xref: /linux/fs/proc/base.c (revision b454cc6636d254fbf6049b73e9560aee76fb04a3)
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/smp_lock.h>
65 #include <linux/rcupdate.h>
66 #include <linux/kallsyms.h>
67 #include <linux/mount.h>
68 #include <linux/security.h>
69 #include <linux/ptrace.h>
70 #include <linux/seccomp.h>
71 #include <linux/cpuset.h>
72 #include <linux/audit.h>
73 #include <linux/poll.h>
74 #include <linux/nsproxy.h>
75 #include <linux/oom.h>
76 #include "internal.h"
77 
78 /* NOTE:
79  *	Implementing inode permission operations in /proc is almost
80  *	certainly an error.  Permission checks need to happen during
81  *	each system call not at open time.  The reason is that most of
82  *	what we wish to check for permissions in /proc varies at runtime.
83  *
84  *	The classic example of a problem is opening file descriptors
85  *	in /proc for a task before it execs a suid executable.
86  */
87 
88 
89 /* Worst case buffer size needed for holding an integer. */
90 #define PROC_NUMBUF 13
91 
92 struct pid_entry {
93 	int len;
94 	char *name;
95 	mode_t mode;
96 	struct inode_operations *iop;
97 	struct file_operations *fop;
98 	union proc_op op;
99 };
100 
101 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
102 	.len  = sizeof(NAME) - 1,			\
103 	.name = (NAME),					\
104 	.mode = MODE,					\
105 	.iop  = IOP,					\
106 	.fop  = FOP,					\
107 	.op   = OP,					\
108 }
109 
110 #define DIR(NAME, MODE, OTYPE)							\
111 	NOD(NAME, (S_IFDIR|(MODE)),						\
112 		&proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,	\
113 		{} )
114 #define LNK(NAME, OTYPE)					\
115 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
116 		&proc_pid_link_inode_operations, NULL,		\
117 		{ .proc_get_link = &proc_##OTYPE##_link } )
118 #define REG(NAME, MODE, OTYPE)				\
119 	NOD(NAME, (S_IFREG|(MODE)), NULL,		\
120 		&proc_##OTYPE##_operations, {})
121 #define INF(NAME, MODE, OTYPE)				\
122 	NOD(NAME, (S_IFREG|(MODE)), 			\
123 		NULL, &proc_info_file_operations,	\
124 		{ .proc_read = &proc_##OTYPE } )
125 
126 static struct fs_struct *get_fs_struct(struct task_struct *task)
127 {
128 	struct fs_struct *fs;
129 	task_lock(task);
130 	fs = task->fs;
131 	if(fs)
132 		atomic_inc(&fs->count);
133 	task_unlock(task);
134 	return fs;
135 }
136 
137 static int get_nr_threads(struct task_struct *tsk)
138 {
139 	/* Must be called with the rcu_read_lock held */
140 	unsigned long flags;
141 	int count = 0;
142 
143 	if (lock_task_sighand(tsk, &flags)) {
144 		count = atomic_read(&tsk->signal->count);
145 		unlock_task_sighand(tsk, &flags);
146 	}
147 	return count;
148 }
149 
150 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
151 {
152 	struct task_struct *task = get_proc_task(inode);
153 	struct fs_struct *fs = NULL;
154 	int result = -ENOENT;
155 
156 	if (task) {
157 		fs = get_fs_struct(task);
158 		put_task_struct(task);
159 	}
160 	if (fs) {
161 		read_lock(&fs->lock);
162 		*mnt = mntget(fs->pwdmnt);
163 		*dentry = dget(fs->pwd);
164 		read_unlock(&fs->lock);
165 		result = 0;
166 		put_fs_struct(fs);
167 	}
168 	return result;
169 }
170 
171 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
172 {
173 	struct task_struct *task = get_proc_task(inode);
174 	struct fs_struct *fs = NULL;
175 	int result = -ENOENT;
176 
177 	if (task) {
178 		fs = get_fs_struct(task);
179 		put_task_struct(task);
180 	}
181 	if (fs) {
182 		read_lock(&fs->lock);
183 		*mnt = mntget(fs->rootmnt);
184 		*dentry = dget(fs->root);
185 		read_unlock(&fs->lock);
186 		result = 0;
187 		put_fs_struct(fs);
188 	}
189 	return result;
190 }
191 
192 #define MAY_PTRACE(task) \
193 	(task == current || \
194 	(task->parent == current && \
195 	(task->ptrace & PT_PTRACED) && \
196 	 (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
197 	 security_ptrace(current,task) == 0))
198 
199 static int proc_pid_environ(struct task_struct *task, char * buffer)
200 {
201 	int res = 0;
202 	struct mm_struct *mm = get_task_mm(task);
203 	if (mm) {
204 		unsigned int len = mm->env_end - mm->env_start;
205 		if (len > PAGE_SIZE)
206 			len = PAGE_SIZE;
207 		res = access_process_vm(task, mm->env_start, buffer, len, 0);
208 		if (!ptrace_may_attach(task))
209 			res = -ESRCH;
210 		mmput(mm);
211 	}
212 	return res;
213 }
214 
215 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
216 {
217 	int res = 0;
218 	unsigned int len;
219 	struct mm_struct *mm = get_task_mm(task);
220 	if (!mm)
221 		goto out;
222 	if (!mm->arg_end)
223 		goto out_mm;	/* Shh! No looking before we're done */
224 
225  	len = mm->arg_end - mm->arg_start;
226 
227 	if (len > PAGE_SIZE)
228 		len = PAGE_SIZE;
229 
230 	res = access_process_vm(task, mm->arg_start, buffer, len, 0);
231 
232 	// If the nul at the end of args has been overwritten, then
233 	// assume application is using setproctitle(3).
234 	if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
235 		len = strnlen(buffer, res);
236 		if (len < res) {
237 		    res = len;
238 		} else {
239 			len = mm->env_end - mm->env_start;
240 			if (len > PAGE_SIZE - res)
241 				len = PAGE_SIZE - res;
242 			res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
243 			res = strnlen(buffer, res);
244 		}
245 	}
246 out_mm:
247 	mmput(mm);
248 out:
249 	return res;
250 }
251 
252 static int proc_pid_auxv(struct task_struct *task, char *buffer)
253 {
254 	int res = 0;
255 	struct mm_struct *mm = get_task_mm(task);
256 	if (mm) {
257 		unsigned int nwords = 0;
258 		do
259 			nwords += 2;
260 		while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
261 		res = nwords * sizeof(mm->saved_auxv[0]);
262 		if (res > PAGE_SIZE)
263 			res = PAGE_SIZE;
264 		memcpy(buffer, mm->saved_auxv, res);
265 		mmput(mm);
266 	}
267 	return res;
268 }
269 
270 
271 #ifdef CONFIG_KALLSYMS
272 /*
273  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
274  * Returns the resolved symbol.  If that fails, simply return the address.
275  */
276 static int proc_pid_wchan(struct task_struct *task, char *buffer)
277 {
278 	char *modname;
279 	const char *sym_name;
280 	unsigned long wchan, size, offset;
281 	char namebuf[KSYM_NAME_LEN+1];
282 
283 	wchan = get_wchan(task);
284 
285 	sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
286 	if (sym_name)
287 		return sprintf(buffer, "%s", sym_name);
288 	return sprintf(buffer, "%lu", wchan);
289 }
290 #endif /* CONFIG_KALLSYMS */
291 
292 #ifdef CONFIG_SCHEDSTATS
293 /*
294  * Provides /proc/PID/schedstat
295  */
296 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
297 {
298 	return sprintf(buffer, "%lu %lu %lu\n",
299 			task->sched_info.cpu_time,
300 			task->sched_info.run_delay,
301 			task->sched_info.pcnt);
302 }
303 #endif
304 
305 /* The badness from the OOM killer */
306 unsigned long badness(struct task_struct *p, unsigned long uptime);
307 static int proc_oom_score(struct task_struct *task, char *buffer)
308 {
309 	unsigned long points;
310 	struct timespec uptime;
311 
312 	do_posix_clock_monotonic_gettime(&uptime);
313 	points = badness(task, uptime.tv_sec);
314 	return sprintf(buffer, "%lu\n", points);
315 }
316 
317 /************************************************************************/
318 /*                       Here the fs part begins                        */
319 /************************************************************************/
320 
321 /* permission checks */
322 static int proc_fd_access_allowed(struct inode *inode)
323 {
324 	struct task_struct *task;
325 	int allowed = 0;
326 	/* Allow access to a task's file descriptors if it is us or we
327 	 * may use ptrace attach to the process and find out that
328 	 * information.
329 	 */
330 	task = get_proc_task(inode);
331 	if (task) {
332 		allowed = ptrace_may_attach(task);
333 		put_task_struct(task);
334 	}
335 	return allowed;
336 }
337 
338 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
339 {
340 	int error;
341 	struct inode *inode = dentry->d_inode;
342 
343 	if (attr->ia_valid & ATTR_MODE)
344 		return -EPERM;
345 
346 	error = inode_change_ok(inode, attr);
347 	if (!error) {
348 		error = security_inode_setattr(dentry, attr);
349 		if (!error)
350 			error = inode_setattr(inode, attr);
351 	}
352 	return error;
353 }
354 
355 static struct inode_operations proc_def_inode_operations = {
356 	.setattr	= proc_setattr,
357 };
358 
359 extern struct seq_operations mounts_op;
360 struct proc_mounts {
361 	struct seq_file m;
362 	int event;
363 };
364 
365 static int mounts_open(struct inode *inode, struct file *file)
366 {
367 	struct task_struct *task = get_proc_task(inode);
368 	struct mnt_namespace *ns = NULL;
369 	struct proc_mounts *p;
370 	int ret = -EINVAL;
371 
372 	if (task) {
373 		task_lock(task);
374 		if (task->nsproxy) {
375 			ns = task->nsproxy->mnt_ns;
376 			if (ns)
377 				get_mnt_ns(ns);
378 		}
379 		task_unlock(task);
380 		put_task_struct(task);
381 	}
382 
383 	if (ns) {
384 		ret = -ENOMEM;
385 		p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
386 		if (p) {
387 			file->private_data = &p->m;
388 			ret = seq_open(file, &mounts_op);
389 			if (!ret) {
390 				p->m.private = ns;
391 				p->event = ns->event;
392 				return 0;
393 			}
394 			kfree(p);
395 		}
396 		put_mnt_ns(ns);
397 	}
398 	return ret;
399 }
400 
401 static int mounts_release(struct inode *inode, struct file *file)
402 {
403 	struct seq_file *m = file->private_data;
404 	struct mnt_namespace *ns = m->private;
405 	put_mnt_ns(ns);
406 	return seq_release(inode, file);
407 }
408 
409 static unsigned mounts_poll(struct file *file, poll_table *wait)
410 {
411 	struct proc_mounts *p = file->private_data;
412 	struct mnt_namespace *ns = p->m.private;
413 	unsigned res = 0;
414 
415 	poll_wait(file, &ns->poll, wait);
416 
417 	spin_lock(&vfsmount_lock);
418 	if (p->event != ns->event) {
419 		p->event = ns->event;
420 		res = POLLERR;
421 	}
422 	spin_unlock(&vfsmount_lock);
423 
424 	return res;
425 }
426 
427 static struct file_operations proc_mounts_operations = {
428 	.open		= mounts_open,
429 	.read		= seq_read,
430 	.llseek		= seq_lseek,
431 	.release	= mounts_release,
432 	.poll		= mounts_poll,
433 };
434 
435 extern struct seq_operations mountstats_op;
436 static int mountstats_open(struct inode *inode, struct file *file)
437 {
438 	int ret = seq_open(file, &mountstats_op);
439 
440 	if (!ret) {
441 		struct seq_file *m = file->private_data;
442 		struct mnt_namespace *mnt_ns = NULL;
443 		struct task_struct *task = get_proc_task(inode);
444 
445 		if (task) {
446 			task_lock(task);
447 			if (task->nsproxy)
448 				mnt_ns = task->nsproxy->mnt_ns;
449 			if (mnt_ns)
450 				get_mnt_ns(mnt_ns);
451 			task_unlock(task);
452 			put_task_struct(task);
453 		}
454 
455 		if (mnt_ns)
456 			m->private = mnt_ns;
457 		else {
458 			seq_release(inode, file);
459 			ret = -EINVAL;
460 		}
461 	}
462 	return ret;
463 }
464 
465 static struct file_operations proc_mountstats_operations = {
466 	.open		= mountstats_open,
467 	.read		= seq_read,
468 	.llseek		= seq_lseek,
469 	.release	= mounts_release,
470 };
471 
472 #define PROC_BLOCK_SIZE	(3*1024)		/* 4K page size but our output routines use some slack for overruns */
473 
474 static ssize_t proc_info_read(struct file * file, char __user * buf,
475 			  size_t count, loff_t *ppos)
476 {
477 	struct inode * inode = file->f_path.dentry->d_inode;
478 	unsigned long page;
479 	ssize_t length;
480 	struct task_struct *task = get_proc_task(inode);
481 
482 	length = -ESRCH;
483 	if (!task)
484 		goto out_no_task;
485 
486 	if (count > PROC_BLOCK_SIZE)
487 		count = PROC_BLOCK_SIZE;
488 
489 	length = -ENOMEM;
490 	if (!(page = __get_free_page(GFP_KERNEL)))
491 		goto out;
492 
493 	length = PROC_I(inode)->op.proc_read(task, (char*)page);
494 
495 	if (length >= 0)
496 		length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
497 	free_page(page);
498 out:
499 	put_task_struct(task);
500 out_no_task:
501 	return length;
502 }
503 
504 static struct file_operations proc_info_file_operations = {
505 	.read		= proc_info_read,
506 };
507 
508 static int mem_open(struct inode* inode, struct file* file)
509 {
510 	file->private_data = (void*)((long)current->self_exec_id);
511 	return 0;
512 }
513 
514 static ssize_t mem_read(struct file * file, char __user * buf,
515 			size_t count, loff_t *ppos)
516 {
517 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
518 	char *page;
519 	unsigned long src = *ppos;
520 	int ret = -ESRCH;
521 	struct mm_struct *mm;
522 
523 	if (!task)
524 		goto out_no_task;
525 
526 	if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
527 		goto out;
528 
529 	ret = -ENOMEM;
530 	page = (char *)__get_free_page(GFP_USER);
531 	if (!page)
532 		goto out;
533 
534 	ret = 0;
535 
536 	mm = get_task_mm(task);
537 	if (!mm)
538 		goto out_free;
539 
540 	ret = -EIO;
541 
542 	if (file->private_data != (void*)((long)current->self_exec_id))
543 		goto out_put;
544 
545 	ret = 0;
546 
547 	while (count > 0) {
548 		int this_len, retval;
549 
550 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
551 		retval = access_process_vm(task, src, page, this_len, 0);
552 		if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
553 			if (!ret)
554 				ret = -EIO;
555 			break;
556 		}
557 
558 		if (copy_to_user(buf, page, retval)) {
559 			ret = -EFAULT;
560 			break;
561 		}
562 
563 		ret += retval;
564 		src += retval;
565 		buf += retval;
566 		count -= retval;
567 	}
568 	*ppos = src;
569 
570 out_put:
571 	mmput(mm);
572 out_free:
573 	free_page((unsigned long) page);
574 out:
575 	put_task_struct(task);
576 out_no_task:
577 	return ret;
578 }
579 
580 #define mem_write NULL
581 
582 #ifndef mem_write
583 /* This is a security hazard */
584 static ssize_t mem_write(struct file * file, const char * buf,
585 			 size_t count, loff_t *ppos)
586 {
587 	int copied;
588 	char *page;
589 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
590 	unsigned long dst = *ppos;
591 
592 	copied = -ESRCH;
593 	if (!task)
594 		goto out_no_task;
595 
596 	if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
597 		goto out;
598 
599 	copied = -ENOMEM;
600 	page = (char *)__get_free_page(GFP_USER);
601 	if (!page)
602 		goto out;
603 
604 	copied = 0;
605 	while (count > 0) {
606 		int this_len, retval;
607 
608 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
609 		if (copy_from_user(page, buf, this_len)) {
610 			copied = -EFAULT;
611 			break;
612 		}
613 		retval = access_process_vm(task, dst, page, this_len, 1);
614 		if (!retval) {
615 			if (!copied)
616 				copied = -EIO;
617 			break;
618 		}
619 		copied += retval;
620 		buf += retval;
621 		dst += retval;
622 		count -= retval;
623 	}
624 	*ppos = dst;
625 	free_page((unsigned long) page);
626 out:
627 	put_task_struct(task);
628 out_no_task:
629 	return copied;
630 }
631 #endif
632 
633 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
634 {
635 	switch (orig) {
636 	case 0:
637 		file->f_pos = offset;
638 		break;
639 	case 1:
640 		file->f_pos += offset;
641 		break;
642 	default:
643 		return -EINVAL;
644 	}
645 	force_successful_syscall_return();
646 	return file->f_pos;
647 }
648 
649 static struct file_operations proc_mem_operations = {
650 	.llseek		= mem_lseek,
651 	.read		= mem_read,
652 	.write		= mem_write,
653 	.open		= mem_open,
654 };
655 
656 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
657 				size_t count, loff_t *ppos)
658 {
659 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
660 	char buffer[PROC_NUMBUF];
661 	size_t len;
662 	int oom_adjust;
663 	loff_t __ppos = *ppos;
664 
665 	if (!task)
666 		return -ESRCH;
667 	oom_adjust = task->oomkilladj;
668 	put_task_struct(task);
669 
670 	len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
671 	if (__ppos >= len)
672 		return 0;
673 	if (count > len-__ppos)
674 		count = len-__ppos;
675 	if (copy_to_user(buf, buffer + __ppos, count))
676 		return -EFAULT;
677 	*ppos = __ppos + count;
678 	return count;
679 }
680 
681 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
682 				size_t count, loff_t *ppos)
683 {
684 	struct task_struct *task;
685 	char buffer[PROC_NUMBUF], *end;
686 	int oom_adjust;
687 
688 	memset(buffer, 0, sizeof(buffer));
689 	if (count > sizeof(buffer) - 1)
690 		count = sizeof(buffer) - 1;
691 	if (copy_from_user(buffer, buf, count))
692 		return -EFAULT;
693 	oom_adjust = simple_strtol(buffer, &end, 0);
694 	if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
695 	     oom_adjust != OOM_DISABLE)
696 		return -EINVAL;
697 	if (*end == '\n')
698 		end++;
699 	task = get_proc_task(file->f_path.dentry->d_inode);
700 	if (!task)
701 		return -ESRCH;
702 	if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
703 		put_task_struct(task);
704 		return -EACCES;
705 	}
706 	task->oomkilladj = oom_adjust;
707 	put_task_struct(task);
708 	if (end - buffer == 0)
709 		return -EIO;
710 	return end - buffer;
711 }
712 
713 static struct file_operations proc_oom_adjust_operations = {
714 	.read		= oom_adjust_read,
715 	.write		= oom_adjust_write,
716 };
717 
718 #ifdef CONFIG_AUDITSYSCALL
719 #define TMPBUFLEN 21
720 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
721 				  size_t count, loff_t *ppos)
722 {
723 	struct inode * inode = file->f_path.dentry->d_inode;
724 	struct task_struct *task = get_proc_task(inode);
725 	ssize_t length;
726 	char tmpbuf[TMPBUFLEN];
727 
728 	if (!task)
729 		return -ESRCH;
730 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
731 				audit_get_loginuid(task->audit_context));
732 	put_task_struct(task);
733 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
734 }
735 
736 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
737 				   size_t count, loff_t *ppos)
738 {
739 	struct inode * inode = file->f_path.dentry->d_inode;
740 	char *page, *tmp;
741 	ssize_t length;
742 	uid_t loginuid;
743 
744 	if (!capable(CAP_AUDIT_CONTROL))
745 		return -EPERM;
746 
747 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
748 		return -EPERM;
749 
750 	if (count >= PAGE_SIZE)
751 		count = PAGE_SIZE - 1;
752 
753 	if (*ppos != 0) {
754 		/* No partial writes. */
755 		return -EINVAL;
756 	}
757 	page = (char*)__get_free_page(GFP_USER);
758 	if (!page)
759 		return -ENOMEM;
760 	length = -EFAULT;
761 	if (copy_from_user(page, buf, count))
762 		goto out_free_page;
763 
764 	page[count] = '\0';
765 	loginuid = simple_strtoul(page, &tmp, 10);
766 	if (tmp == page) {
767 		length = -EINVAL;
768 		goto out_free_page;
769 
770 	}
771 	length = audit_set_loginuid(current, loginuid);
772 	if (likely(length == 0))
773 		length = count;
774 
775 out_free_page:
776 	free_page((unsigned long) page);
777 	return length;
778 }
779 
780 static struct file_operations proc_loginuid_operations = {
781 	.read		= proc_loginuid_read,
782 	.write		= proc_loginuid_write,
783 };
784 #endif
785 
786 #ifdef CONFIG_SECCOMP
787 static ssize_t seccomp_read(struct file *file, char __user *buf,
788 			    size_t count, loff_t *ppos)
789 {
790 	struct task_struct *tsk = get_proc_task(file->f_dentry->d_inode);
791 	char __buf[20];
792 	loff_t __ppos = *ppos;
793 	size_t len;
794 
795 	if (!tsk)
796 		return -ESRCH;
797 	/* no need to print the trailing zero, so use only len */
798 	len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
799 	put_task_struct(tsk);
800 	if (__ppos >= len)
801 		return 0;
802 	if (count > len - __ppos)
803 		count = len - __ppos;
804 	if (copy_to_user(buf, __buf + __ppos, count))
805 		return -EFAULT;
806 	*ppos = __ppos + count;
807 	return count;
808 }
809 
810 static ssize_t seccomp_write(struct file *file, const char __user *buf,
811 			     size_t count, loff_t *ppos)
812 {
813 	struct task_struct *tsk = get_proc_task(file->f_dentry->d_inode);
814 	char __buf[20], *end;
815 	unsigned int seccomp_mode;
816 	ssize_t result;
817 
818 	result = -ESRCH;
819 	if (!tsk)
820 		goto out_no_task;
821 
822 	/* can set it only once to be even more secure */
823 	result = -EPERM;
824 	if (unlikely(tsk->seccomp.mode))
825 		goto out;
826 
827 	result = -EFAULT;
828 	memset(__buf, 0, sizeof(__buf));
829 	count = min(count, sizeof(__buf) - 1);
830 	if (copy_from_user(__buf, buf, count))
831 		goto out;
832 
833 	seccomp_mode = simple_strtoul(__buf, &end, 0);
834 	if (*end == '\n')
835 		end++;
836 	result = -EINVAL;
837 	if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
838 		tsk->seccomp.mode = seccomp_mode;
839 		set_tsk_thread_flag(tsk, TIF_SECCOMP);
840 	} else
841 		goto out;
842 	result = -EIO;
843 	if (unlikely(!(end - __buf)))
844 		goto out;
845 	result = end - __buf;
846 out:
847 	put_task_struct(tsk);
848 out_no_task:
849 	return result;
850 }
851 
852 static struct file_operations proc_seccomp_operations = {
853 	.read		= seccomp_read,
854 	.write		= seccomp_write,
855 };
856 #endif /* CONFIG_SECCOMP */
857 
858 #ifdef CONFIG_FAULT_INJECTION
859 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
860 				      size_t count, loff_t *ppos)
861 {
862 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
863 	char buffer[PROC_NUMBUF];
864 	size_t len;
865 	int make_it_fail;
866 	loff_t __ppos = *ppos;
867 
868 	if (!task)
869 		return -ESRCH;
870 	make_it_fail = task->make_it_fail;
871 	put_task_struct(task);
872 
873 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
874 	if (__ppos >= len)
875 		return 0;
876 	if (count > len-__ppos)
877 		count = len-__ppos;
878 	if (copy_to_user(buf, buffer + __ppos, count))
879 		return -EFAULT;
880 	*ppos = __ppos + count;
881 	return count;
882 }
883 
884 static ssize_t proc_fault_inject_write(struct file * file,
885 			const char __user * buf, size_t count, loff_t *ppos)
886 {
887 	struct task_struct *task;
888 	char buffer[PROC_NUMBUF], *end;
889 	int make_it_fail;
890 
891 	if (!capable(CAP_SYS_RESOURCE))
892 		return -EPERM;
893 	memset(buffer, 0, sizeof(buffer));
894 	if (count > sizeof(buffer) - 1)
895 		count = sizeof(buffer) - 1;
896 	if (copy_from_user(buffer, buf, count))
897 		return -EFAULT;
898 	make_it_fail = simple_strtol(buffer, &end, 0);
899 	if (*end == '\n')
900 		end++;
901 	task = get_proc_task(file->f_dentry->d_inode);
902 	if (!task)
903 		return -ESRCH;
904 	task->make_it_fail = make_it_fail;
905 	put_task_struct(task);
906 	if (end - buffer == 0)
907 		return -EIO;
908 	return end - buffer;
909 }
910 
911 static struct file_operations proc_fault_inject_operations = {
912 	.read		= proc_fault_inject_read,
913 	.write		= proc_fault_inject_write,
914 };
915 #endif
916 
917 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
918 {
919 	struct inode *inode = dentry->d_inode;
920 	int error = -EACCES;
921 
922 	/* We don't need a base pointer in the /proc filesystem */
923 	path_release(nd);
924 
925 	/* Are we allowed to snoop on the tasks file descriptors? */
926 	if (!proc_fd_access_allowed(inode))
927 		goto out;
928 
929 	error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
930 	nd->last_type = LAST_BIND;
931 out:
932 	return ERR_PTR(error);
933 }
934 
935 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
936 			    char __user *buffer, int buflen)
937 {
938 	struct inode * inode;
939 	char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
940 	int len;
941 
942 	if (!tmp)
943 		return -ENOMEM;
944 
945 	inode = dentry->d_inode;
946 	path = d_path(dentry, mnt, tmp, PAGE_SIZE);
947 	len = PTR_ERR(path);
948 	if (IS_ERR(path))
949 		goto out;
950 	len = tmp + PAGE_SIZE - 1 - path;
951 
952 	if (len > buflen)
953 		len = buflen;
954 	if (copy_to_user(buffer, path, len))
955 		len = -EFAULT;
956  out:
957 	free_page((unsigned long)tmp);
958 	return len;
959 }
960 
961 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
962 {
963 	int error = -EACCES;
964 	struct inode *inode = dentry->d_inode;
965 	struct dentry *de;
966 	struct vfsmount *mnt = NULL;
967 
968 	/* Are we allowed to snoop on the tasks file descriptors? */
969 	if (!proc_fd_access_allowed(inode))
970 		goto out;
971 
972 	error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
973 	if (error)
974 		goto out;
975 
976 	error = do_proc_readlink(de, mnt, buffer, buflen);
977 	dput(de);
978 	mntput(mnt);
979 out:
980 	return error;
981 }
982 
983 static struct inode_operations proc_pid_link_inode_operations = {
984 	.readlink	= proc_pid_readlink,
985 	.follow_link	= proc_pid_follow_link,
986 	.setattr	= proc_setattr,
987 };
988 
989 
990 /* building an inode */
991 
992 static int task_dumpable(struct task_struct *task)
993 {
994 	int dumpable = 0;
995 	struct mm_struct *mm;
996 
997 	task_lock(task);
998 	mm = task->mm;
999 	if (mm)
1000 		dumpable = mm->dumpable;
1001 	task_unlock(task);
1002 	if(dumpable == 1)
1003 		return 1;
1004 	return 0;
1005 }
1006 
1007 
1008 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1009 {
1010 	struct inode * inode;
1011 	struct proc_inode *ei;
1012 
1013 	/* We need a new inode */
1014 
1015 	inode = new_inode(sb);
1016 	if (!inode)
1017 		goto out;
1018 
1019 	/* Common stuff */
1020 	ei = PROC_I(inode);
1021 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1022 	inode->i_op = &proc_def_inode_operations;
1023 
1024 	/*
1025 	 * grab the reference to task.
1026 	 */
1027 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1028 	if (!ei->pid)
1029 		goto out_unlock;
1030 
1031 	inode->i_uid = 0;
1032 	inode->i_gid = 0;
1033 	if (task_dumpable(task)) {
1034 		inode->i_uid = task->euid;
1035 		inode->i_gid = task->egid;
1036 	}
1037 	security_task_to_inode(task, inode);
1038 
1039 out:
1040 	return inode;
1041 
1042 out_unlock:
1043 	iput(inode);
1044 	return NULL;
1045 }
1046 
1047 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1048 {
1049 	struct inode *inode = dentry->d_inode;
1050 	struct task_struct *task;
1051 	generic_fillattr(inode, stat);
1052 
1053 	rcu_read_lock();
1054 	stat->uid = 0;
1055 	stat->gid = 0;
1056 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1057 	if (task) {
1058 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1059 		    task_dumpable(task)) {
1060 			stat->uid = task->euid;
1061 			stat->gid = task->egid;
1062 		}
1063 	}
1064 	rcu_read_unlock();
1065 	return 0;
1066 }
1067 
1068 /* dentry stuff */
1069 
1070 /*
1071  *	Exceptional case: normally we are not allowed to unhash a busy
1072  * directory. In this case, however, we can do it - no aliasing problems
1073  * due to the way we treat inodes.
1074  *
1075  * Rewrite the inode's ownerships here because the owning task may have
1076  * performed a setuid(), etc.
1077  *
1078  * Before the /proc/pid/status file was created the only way to read
1079  * the effective uid of a /process was to stat /proc/pid.  Reading
1080  * /proc/pid/status is slow enough that procps and other packages
1081  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1082  * made this apply to all per process world readable and executable
1083  * directories.
1084  */
1085 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1086 {
1087 	struct inode *inode = dentry->d_inode;
1088 	struct task_struct *task = get_proc_task(inode);
1089 	if (task) {
1090 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1091 		    task_dumpable(task)) {
1092 			inode->i_uid = task->euid;
1093 			inode->i_gid = task->egid;
1094 		} else {
1095 			inode->i_uid = 0;
1096 			inode->i_gid = 0;
1097 		}
1098 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1099 		security_task_to_inode(task, inode);
1100 		put_task_struct(task);
1101 		return 1;
1102 	}
1103 	d_drop(dentry);
1104 	return 0;
1105 }
1106 
1107 static int pid_delete_dentry(struct dentry * dentry)
1108 {
1109 	/* Is the task we represent dead?
1110 	 * If so, then don't put the dentry on the lru list,
1111 	 * kill it immediately.
1112 	 */
1113 	return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1114 }
1115 
1116 static struct dentry_operations pid_dentry_operations =
1117 {
1118 	.d_revalidate	= pid_revalidate,
1119 	.d_delete	= pid_delete_dentry,
1120 };
1121 
1122 /* Lookups */
1123 
1124 typedef struct dentry *instantiate_t(struct inode *, struct dentry *, struct task_struct *, void *);
1125 
1126 /*
1127  * Fill a directory entry.
1128  *
1129  * If possible create the dcache entry and derive our inode number and
1130  * file type from dcache entry.
1131  *
1132  * Since all of the proc inode numbers are dynamically generated, the inode
1133  * numbers do not exist until the inode is cache.  This means creating the
1134  * the dcache entry in readdir is necessary to keep the inode numbers
1135  * reported by readdir in sync with the inode numbers reported
1136  * by stat.
1137  */
1138 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1139 	char *name, int len,
1140 	instantiate_t instantiate, struct task_struct *task, void *ptr)
1141 {
1142 	struct dentry *child, *dir = filp->f_path.dentry;
1143 	struct inode *inode;
1144 	struct qstr qname;
1145 	ino_t ino = 0;
1146 	unsigned type = DT_UNKNOWN;
1147 
1148 	qname.name = name;
1149 	qname.len  = len;
1150 	qname.hash = full_name_hash(name, len);
1151 
1152 	child = d_lookup(dir, &qname);
1153 	if (!child) {
1154 		struct dentry *new;
1155 		new = d_alloc(dir, &qname);
1156 		if (new) {
1157 			child = instantiate(dir->d_inode, new, task, ptr);
1158 			if (child)
1159 				dput(new);
1160 			else
1161 				child = new;
1162 		}
1163 	}
1164 	if (!child || IS_ERR(child) || !child->d_inode)
1165 		goto end_instantiate;
1166 	inode = child->d_inode;
1167 	if (inode) {
1168 		ino = inode->i_ino;
1169 		type = inode->i_mode >> 12;
1170 	}
1171 	dput(child);
1172 end_instantiate:
1173 	if (!ino)
1174 		ino = find_inode_number(dir, &qname);
1175 	if (!ino)
1176 		ino = 1;
1177 	return filldir(dirent, name, len, filp->f_pos, ino, type);
1178 }
1179 
1180 static unsigned name_to_int(struct dentry *dentry)
1181 {
1182 	const char *name = dentry->d_name.name;
1183 	int len = dentry->d_name.len;
1184 	unsigned n = 0;
1185 
1186 	if (len > 1 && *name == '0')
1187 		goto out;
1188 	while (len-- > 0) {
1189 		unsigned c = *name++ - '0';
1190 		if (c > 9)
1191 			goto out;
1192 		if (n >= (~0U-9)/10)
1193 			goto out;
1194 		n *= 10;
1195 		n += c;
1196 	}
1197 	return n;
1198 out:
1199 	return ~0U;
1200 }
1201 
1202 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
1203 {
1204 	struct task_struct *task = get_proc_task(inode);
1205 	struct files_struct *files = NULL;
1206 	struct file *file;
1207 	int fd = proc_fd(inode);
1208 
1209 	if (task) {
1210 		files = get_files_struct(task);
1211 		put_task_struct(task);
1212 	}
1213 	if (files) {
1214 		/*
1215 		 * We are not taking a ref to the file structure, so we must
1216 		 * hold ->file_lock.
1217 		 */
1218 		spin_lock(&files->file_lock);
1219 		file = fcheck_files(files, fd);
1220 		if (file) {
1221 			*mnt = mntget(file->f_path.mnt);
1222 			*dentry = dget(file->f_path.dentry);
1223 			spin_unlock(&files->file_lock);
1224 			put_files_struct(files);
1225 			return 0;
1226 		}
1227 		spin_unlock(&files->file_lock);
1228 		put_files_struct(files);
1229 	}
1230 	return -ENOENT;
1231 }
1232 
1233 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1234 {
1235 	struct inode *inode = dentry->d_inode;
1236 	struct task_struct *task = get_proc_task(inode);
1237 	int fd = proc_fd(inode);
1238 	struct files_struct *files;
1239 
1240 	if (task) {
1241 		files = get_files_struct(task);
1242 		if (files) {
1243 			rcu_read_lock();
1244 			if (fcheck_files(files, fd)) {
1245 				rcu_read_unlock();
1246 				put_files_struct(files);
1247 				if (task_dumpable(task)) {
1248 					inode->i_uid = task->euid;
1249 					inode->i_gid = task->egid;
1250 				} else {
1251 					inode->i_uid = 0;
1252 					inode->i_gid = 0;
1253 				}
1254 				inode->i_mode &= ~(S_ISUID | S_ISGID);
1255 				security_task_to_inode(task, inode);
1256 				put_task_struct(task);
1257 				return 1;
1258 			}
1259 			rcu_read_unlock();
1260 			put_files_struct(files);
1261 		}
1262 		put_task_struct(task);
1263 	}
1264 	d_drop(dentry);
1265 	return 0;
1266 }
1267 
1268 static struct dentry_operations tid_fd_dentry_operations =
1269 {
1270 	.d_revalidate	= tid_fd_revalidate,
1271 	.d_delete	= pid_delete_dentry,
1272 };
1273 
1274 static struct dentry *proc_fd_instantiate(struct inode *dir,
1275 	struct dentry *dentry, struct task_struct *task, void *ptr)
1276 {
1277 	unsigned fd = *(unsigned *)ptr;
1278 	struct file *file;
1279 	struct files_struct *files;
1280  	struct inode *inode;
1281  	struct proc_inode *ei;
1282 	struct dentry *error = ERR_PTR(-ENOENT);
1283 
1284 	inode = proc_pid_make_inode(dir->i_sb, task);
1285 	if (!inode)
1286 		goto out;
1287 	ei = PROC_I(inode);
1288 	ei->fd = fd;
1289 	files = get_files_struct(task);
1290 	if (!files)
1291 		goto out_iput;
1292 	inode->i_mode = S_IFLNK;
1293 
1294 	/*
1295 	 * We are not taking a ref to the file structure, so we must
1296 	 * hold ->file_lock.
1297 	 */
1298 	spin_lock(&files->file_lock);
1299 	file = fcheck_files(files, fd);
1300 	if (!file)
1301 		goto out_unlock;
1302 	if (file->f_mode & 1)
1303 		inode->i_mode |= S_IRUSR | S_IXUSR;
1304 	if (file->f_mode & 2)
1305 		inode->i_mode |= S_IWUSR | S_IXUSR;
1306 	spin_unlock(&files->file_lock);
1307 	put_files_struct(files);
1308 
1309 	inode->i_op = &proc_pid_link_inode_operations;
1310 	inode->i_size = 64;
1311 	ei->op.proc_get_link = proc_fd_link;
1312 	dentry->d_op = &tid_fd_dentry_operations;
1313 	d_add(dentry, inode);
1314 	/* Close the race of the process dying before we return the dentry */
1315 	if (tid_fd_revalidate(dentry, NULL))
1316 		error = NULL;
1317 
1318  out:
1319 	return error;
1320 out_unlock:
1321 	spin_unlock(&files->file_lock);
1322 	put_files_struct(files);
1323 out_iput:
1324 	iput(inode);
1325 	goto out;
1326 }
1327 
1328 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1329 {
1330 	struct task_struct *task = get_proc_task(dir);
1331 	unsigned fd = name_to_int(dentry);
1332 	struct dentry *result = ERR_PTR(-ENOENT);
1333 
1334 	if (!task)
1335 		goto out_no_task;
1336 	if (fd == ~0U)
1337 		goto out;
1338 
1339 	result = proc_fd_instantiate(dir, dentry, task, &fd);
1340 out:
1341 	put_task_struct(task);
1342 out_no_task:
1343 	return result;
1344 }
1345 
1346 static int proc_fd_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1347 	struct task_struct *task, int fd)
1348 {
1349 	char name[PROC_NUMBUF];
1350 	int len = snprintf(name, sizeof(name), "%d", fd);
1351 	return proc_fill_cache(filp, dirent, filldir, name, len,
1352 				proc_fd_instantiate, task, &fd);
1353 }
1354 
1355 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1356 {
1357 	struct dentry *dentry = filp->f_path.dentry;
1358 	struct inode *inode = dentry->d_inode;
1359 	struct task_struct *p = get_proc_task(inode);
1360 	unsigned int fd, tid, ino;
1361 	int retval;
1362 	struct files_struct * files;
1363 	struct fdtable *fdt;
1364 
1365 	retval = -ENOENT;
1366 	if (!p)
1367 		goto out_no_task;
1368 	retval = 0;
1369 	tid = p->pid;
1370 
1371 	fd = filp->f_pos;
1372 	switch (fd) {
1373 		case 0:
1374 			if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1375 				goto out;
1376 			filp->f_pos++;
1377 		case 1:
1378 			ino = parent_ino(dentry);
1379 			if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1380 				goto out;
1381 			filp->f_pos++;
1382 		default:
1383 			files = get_files_struct(p);
1384 			if (!files)
1385 				goto out;
1386 			rcu_read_lock();
1387 			fdt = files_fdtable(files);
1388 			for (fd = filp->f_pos-2;
1389 			     fd < fdt->max_fds;
1390 			     fd++, filp->f_pos++) {
1391 
1392 				if (!fcheck_files(files, fd))
1393 					continue;
1394 				rcu_read_unlock();
1395 
1396 				if (proc_fd_fill_cache(filp, dirent, filldir, p, fd) < 0) {
1397 					rcu_read_lock();
1398 					break;
1399 				}
1400 				rcu_read_lock();
1401 			}
1402 			rcu_read_unlock();
1403 			put_files_struct(files);
1404 	}
1405 out:
1406 	put_task_struct(p);
1407 out_no_task:
1408 	return retval;
1409 }
1410 
1411 static struct file_operations proc_fd_operations = {
1412 	.read		= generic_read_dir,
1413 	.readdir	= proc_readfd,
1414 };
1415 
1416 /*
1417  * proc directories can do almost nothing..
1418  */
1419 static struct inode_operations proc_fd_inode_operations = {
1420 	.lookup		= proc_lookupfd,
1421 	.setattr	= proc_setattr,
1422 };
1423 
1424 static struct dentry *proc_pident_instantiate(struct inode *dir,
1425 	struct dentry *dentry, struct task_struct *task, void *ptr)
1426 {
1427 	struct pid_entry *p = ptr;
1428 	struct inode *inode;
1429 	struct proc_inode *ei;
1430 	struct dentry *error = ERR_PTR(-EINVAL);
1431 
1432 	inode = proc_pid_make_inode(dir->i_sb, task);
1433 	if (!inode)
1434 		goto out;
1435 
1436 	ei = PROC_I(inode);
1437 	inode->i_mode = p->mode;
1438 	if (S_ISDIR(inode->i_mode))
1439 		inode->i_nlink = 2;	/* Use getattr to fix if necessary */
1440 	if (p->iop)
1441 		inode->i_op = p->iop;
1442 	if (p->fop)
1443 		inode->i_fop = p->fop;
1444 	ei->op = p->op;
1445 	dentry->d_op = &pid_dentry_operations;
1446 	d_add(dentry, inode);
1447 	/* Close the race of the process dying before we return the dentry */
1448 	if (pid_revalidate(dentry, NULL))
1449 		error = NULL;
1450 out:
1451 	return error;
1452 }
1453 
1454 static struct dentry *proc_pident_lookup(struct inode *dir,
1455 					 struct dentry *dentry,
1456 					 struct pid_entry *ents,
1457 					 unsigned int nents)
1458 {
1459 	struct inode *inode;
1460 	struct dentry *error;
1461 	struct task_struct *task = get_proc_task(dir);
1462 	struct pid_entry *p, *last;
1463 
1464 	error = ERR_PTR(-ENOENT);
1465 	inode = NULL;
1466 
1467 	if (!task)
1468 		goto out_no_task;
1469 
1470 	/*
1471 	 * Yes, it does not scale. And it should not. Don't add
1472 	 * new entries into /proc/<tgid>/ without very good reasons.
1473 	 */
1474 	last = &ents[nents - 1];
1475 	for (p = ents; p <= last; p++) {
1476 		if (p->len != dentry->d_name.len)
1477 			continue;
1478 		if (!memcmp(dentry->d_name.name, p->name, p->len))
1479 			break;
1480 	}
1481 	if (p > last)
1482 		goto out;
1483 
1484 	error = proc_pident_instantiate(dir, dentry, task, p);
1485 out:
1486 	put_task_struct(task);
1487 out_no_task:
1488 	return error;
1489 }
1490 
1491 static int proc_pident_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1492 	struct task_struct *task, struct pid_entry *p)
1493 {
1494 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1495 				proc_pident_instantiate, task, p);
1496 }
1497 
1498 static int proc_pident_readdir(struct file *filp,
1499 		void *dirent, filldir_t filldir,
1500 		struct pid_entry *ents, unsigned int nents)
1501 {
1502 	int i;
1503 	int pid;
1504 	struct dentry *dentry = filp->f_path.dentry;
1505 	struct inode *inode = dentry->d_inode;
1506 	struct task_struct *task = get_proc_task(inode);
1507 	struct pid_entry *p, *last;
1508 	ino_t ino;
1509 	int ret;
1510 
1511 	ret = -ENOENT;
1512 	if (!task)
1513 		goto out_no_task;
1514 
1515 	ret = 0;
1516 	pid = task->pid;
1517 	i = filp->f_pos;
1518 	switch (i) {
1519 	case 0:
1520 		ino = inode->i_ino;
1521 		if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1522 			goto out;
1523 		i++;
1524 		filp->f_pos++;
1525 		/* fall through */
1526 	case 1:
1527 		ino = parent_ino(dentry);
1528 		if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1529 			goto out;
1530 		i++;
1531 		filp->f_pos++;
1532 		/* fall through */
1533 	default:
1534 		i -= 2;
1535 		if (i >= nents) {
1536 			ret = 1;
1537 			goto out;
1538 		}
1539 		p = ents + i;
1540 		last = &ents[nents - 1];
1541 		while (p <= last) {
1542 			if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1543 				goto out;
1544 			filp->f_pos++;
1545 			p++;
1546 		}
1547 	}
1548 
1549 	ret = 1;
1550 out:
1551 	put_task_struct(task);
1552 out_no_task:
1553 	return ret;
1554 }
1555 
1556 #ifdef CONFIG_SECURITY
1557 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1558 				  size_t count, loff_t *ppos)
1559 {
1560 	struct inode * inode = file->f_path.dentry->d_inode;
1561 	unsigned long page;
1562 	ssize_t length;
1563 	struct task_struct *task = get_proc_task(inode);
1564 
1565 	length = -ESRCH;
1566 	if (!task)
1567 		goto out_no_task;
1568 
1569 	if (count > PAGE_SIZE)
1570 		count = PAGE_SIZE;
1571 	length = -ENOMEM;
1572 	if (!(page = __get_free_page(GFP_KERNEL)))
1573 		goto out;
1574 
1575 	length = security_getprocattr(task,
1576 				      (char*)file->f_path.dentry->d_name.name,
1577 				      (void*)page, count);
1578 	if (length >= 0)
1579 		length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1580 	free_page(page);
1581 out:
1582 	put_task_struct(task);
1583 out_no_task:
1584 	return length;
1585 }
1586 
1587 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1588 				   size_t count, loff_t *ppos)
1589 {
1590 	struct inode * inode = file->f_path.dentry->d_inode;
1591 	char *page;
1592 	ssize_t length;
1593 	struct task_struct *task = get_proc_task(inode);
1594 
1595 	length = -ESRCH;
1596 	if (!task)
1597 		goto out_no_task;
1598 	if (count > PAGE_SIZE)
1599 		count = PAGE_SIZE;
1600 
1601 	/* No partial writes. */
1602 	length = -EINVAL;
1603 	if (*ppos != 0)
1604 		goto out;
1605 
1606 	length = -ENOMEM;
1607 	page = (char*)__get_free_page(GFP_USER);
1608 	if (!page)
1609 		goto out;
1610 
1611 	length = -EFAULT;
1612 	if (copy_from_user(page, buf, count))
1613 		goto out_free;
1614 
1615 	length = security_setprocattr(task,
1616 				      (char*)file->f_path.dentry->d_name.name,
1617 				      (void*)page, count);
1618 out_free:
1619 	free_page((unsigned long) page);
1620 out:
1621 	put_task_struct(task);
1622 out_no_task:
1623 	return length;
1624 }
1625 
1626 static struct file_operations proc_pid_attr_operations = {
1627 	.read		= proc_pid_attr_read,
1628 	.write		= proc_pid_attr_write,
1629 };
1630 
1631 static struct pid_entry attr_dir_stuff[] = {
1632 	REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1633 	REG("prev",       S_IRUGO,	   pid_attr),
1634 	REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1635 	REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1636 	REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1637 	REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1638 };
1639 
1640 static int proc_attr_dir_readdir(struct file * filp,
1641 			     void * dirent, filldir_t filldir)
1642 {
1643 	return proc_pident_readdir(filp,dirent,filldir,
1644 				   attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1645 }
1646 
1647 static struct file_operations proc_attr_dir_operations = {
1648 	.read		= generic_read_dir,
1649 	.readdir	= proc_attr_dir_readdir,
1650 };
1651 
1652 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
1653 				struct dentry *dentry, struct nameidata *nd)
1654 {
1655 	return proc_pident_lookup(dir, dentry,
1656 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
1657 }
1658 
1659 static struct inode_operations proc_attr_dir_inode_operations = {
1660 	.lookup		= proc_attr_dir_lookup,
1661 	.getattr	= pid_getattr,
1662 	.setattr	= proc_setattr,
1663 };
1664 
1665 #endif
1666 
1667 /*
1668  * /proc/self:
1669  */
1670 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1671 			      int buflen)
1672 {
1673 	char tmp[PROC_NUMBUF];
1674 	sprintf(tmp, "%d", current->tgid);
1675 	return vfs_readlink(dentry,buffer,buflen,tmp);
1676 }
1677 
1678 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1679 {
1680 	char tmp[PROC_NUMBUF];
1681 	sprintf(tmp, "%d", current->tgid);
1682 	return ERR_PTR(vfs_follow_link(nd,tmp));
1683 }
1684 
1685 static struct inode_operations proc_self_inode_operations = {
1686 	.readlink	= proc_self_readlink,
1687 	.follow_link	= proc_self_follow_link,
1688 };
1689 
1690 /*
1691  * proc base
1692  *
1693  * These are the directory entries in the root directory of /proc
1694  * that properly belong to the /proc filesystem, as they describe
1695  * describe something that is process related.
1696  */
1697 static struct pid_entry proc_base_stuff[] = {
1698 	NOD("self", S_IFLNK|S_IRWXUGO,
1699 		&proc_self_inode_operations, NULL, {}),
1700 };
1701 
1702 /*
1703  *	Exceptional case: normally we are not allowed to unhash a busy
1704  * directory. In this case, however, we can do it - no aliasing problems
1705  * due to the way we treat inodes.
1706  */
1707 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
1708 {
1709 	struct inode *inode = dentry->d_inode;
1710 	struct task_struct *task = get_proc_task(inode);
1711 	if (task) {
1712 		put_task_struct(task);
1713 		return 1;
1714 	}
1715 	d_drop(dentry);
1716 	return 0;
1717 }
1718 
1719 static struct dentry_operations proc_base_dentry_operations =
1720 {
1721 	.d_revalidate	= proc_base_revalidate,
1722 	.d_delete	= pid_delete_dentry,
1723 };
1724 
1725 static struct dentry *proc_base_instantiate(struct inode *dir,
1726 	struct dentry *dentry, struct task_struct *task, void *ptr)
1727 {
1728 	struct pid_entry *p = ptr;
1729 	struct inode *inode;
1730 	struct proc_inode *ei;
1731 	struct dentry *error = ERR_PTR(-EINVAL);
1732 
1733 	/* Allocate the inode */
1734 	error = ERR_PTR(-ENOMEM);
1735 	inode = new_inode(dir->i_sb);
1736 	if (!inode)
1737 		goto out;
1738 
1739 	/* Initialize the inode */
1740 	ei = PROC_I(inode);
1741 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1742 
1743 	/*
1744 	 * grab the reference to the task.
1745 	 */
1746 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1747 	if (!ei->pid)
1748 		goto out_iput;
1749 
1750 	inode->i_uid = 0;
1751 	inode->i_gid = 0;
1752 	inode->i_mode = p->mode;
1753 	if (S_ISDIR(inode->i_mode))
1754 		inode->i_nlink = 2;
1755 	if (S_ISLNK(inode->i_mode))
1756 		inode->i_size = 64;
1757 	if (p->iop)
1758 		inode->i_op = p->iop;
1759 	if (p->fop)
1760 		inode->i_fop = p->fop;
1761 	ei->op = p->op;
1762 	dentry->d_op = &proc_base_dentry_operations;
1763 	d_add(dentry, inode);
1764 	error = NULL;
1765 out:
1766 	return error;
1767 out_iput:
1768 	iput(inode);
1769 	goto out;
1770 }
1771 
1772 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
1773 {
1774 	struct dentry *error;
1775 	struct task_struct *task = get_proc_task(dir);
1776 	struct pid_entry *p, *last;
1777 
1778 	error = ERR_PTR(-ENOENT);
1779 
1780 	if (!task)
1781 		goto out_no_task;
1782 
1783 	/* Lookup the directory entry */
1784 	last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
1785 	for (p = proc_base_stuff; p <= last; p++) {
1786 		if (p->len != dentry->d_name.len)
1787 			continue;
1788 		if (!memcmp(dentry->d_name.name, p->name, p->len))
1789 			break;
1790 	}
1791 	if (p > last)
1792 		goto out;
1793 
1794 	error = proc_base_instantiate(dir, dentry, task, p);
1795 
1796 out:
1797 	put_task_struct(task);
1798 out_no_task:
1799 	return error;
1800 }
1801 
1802 static int proc_base_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1803 	struct task_struct *task, struct pid_entry *p)
1804 {
1805 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1806 				proc_base_instantiate, task, p);
1807 }
1808 
1809 #ifdef CONFIG_TASK_IO_ACCOUNTING
1810 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
1811 {
1812 	return sprintf(buffer,
1813 			"rchar: %llu\n"
1814 			"wchar: %llu\n"
1815 			"syscr: %llu\n"
1816 			"syscw: %llu\n"
1817 			"read_bytes: %llu\n"
1818 			"write_bytes: %llu\n"
1819 			"cancelled_write_bytes: %llu\n",
1820 			(unsigned long long)task->rchar,
1821 			(unsigned long long)task->wchar,
1822 			(unsigned long long)task->syscr,
1823 			(unsigned long long)task->syscw,
1824 			(unsigned long long)task->ioac.read_bytes,
1825 			(unsigned long long)task->ioac.write_bytes,
1826 			(unsigned long long)task->ioac.cancelled_write_bytes);
1827 }
1828 #endif
1829 
1830 /*
1831  * Thread groups
1832  */
1833 static struct file_operations proc_task_operations;
1834 static struct inode_operations proc_task_inode_operations;
1835 
1836 static struct pid_entry tgid_base_stuff[] = {
1837 	DIR("task",       S_IRUGO|S_IXUGO, task),
1838 	DIR("fd",         S_IRUSR|S_IXUSR, fd),
1839 	INF("environ",    S_IRUSR, pid_environ),
1840 	INF("auxv",       S_IRUSR, pid_auxv),
1841 	INF("status",     S_IRUGO, pid_status),
1842 	INF("cmdline",    S_IRUGO, pid_cmdline),
1843 	INF("stat",       S_IRUGO, tgid_stat),
1844 	INF("statm",      S_IRUGO, pid_statm),
1845 	REG("maps",       S_IRUGO, maps),
1846 #ifdef CONFIG_NUMA
1847 	REG("numa_maps",  S_IRUGO, numa_maps),
1848 #endif
1849 	REG("mem",        S_IRUSR|S_IWUSR, mem),
1850 #ifdef CONFIG_SECCOMP
1851 	REG("seccomp",    S_IRUSR|S_IWUSR, seccomp),
1852 #endif
1853 	LNK("cwd",        cwd),
1854 	LNK("root",       root),
1855 	LNK("exe",        exe),
1856 	REG("mounts",     S_IRUGO, mounts),
1857 	REG("mountstats", S_IRUSR, mountstats),
1858 #ifdef CONFIG_MMU
1859 	REG("smaps",      S_IRUGO, smaps),
1860 #endif
1861 #ifdef CONFIG_SECURITY
1862 	DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
1863 #endif
1864 #ifdef CONFIG_KALLSYMS
1865 	INF("wchan",      S_IRUGO, pid_wchan),
1866 #endif
1867 #ifdef CONFIG_SCHEDSTATS
1868 	INF("schedstat",  S_IRUGO, pid_schedstat),
1869 #endif
1870 #ifdef CONFIG_CPUSETS
1871 	REG("cpuset",     S_IRUGO, cpuset),
1872 #endif
1873 	INF("oom_score",  S_IRUGO, oom_score),
1874 	REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
1875 #ifdef CONFIG_AUDITSYSCALL
1876 	REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
1877 #endif
1878 #ifdef CONFIG_FAULT_INJECTION
1879 	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
1880 #endif
1881 #ifdef CONFIG_TASK_IO_ACCOUNTING
1882 	INF("io",	S_IRUGO, pid_io_accounting),
1883 #endif
1884 };
1885 
1886 static int proc_tgid_base_readdir(struct file * filp,
1887 			     void * dirent, filldir_t filldir)
1888 {
1889 	return proc_pident_readdir(filp,dirent,filldir,
1890 				   tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1891 }
1892 
1893 static struct file_operations proc_tgid_base_operations = {
1894 	.read		= generic_read_dir,
1895 	.readdir	= proc_tgid_base_readdir,
1896 };
1897 
1898 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1899 	return proc_pident_lookup(dir, dentry,
1900 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
1901 }
1902 
1903 static struct inode_operations proc_tgid_base_inode_operations = {
1904 	.lookup		= proc_tgid_base_lookup,
1905 	.getattr	= pid_getattr,
1906 	.setattr	= proc_setattr,
1907 };
1908 
1909 /**
1910  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
1911  *
1912  * @task: task that should be flushed.
1913  *
1914  * Looks in the dcache for
1915  * /proc/@pid
1916  * /proc/@tgid/task/@pid
1917  * if either directory is present flushes it and all of it'ts children
1918  * from the dcache.
1919  *
1920  * It is safe and reasonable to cache /proc entries for a task until
1921  * that task exits.  After that they just clog up the dcache with
1922  * useless entries, possibly causing useful dcache entries to be
1923  * flushed instead.  This routine is proved to flush those useless
1924  * dcache entries at process exit time.
1925  *
1926  * NOTE: This routine is just an optimization so it does not guarantee
1927  *       that no dcache entries will exist at process exit time it
1928  *       just makes it very unlikely that any will persist.
1929  */
1930 void proc_flush_task(struct task_struct *task)
1931 {
1932 	struct dentry *dentry, *leader, *dir;
1933 	char buf[PROC_NUMBUF];
1934 	struct qstr name;
1935 
1936 	name.name = buf;
1937 	name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
1938 	dentry = d_hash_and_lookup(proc_mnt->mnt_root, &name);
1939 	if (dentry) {
1940 		shrink_dcache_parent(dentry);
1941 		d_drop(dentry);
1942 		dput(dentry);
1943 	}
1944 
1945 	if (thread_group_leader(task))
1946 		goto out;
1947 
1948 	name.name = buf;
1949 	name.len = snprintf(buf, sizeof(buf), "%d", task->tgid);
1950 	leader = d_hash_and_lookup(proc_mnt->mnt_root, &name);
1951 	if (!leader)
1952 		goto out;
1953 
1954 	name.name = "task";
1955 	name.len = strlen(name.name);
1956 	dir = d_hash_and_lookup(leader, &name);
1957 	if (!dir)
1958 		goto out_put_leader;
1959 
1960 	name.name = buf;
1961 	name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
1962 	dentry = d_hash_and_lookup(dir, &name);
1963 	if (dentry) {
1964 		shrink_dcache_parent(dentry);
1965 		d_drop(dentry);
1966 		dput(dentry);
1967 	}
1968 
1969 	dput(dir);
1970 out_put_leader:
1971 	dput(leader);
1972 out:
1973 	return;
1974 }
1975 
1976 static struct dentry *proc_pid_instantiate(struct inode *dir,
1977 					   struct dentry * dentry,
1978 					   struct task_struct *task, void *ptr)
1979 {
1980 	struct dentry *error = ERR_PTR(-ENOENT);
1981 	struct inode *inode;
1982 
1983 	inode = proc_pid_make_inode(dir->i_sb, task);
1984 	if (!inode)
1985 		goto out;
1986 
1987 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
1988 	inode->i_op = &proc_tgid_base_inode_operations;
1989 	inode->i_fop = &proc_tgid_base_operations;
1990 	inode->i_flags|=S_IMMUTABLE;
1991 	inode->i_nlink = 4;
1992 #ifdef CONFIG_SECURITY
1993 	inode->i_nlink += 1;
1994 #endif
1995 
1996 	dentry->d_op = &pid_dentry_operations;
1997 
1998 	d_add(dentry, inode);
1999 	/* Close the race of the process dying before we return the dentry */
2000 	if (pid_revalidate(dentry, NULL))
2001 		error = NULL;
2002 out:
2003 	return error;
2004 }
2005 
2006 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2007 {
2008 	struct dentry *result = ERR_PTR(-ENOENT);
2009 	struct task_struct *task;
2010 	unsigned tgid;
2011 
2012 	result = proc_base_lookup(dir, dentry);
2013 	if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2014 		goto out;
2015 
2016 	tgid = name_to_int(dentry);
2017 	if (tgid == ~0U)
2018 		goto out;
2019 
2020 	rcu_read_lock();
2021 	task = find_task_by_pid(tgid);
2022 	if (task)
2023 		get_task_struct(task);
2024 	rcu_read_unlock();
2025 	if (!task)
2026 		goto out;
2027 
2028 	result = proc_pid_instantiate(dir, dentry, task, NULL);
2029 	put_task_struct(task);
2030 out:
2031 	return result;
2032 }
2033 
2034 /*
2035  * Find the first task with tgid >= tgid
2036  *
2037  */
2038 static struct task_struct *next_tgid(unsigned int tgid)
2039 {
2040 	struct task_struct *task;
2041 	struct pid *pid;
2042 
2043 	rcu_read_lock();
2044 retry:
2045 	task = NULL;
2046 	pid = find_ge_pid(tgid);
2047 	if (pid) {
2048 		tgid = pid->nr + 1;
2049 		task = pid_task(pid, PIDTYPE_PID);
2050 		/* What we to know is if the pid we have find is the
2051 		 * pid of a thread_group_leader.  Testing for task
2052 		 * being a thread_group_leader is the obvious thing
2053 		 * todo but there is a window when it fails, due to
2054 		 * the pid transfer logic in de_thread.
2055 		 *
2056 		 * So we perform the straight forward test of seeing
2057 		 * if the pid we have found is the pid of a thread
2058 		 * group leader, and don't worry if the task we have
2059 		 * found doesn't happen to be a thread group leader.
2060 		 * As we don't care in the case of readdir.
2061 		 */
2062 		if (!task || !has_group_leader_pid(task))
2063 			goto retry;
2064 		get_task_struct(task);
2065 	}
2066 	rcu_read_unlock();
2067 	return task;
2068 }
2069 
2070 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2071 
2072 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2073 	struct task_struct *task, int tgid)
2074 {
2075 	char name[PROC_NUMBUF];
2076 	int len = snprintf(name, sizeof(name), "%d", tgid);
2077 	return proc_fill_cache(filp, dirent, filldir, name, len,
2078 				proc_pid_instantiate, task, NULL);
2079 }
2080 
2081 /* for the /proc/ directory itself, after non-process stuff has been done */
2082 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2083 {
2084 	unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2085 	struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2086 	struct task_struct *task;
2087 	int tgid;
2088 
2089 	if (!reaper)
2090 		goto out_no_task;
2091 
2092 	for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2093 		struct pid_entry *p = &proc_base_stuff[nr];
2094 		if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2095 			goto out;
2096 	}
2097 
2098 	tgid = filp->f_pos - TGID_OFFSET;
2099 	for (task = next_tgid(tgid);
2100 	     task;
2101 	     put_task_struct(task), task = next_tgid(tgid + 1)) {
2102 		tgid = task->pid;
2103 		filp->f_pos = tgid + TGID_OFFSET;
2104 		if (proc_pid_fill_cache(filp, dirent, filldir, task, tgid) < 0) {
2105 			put_task_struct(task);
2106 			goto out;
2107 		}
2108 	}
2109 	filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2110 out:
2111 	put_task_struct(reaper);
2112 out_no_task:
2113 	return 0;
2114 }
2115 
2116 /*
2117  * Tasks
2118  */
2119 static struct pid_entry tid_base_stuff[] = {
2120 	DIR("fd",        S_IRUSR|S_IXUSR, fd),
2121 	INF("environ",   S_IRUSR, pid_environ),
2122 	INF("auxv",      S_IRUSR, pid_auxv),
2123 	INF("status",    S_IRUGO, pid_status),
2124 	INF("cmdline",   S_IRUGO, pid_cmdline),
2125 	INF("stat",      S_IRUGO, tid_stat),
2126 	INF("statm",     S_IRUGO, pid_statm),
2127 	REG("maps",      S_IRUGO, maps),
2128 #ifdef CONFIG_NUMA
2129 	REG("numa_maps", S_IRUGO, numa_maps),
2130 #endif
2131 	REG("mem",       S_IRUSR|S_IWUSR, mem),
2132 #ifdef CONFIG_SECCOMP
2133 	REG("seccomp",   S_IRUSR|S_IWUSR, seccomp),
2134 #endif
2135 	LNK("cwd",       cwd),
2136 	LNK("root",      root),
2137 	LNK("exe",       exe),
2138 	REG("mounts",    S_IRUGO, mounts),
2139 #ifdef CONFIG_MMU
2140 	REG("smaps",     S_IRUGO, smaps),
2141 #endif
2142 #ifdef CONFIG_SECURITY
2143 	DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2144 #endif
2145 #ifdef CONFIG_KALLSYMS
2146 	INF("wchan",     S_IRUGO, pid_wchan),
2147 #endif
2148 #ifdef CONFIG_SCHEDSTATS
2149 	INF("schedstat", S_IRUGO, pid_schedstat),
2150 #endif
2151 #ifdef CONFIG_CPUSETS
2152 	REG("cpuset",    S_IRUGO, cpuset),
2153 #endif
2154 	INF("oom_score", S_IRUGO, oom_score),
2155 	REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2156 #ifdef CONFIG_AUDITSYSCALL
2157 	REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2158 #endif
2159 #ifdef CONFIG_FAULT_INJECTION
2160 	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2161 #endif
2162 };
2163 
2164 static int proc_tid_base_readdir(struct file * filp,
2165 			     void * dirent, filldir_t filldir)
2166 {
2167 	return proc_pident_readdir(filp,dirent,filldir,
2168 				   tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2169 }
2170 
2171 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2172 	return proc_pident_lookup(dir, dentry,
2173 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2174 }
2175 
2176 static struct file_operations proc_tid_base_operations = {
2177 	.read		= generic_read_dir,
2178 	.readdir	= proc_tid_base_readdir,
2179 };
2180 
2181 static struct inode_operations proc_tid_base_inode_operations = {
2182 	.lookup		= proc_tid_base_lookup,
2183 	.getattr	= pid_getattr,
2184 	.setattr	= proc_setattr,
2185 };
2186 
2187 static struct dentry *proc_task_instantiate(struct inode *dir,
2188 	struct dentry *dentry, struct task_struct *task, void *ptr)
2189 {
2190 	struct dentry *error = ERR_PTR(-ENOENT);
2191 	struct inode *inode;
2192 	inode = proc_pid_make_inode(dir->i_sb, task);
2193 
2194 	if (!inode)
2195 		goto out;
2196 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2197 	inode->i_op = &proc_tid_base_inode_operations;
2198 	inode->i_fop = &proc_tid_base_operations;
2199 	inode->i_flags|=S_IMMUTABLE;
2200 	inode->i_nlink = 3;
2201 #ifdef CONFIG_SECURITY
2202 	inode->i_nlink += 1;
2203 #endif
2204 
2205 	dentry->d_op = &pid_dentry_operations;
2206 
2207 	d_add(dentry, inode);
2208 	/* Close the race of the process dying before we return the dentry */
2209 	if (pid_revalidate(dentry, NULL))
2210 		error = NULL;
2211 out:
2212 	return error;
2213 }
2214 
2215 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2216 {
2217 	struct dentry *result = ERR_PTR(-ENOENT);
2218 	struct task_struct *task;
2219 	struct task_struct *leader = get_proc_task(dir);
2220 	unsigned tid;
2221 
2222 	if (!leader)
2223 		goto out_no_task;
2224 
2225 	tid = name_to_int(dentry);
2226 	if (tid == ~0U)
2227 		goto out;
2228 
2229 	rcu_read_lock();
2230 	task = find_task_by_pid(tid);
2231 	if (task)
2232 		get_task_struct(task);
2233 	rcu_read_unlock();
2234 	if (!task)
2235 		goto out;
2236 	if (leader->tgid != task->tgid)
2237 		goto out_drop_task;
2238 
2239 	result = proc_task_instantiate(dir, dentry, task, NULL);
2240 out_drop_task:
2241 	put_task_struct(task);
2242 out:
2243 	put_task_struct(leader);
2244 out_no_task:
2245 	return result;
2246 }
2247 
2248 /*
2249  * Find the first tid of a thread group to return to user space.
2250  *
2251  * Usually this is just the thread group leader, but if the users
2252  * buffer was too small or there was a seek into the middle of the
2253  * directory we have more work todo.
2254  *
2255  * In the case of a short read we start with find_task_by_pid.
2256  *
2257  * In the case of a seek we start with the leader and walk nr
2258  * threads past it.
2259  */
2260 static struct task_struct *first_tid(struct task_struct *leader,
2261 					int tid, int nr)
2262 {
2263 	struct task_struct *pos;
2264 
2265 	rcu_read_lock();
2266 	/* Attempt to start with the pid of a thread */
2267 	if (tid && (nr > 0)) {
2268 		pos = find_task_by_pid(tid);
2269 		if (pos && (pos->group_leader == leader))
2270 			goto found;
2271 	}
2272 
2273 	/* If nr exceeds the number of threads there is nothing todo */
2274 	pos = NULL;
2275 	if (nr && nr >= get_nr_threads(leader))
2276 		goto out;
2277 
2278 	/* If we haven't found our starting place yet start
2279 	 * with the leader and walk nr threads forward.
2280 	 */
2281 	for (pos = leader; nr > 0; --nr) {
2282 		pos = next_thread(pos);
2283 		if (pos == leader) {
2284 			pos = NULL;
2285 			goto out;
2286 		}
2287 	}
2288 found:
2289 	get_task_struct(pos);
2290 out:
2291 	rcu_read_unlock();
2292 	return pos;
2293 }
2294 
2295 /*
2296  * Find the next thread in the thread list.
2297  * Return NULL if there is an error or no next thread.
2298  *
2299  * The reference to the input task_struct is released.
2300  */
2301 static struct task_struct *next_tid(struct task_struct *start)
2302 {
2303 	struct task_struct *pos = NULL;
2304 	rcu_read_lock();
2305 	if (pid_alive(start)) {
2306 		pos = next_thread(start);
2307 		if (thread_group_leader(pos))
2308 			pos = NULL;
2309 		else
2310 			get_task_struct(pos);
2311 	}
2312 	rcu_read_unlock();
2313 	put_task_struct(start);
2314 	return pos;
2315 }
2316 
2317 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2318 	struct task_struct *task, int tid)
2319 {
2320 	char name[PROC_NUMBUF];
2321 	int len = snprintf(name, sizeof(name), "%d", tid);
2322 	return proc_fill_cache(filp, dirent, filldir, name, len,
2323 				proc_task_instantiate, task, NULL);
2324 }
2325 
2326 /* for the /proc/TGID/task/ directories */
2327 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2328 {
2329 	struct dentry *dentry = filp->f_path.dentry;
2330 	struct inode *inode = dentry->d_inode;
2331 	struct task_struct *leader = NULL;
2332 	struct task_struct *task;
2333 	int retval = -ENOENT;
2334 	ino_t ino;
2335 	int tid;
2336 	unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2337 
2338 	task = get_proc_task(inode);
2339 	if (!task)
2340 		goto out_no_task;
2341 	rcu_read_lock();
2342 	if (pid_alive(task)) {
2343 		leader = task->group_leader;
2344 		get_task_struct(leader);
2345 	}
2346 	rcu_read_unlock();
2347 	put_task_struct(task);
2348 	if (!leader)
2349 		goto out_no_task;
2350 	retval = 0;
2351 
2352 	switch (pos) {
2353 	case 0:
2354 		ino = inode->i_ino;
2355 		if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2356 			goto out;
2357 		pos++;
2358 		/* fall through */
2359 	case 1:
2360 		ino = parent_ino(dentry);
2361 		if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2362 			goto out;
2363 		pos++;
2364 		/* fall through */
2365 	}
2366 
2367 	/* f_version caches the tgid value that the last readdir call couldn't
2368 	 * return. lseek aka telldir automagically resets f_version to 0.
2369 	 */
2370 	tid = filp->f_version;
2371 	filp->f_version = 0;
2372 	for (task = first_tid(leader, tid, pos - 2);
2373 	     task;
2374 	     task = next_tid(task), pos++) {
2375 		tid = task->pid;
2376 		if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2377 			/* returning this tgid failed, save it as the first
2378 			 * pid for the next readir call */
2379 			filp->f_version = tid;
2380 			put_task_struct(task);
2381 			break;
2382 		}
2383 	}
2384 out:
2385 	filp->f_pos = pos;
2386 	put_task_struct(leader);
2387 out_no_task:
2388 	return retval;
2389 }
2390 
2391 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2392 {
2393 	struct inode *inode = dentry->d_inode;
2394 	struct task_struct *p = get_proc_task(inode);
2395 	generic_fillattr(inode, stat);
2396 
2397 	if (p) {
2398 		rcu_read_lock();
2399 		stat->nlink += get_nr_threads(p);
2400 		rcu_read_unlock();
2401 		put_task_struct(p);
2402 	}
2403 
2404 	return 0;
2405 }
2406 
2407 static struct inode_operations proc_task_inode_operations = {
2408 	.lookup		= proc_task_lookup,
2409 	.getattr	= proc_task_getattr,
2410 	.setattr	= proc_setattr,
2411 };
2412 
2413 static struct file_operations proc_task_operations = {
2414 	.read		= generic_read_dir,
2415 	.readdir	= proc_task_readdir,
2416 };
2417