
310 }
311 #endif
312 
313 #ifdef CONFIG_LATENCYTOP
314 static int lstats_show_proc(struct seq_file *m, void *v)
315 {
316 	int i;
317 	struct task_struct *task = m->private;
318 	seq_puts(m, "Latency Top version : v0.1\n");
319 
320 	for (i = 0; i < 32; i++) {
321 		if (task->latency_record[i].backtrace[0]) {
322 			int q;
323 			seq_printf(m, "%i %li %li ",
324 				task->latency_record[i].count,
325 				task->latency_record[i].time,
326 				task->latency_record[i].max);
327 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
328 				char sym[KSYM_NAME_LEN];
329 				char *c;
330 				if (!task->latency_record[i].backtrace[q])
331 					break;
332 				if (task->latency_record[i].backtrace[q] == ULONG_MAX)
333 					break;
334 				sprint_symbol(sym, task->latency_record[i].backtrace[q]);
335 				c = strchr(sym, '+');
336 				if (c)
337 					*c = 0;
338 				seq_printf(m, "%s ", sym);
339 			}
340 			seq_printf(m, "\n");
341 		}
342 
343 	}
344 	return 0;
345 }
346 
347 static int lstats_open(struct inode *inode, struct file *file)
348 {
349 	int ret;
350 	struct seq_file *m;
351 	struct task_struct *task = get_proc_task(inode);
352 
353 	ret = single_open(file, lstats_show_proc, NULL);
354 	if (!ret) {
355 		m = file->private_data;
356 		m->private = task;
357 	}
358 	return ret;
359 }
360 
361 static ssize_t lstats_write(struct file *file, const char __user *buf,
362 			    size_t count, loff_t *offs)
363 {
364 	struct seq_file *m;
365 	struct task_struct *task;
366 
367 	m = file->private_data;
368 	task = m->private;
369 	clear_all_latency_tracing(task);
370 
371 	return count;
372 }
373 
374 static const struct file_operations proc_lstats_operations = {
375 	.open		= lstats_open,
376 	.read		= seq_read,
377 	.write		= lstats_write,
378 	.llseek		= seq_lseek,
379 	.release	= single_release,
380 };
381 
382 #endif
383 
384 /* The badness from the OOM killer */
385 unsigned long badness(struct task_struct *p, unsigned long uptime);
386 static int proc_oom_score(struct task_struct *task, char *buffer)
387 {
388 	unsigned long points;
389 	struct timespec uptime;
390 
391 	do_posix_clock_monotonic_gettime(&uptime);
392 	read_lock(&tasklist_lock);
393 	points = badness(task, uptime.tv_sec);
394 	read_unlock(&tasklist_lock);
395 	return sprintf(buffer, "%lu\n", points);
396 }
397 
398 struct limit_names {
399 	char *name;
400 	char *unit;
401 };
402 
403 static const struct limit_names lnames[RLIM_NLIMITS] = {
404 	[RLIMIT_CPU] = {"Max cpu time", "ms"},
405 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
406 	[RLIMIT_DATA] = {"Max data size", "bytes"},
407 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
408 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
409 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
410 	[RLIMIT_NPROC] = {"Max processes", "processes"},
411 	[RLIMIT_NOFILE] = {"Max open files", "files"},
412 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
413 	[RLIMIT_AS] = {"Max address space", "bytes"},
414 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
415 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
416 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
417 	[RLIMIT_NICE] = {"Max nice priority", NULL},
418 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
419 };
420 
421 /* Display limits for a process */
422 static int proc_pid_limits(struct task_struct *task, char *buffer)
423 {
424 	unsigned int i;
425 	int count = 0;
426 	unsigned long flags;
427 	char *bufptr = buffer;
428 
429 	struct rlimit rlim[RLIM_NLIMITS];
430 
431 	rcu_read_lock();
432 	if (!lock_task_sighand(task,&flags)) {
433 		rcu_read_unlock();
434 		return 0;
435 	}
436 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
437 	unlock_task_sighand(task, &flags);
438 	rcu_read_unlock();
439 
440 	/*
441 	 * print the file header
442 	 */
443 	count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
444 			"Limit", "Soft Limit", "Hard Limit", "Units");
445 
446 	for (i = 0; i < RLIM_NLIMITS; i++) {
447 		if (rlim[i].rlim_cur == RLIM_INFINITY)
448 			count += sprintf(&bufptr[count], "%-25s %-20s ",
449 					 lnames[i].name, "unlimited");
450 		else
451 			count += sprintf(&bufptr[count], "%-25s %-20lu ",
452 					 lnames[i].name, rlim[i].rlim_cur);
453 
454 		if (rlim[i].rlim_max == RLIM_INFINITY)
455 			count += sprintf(&bufptr[count], "%-20s ", "unlimited");
456 		else
457 			count += sprintf(&bufptr[count], "%-20lu ",
458 					 rlim[i].rlim_max);
459 
460 		if (lnames[i].unit)
461 			count += sprintf(&bufptr[count], "%-10s\n",
462 					 lnames[i].unit);
463 		else
464 			count += sprintf(&bufptr[count], "\n");
465 	}
466 
467 	return count;
468 }
469 
470 /************************************************************************/
471 /*                       Here the fs part begins                        */
472 /************************************************************************/
473 
474 /* permission checks */
475 static int proc_fd_access_allowed(struct inode *inode)
476 {
477 	struct task_struct *task;
478 	int allowed = 0;
479 	/* Allow access to a task's file descriptors if it is us or we
480 	 * may use ptrace attach to the process and find out that
481 	 * information.
482 	 */
483 	task = get_proc_task(inode);
484 	if (task) {
485 		allowed = ptrace_may_attach(task);
486 		put_task_struct(task);
487 	}
488 	return allowed;
489 }
490 
491 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
492 {
493 	int error;
494 	struct inode *inode = dentry->d_inode;
495 
496 	if (attr->ia_valid & ATTR_MODE)
497 		return -EPERM;
498 
499 	error = inode_change_ok(inode, attr);
500 	if (!error)
501 		error = inode_setattr(inode, attr);
502 	return error;
503 }
504 
505 static const struct inode_operations proc_def_inode_operations = {
506 	.setattr	= proc_setattr,
507 };
508 
509 extern const struct seq_operations mounts_op;
510 struct proc_mounts {
511 	struct seq_file m;
512 	int event;
513 };
514 
515 static int mounts_open(struct inode *inode, struct file *file)
516 {
517 	struct task_struct *task = get_proc_task(inode);
518 	struct nsproxy *nsp;
519 	struct mnt_namespace *ns = NULL;
520 	struct proc_mounts *p;
521 	int ret = -EINVAL;
522 
523 	if (task) {
524 		rcu_read_lock();
525 		nsp = task_nsproxy(task);
526 		if (nsp) {
527 			ns = nsp->mnt_ns;
528 			if (ns)
529 				get_mnt_ns(ns);
530 		}
531 		rcu_read_unlock();
532 
533 		put_task_struct(task);
534 	}
535 
536 	if (ns) {
537 		ret = -ENOMEM;
538 		p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
539 		if (p) {
540 			file->private_data = &p->m;
541 			ret = seq_open(file, &mounts_op);
542 			if (!ret) {
543 				p->m.private = ns;
544 				p->event = ns->event;
545 				return 0;
546 			}
547 			kfree(p);
548 		}
549 		put_mnt_ns(ns);
550 	}
551 	return ret;
552 }
553 
554 static int mounts_release(struct inode *inode, struct file *file)
555 {
556 	struct seq_file *m = file->private_data;
557 	struct mnt_namespace *ns = m->private;
558 	put_mnt_ns(ns);
559 	return seq_release(inode, file);
560 }
561 
562 static unsigned mounts_poll(struct file *file, poll_table *wait)
563 {
564 	struct proc_mounts *p = file->private_data;
565 	struct mnt_namespace *ns = p->m.private;
566 	unsigned res = 0;
567 
568 	poll_wait(file, &ns->poll, wait);
569 
570 	spin_lock(&vfsmount_lock);
571 	if (p->event != ns->event) {
572 		p->event = ns->event;
573 		res = POLLERR;
574 	}
575 	spin_unlock(&vfsmount_lock);
576 
577 	return res;
578 }
579 
580 static const struct file_operations proc_mounts_operations = {
581 	.open		= mounts_open,
582 	.read		= seq_read,
583 	.llseek		= seq_lseek,
584 	.release	= mounts_release,
585 	.poll		= mounts_poll,
586 };
587 
588 extern const struct seq_operations mountstats_op;
589 static int mountstats_open(struct inode *inode, struct file *file)
590 {
591 	int ret = seq_open(file, &mountstats_op);
592 
593 	if (!ret) {
594 		struct seq_file *m = file->private_data;
595 		struct nsproxy *nsp;
596 		struct mnt_namespace *mnt_ns = NULL;
597 		struct task_struct *task = get_proc_task(inode);
598 
599 		if (task) {
600 			rcu_read_lock();
601 			nsp = task_nsproxy(task);
602 			if (nsp) {
603 				mnt_ns = nsp->mnt_ns;
604 				if (mnt_ns)
605 					get_mnt_ns(mnt_ns);
606 			}
607 			rcu_read_unlock();
608 
609 			put_task_struct(task);
610 		}
611 
612 		if (mnt_ns)
613 			m->private = mnt_ns;
614 		else {
615 			seq_release(inode, file);
616 			ret = -EINVAL;
617 		}
618 	}
619 	return ret;
620 }
621 
622 static const struct file_operations proc_mountstats_operations = {
623 	.open		= mountstats_open,
624 	.read		= seq_read,
625 	.llseek		= seq_lseek,
626 	.release	= mounts_release,
627 };
628 
629 #define PROC_BLOCK_SIZE	(3*1024)		/* 4K page size but our output routines use some slack for overruns */
630 
631 static ssize_t proc_info_read(struct file * file, char __user * buf,
632 			  size_t count, loff_t *ppos)
633 {
634 	struct inode * inode = file->f_path.dentry->d_inode;
635 	unsigned long page;
636 	ssize_t length;
637 	struct task_struct *task = get_proc_task(inode);
638 
639 	length = -ESRCH;
640 	if (!task)
641 		goto out_no_task;
642 
643 	if (count > PROC_BLOCK_SIZE)
644 		count = PROC_BLOCK_SIZE;
645 
646 	length = -ENOMEM;
647 	if (!(page = __get_free_page(GFP_TEMPORARY)))
648 		goto out;
649 
650 	length = PROC_I(inode)->op.proc_read(task, (char*)page);
651 
652 	if (length >= 0)
653 		length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
654 	free_page(page);
655 out:
656 	put_task_struct(task);
657 out_no_task:
658 	return length;
659 }
660 
661 static const struct file_operations proc_info_file_operations = {
662 	.read		= proc_info_read,
663 };
664 
665 static int proc_single_show(struct seq_file *m, void *v)
666 {
667 	struct inode *inode = m->private;
668 	struct pid_namespace *ns;
669 	struct pid *pid;
670 	struct task_struct *task;
671 	int ret;
672 
673 	ns = inode->i_sb->s_fs_info;
674 	pid = proc_pid(inode);
675 	task = get_pid_task(pid, PIDTYPE_PID);
676 	if (!task)
677 		return -ESRCH;
678 
679 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
680 
681 	put_task_struct(task);
682 	return ret;
683 }
684 
685 static int proc_single_open(struct inode *inode, struct file *filp)
686 {
687 	int ret;
688 	ret = single_open(filp, proc_single_show, NULL);
689 	if (!ret) {
690 		struct seq_file *m = filp->private_data;
691 
692 		m->private = inode;
693 	}
694 	return ret;
695 }
696 
697 static const struct file_operations proc_single_file_operations = {
698 	.open		= proc_single_open,
699 	.read		= seq_read,
700 	.llseek		= seq_lseek,
701 	.release	= single_release,
702 };
703 
704 static int mem_open(struct inode* inode, struct file* file)
705 {
706 	file->private_data = (void*)((long)current->self_exec_id);
707 	return 0;
708 }
709 
710 static ssize_t mem_read(struct file * file, char __user * buf,
711 			size_t count, loff_t *ppos)
712 {
713 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
714 	char *page;
715 	unsigned long src = *ppos;
716 	int ret = -ESRCH;
717 	struct mm_struct *mm;
718 
719 	if (!task)
720 		goto out_no_task;
721 
722 	if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
723 		goto out;
724 
725 	ret = -ENOMEM;
726 	page = (char *)__get_free_page(GFP_TEMPORARY);
727 	if (!page)
728 		goto out;
729 
730 	ret = 0;
731 
732 	mm = get_task_mm(task);
733 	if (!mm)
734 		goto out_free;
735 
736 	ret = -EIO;
737 
738 	if (file->private_data != (void*)((long)current->self_exec_id))
739 		goto out_put;
740 
741 	ret = 0;
742 
743 	while (count > 0) {
744 		int this_len, retval;
745 
746 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
747 		retval = access_process_vm(task, src, page, this_len, 0);
748 		if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
749 			if (!ret)
750 				ret = -EIO;
751 			break;
752 		}
753 
754 		if (copy_to_user(buf, page, retval)) {
755 			ret = -EFAULT;
756 			break;
757 		}
758 
759 		ret += retval;
760 		src += retval;
761 		buf += retval;
762 		count -= retval;
763 	}
764 	*ppos = src;
765 
766 out_put:
767 	mmput(mm);
768 out_free:
769 	free_page((unsigned long) page);
770 out:
771 	put_task_struct(task);
772 out_no_task:
773 	return ret;
774 }
775 
776 #define mem_write NULL
777 
778 #ifndef mem_write
779 /* This is a security hazard */
780 static ssize_t mem_write(struct file * file, const char __user *buf,
781 			 size_t count, loff_t *ppos)
782 {
783 	int copied;
784 	char *page;
785 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
786 	unsigned long dst = *ppos;
787 
788 	copied = -ESRCH;
789 	if (!task)
790 		goto out_no_task;
791 
792 	if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
793 		goto out;
794 
795 	copied = -ENOMEM;
796 	page = (char *)__get_free_page(GFP_TEMPORARY);
797 	if (!page)
798 		goto out;
799 
800 	copied = 0;
801 	while (count > 0) {
802 		int this_len, retval;
803 
804 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
805 		if (copy_from_user(page, buf, this_len)) {
806 			copied = -EFAULT;
807 			break;
808 		}
809 		retval = access_process_vm(task, dst, page, this_len, 1);
810 		if (!retval) {
811 			if (!copied)
812 				copied = -EIO;
813 			break;
814 		}
815 		copied += retval;
816 		buf += retval;
817 		dst += retval;
818 		count -= retval;
819 	}
820 	*ppos = dst;
821 	free_page((unsigned long) page);
822 out:
823 	put_task_struct(task);
824 out_no_task:
825 	return copied;
826 }
827 #endif
828 
829 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
830 {
831 	switch (orig) {
832 	case 0:
833 		file->f_pos = offset;
834 		break;
835 	case 1:
836 		file->f_pos += offset;
837 		break;
838 	default:
839 		return -EINVAL;
840 	}
841 	force_successful_syscall_return();
842 	return file->f_pos;
843 }
844 
845 static const struct file_operations proc_mem_operations = {
846 	.llseek		= mem_lseek,
847 	.read		= mem_read,
848 	.write		= mem_write,
849 	.open		= mem_open,
850 };
851 
852 static ssize_t environ_read(struct file *file, char __user *buf,
853 			size_t count, loff_t *ppos)
854 {
855 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
856 	char *page;
857 	unsigned long src = *ppos;
858 	int ret = -ESRCH;
859 	struct mm_struct *mm;
860 
861 	if (!task)
862 		goto out_no_task;
863 
864 	if (!ptrace_may_attach(task))
865 		goto out;
866 
867 	ret = -ENOMEM;
868 	page = (char *)__get_free_page(GFP_TEMPORARY);
869 	if (!page)
870 		goto out;
871 
872 	ret = 0;
873 
874 	mm = get_task_mm(task);
875 	if (!mm)
876 		goto out_free;
877 
878 	while (count > 0) {
879 		int this_len, retval, max_len;
880 
881 		this_len = mm->env_end - (mm->env_start + src);
882 
883 		if (this_len <= 0)
884 			break;
885 
886 		max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
887 		this_len = (this_len > max_len) ? max_len : this_len;
888 
889 		retval = access_process_vm(task, (mm->env_start + src),
890 			page, this_len, 0);
891 
892 		if (retval <= 0) {
893 			ret = retval;
894 			break;
895 		}
896 
897 		if (copy_to_user(buf, page, retval)) {
898 			ret = -EFAULT;
899 			break;
900 		}
901 
902 		ret += retval;
903 		src += retval;
904 		buf += retval;
905 		count -= retval;
906 	}
907 	*ppos = src;
908 
909 	mmput(mm);
910 out_free:
911 	free_page((unsigned long) page);
912 out:
913 	put_task_struct(task);
914 out_no_task:
915 	return ret;
916 }
917 
918 static const struct file_operations proc_environ_operations = {
919 	.read		= environ_read,
920 };
921 
922 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
923 				size_t count, loff_t *ppos)
924 {
925 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
926 	char buffer[PROC_NUMBUF];
927 	size_t len;
928 	int oom_adjust;
929 
930 	if (!task)
931 		return -ESRCH;
932 	oom_adjust = task->oomkilladj;
933 	put_task_struct(task);
934 
935 	len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
936 
937 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
938 }
939 
940 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
941 				size_t count, loff_t *ppos)
942 {
943 	struct task_struct *task;
944 	char buffer[PROC_NUMBUF], *end;
945 	int oom_adjust;
946 
947 	memset(buffer, 0, sizeof(buffer));
948 	if (count > sizeof(buffer) - 1)
949 		count = sizeof(buffer) - 1;
950 	if (copy_from_user(buffer, buf, count))
951 		return -EFAULT;
952 	oom_adjust = simple_strtol(buffer, &end, 0);
953 	if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
954 	     oom_adjust != OOM_DISABLE)
955 		return -EINVAL;
956 	if (*end == '\n')
957 		end++;
958 	task = get_proc_task(file->f_path.dentry->d_inode);
959 	if (!task)
960 		return -ESRCH;
961 	if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
962 		put_task_struct(task);
963 		return -EACCES;
964 	}
965 	task->oomkilladj = oom_adjust;
966 	put_task_struct(task);
967 	if (end - buffer == 0)
968 		return -EIO;
969 	return end - buffer;
970 }
971 
972 static const struct file_operations proc_oom_adjust_operations = {
973 	.read		= oom_adjust_read,
974 	.write		= oom_adjust_write,
975 };
976 
977 #ifdef CONFIG_AUDITSYSCALL
978 #define TMPBUFLEN 21
979 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
980 				  size_t count, loff_t *ppos)
981 {
982 	struct inode * inode = file->f_path.dentry->d_inode;
983 	struct task_struct *task = get_proc_task(inode);
984 	ssize_t length;
985 	char tmpbuf[TMPBUFLEN];
986 
987 	if (!task)
988 		return -ESRCH;
989 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
990 				audit_get_loginuid(task));
991 	put_task_struct(task);
992 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
993 }
994 
995 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
996 				   size_t count, loff_t *ppos)
997 {
998 	struct inode * inode = file->f_path.dentry->d_inode;
999 	char *page, *tmp;
1000 	ssize_t length;
1001 	uid_t loginuid;
1002 
1003 	if (!capable(CAP_AUDIT_CONTROL))
1004 		return -EPERM;
1005 
1006 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1007 		return -EPERM;
1008 
1009 	if (count >= PAGE_SIZE)
1010 		count = PAGE_SIZE - 1;
1011 
1012 	if (*ppos != 0) {
1013 		/* No partial writes. */
1014 		return -EINVAL;
1015 	}
1016 	page = (char*)__get_free_page(GFP_TEMPORARY);
1017 	if (!page)
1018 		return -ENOMEM;
1019 	length = -EFAULT;
1020 	if (copy_from_user(page, buf, count))
1021 		goto out_free_page;
1022 
1023 	page[count] = '\0';
1024 	loginuid = simple_strtoul(page, &tmp, 10);
1025 	if (tmp == page) {
1026 		length = -EINVAL;
1027 		goto out_free_page;
1028 
1029 	}
1030 	length = audit_set_loginuid(current, loginuid);
1031 	if (likely(length == 0))
1032 		length = count;
1033 
1034 out_free_page:
1035 	free_page((unsigned long) page);
1036 	return length;
1037 }
1038 
1039 static const struct file_operations proc_loginuid_operations = {
1040 	.read		= proc_loginuid_read,
1041 	.write		= proc_loginuid_write,
1042 };
1043 #endif
1044 
1045 #ifdef CONFIG_FAULT_INJECTION
1046 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1047 				      size_t count, loff_t *ppos)
1048 {
1049 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1050 	char buffer[PROC_NUMBUF];
1051 	size_t len;
1052 	int make_it_fail;
1053 
1054 	if (!task)
1055 		return -ESRCH;
1056 	make_it_fail = task->make_it_fail;
1057 	put_task_struct(task);
1058 
1059 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1060 
1061 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1062 }
1063 
1064 static ssize_t proc_fault_inject_write(struct file * file,
1065 			const char __user * buf, size_t count, loff_t *ppos)
1066 {
1067 	struct task_struct *task;
1068 	char buffer[PROC_NUMBUF], *end;
1069 	int make_it_fail;
1070 
1071 	if (!capable(CAP_SYS_RESOURCE))
1072 		return -EPERM;
1073 	memset(buffer, 0, sizeof(buffer));
1074 	if (count > sizeof(buffer) - 1)
1075 		count = sizeof(buffer) - 1;
1076 	if (copy_from_user(buffer, buf, count))
1077 		return -EFAULT;
1078 	make_it_fail = simple_strtol(buffer, &end, 0);
1079 	if (*end == '\n')
1080 		end++;
1081 	task = get_proc_task(file->f_dentry->d_inode);
1082 	if (!task)
1083 		return -ESRCH;
1084 	task->make_it_fail = make_it_fail;
1085 	put_task_struct(task);
1086 	if (end - buffer == 0)
1087 		return -EIO;
1088 	return end - buffer;
1089 }
1090 
1091 static const struct file_operations proc_fault_inject_operations = {
1092 	.read		= proc_fault_inject_read,
1093 	.write		= proc_fault_inject_write,
1094 };
1095 #endif
1096 
1097 
1098 #ifdef CONFIG_SCHED_DEBUG
1099 /*
1100  * Print out various scheduling related per-task fields:
1101  */
1102 static int sched_show(struct seq_file *m, void *v)
1103 {
1104 	struct inode *inode = m->private;
1105 	struct task_struct *p;
1106 
1107 	WARN_ON(!inode);
1108 
1109 	p = get_proc_task(inode);
1110 	if (!p)
1111 		return -ESRCH;
1112 	proc_sched_show_task(p, m);
1113 
1114 	put_task_struct(p);
1115 
1116 	return 0;
1117 }
1118 
1119 static ssize_t
1120 sched_write(struct file *file, const char __user *buf,
1121 	    size_t count, loff_t *offset)
1122 {
1123 	struct inode *inode = file->f_path.dentry->d_inode;
1124 	struct task_struct *p;
1125 
1126 	WARN_ON(!inode);
1127 
1128 	p = get_proc_task(inode);
1129 	if (!p)
1130 		return -ESRCH;
1131 	proc_sched_set_task(p);
1132 
1133 	put_task_struct(p);
1134 
1135 	return count;
1136 }
1137 
1138 static int sched_open(struct inode *inode, struct file *filp)
1139 {
1140 	int ret;
1141 
1142 	ret = single_open(filp, sched_show, NULL);
1143 	if (!ret) {
1144 		struct seq_file *m = filp->private_data;
1145 
1146 		m->private = inode;
1147 	}
1148 	return ret;
1149 }
1150 
1151 static const struct file_operations proc_pid_sched_operations = {
1152 	.open		= sched_open,
1153 	.read		= seq_read,
1154 	.write		= sched_write,
1155 	.llseek		= seq_lseek,
1156 	.release	= single_release,
1157 };
1158 
1159 #endif
1160 
1161 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1162 {
1163 	struct inode *inode = dentry->d_inode;
1164 	int error = -EACCES;
1165 
1166 	/* We don't need a base pointer in the /proc filesystem */
1167 	path_release(nd);
1168 
1169 	/* Are we allowed to snoop on the tasks file descriptors? */
1170 	if (!proc_fd_access_allowed(inode))
1171 		goto out;
1172 
1173 	error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1174 	nd->last_type = LAST_BIND;
1175 out:
1176 	return ERR_PTR(error);
1177 }
1178 
1179 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1180 			    char __user *buffer, int buflen)
1181 {
1182 	struct inode * inode;
1183 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1184 	char *path;
1185 	int len;
1186 
1187 	if (!tmp)
1188 		return -ENOMEM;
1189 
1190 	inode = dentry->d_inode;
1191 	path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1192 	len = PTR_ERR(path);
1193 	if (IS_ERR(path))
1194 		goto out;
1195 	len = tmp + PAGE_SIZE - 1 - path;
1196 
1197 	if (len > buflen)
1198 		len = buflen;
1199 	if (copy_to_user(buffer, path, len))
1200 		len = -EFAULT;
1201  out:
1202 	free_page((unsigned long)tmp);
1203 	return len;
1204 }
1205 
1206 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1207 {
1208 	int error = -EACCES;
1209 	struct inode *inode = dentry->d_inode;
1210 	struct dentry *de;
1211 	struct vfsmount *mnt = NULL;
1212 
1213 	/* Are we allowed to snoop on the tasks file descriptors? */
1214 	if (!proc_fd_access_allowed(inode))
1215 		goto out;
1216 
1217 	error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1218 	if (error)
1219 		goto out;
1220 
1221 	error = do_proc_readlink(de, mnt, buffer, buflen);
1222 	dput(de);
1223 	mntput(mnt);
1224 out:
1225 	return error;
1226 }
1227 
1228 static const struct inode_operations proc_pid_link_inode_operations = {
1229 	.readlink	= proc_pid_readlink,
1230 	.follow_link	= proc_pid_follow_link,
1231 	.setattr	= proc_setattr,
1232 };
1233 
1234 
1235 /* building an inode */
1236 
1237 static int task_dumpable(struct task_struct *task)
1238 {
1239 	int dumpable = 0;
1240 	struct mm_struct *mm;
1241 
1242 	task_lock(task);
1243 	mm = task->mm;
1244 	if (mm)
1245 		dumpable = get_dumpable(mm);
1246 	task_unlock(task);
1247 	if(dumpable == 1)
1248 		return 1;
1249 	return 0;
1250 }
1251 
1252 
1253 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1254 {
1255 	struct inode * inode;
1256 	struct proc_inode *ei;
1257 
1258 	/* We need a new inode */
1259 
1260 	inode = new_inode(sb);
1261 	if (!inode)
1262 		goto out;
1263 
1264 	/* Common stuff */
1265 	ei = PROC_I(inode);
1266 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1267 	inode->i_op = &proc_def_inode_operations;
1268 
1269 	/*
1270 	 * grab the reference to task.
1271 	 */
1272 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1273 	if (!ei->pid)
1274 		goto out_unlock;
1275 
1276 	inode->i_uid = 0;
1277 	inode->i_gid = 0;
1278 	if (task_dumpable(task)) {
1279 		inode->i_uid = task->euid;
1280 		inode->i_gid = task->egid;
1281 	}
1282 	security_task_to_inode(task, inode);
1283 
1284 out:
1285 	return inode;
1286 
1287 out_unlock:
1288 	iput(inode);
1289 	return NULL;
1290 }
1291 
1292 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1293 {
1294 	struct inode *inode = dentry->d_inode;
1295 	struct task_struct *task;
1296 	generic_fillattr(inode, stat);
1297 
1298 	rcu_read_lock();
1299 	stat->uid = 0;
1300 	stat->gid = 0;
1301 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1302 	if (task) {
1303 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1304 		    task_dumpable(task)) {
1305 			stat->uid = task->euid;
1306 			stat->gid = task->egid;
1307 		}
1308 	}
1309 	rcu_read_unlock();
1310 	return 0;
1311 }
1312 
1313 /* dentry stuff */
1314 
1315 /*
1316  *	Exceptional case: normally we are not allowed to unhash a busy
1317  * directory. In this case, however, we can do it - no aliasing problems
1318  * due to the way we treat inodes.
1319  *
1320  * Rewrite the inode's ownerships here because the owning task may have
1321  * performed a setuid(), etc.
1322  *
1323  * Before the /proc/pid/status file was created the only way to read
1324  * the effective uid of a /process was to stat /proc/pid.  Reading
1325  * /proc/pid/status is slow enough that procps and other packages
1326  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1327  * made this apply to all per process world readable and executable
1328  * directories.
1329  */
1330 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1331 {
1332 	struct inode *inode = dentry->d_inode;
1333 	struct task_struct *task = get_proc_task(inode);
1334 	if (task) {
1335 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1336 		    task_dumpable(task)) {
1337 			inode->i_uid = task->euid;
1338 			inode->i_gid = task->egid;
1339 		} else {
1340 			inode->i_uid = 0;
1341 			inode->i_gid = 0;
1342 		}
1343 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1344 		security_task_to_inode(task, inode);
1345 		put_task_struct(task);
1346 		return 1;
1347 	}
1348 	d_drop(dentry);
1349 	return 0;
1350 }
1351 
1352 static int pid_delete_dentry(struct dentry * dentry)
1353 {
1354 	/* Is the task we represent dead?
1355 	 * If so, then don't put the dentry on the lru list,
1356 	 * kill it immediately.
1357 	 */
1358 	return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1359 }
1360 
1361 static struct dentry_operations pid_dentry_operations =
1362 {
1363 	.d_revalidate	= pid_revalidate,
1364 	.d_delete	= pid_delete_dentry,
1365 };
1366 
1367 /* Lookups */
1368 
1369 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1370 				struct task_struct *, const void *);
1371 
1372 /*
1373  * Fill a directory entry.
1374  *
1375  * If possible create the dcache entry and derive our inode number and
1376  * file type from dcache entry.
1377  *
1378  * Since all of the proc inode numbers are dynamically generated, the inode
1379  * numbers do not exist until the inode is cache.  This means creating the
1380  * the dcache entry in readdir is necessary to keep the inode numbers
1381  * reported by readdir in sync with the inode numbers reported
1382  * by stat.
1383  */
1384 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1385 	char *name, int len,
1386 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1387 {
1388 	struct dentry *child, *dir = filp->f_path.dentry;
1389 	struct inode *inode;
1390 	struct qstr qname;
1391 	ino_t ino = 0;
1392 	unsigned type = DT_UNKNOWN;
1393 
1394 	qname.name = name;
1395 	qname.len  = len;
1396 	qname.hash = full_name_hash(name, len);
1397 
1398 	child = d_lookup(dir, &qname);
1399 	if (!child) {
1400 		struct dentry *new;
1401 		new = d_alloc(dir, &qname);
1402 		if (new) {
1403 			child = instantiate(dir->d_inode, new, task, ptr);
1404 			if (child)
1405 				dput(new);
1406 			else
1407 				child = new;
1408 		}
1409 	}
1410 	if (!child || IS_ERR(child) || !child->d_inode)
1411 		goto end_instantiate;
1412 	inode = child->d_inode;
1413 	if (inode) {
1414 		ino = inode->i_ino;
1415 		type = inode->i_mode >> 12;
1416 	}
1417 	dput(child);
1418 end_instantiate:
1419 	if (!ino)
1420 		ino = find_inode_number(dir, &qname);
1421 	if (!ino)
1422 		ino = 1;
1423 	return filldir(dirent, name, len, filp->f_pos, ino, type);
1424 }
1425 
1426 static unsigned name_to_int(struct dentry *dentry)
1427 {
1428 	const char *name = dentry->d_name.name;
1429 	int len = dentry->d_name.len;
1430 	unsigned n = 0;
1431 
1432 	if (len > 1 && *name == '0')
1433 		goto out;
1434 	while (len-- > 0) {
1435 		unsigned c = *name++ - '0';
1436 		if (c > 9)
1437 			goto out;
1438 		if (n >= (~0U-9)/10)
1439 			goto out;
1440 		n *= 10;
1441 		n += c;
1442 	}
1443 	return n;
1444 out:
1445 	return ~0U;
1446 }
1447 
1448 #define PROC_FDINFO_MAX 64
1449 
1450 static int proc_fd_info(struct inode *inode, struct dentry **dentry,
1451 			struct vfsmount **mnt, char *info)
1452 {
1453 	struct task_struct *task = get_proc_task(inode);
1454 	struct files_struct *files = NULL;
1455 	struct file *file;
1456 	int fd = proc_fd(inode);
1457 
1458 	if (task) {
1459 		files = get_files_struct(task);
1460 		put_task_struct(task);
1461 	}
1462 	if (files) {
1463 		/*
1464 		 * We are not taking a ref to the file structure, so we must
1465 		 * hold ->file_lock.
1466 		 */
1467 		spin_lock(&files->file_lock);
1468 		file = fcheck_files(files, fd);
1469 		if (file) {
1470 			if (mnt)
1471 				*mnt = mntget(file->f_path.mnt);
1472 			if (dentry)
1473 				*dentry = dget(file->f_path.dentry);
1474 			if (info)
1475 				snprintf(info, PROC_FDINFO_MAX,
1476 					 "pos:\t%lli\n"
1477 					 "flags:\t0%o\n",
1478 					 (long long) file->f_pos,
1479 					 file->f_flags);
1480 			spin_unlock(&files->file_lock);
1481 			put_files_struct(files);
1482 			return 0;
1483 		}
1484 		spin_unlock(&files->file_lock);
1485 		put_files_struct(files);
1486 	}
1487 	return -ENOENT;
1488 }
1489 
1490 static int proc_fd_link(struct inode *inode, struct dentry **dentry,
1491 			struct vfsmount **mnt)
1492 {
1493 	return proc_fd_info(inode, dentry, mnt, NULL);
1494 }
1495 
1496 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1497 {
1498 	struct inode *inode = dentry->d_inode;
1499 	struct task_struct *task = get_proc_task(inode);
1500 	int fd = proc_fd(inode);
1501 	struct files_struct *files;
1502 
1503 	if (task) {
1504 		files = get_files_struct(task);
1505 		if (files) {
1506 			rcu_read_lock();
1507 			if (fcheck_files(files, fd)) {
1508 				rcu_read_unlock();
1509 				put_files_struct(files);
1510 				if (task_dumpable(task)) {
1511 					inode->i_uid = task->euid;
1512 					inode->i_gid = task->egid;
1513 				} else {
1514 					inode->i_uid = 0;
1515 					inode->i_gid = 0;
1516 				}
1517 				inode->i_mode &= ~(S_ISUID | S_ISGID);
1518 				security_task_to_inode(task, inode);
1519 				put_task_struct(task);
1520 				return 1;
1521 			}
1522 			rcu_read_unlock();
1523 			put_files_struct(files);
1524 		}
1525 		put_task_struct(task);
1526 	}
1527 	d_drop(dentry);
1528 	return 0;
1529 }
1530 
1531 static struct dentry_operations tid_fd_dentry_operations =
1532 {
1533 	.d_revalidate	= tid_fd_revalidate,
1534 	.d_delete	= pid_delete_dentry,
1535 };
1536 
1537 static struct dentry *proc_fd_instantiate(struct inode *dir,
1538 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1539 {
1540 	unsigned fd = *(const unsigned *)ptr;
1541 	struct file *file;
1542 	struct files_struct *files;
1543  	struct inode *inode;
1544  	struct proc_inode *ei;
1545 	struct dentry *error = ERR_PTR(-ENOENT);
1546 
1547 	inode = proc_pid_make_inode(dir->i_sb, task);
1548 	if (!inode)
1549 		goto out;
1550 	ei = PROC_I(inode);
1551 	ei->fd = fd;
1552 	files = get_files_struct(task);
1553 	if (!files)
1554 		goto out_iput;
1555 	inode->i_mode = S_IFLNK;
1556 
1557 	/*
1558 	 * We are not taking a ref to the file structure, so we must
1559 	 * hold ->file_lock.
1560 	 */
1561 	spin_lock(&files->file_lock);
1562 	file = fcheck_files(files, fd);
1563 	if (!file)
1564 		goto out_unlock;
1565 	if (file->f_mode & 1)
1566 		inode->i_mode |= S_IRUSR | S_IXUSR;
1567 	if (file->f_mode & 2)
1568 		inode->i_mode |= S_IWUSR | S_IXUSR;
1569 	spin_unlock(&files->file_lock);
1570 	put_files_struct(files);
1571 
1572 	inode->i_op = &proc_pid_link_inode_operations;
1573 	inode->i_size = 64;
1574 	ei->op.proc_get_link = proc_fd_link;
1575 	dentry->d_op = &tid_fd_dentry_operations;
1576 	d_add(dentry, inode);
1577 	/* Close the race of the process dying before we return the dentry */
1578 	if (tid_fd_revalidate(dentry, NULL))
1579 		error = NULL;
1580 
1581  out:
1582 	return error;
1583 out_unlock:
1584 	spin_unlock(&files->file_lock);
1585 	put_files_struct(files);
1586 out_iput:
1587 	iput(inode);
1588 	goto out;
1589 }
1590 
1591 static struct dentry *proc_lookupfd_common(struct inode *dir,
1592 					   struct dentry *dentry,
1593 					   instantiate_t instantiate)
1594 {
1595 	struct task_struct *task = get_proc_task(dir);
1596 	unsigned fd = name_to_int(dentry);
1597 	struct dentry *result = ERR_PTR(-ENOENT);
1598 
1599 	if (!task)
1600 		goto out_no_task;
1601 	if (fd == ~0U)
1602 		goto out;
1603 
1604 	result = instantiate(dir, dentry, task, &fd);
1605 out:
1606 	put_task_struct(task);
1607 out_no_task:
1608 	return result;
1609 }
1610 
1611 static int proc_readfd_common(struct file * filp, void * dirent,
1612 			      filldir_t filldir, instantiate_t instantiate)
1613 {
1614 	struct dentry *dentry = filp->f_path.dentry;
1615 	struct inode *inode = dentry->d_inode;
1616 	struct task_struct *p = get_proc_task(inode);
1617 	unsigned int fd, ino;
1618 	int retval;
1619 	struct files_struct * files;
1620 	struct fdtable *fdt;
1621 
1622 	retval = -ENOENT;
1623 	if (!p)
1624 		goto out_no_task;
1625 	retval = 0;
1626 
1627 	fd = filp->f_pos;
1628 	switch (fd) {
1629 		case 0:
1630 			if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1631 				goto out;
1632 			filp->f_pos++;
1633 		case 1:
1634 			ino = parent_ino(dentry);
1635 			if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1636 				goto out;
1637 			filp->f_pos++;
1638 		default:
1639 			files = get_files_struct(p);
1640 			if (!files)
1641 				goto out;
1642 			rcu_read_lock();
1643 			fdt = files_fdtable(files);
1644 			for (fd = filp->f_pos-2;
1645 			     fd < fdt->max_fds;
1646 			     fd++, filp->f_pos++) {
1647 				char name[PROC_NUMBUF];
1648 				int len;
1649 
1650 				if (!fcheck_files(files, fd))
1651 					continue;
1652 				rcu_read_unlock();
1653 
1654 				len = snprintf(name, sizeof(name), "%d", fd);
1655 				if (proc_fill_cache(filp, dirent, filldir,
1656 						    name, len, instantiate,
1657 						    p, &fd) < 0) {
1658 					rcu_read_lock();
1659 					break;
1660 				}
1661 				rcu_read_lock();
1662 			}
1663 			rcu_read_unlock();
1664 			put_files_struct(files);
1665 	}
1666 out:
1667 	put_task_struct(p);
1668 out_no_task:
1669 	return retval;
1670 }
1671 
1672 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1673 				    struct nameidata *nd)
1674 {
1675 	return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1676 }
1677 
1678 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1679 {
1680 	return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1681 }
1682 
1683 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1684 				      size_t len, loff_t *ppos)
1685 {
1686 	char tmp[PROC_FDINFO_MAX];
1687 	int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, NULL, tmp);
1688 	if (!err)
1689 		err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1690 	return err;
1691 }
1692 
1693 static const struct file_operations proc_fdinfo_file_operations = {
1694 	.open		= nonseekable_open,
1695 	.read		= proc_fdinfo_read,
1696 };
1697 
1698 static const struct file_operations proc_fd_operations = {
1699 	.read		= generic_read_dir,
1700 	.readdir	= proc_readfd,
1701 };
1702 
1703 /*
1704  * /proc/pid/fd needs a special permission handler so that a process can still
1705  * access /proc/self/fd after it has executed a setuid().
1706  */
1707 static int proc_fd_permission(struct inode *inode, int mask,
1708 				struct nameidata *nd)
1709 {
1710 	int rv;
1711 
1712 	rv = generic_permission(inode, mask, NULL);
1713 	if (rv == 0)
1714 		return 0;
1715 	if (task_pid(current) == proc_pid(inode))
1716 		rv = 0;
1717 	return rv;
1718 }
1719 
1720 /*
1721  * proc directories can do almost nothing..
1722  */
1723 static const struct inode_operations proc_fd_inode_operations = {
1724 	.lookup		= proc_lookupfd,
1725 	.permission	= proc_fd_permission,
1726 	.setattr	= proc_setattr,
1727 };
1728 
1729 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1730 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1731 {
1732 	unsigned fd = *(unsigned *)ptr;
1733  	struct inode *inode;
1734  	struct proc_inode *ei;
1735 	struct dentry *error = ERR_PTR(-ENOENT);
1736 
1737 	inode = proc_pid_make_inode(dir->i_sb, task);
1738 	if (!inode)
1739 		goto out;
1740 	ei = PROC_I(inode);
1741 	ei->fd = fd;
1742 	inode->i_mode = S_IFREG | S_IRUSR;
1743 	inode->i_fop = &proc_fdinfo_file_operations;
1744 	dentry->d_op = &tid_fd_dentry_operations;
1745 	d_add(dentry, inode);
1746 	/* Close the race of the process dying before we return the dentry */
1747 	if (tid_fd_revalidate(dentry, NULL))
1748 		error = NULL;
1749 
1750  out:
1751 	return error;
1752 }
1753 
1754 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1755 					struct dentry *dentry,
1756 					struct nameidata *nd)
1757 {
1758 	return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1759 }
1760 
1761 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1762 {
1763 	return proc_readfd_common(filp, dirent, filldir,
1764 				  proc_fdinfo_instantiate);
1765 }
1766 
1767 static const struct file_operations proc_fdinfo_operations = {
1768 	.read		= generic_read_dir,
1769 	.readdir	= proc_readfdinfo,
1770 };
1771 
1772 /*
1773  * proc directories can do almost nothing..
1774  */
1775 static const struct inode_operations proc_fdinfo_inode_operations = {
1776 	.lookup		= proc_lookupfdinfo,
1777 	.setattr	= proc_setattr,
1778 };
1779 
1780 
1781 static struct dentry *proc_pident_instantiate(struct inode *dir,
1782 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1783 {
1784 	const struct pid_entry *p = ptr;
1785 	struct inode *inode;
1786 	struct proc_inode *ei;
1787 	struct dentry *error = ERR_PTR(-EINVAL);
1788 
1789 	inode = proc_pid_make_inode(dir->i_sb, task);
1790 	if (!inode)
1791 		goto out;
1792 
1793 	ei = PROC_I(inode);
1794 	inode->i_mode = p->mode;
1795 	if (S_ISDIR(inode->i_mode))
1796 		inode->i_nlink = 2;	/* Use getattr to fix if necessary */
1797 	if (p->iop)
1798 		inode->i_op = p->iop;
1799 	if (p->fop)
1800 		inode->i_fop = p->fop;
1801 	ei->op = p->op;
1802 	dentry->d_op = &pid_dentry_operations;
1803 	d_add(dentry, inode);
1804 	/* Close the race of the process dying before we return the dentry */
1805 	if (pid_revalidate(dentry, NULL))
1806 		error = NULL;
1807 out:
1808 	return error;
1809 }
1810 
1811 static struct dentry *proc_pident_lookup(struct inode *dir,
1812 					 struct dentry *dentry,
1813 					 const struct pid_entry *ents,
1814 					 unsigned int nents)
1815 {
1816 	struct inode *inode;
1817 	struct dentry *error;
1818 	struct task_struct *task = get_proc_task(dir);
1819 	const struct pid_entry *p, *last;
1820 
1821 	error = ERR_PTR(-ENOENT);
1822 	inode = NULL;
1823 
1824 	if (!task)
1825 		goto out_no_task;
1826 
1827 	/*
1828 	 * Yes, it does not scale. And it should not. Don't add
1829 	 * new entries into /proc/<tgid>/ without very good reasons.
1830 	 */
1831 	last = &ents[nents - 1];
1832 	for (p = ents; p <= last; p++) {
1833 		if (p->len != dentry->d_name.len)
1834 			continue;
1835 		if (!memcmp(dentry->d_name.name, p->name, p->len))
1836 			break;
1837 	}
1838 	if (p > last)
1839 		goto out;
1840 
1841 	error = proc_pident_instantiate(dir, dentry, task, p);
1842 out:
1843 	put_task_struct(task);
1844 out_no_task:
1845 	return error;
1846 }
1847 
1848 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1849 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1850 {
1851 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1852 				proc_pident_instantiate, task, p);
1853 }
1854 
1855 static int proc_pident_readdir(struct file *filp,
1856 		void *dirent, filldir_t filldir,
1857 		const struct pid_entry *ents, unsigned int nents)
1858 {
1859 	int i;
1860 	struct dentry *dentry = filp->f_path.dentry;
1861 	struct inode *inode = dentry->d_inode;
1862 	struct task_struct *task = get_proc_task(inode);
1863 	const struct pid_entry *p, *last;
1864 	ino_t ino;
1865 	int ret;
1866 
1867 	ret = -ENOENT;
1868 	if (!task)
1869 		goto out_no_task;
1870 
1871 	ret = 0;
1872 	i = filp->f_pos;
1873 	switch (i) {
1874 	case 0:
1875 		ino = inode->i_ino;
1876 		if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1877 			goto out;
1878 		i++;
1879 		filp->f_pos++;
1880 		/* fall through */
1881 	case 1:
1882 		ino = parent_ino(dentry);
1883 		if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1884 			goto out;
1885 		i++;
1886 		filp->f_pos++;
1887 		/* fall through */
1888 	default:
1889 		i -= 2;
1890 		if (i >= nents) {
1891 			ret = 1;
1892 			goto out;
1893 		}
1894 		p = ents + i;
1895 		last = &ents[nents - 1];
1896 		while (p <= last) {
1897 			if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1898 				goto out;
1899 			filp->f_pos++;
1900 			p++;
1901 		}
1902 	}
1903 
1904 	ret = 1;
1905 out:
1906 	put_task_struct(task);
1907 out_no_task:
1908 	return ret;
1909 }
1910 
1911 #ifdef CONFIG_SECURITY
1912 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1913 				  size_t count, loff_t *ppos)
1914 {
1915 	struct inode * inode = file->f_path.dentry->d_inode;
1916 	char *p = NULL;
1917 	ssize_t length;
1918 	struct task_struct *task = get_proc_task(inode);
1919 
1920 	if (!task)
1921 		return -ESRCH;
1922 
1923 	length = security_getprocattr(task,
1924 				      (char*)file->f_path.dentry->d_name.name,
1925 				      &p);
1926 	put_task_struct(task);
1927 	if (length > 0)
1928 		length = simple_read_from_buffer(buf, count, ppos, p, length);
1929 	kfree(p);
1930 	return length;
1931 }
1932 
1933 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1934 				   size_t count, loff_t *ppos)
1935 {
1936 	struct inode * inode = file->f_path.dentry->d_inode;
1937 	char *page;
1938 	ssize_t length;
1939 	struct task_struct *task = get_proc_task(inode);
1940 
1941 	length = -ESRCH;
1942 	if (!task)
1943 		goto out_no_task;
1944 	if (count > PAGE_SIZE)
1945 		count = PAGE_SIZE;
1946 
1947 	/* No partial writes. */
1948 	length = -EINVAL;
1949 	if (*ppos != 0)
1950 		goto out;
1951 
1952 	length = -ENOMEM;
1953 	page = (char*)__get_free_page(GFP_TEMPORARY);
1954 	if (!page)
1955 		goto out;
1956 
1957 	length = -EFAULT;
1958 	if (copy_from_user(page, buf, count))
1959 		goto out_free;
1960 
1961 	length = security_setprocattr(task,
1962 				      (char*)file->f_path.dentry->d_name.name,
1963 				      (void*)page, count);
1964 out_free:
1965 	free_page((unsigned long) page);
1966 out:
1967 	put_task_struct(task);
1968 out_no_task:
1969 	return length;
1970 }
1971 
1972 static const struct file_operations proc_pid_attr_operations = {
1973 	.read		= proc_pid_attr_read,
1974 	.write		= proc_pid_attr_write,
1975 };
1976 
1977 static const struct pid_entry attr_dir_stuff[] = {
1978 	REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1979 	REG("prev",       S_IRUGO,	   pid_attr),
1980 	REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1981 	REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1982 	REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1983 	REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1984 };
1985 
1986 static int proc_attr_dir_readdir(struct file * filp,
1987 			     void * dirent, filldir_t filldir)
1988 {
1989 	return proc_pident_readdir(filp,dirent,filldir,
1990 				   attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1991 }
1992 
1993 static const struct file_operations proc_attr_dir_operations = {
1994 	.read		= generic_read_dir,
1995 	.readdir	= proc_attr_dir_readdir,
1996 };
1997 
1998 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
1999 				struct dentry *dentry, struct nameidata *nd)
2000 {
2001 	return proc_pident_lookup(dir, dentry,
2002 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2003 }
2004 
2005 static const struct inode_operations proc_attr_dir_inode_operations = {
2006 	.lookup		= proc_attr_dir_lookup,
2007 	.getattr	= pid_getattr,
2008 	.setattr	= proc_setattr,
2009 };
2010 
2011 #endif
2012 
2013 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2014 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2015 					 size_t count, loff_t *ppos)
2016 {
2017 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2018 	struct mm_struct *mm;
2019 	char buffer[PROC_NUMBUF];
2020 	size_t len;
2021 	int ret;
2022 
2023 	if (!task)
2024 		return -ESRCH;
2025 
2026 	ret = 0;
2027 	mm = get_task_mm(task);
2028 	if (mm) {
2029 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2030 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2031 				MMF_DUMP_FILTER_SHIFT));
2032 		mmput(mm);
2033 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2034 	}
2035 
2036 	put_task_struct(task);
2037 
2038 	return ret;
2039 }
2040 
2041 static ssize_t proc_coredump_filter_write(struct file *file,
2042 					  const char __user *buf,
2043 					  size_t count,
2044 					  loff_t *ppos)
2045 {
2046 	struct task_struct *task;
2047 	struct mm_struct *mm;
2048 	char buffer[PROC_NUMBUF], *end;
2049 	unsigned int val;
2050 	int ret;
2051 	int i;
2052 	unsigned long mask;
2053 
2054 	ret = -EFAULT;
2055 	memset(buffer, 0, sizeof(buffer));
2056 	if (count > sizeof(buffer) - 1)
2057 		count = sizeof(buffer) - 1;
2058 	if (copy_from_user(buffer, buf, count))
2059 		goto out_no_task;
2060 
2061 	ret = -EINVAL;
2062 	val = (unsigned int)simple_strtoul(buffer, &end, 0);
2063 	if (*end == '\n')
2064 		end++;
2065 	if (end - buffer == 0)
2066 		goto out_no_task;
2067 
2068 	ret = -ESRCH;
2069 	task = get_proc_task(file->f_dentry->d_inode);
2070 	if (!task)
2071 		goto out_no_task;
2072 
2073 	ret = end - buffer;
2074 	mm = get_task_mm(task);
2075 	if (!mm)
2076 		goto out_no_mm;
2077 
2078 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2079 		if (val & mask)
2080 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2081 		else
2082 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2083 	}
2084 
2085 	mmput(mm);
2086  out_no_mm:
2087 	put_task_struct(task);
2088  out_no_task:
2089 	return ret;
2090 }
2091 
2092 static const struct file_operations proc_coredump_filter_operations = {
2093 	.read		= proc_coredump_filter_read,
2094 	.write		= proc_coredump_filter_write,
2095 };
2096 #endif
2097 
2098 /*
2099  * /proc/self:
2100  */
2101 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2102 			      int buflen)
2103 {
2104 	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2105 	pid_t tgid = task_tgid_nr_ns(current, ns);
2106 	char tmp[PROC_NUMBUF];
2107 	if (!tgid)
2108 		return -ENOENT;
2109 	sprintf(tmp, "%d", tgid);
2110 	return vfs_readlink(dentry,buffer,buflen,tmp);
2111 }
2112 
2113 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2114 {
2115 	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2116 	pid_t tgid = task_tgid_nr_ns(current, ns);
2117 	char tmp[PROC_NUMBUF];
2118 	if (!tgid)
2119 		return ERR_PTR(-ENOENT);
2120 	sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2121 	return ERR_PTR(vfs_follow_link(nd,tmp));
2122 }
2123 
2124 static const struct inode_operations proc_self_inode_operations = {
2125 	.readlink	= proc_self_readlink,
2126 	.follow_link	= proc_self_follow_link,
2127 };
2128 
2129 /*
2130  * proc base
2131  *
2132  * These are the directory entries in the root directory of /proc
2133  * that properly belong to the /proc filesystem, as they describe
2134  * describe something that is process related.
2135  */
2136 static const struct pid_entry proc_base_stuff[] = {
2137 	NOD("self", S_IFLNK|S_IRWXUGO,
2138 		&proc_self_inode_operations, NULL, {}),
2139 };
2140 
2141 /*
2142  *	Exceptional case: normally we are not allowed to unhash a busy
2143  * directory. In this case, however, we can do it - no aliasing problems
2144  * due to the way we treat inodes.
2145  */
2146 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2147 {
2148 	struct inode *inode = dentry->d_inode;
2149 	struct task_struct *task = get_proc_task(inode);
2150 	if (task) {
2151 		put_task_struct(task);
2152 		return 1;
2153 	}
2154 	d_drop(dentry);
2155 	return 0;
2156 }
2157 
2158 static struct dentry_operations proc_base_dentry_operations =
2159 {
2160 	.d_revalidate	= proc_base_revalidate,
2161 	.d_delete	= pid_delete_dentry,
2162 };
2163 
2164 static struct dentry *proc_base_instantiate(struct inode *dir,
2165 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2166 {
2167 	const struct pid_entry *p = ptr;
2168 	struct inode *inode;
2169 	struct proc_inode *ei;
2170 	struct dentry *error = ERR_PTR(-EINVAL);
2171 
2172 	/* Allocate the inode */
2173 	error = ERR_PTR(-ENOMEM);
2174 	inode = new_inode(dir->i_sb);
2175 	if (!inode)
2176 		goto out;
2177 
2178 	/* Initialize the inode */
2179 	ei = PROC_I(inode);
2180 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2181 
2182 	/*
2183 	 * grab the reference to the task.
2184 	 */
2185 	ei->pid = get_task_pid(task, PIDTYPE_PID);
2186 	if (!ei->pid)
2187 		goto out_iput;
2188 
2189 	inode->i_uid = 0;
2190 	inode->i_gid = 0;
2191 	inode->i_mode = p->mode;
2192 	if (S_ISDIR(inode->i_mode))
2193 		inode->i_nlink = 2;
2194 	if (S_ISLNK(inode->i_mode))
2195 		inode->i_size = 64;
2196 	if (p->iop)
2197 		inode->i_op = p->iop;
2198 	if (p->fop)
2199 		inode->i_fop = p->fop;
2200 	ei->op = p->op;
2201 	dentry->d_op = &proc_base_dentry_operations;
2202 	d_add(dentry, inode);
2203 	error = NULL;
2204 out:
2205 	return error;
2206 out_iput:
2207 	iput(inode);
2208 	goto out;
2209 }
2210 
2211 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2212 {
2213 	struct dentry *error;
2214 	struct task_struct *task = get_proc_task(dir);
2215 	const struct pid_entry *p, *last;
2216 
2217 	error = ERR_PTR(-ENOENT);
2218 
2219 	if (!task)
2220 		goto out_no_task;
2221 
2222 	/* Lookup the directory entry */
2223 	last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2224 	for (p = proc_base_stuff; p <= last; p++) {
2225 		if (p->len != dentry->d_name.len)
2226 			continue;
2227 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2228 			break;
2229 	}
2230 	if (p > last)
2231 		goto out;
2232 
2233 	error = proc_base_instantiate(dir, dentry, task, p);
2234 
2235 out:
2236 	put_task_struct(task);
2237 out_no_task:
2238 	return error;
2239 }
2240 
2241 static int proc_base_fill_cache(struct file *filp, void *dirent,
2242 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2243 {
2244 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2245 				proc_base_instantiate, task, p);
2246 }
2247 
2248 #ifdef CONFIG_TASK_IO_ACCOUNTING
2249 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2250 {
2251 	return sprintf(buffer,
2252 #ifdef CONFIG_TASK_XACCT
2253 			"rchar: %llu\n"
2254 			"wchar: %llu\n"
2255 			"syscr: %llu\n"
2256 			"syscw: %llu\n"
2257 #endif
2258 			"read_bytes: %llu\n"
2259 			"write_bytes: %llu\n"
2260 			"cancelled_write_bytes: %llu\n",
2261 #ifdef CONFIG_TASK_XACCT
2262 			(unsigned long long)task->rchar,
2263 			(unsigned long long)task->wchar,
2264 			(unsigned long long)task->syscr,
2265 			(unsigned long long)task->syscw,
2266 #endif
2267 			(unsigned long long)task->ioac.read_bytes,
2268 			(unsigned long long)task->ioac.write_bytes,
2269 			(unsigned long long)task->ioac.cancelled_write_bytes);
2270 }
2271 #endif
2272 
2273 /*
2274  * Thread groups
2275  */
2276 static const struct file_operations proc_task_operations;
2277 static const struct inode_operations proc_task_inode_operations;
2278 
2279 static const struct pid_entry tgid_base_stuff[] = {
2280 	DIR("task",       S_IRUGO|S_IXUGO, task),
2281 	DIR("fd",         S_IRUSR|S_IXUSR, fd),
2282 	DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2283 	REG("environ",    S_IRUSR, environ),
2284 	INF("auxv",       S_IRUSR, pid_auxv),
2285 	ONE("status",     S_IRUGO, pid_status),
2286 	INF("limits",	  S_IRUSR, pid_limits),
2287 #ifdef CONFIG_SCHED_DEBUG
2288 	REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2289 #endif
2290 	INF("cmdline",    S_IRUGO, pid_cmdline),
2291 	ONE("stat",       S_IRUGO, tgid_stat),
2292 	ONE("statm",      S_IRUGO, pid_statm),
2293 	REG("maps",       S_IRUGO, maps),
2294 #ifdef CONFIG_NUMA
2295 	REG("numa_maps",  S_IRUGO, numa_maps),
2296 #endif
2297 	REG("mem",        S_IRUSR|S_IWUSR, mem),
2298 	LNK("cwd",        cwd),
2299 	LNK("root",       root),
2300 	LNK("exe",        exe),
2301 	REG("mounts",     S_IRUGO, mounts),
2302 	REG("mountstats", S_IRUSR, mountstats),
2303 #ifdef CONFIG_PROC_PAGE_MONITOR
2304 	REG("clear_refs", S_IWUSR, clear_refs),
2305 	REG("smaps",      S_IRUGO, smaps),
2306 	REG("pagemap",    S_IRUSR, pagemap),
2307 #endif
2308 #ifdef CONFIG_SECURITY
2309 	DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2310 #endif
2311 #ifdef CONFIG_KALLSYMS
2312 	INF("wchan",      S_IRUGO, pid_wchan),
2313 #endif
2314 #ifdef CONFIG_SCHEDSTATS
2315 	INF("schedstat",  S_IRUGO, pid_schedstat),
2316 #endif
2317 #ifdef CONFIG_LATENCYTOP
2318 	REG("latency",  S_IRUGO, lstats),
2319 #endif
2320 #ifdef CONFIG_PROC_PID_CPUSET
2321 	REG("cpuset",     S_IRUGO, cpuset),
2322 #endif
2323 #ifdef CONFIG_CGROUPS
2324 	REG("cgroup",  S_IRUGO, cgroup),
2325 #endif
2326 	INF("oom_score",  S_IRUGO, oom_score),
2327 	REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2328 #ifdef CONFIG_AUDITSYSCALL
2329 	REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2330 #endif
2331 #ifdef CONFIG_FAULT_INJECTION
2332 	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2333 #endif
2334 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2335 	REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2336 #endif
2337 #ifdef CONFIG_TASK_IO_ACCOUNTING
2338 	INF("io",	S_IRUGO, pid_io_accounting),
2339 #endif
2340 };
2341 
2342 static int proc_tgid_base_readdir(struct file * filp,
2343 			     void * dirent, filldir_t filldir)
2344 {
2345 	return proc_pident_readdir(filp,dirent,filldir,
2346 				   tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2347 }
2348 
2349 static const struct file_operations proc_tgid_base_operations = {
2350 	.read		= generic_read_dir,
2351 	.readdir	= proc_tgid_base_readdir,
2352 };
2353 
2354 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2355 	return proc_pident_lookup(dir, dentry,
2356 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2357 }
2358 
2359 static const struct inode_operations proc_tgid_base_inode_operations = {
2360 	.lookup		= proc_tgid_base_lookup,
2361 	.getattr	= pid_getattr,
2362 	.setattr	= proc_setattr,
2363 };
2364 
2365 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2366 {
2367 	struct dentry *dentry, *leader, *dir;
2368 	char buf[PROC_NUMBUF];
2369 	struct qstr name;
2370 
2371 	name.name = buf;
2372 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2373 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2374 	if (dentry) {
2375 		if (!(current->flags & PF_EXITING))
2376 			shrink_dcache_parent(dentry);
2377 		d_drop(dentry);
2378 		dput(dentry);
2379 	}
2380 
2381 	if (tgid == 0)
2382 		goto out;
2383 
2384 	name.name = buf;
2385 	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2386 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2387 	if (!leader)
2388 		goto out;
2389 
2390 	name.name = "task";
2391 	name.len = strlen(name.name);
2392 	dir = d_hash_and_lookup(leader, &name);
2393 	if (!dir)
2394 		goto out_put_leader;
2395 
2396 	name.name = buf;
2397 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2398 	dentry = d_hash_and_lookup(dir, &name);
2399 	if (dentry) {
2400 		shrink_dcache_parent(dentry);
2401 		d_drop(dentry);
2402 		dput(dentry);
2403 	}
2404 
2405 	dput(dir);
2406 out_put_leader:
2407 	dput(leader);
2408 out:
2409 	return;
2410 }
2411 
2412 /**
2413  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2414  * @task: task that should be flushed.
2415  *
2416  * When flushing dentries from proc, one needs to flush them from global
2417  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2418  * in. This call is supposed to do all of this job.
2419  *
2420  * Looks in the dcache for
2421  * /proc/@pid
2422  * /proc/@tgid/task/@pid
2423  * if either directory is present flushes it and all of it'ts children
2424  * from the dcache.
2425  *
2426  * It is safe and reasonable to cache /proc entries for a task until
2427  * that task exits.  After that they just clog up the dcache with
2428  * useless entries, possibly causing useful dcache entries to be
2429  * flushed instead.  This routine is proved to flush those useless
2430  * dcache entries at process exit time.
2431  *
2432  * NOTE: This routine is just an optimization so it does not guarantee
2433  *       that no dcache entries will exist at process exit time it
2434  *       just makes it very unlikely that any will persist.
2435  */
2436 
2437 void proc_flush_task(struct task_struct *task)
2438 {
2439 	int i;
2440 	struct pid *pid, *tgid = NULL;
2441 	struct upid *upid;
2442 
2443 	pid = task_pid(task);
2444 	if (thread_group_leader(task))
2445 		tgid = task_tgid(task);
2446 
2447 	for (i = 0; i <= pid->level; i++) {
2448 		upid = &pid->numbers[i];
2449 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2450 			tgid ? tgid->numbers[i].nr : 0);
2451 	}
2452 
2453 	upid = &pid->numbers[pid->level];
2454 	if (upid->nr == 1)
2455 		pid_ns_release_proc(upid->ns);
2456 }
2457 
2458 static struct dentry *proc_pid_instantiate(struct inode *dir,
2459 					   struct dentry * dentry,
2460 					   struct task_struct *task, const void *ptr)
2461 {
2462 	struct dentry *error = ERR_PTR(-ENOENT);
2463 	struct inode *inode;
2464 
2465 	inode = proc_pid_make_inode(dir->i_sb, task);
2466 	if (!inode)
2467 		goto out;
2468 
2469 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2470 	inode->i_op = &proc_tgid_base_inode_operations;
2471 	inode->i_fop = &proc_tgid_base_operations;
2472 	inode->i_flags|=S_IMMUTABLE;
2473 	inode->i_nlink = 5;
2474 #ifdef CONFIG_SECURITY
2475 	inode->i_nlink += 1;
2476 #endif
2477 
2478 	dentry->d_op = &pid_dentry_operations;
2479 
2480 	d_add(dentry, inode);
2481 	/* Close the race of the process dying before we return the dentry */
2482 	if (pid_revalidate(dentry, NULL))
2483 		error = NULL;
2484 out:
2485 	return error;
2486 }
2487 
2488 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2489 {
2490 	struct dentry *result = ERR_PTR(-ENOENT);
2491 	struct task_struct *task;
2492 	unsigned tgid;
2493 	struct pid_namespace *ns;
2494 
2495 	result = proc_base_lookup(dir, dentry);
2496 	if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2497 		goto out;
2498 
2499 	tgid = name_to_int(dentry);
2500 	if (tgid == ~0U)
2501 		goto out;
2502 
2503 	ns = dentry->d_sb->s_fs_info;
2504 	rcu_read_lock();
2505 	task = find_task_by_pid_ns(tgid, ns);
2506 	if (task)
2507 		get_task_struct(task);
2508 	rcu_read_unlock();
2509 	if (!task)
2510 		goto out;
2511 
2512 	result = proc_pid_instantiate(dir, dentry, task, NULL);
2513 	put_task_struct(task);
2514 out:
2515 	return result;
2516 }
2517 
2518 /*
2519  * Find the first task with tgid >= tgid
2520  *
2521  */
2522 struct tgid_iter {
2523 	unsigned int tgid;
2524 	struct task_struct *task;
2525 };
2526 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2527 {
2528 	struct pid *pid;
2529 
2530 	if (iter.task)
2531 		put_task_struct(iter.task);
2532 	rcu_read_lock();
2533 retry:
2534 	iter.task = NULL;
2535 	pid = find_ge_pid(iter.tgid, ns);
2536 	if (pid) {
2537 		iter.tgid = pid_nr_ns(pid, ns);
2538 		iter.task = pid_task(pid, PIDTYPE_PID);
2539 		/* What we to know is if the pid we have find is the
2540 		 * pid of a thread_group_leader.  Testing for task
2541 		 * being a thread_group_leader is the obvious thing
2542 		 * todo but there is a window when it fails, due to
2543 		 * the pid transfer logic in de_thread.
2544 		 *
2545 		 * So we perform the straight forward test of seeing
2546 		 * if the pid we have found is the pid of a thread
2547 		 * group leader, and don't worry if the task we have
2548 		 * found doesn't happen to be a thread group leader.
2549 		 * As we don't care in the case of readdir.
2550 		 */
2551 		if (!iter.task || !has_group_leader_pid(iter.task)) {
2552 			iter.tgid += 1;
2553 			goto retry;
2554 		}
2555 		get_task_struct(iter.task);
2556 	}
2557 	rcu_read_unlock();
2558 	return iter;
2559 }
2560 
2561 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2562 
2563 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2564 	struct tgid_iter iter)
2565 {
2566 	char name[PROC_NUMBUF];
2567 	int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2568 	return proc_fill_cache(filp, dirent, filldir, name, len,
2569 				proc_pid_instantiate, iter.task, NULL);
2570 }
2571 
2572 /* for the /proc/ directory itself, after non-process stuff has been done */
2573 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2574 {
2575 	unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2576 	struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2577 	struct tgid_iter iter;
2578 	struct pid_namespace *ns;
2579 
2580 	if (!reaper)
2581 		goto out_no_task;
2582 
2583 	for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2584 		const struct pid_entry *p = &proc_base_stuff[nr];
2585 		if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2586 			goto out;
2587 	}
2588 
2589 	ns = filp->f_dentry->d_sb->s_fs_info;
2590 	iter.task = NULL;
2591 	iter.tgid = filp->f_pos - TGID_OFFSET;
2592 	for (iter = next_tgid(ns, iter);
2593 	     iter.task;
2594 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
2595 		filp->f_pos = iter.tgid + TGID_OFFSET;
2596 		if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2597 			put_task_struct(iter.task);
2598 			goto out;
2599 		}
2600 	}
2601 	filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2602 out:
2603 	put_task_struct(reaper);
2604 out_no_task:
2605 	return 0;
2606 }
2607 
2608 /*
2609  * Tasks
2610  */
2611 static const struct pid_entry tid_base_stuff[] = {
2612 	DIR("fd",        S_IRUSR|S_IXUSR, fd),
2613 	DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2614 	REG("environ",   S_IRUSR, environ),
2615 	INF("auxv",      S_IRUSR, pid_auxv),
2616 	ONE("status",    S_IRUGO, pid_status),
2617 	INF("limits",	 S_IRUSR, pid_limits),
2618 #ifdef CONFIG_SCHED_DEBUG
2619 	REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2620 #endif
2621 	INF("cmdline",   S_IRUGO, pid_cmdline),
2622 	ONE("stat",      S_IRUGO, tid_stat),
2623 	ONE("statm",     S_IRUGO, pid_statm),
2624 	REG("maps",      S_IRUGO, maps),
2625 #ifdef CONFIG_NUMA
2626 	REG("numa_maps", S_IRUGO, numa_maps),
2627 #endif
2628 	REG("mem",       S_IRUSR|S_IWUSR, mem),
2629 	LNK("cwd",       cwd),
2630 	LNK("root",      root),
2631 	LNK("exe",       exe),
2632 	REG("mounts",    S_IRUGO, mounts),
2633 #ifdef CONFIG_PROC_PAGE_MONITOR
2634 	REG("clear_refs", S_IWUSR, clear_refs),
2635 	REG("smaps",     S_IRUGO, smaps),
2636 	REG("pagemap",    S_IRUSR, pagemap),
2637 #endif
2638 #ifdef CONFIG_SECURITY
2639 	DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2640 #endif
2641 #ifdef CONFIG_KALLSYMS
2642 	INF("wchan",     S_IRUGO, pid_wchan),
2643 #endif
2644 #ifdef CONFIG_SCHEDSTATS
2645 	INF("schedstat", S_IRUGO, pid_schedstat),
2646 #endif
2647 #ifdef CONFIG_LATENCYTOP
2648 	REG("latency",  S_IRUGO, lstats),
2649 #endif
2650 #ifdef CONFIG_PROC_PID_CPUSET
2651 	REG("cpuset",    S_IRUGO, cpuset),
2652 #endif
2653 #ifdef CONFIG_CGROUPS
2654 	REG("cgroup",  S_IRUGO, cgroup),
2655 #endif
2656 	INF("oom_score", S_IRUGO, oom_score),
2657 	REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2658 #ifdef CONFIG_AUDITSYSCALL
2659 	REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2660 #endif
2661 #ifdef CONFIG_FAULT_INJECTION
2662 	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2663 #endif
2664 };
2665 
2666 static int proc_tid_base_readdir(struct file * filp,
2667 			     void * dirent, filldir_t filldir)
2668 {
2669 	return proc_pident_readdir(filp,dirent,filldir,
2670 				   tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2671 }
2672 
2673 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2674 	return proc_pident_lookup(dir, dentry,
2675 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2676 }
2677 
2678 static const struct file_operations proc_tid_base_operations = {
2679 	.read		= generic_read_dir,
2680 	.readdir	= proc_tid_base_readdir,
2681 };
2682 
2683 static const struct inode_operations proc_tid_base_inode_operations = {
2684 	.lookup		= proc_tid_base_lookup,
2685 	.getattr	= pid_getattr,
2686 	.setattr	= proc_setattr,
2687 };
2688 
2689 static struct dentry *proc_task_instantiate(struct inode *dir,
2690 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2691 {
2692 	struct dentry *error = ERR_PTR(-ENOENT);
2693 	struct inode *inode;
2694 	inode = proc_pid_make_inode(dir->i_sb, task);
2695 
2696 	if (!inode)
2697 		goto out;
2698 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2699 	inode->i_op = &proc_tid_base_inode_operations;
2700 	inode->i_fop = &proc_tid_base_operations;
2701 	inode->i_flags|=S_IMMUTABLE;
2702 	inode->i_nlink = 4;
2703 #ifdef CONFIG_SECURITY
2704 	inode->i_nlink += 1;
2705 #endif
2706 
2707 	dentry->d_op = &pid_dentry_operations;
2708 
2709 	d_add(dentry, inode);
2710 	/* Close the race of the process dying before we return the dentry */
2711 	if (pid_revalidate(dentry, NULL))
2712 		error = NULL;
2713 out:
2714 	return error;
2715 }
2716 
2717 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2718 {
2719 	struct dentry *result = ERR_PTR(-ENOENT);
2720 	struct task_struct *task;
2721 	struct task_struct *leader = get_proc_task(dir);
2722 	unsigned tid;
2723 	struct pid_namespace *ns;
2724 
2725 	if (!leader)
2726 		goto out_no_task;
2727 
2728 	tid = name_to_int(dentry);
2729 	if (tid == ~0U)
2730 		goto out;
2731 
2732 	ns = dentry->d_sb->s_fs_info;
2733 	rcu_read_lock();
2734 	task = find_task_by_pid_ns(tid, ns);
2735 	if (task)
2736 		get_task_struct(task);
2737 	rcu_read_unlock();
2738 	if (!task)
2739 		goto out;
2740 	if (!same_thread_group(leader, task))
2741 		goto out_drop_task;
2742 
2743 	result = proc_task_instantiate(dir, dentry, task, NULL);
2744 out_drop_task:
2745 	put_task_struct(task);
2746 out:
2747 	put_task_struct(leader);
2748 out_no_task:
2749 	return result;
2750 }
2751 
2752 /*
2753  * Find the first tid of a thread group to return to user space.
2754  *
2755  * Usually this is just the thread group leader, but if the users
2756  * buffer was too small or there was a seek into the middle of the
2757  * directory we have more work todo.
2758  *
2759  * In the case of a short read we start with find_task_by_pid.
2760  *
2761  * In the case of a seek we start with the leader and walk nr
2762  * threads past it.
2763  */
2764 static struct task_struct *first_tid(struct task_struct *leader,
2765 		int tid, int nr, struct pid_namespace *ns)
2766 {
2767 	struct task_struct *pos;
2768 
2769 	rcu_read_lock();
2770 	/* Attempt to start with the pid of a thread */
2771 	if (tid && (nr > 0)) {
2772 		pos = find_task_by_pid_ns(tid, ns);
2773 		if (pos && (pos->group_leader == leader))
2774 			goto found;
2775 	}
2776 
2777 	/* If nr exceeds the number of threads there is nothing todo */
2778 	pos = NULL;
2779 	if (nr && nr >= get_nr_threads(leader))
2780 		goto out;
2781 
2782 	/* If we haven't found our starting place yet start
2783 	 * with the leader and walk nr threads forward.
2784 	 */
2785 	for (pos = leader; nr > 0; --nr) {
2786 		pos = next_thread(pos);
2787 		if (pos == leader) {
2788 			pos = NULL;
2789 			goto out;
2790 		}
2791 	}
2792 found:
2793 	get_task_struct(pos);
2794 out:
2795 	rcu_read_unlock();
2796 	return pos;
2797 }
2798 
2799 /*
2800  * Find the next thread in the thread list.
2801  * Return NULL if there is an error or no next thread.
2802  *
2803  * The reference to the input task_struct is released.
2804  */
2805 static struct task_struct *next_tid(struct task_struct *start)
2806 {
2807 	struct task_struct *pos = NULL;
2808 	rcu_read_lock();
2809 	if (pid_alive(start)) {
2810 		pos = next_thread(start);
2811 		if (thread_group_leader(pos))
2812 			pos = NULL;
2813 		else
2814 			get_task_struct(pos);
2815 	}
2816 	rcu_read_unlock();
2817 	put_task_struct(start);
2818 	return pos;
2819 }
2820 
2821 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2822 	struct task_struct *task, int tid)
2823 {
2824 	char name[PROC_NUMBUF];
2825 	int len = snprintf(name, sizeof(name), "%d", tid);
2826 	return proc_fill_cache(filp, dirent, filldir, name, len,
2827 				proc_task_instantiate, task, NULL);
2828 }
2829 
2830 /* for the /proc/TGID/task/ directories */
2831 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2832 {
2833 	struct dentry *dentry = filp->f_path.dentry;
2834 	struct inode *inode = dentry->d_inode;
2835 	struct task_struct *leader = NULL;
2836 	struct task_struct *task;
2837 	int retval = -ENOENT;
2838 	ino_t ino;
2839 	int tid;
2840 	unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2841 	struct pid_namespace *ns;
2842 
2843 	task = get_proc_task(inode);
2844 	if (!task)
2845 		goto out_no_task;
2846 	rcu_read_lock();
2847 	if (pid_alive(task)) {
2848 		leader = task->group_leader;
2849 		get_task_struct(leader);
2850 	}
2851 	rcu_read_unlock();
2852 	put_task_struct(task);
2853 	if (!leader)
2854 		goto out_no_task;
2855 	retval = 0;
2856 
2857 	switch (pos) {
2858 	case 0:
2859 		ino = inode->i_ino;
2860 		if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2861 			goto out;
2862 		pos++;
2863 		/* fall through */
2864 	case 1:
2865 		ino = parent_ino(dentry);
2866 		if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2867 			goto out;
2868 		pos++;
2869 		/* fall through */
2870 	}
2871 
2872 	/* f_version caches the tgid value that the last readdir call couldn't
2873 	 * return. lseek aka telldir automagically resets f_version to 0.
2874 	 */
2875 	ns = filp->f_dentry->d_sb->s_fs_info;
2876 	tid = (int)filp->f_version;
2877 	filp->f_version = 0;
2878 	for (task = first_tid(leader, tid, pos - 2, ns);
2879 	     task;
2880 	     task = next_tid(task), pos++) {
2881 		tid = task_pid_nr_ns(task, ns);
2882 		if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2883 			/* returning this tgid failed, save it as the first
2884 			 * pid for the next readir call */
2885 			filp->f_version = (u64)tid;
2886 			put_task_struct(task);
2887 			break;
2888 		}
2889 	}
2890 out:
2891 	filp->f_pos = pos;
2892 	put_task_struct(leader);
2893 out_no_task:
2894 	return retval;
2895 }
2896 
2897 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2898 {
2899 	struct inode *inode = dentry->d_inode;
2900 	struct task_struct *p = get_proc_task(inode);
2901 	generic_fillattr(inode, stat);
2902 
2903 	if (p) {
2904 		rcu_read_lock();
2905 		stat->nlink += get_nr_threads(p);
2906 		rcu_read_unlock();
2907 		put_task_struct(p);
2908 	}
2909 
2910 	return 0;
2911 }
2912 
2913 static const struct inode_operations proc_task_inode_operations = {
2914 	.lookup		= proc_task_lookup,
2915 	.getattr	= proc_task_getattr,
2916 	.setattr	= proc_setattr,
2917 };
2918 
2919 static const struct file_operations proc_task_operations = {
2920 	.read		= generic_read_dir,
2921 	.readdir	= proc_task_readdir,
2922 };
2923