xref: /linux/fs/proc/base.c (revision 42fda66387daa53538ae13a2c858396aaf037158)
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/module.h>
67 #include <linux/mount.h>
68 #include <linux/security.h>
69 #include <linux/ptrace.h>
70 #include <linux/cpuset.h>
71 #include <linux/audit.h>
72 #include <linux/poll.h>
73 #include <linux/nsproxy.h>
74 #include <linux/oom.h>
75 #include <linux/elf.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 	char *name;
94 	int len;
95 	mode_t mode;
96 	const struct inode_operations *iop;
97 	const struct file_operations *fop;
98 	union proc_op op;
99 };
100 
101 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
102 	.name = (NAME),					\
103 	.len  = sizeof(NAME) - 1,			\
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 int maps_protect;
127 EXPORT_SYMBOL(maps_protect);
128 
129 static struct fs_struct *get_fs_struct(struct task_struct *task)
130 {
131 	struct fs_struct *fs;
132 	task_lock(task);
133 	fs = task->fs;
134 	if(fs)
135 		atomic_inc(&fs->count);
136 	task_unlock(task);
137 	return fs;
138 }
139 
140 static int get_nr_threads(struct task_struct *tsk)
141 {
142 	/* Must be called with the rcu_read_lock held */
143 	unsigned long flags;
144 	int count = 0;
145 
146 	if (lock_task_sighand(tsk, &flags)) {
147 		count = atomic_read(&tsk->signal->count);
148 		unlock_task_sighand(tsk, &flags);
149 	}
150 	return count;
151 }
152 
153 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
154 {
155 	struct task_struct *task = get_proc_task(inode);
156 	struct fs_struct *fs = NULL;
157 	int result = -ENOENT;
158 
159 	if (task) {
160 		fs = get_fs_struct(task);
161 		put_task_struct(task);
162 	}
163 	if (fs) {
164 		read_lock(&fs->lock);
165 		*mnt = mntget(fs->pwdmnt);
166 		*dentry = dget(fs->pwd);
167 		read_unlock(&fs->lock);
168 		result = 0;
169 		put_fs_struct(fs);
170 	}
171 	return result;
172 }
173 
174 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
175 {
176 	struct task_struct *task = get_proc_task(inode);
177 	struct fs_struct *fs = NULL;
178 	int result = -ENOENT;
179 
180 	if (task) {
181 		fs = get_fs_struct(task);
182 		put_task_struct(task);
183 	}
184 	if (fs) {
185 		read_lock(&fs->lock);
186 		*mnt = mntget(fs->rootmnt);
187 		*dentry = dget(fs->root);
188 		read_unlock(&fs->lock);
189 		result = 0;
190 		put_fs_struct(fs);
191 	}
192 	return result;
193 }
194 
195 #define MAY_PTRACE(task) \
196 	(task == current || \
197 	(task->parent == current && \
198 	(task->ptrace & PT_PTRACED) && \
199 	 (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
200 	 security_ptrace(current,task) == 0))
201 
202 static int proc_pid_environ(struct task_struct *task, char * buffer)
203 {
204 	int res = 0;
205 	struct mm_struct *mm = get_task_mm(task);
206 	if (mm) {
207 		unsigned int len;
208 
209 		res = -ESRCH;
210 		if (!ptrace_may_attach(task))
211 			goto out;
212 
213 		len  = mm->env_end - mm->env_start;
214 		if (len > PAGE_SIZE)
215 			len = PAGE_SIZE;
216 		res = access_process_vm(task, mm->env_start, buffer, len, 0);
217 out:
218 		mmput(mm);
219 	}
220 	return res;
221 }
222 
223 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
224 {
225 	int res = 0;
226 	unsigned int len;
227 	struct mm_struct *mm = get_task_mm(task);
228 	if (!mm)
229 		goto out;
230 	if (!mm->arg_end)
231 		goto out_mm;	/* Shh! No looking before we're done */
232 
233  	len = mm->arg_end - mm->arg_start;
234 
235 	if (len > PAGE_SIZE)
236 		len = PAGE_SIZE;
237 
238 	res = access_process_vm(task, mm->arg_start, buffer, len, 0);
239 
240 	// If the nul at the end of args has been overwritten, then
241 	// assume application is using setproctitle(3).
242 	if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
243 		len = strnlen(buffer, res);
244 		if (len < res) {
245 		    res = len;
246 		} else {
247 			len = mm->env_end - mm->env_start;
248 			if (len > PAGE_SIZE - res)
249 				len = PAGE_SIZE - res;
250 			res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
251 			res = strnlen(buffer, res);
252 		}
253 	}
254 out_mm:
255 	mmput(mm);
256 out:
257 	return res;
258 }
259 
260 static int proc_pid_auxv(struct task_struct *task, char *buffer)
261 {
262 	int res = 0;
263 	struct mm_struct *mm = get_task_mm(task);
264 	if (mm) {
265 		unsigned int nwords = 0;
266 		do
267 			nwords += 2;
268 		while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
269 		res = nwords * sizeof(mm->saved_auxv[0]);
270 		if (res > PAGE_SIZE)
271 			res = PAGE_SIZE;
272 		memcpy(buffer, mm->saved_auxv, res);
273 		mmput(mm);
274 	}
275 	return res;
276 }
277 
278 
279 #ifdef CONFIG_KALLSYMS
280 /*
281  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
282  * Returns the resolved symbol.  If that fails, simply return the address.
283  */
284 static int proc_pid_wchan(struct task_struct *task, char *buffer)
285 {
286 	unsigned long wchan;
287 	char symname[KSYM_NAME_LEN];
288 
289 	wchan = get_wchan(task);
290 
291 	if (lookup_symbol_name(wchan, symname) < 0)
292 		return sprintf(buffer, "%lu", wchan);
293 	else
294 		return sprintf(buffer, "%s", symname);
295 }
296 #endif /* CONFIG_KALLSYMS */
297 
298 #ifdef CONFIG_SCHEDSTATS
299 /*
300  * Provides /proc/PID/schedstat
301  */
302 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
303 {
304 	return sprintf(buffer, "%llu %llu %lu\n",
305 			task->sched_info.cpu_time,
306 			task->sched_info.run_delay,
307 			task->sched_info.pcount);
308 }
309 #endif
310 
311 /* The badness from the OOM killer */
312 unsigned long badness(struct task_struct *p, unsigned long uptime);
313 static int proc_oom_score(struct task_struct *task, char *buffer)
314 {
315 	unsigned long points;
316 	struct timespec uptime;
317 
318 	do_posix_clock_monotonic_gettime(&uptime);
319 	read_lock(&tasklist_lock);
320 	points = badness(task, uptime.tv_sec);
321 	read_unlock(&tasklist_lock);
322 	return sprintf(buffer, "%lu\n", points);
323 }
324 
325 /************************************************************************/
326 /*                       Here the fs part begins                        */
327 /************************************************************************/
328 
329 /* permission checks */
330 static int proc_fd_access_allowed(struct inode *inode)
331 {
332 	struct task_struct *task;
333 	int allowed = 0;
334 	/* Allow access to a task's file descriptors if it is us or we
335 	 * may use ptrace attach to the process and find out that
336 	 * information.
337 	 */
338 	task = get_proc_task(inode);
339 	if (task) {
340 		allowed = ptrace_may_attach(task);
341 		put_task_struct(task);
342 	}
343 	return allowed;
344 }
345 
346 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
347 {
348 	int error;
349 	struct inode *inode = dentry->d_inode;
350 
351 	if (attr->ia_valid & ATTR_MODE)
352 		return -EPERM;
353 
354 	error = inode_change_ok(inode, attr);
355 	if (!error)
356 		error = inode_setattr(inode, attr);
357 	return error;
358 }
359 
360 static const struct inode_operations proc_def_inode_operations = {
361 	.setattr	= proc_setattr,
362 };
363 
364 extern struct seq_operations mounts_op;
365 struct proc_mounts {
366 	struct seq_file m;
367 	int event;
368 };
369 
370 static int mounts_open(struct inode *inode, struct file *file)
371 {
372 	struct task_struct *task = get_proc_task(inode);
373 	struct mnt_namespace *ns = NULL;
374 	struct proc_mounts *p;
375 	int ret = -EINVAL;
376 
377 	if (task) {
378 		task_lock(task);
379 		if (task->nsproxy) {
380 			ns = task->nsproxy->mnt_ns;
381 			if (ns)
382 				get_mnt_ns(ns);
383 		}
384 		task_unlock(task);
385 		put_task_struct(task);
386 	}
387 
388 	if (ns) {
389 		ret = -ENOMEM;
390 		p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
391 		if (p) {
392 			file->private_data = &p->m;
393 			ret = seq_open(file, &mounts_op);
394 			if (!ret) {
395 				p->m.private = ns;
396 				p->event = ns->event;
397 				return 0;
398 			}
399 			kfree(p);
400 		}
401 		put_mnt_ns(ns);
402 	}
403 	return ret;
404 }
405 
406 static int mounts_release(struct inode *inode, struct file *file)
407 {
408 	struct seq_file *m = file->private_data;
409 	struct mnt_namespace *ns = m->private;
410 	put_mnt_ns(ns);
411 	return seq_release(inode, file);
412 }
413 
414 static unsigned mounts_poll(struct file *file, poll_table *wait)
415 {
416 	struct proc_mounts *p = file->private_data;
417 	struct mnt_namespace *ns = p->m.private;
418 	unsigned res = 0;
419 
420 	poll_wait(file, &ns->poll, wait);
421 
422 	spin_lock(&vfsmount_lock);
423 	if (p->event != ns->event) {
424 		p->event = ns->event;
425 		res = POLLERR;
426 	}
427 	spin_unlock(&vfsmount_lock);
428 
429 	return res;
430 }
431 
432 static const struct file_operations proc_mounts_operations = {
433 	.open		= mounts_open,
434 	.read		= seq_read,
435 	.llseek		= seq_lseek,
436 	.release	= mounts_release,
437 	.poll		= mounts_poll,
438 };
439 
440 extern struct seq_operations mountstats_op;
441 static int mountstats_open(struct inode *inode, struct file *file)
442 {
443 	int ret = seq_open(file, &mountstats_op);
444 
445 	if (!ret) {
446 		struct seq_file *m = file->private_data;
447 		struct mnt_namespace *mnt_ns = NULL;
448 		struct task_struct *task = get_proc_task(inode);
449 
450 		if (task) {
451 			task_lock(task);
452 			if (task->nsproxy)
453 				mnt_ns = task->nsproxy->mnt_ns;
454 			if (mnt_ns)
455 				get_mnt_ns(mnt_ns);
456 			task_unlock(task);
457 			put_task_struct(task);
458 		}
459 
460 		if (mnt_ns)
461 			m->private = mnt_ns;
462 		else {
463 			seq_release(inode, file);
464 			ret = -EINVAL;
465 		}
466 	}
467 	return ret;
468 }
469 
470 static const struct file_operations proc_mountstats_operations = {
471 	.open		= mountstats_open,
472 	.read		= seq_read,
473 	.llseek		= seq_lseek,
474 	.release	= mounts_release,
475 };
476 
477 #define PROC_BLOCK_SIZE	(3*1024)		/* 4K page size but our output routines use some slack for overruns */
478 
479 static ssize_t proc_info_read(struct file * file, char __user * buf,
480 			  size_t count, loff_t *ppos)
481 {
482 	struct inode * inode = file->f_path.dentry->d_inode;
483 	unsigned long page;
484 	ssize_t length;
485 	struct task_struct *task = get_proc_task(inode);
486 
487 	length = -ESRCH;
488 	if (!task)
489 		goto out_no_task;
490 
491 	if (count > PROC_BLOCK_SIZE)
492 		count = PROC_BLOCK_SIZE;
493 
494 	length = -ENOMEM;
495 	if (!(page = __get_free_page(GFP_TEMPORARY)))
496 		goto out;
497 
498 	length = PROC_I(inode)->op.proc_read(task, (char*)page);
499 
500 	if (length >= 0)
501 		length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
502 	free_page(page);
503 out:
504 	put_task_struct(task);
505 out_no_task:
506 	return length;
507 }
508 
509 static const struct file_operations proc_info_file_operations = {
510 	.read		= proc_info_read,
511 };
512 
513 static int mem_open(struct inode* inode, struct file* file)
514 {
515 	file->private_data = (void*)((long)current->self_exec_id);
516 	return 0;
517 }
518 
519 static ssize_t mem_read(struct file * file, char __user * buf,
520 			size_t count, loff_t *ppos)
521 {
522 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
523 	char *page;
524 	unsigned long src = *ppos;
525 	int ret = -ESRCH;
526 	struct mm_struct *mm;
527 
528 	if (!task)
529 		goto out_no_task;
530 
531 	if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
532 		goto out;
533 
534 	ret = -ENOMEM;
535 	page = (char *)__get_free_page(GFP_TEMPORARY);
536 	if (!page)
537 		goto out;
538 
539 	ret = 0;
540 
541 	mm = get_task_mm(task);
542 	if (!mm)
543 		goto out_free;
544 
545 	ret = -EIO;
546 
547 	if (file->private_data != (void*)((long)current->self_exec_id))
548 		goto out_put;
549 
550 	ret = 0;
551 
552 	while (count > 0) {
553 		int this_len, retval;
554 
555 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
556 		retval = access_process_vm(task, src, page, this_len, 0);
557 		if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
558 			if (!ret)
559 				ret = -EIO;
560 			break;
561 		}
562 
563 		if (copy_to_user(buf, page, retval)) {
564 			ret = -EFAULT;
565 			break;
566 		}
567 
568 		ret += retval;
569 		src += retval;
570 		buf += retval;
571 		count -= retval;
572 	}
573 	*ppos = src;
574 
575 out_put:
576 	mmput(mm);
577 out_free:
578 	free_page((unsigned long) page);
579 out:
580 	put_task_struct(task);
581 out_no_task:
582 	return ret;
583 }
584 
585 #define mem_write NULL
586 
587 #ifndef mem_write
588 /* This is a security hazard */
589 static ssize_t mem_write(struct file * file, const char __user *buf,
590 			 size_t count, loff_t *ppos)
591 {
592 	int copied;
593 	char *page;
594 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
595 	unsigned long dst = *ppos;
596 
597 	copied = -ESRCH;
598 	if (!task)
599 		goto out_no_task;
600 
601 	if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
602 		goto out;
603 
604 	copied = -ENOMEM;
605 	page = (char *)__get_free_page(GFP_TEMPORARY);
606 	if (!page)
607 		goto out;
608 
609 	copied = 0;
610 	while (count > 0) {
611 		int this_len, retval;
612 
613 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
614 		if (copy_from_user(page, buf, this_len)) {
615 			copied = -EFAULT;
616 			break;
617 		}
618 		retval = access_process_vm(task, dst, page, this_len, 1);
619 		if (!retval) {
620 			if (!copied)
621 				copied = -EIO;
622 			break;
623 		}
624 		copied += retval;
625 		buf += retval;
626 		dst += retval;
627 		count -= retval;
628 	}
629 	*ppos = dst;
630 	free_page((unsigned long) page);
631 out:
632 	put_task_struct(task);
633 out_no_task:
634 	return copied;
635 }
636 #endif
637 
638 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
639 {
640 	switch (orig) {
641 	case 0:
642 		file->f_pos = offset;
643 		break;
644 	case 1:
645 		file->f_pos += offset;
646 		break;
647 	default:
648 		return -EINVAL;
649 	}
650 	force_successful_syscall_return();
651 	return file->f_pos;
652 }
653 
654 static const struct file_operations proc_mem_operations = {
655 	.llseek		= mem_lseek,
656 	.read		= mem_read,
657 	.write		= mem_write,
658 	.open		= mem_open,
659 };
660 
661 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
662 				size_t count, loff_t *ppos)
663 {
664 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
665 	char buffer[PROC_NUMBUF];
666 	size_t len;
667 	int oom_adjust;
668 
669 	if (!task)
670 		return -ESRCH;
671 	oom_adjust = task->oomkilladj;
672 	put_task_struct(task);
673 
674 	len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
675 
676 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
677 }
678 
679 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
680 				size_t count, loff_t *ppos)
681 {
682 	struct task_struct *task;
683 	char buffer[PROC_NUMBUF], *end;
684 	int oom_adjust;
685 
686 	memset(buffer, 0, sizeof(buffer));
687 	if (count > sizeof(buffer) - 1)
688 		count = sizeof(buffer) - 1;
689 	if (copy_from_user(buffer, buf, count))
690 		return -EFAULT;
691 	oom_adjust = simple_strtol(buffer, &end, 0);
692 	if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
693 	     oom_adjust != OOM_DISABLE)
694 		return -EINVAL;
695 	if (*end == '\n')
696 		end++;
697 	task = get_proc_task(file->f_path.dentry->d_inode);
698 	if (!task)
699 		return -ESRCH;
700 	if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
701 		put_task_struct(task);
702 		return -EACCES;
703 	}
704 	task->oomkilladj = oom_adjust;
705 	put_task_struct(task);
706 	if (end - buffer == 0)
707 		return -EIO;
708 	return end - buffer;
709 }
710 
711 static const struct file_operations proc_oom_adjust_operations = {
712 	.read		= oom_adjust_read,
713 	.write		= oom_adjust_write,
714 };
715 
716 #ifdef CONFIG_MMU
717 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
718 				size_t count, loff_t *ppos)
719 {
720 	struct task_struct *task;
721 	char buffer[PROC_NUMBUF], *end;
722 	struct mm_struct *mm;
723 
724 	memset(buffer, 0, sizeof(buffer));
725 	if (count > sizeof(buffer) - 1)
726 		count = sizeof(buffer) - 1;
727 	if (copy_from_user(buffer, buf, count))
728 		return -EFAULT;
729 	if (!simple_strtol(buffer, &end, 0))
730 		return -EINVAL;
731 	if (*end == '\n')
732 		end++;
733 	task = get_proc_task(file->f_path.dentry->d_inode);
734 	if (!task)
735 		return -ESRCH;
736 	mm = get_task_mm(task);
737 	if (mm) {
738 		clear_refs_smap(mm);
739 		mmput(mm);
740 	}
741 	put_task_struct(task);
742 	if (end - buffer == 0)
743 		return -EIO;
744 	return end - buffer;
745 }
746 
747 static struct file_operations proc_clear_refs_operations = {
748 	.write		= clear_refs_write,
749 };
750 #endif
751 
752 #ifdef CONFIG_AUDITSYSCALL
753 #define TMPBUFLEN 21
754 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
755 				  size_t count, loff_t *ppos)
756 {
757 	struct inode * inode = file->f_path.dentry->d_inode;
758 	struct task_struct *task = get_proc_task(inode);
759 	ssize_t length;
760 	char tmpbuf[TMPBUFLEN];
761 
762 	if (!task)
763 		return -ESRCH;
764 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
765 				audit_get_loginuid(task->audit_context));
766 	put_task_struct(task);
767 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
768 }
769 
770 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
771 				   size_t count, loff_t *ppos)
772 {
773 	struct inode * inode = file->f_path.dentry->d_inode;
774 	char *page, *tmp;
775 	ssize_t length;
776 	uid_t loginuid;
777 
778 	if (!capable(CAP_AUDIT_CONTROL))
779 		return -EPERM;
780 
781 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
782 		return -EPERM;
783 
784 	if (count >= PAGE_SIZE)
785 		count = PAGE_SIZE - 1;
786 
787 	if (*ppos != 0) {
788 		/* No partial writes. */
789 		return -EINVAL;
790 	}
791 	page = (char*)__get_free_page(GFP_TEMPORARY);
792 	if (!page)
793 		return -ENOMEM;
794 	length = -EFAULT;
795 	if (copy_from_user(page, buf, count))
796 		goto out_free_page;
797 
798 	page[count] = '\0';
799 	loginuid = simple_strtoul(page, &tmp, 10);
800 	if (tmp == page) {
801 		length = -EINVAL;
802 		goto out_free_page;
803 
804 	}
805 	length = audit_set_loginuid(current, loginuid);
806 	if (likely(length == 0))
807 		length = count;
808 
809 out_free_page:
810 	free_page((unsigned long) page);
811 	return length;
812 }
813 
814 static const struct file_operations proc_loginuid_operations = {
815 	.read		= proc_loginuid_read,
816 	.write		= proc_loginuid_write,
817 };
818 #endif
819 
820 #ifdef CONFIG_FAULT_INJECTION
821 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
822 				      size_t count, loff_t *ppos)
823 {
824 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
825 	char buffer[PROC_NUMBUF];
826 	size_t len;
827 	int make_it_fail;
828 
829 	if (!task)
830 		return -ESRCH;
831 	make_it_fail = task->make_it_fail;
832 	put_task_struct(task);
833 
834 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
835 
836 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
837 }
838 
839 static ssize_t proc_fault_inject_write(struct file * file,
840 			const char __user * buf, size_t count, loff_t *ppos)
841 {
842 	struct task_struct *task;
843 	char buffer[PROC_NUMBUF], *end;
844 	int make_it_fail;
845 
846 	if (!capable(CAP_SYS_RESOURCE))
847 		return -EPERM;
848 	memset(buffer, 0, sizeof(buffer));
849 	if (count > sizeof(buffer) - 1)
850 		count = sizeof(buffer) - 1;
851 	if (copy_from_user(buffer, buf, count))
852 		return -EFAULT;
853 	make_it_fail = simple_strtol(buffer, &end, 0);
854 	if (*end == '\n')
855 		end++;
856 	task = get_proc_task(file->f_dentry->d_inode);
857 	if (!task)
858 		return -ESRCH;
859 	task->make_it_fail = make_it_fail;
860 	put_task_struct(task);
861 	if (end - buffer == 0)
862 		return -EIO;
863 	return end - buffer;
864 }
865 
866 static const struct file_operations proc_fault_inject_operations = {
867 	.read		= proc_fault_inject_read,
868 	.write		= proc_fault_inject_write,
869 };
870 #endif
871 
872 #ifdef CONFIG_SCHED_DEBUG
873 /*
874  * Print out various scheduling related per-task fields:
875  */
876 static int sched_show(struct seq_file *m, void *v)
877 {
878 	struct inode *inode = m->private;
879 	struct task_struct *p;
880 
881 	WARN_ON(!inode);
882 
883 	p = get_proc_task(inode);
884 	if (!p)
885 		return -ESRCH;
886 	proc_sched_show_task(p, m);
887 
888 	put_task_struct(p);
889 
890 	return 0;
891 }
892 
893 static ssize_t
894 sched_write(struct file *file, const char __user *buf,
895 	    size_t count, loff_t *offset)
896 {
897 	struct inode *inode = file->f_path.dentry->d_inode;
898 	struct task_struct *p;
899 
900 	WARN_ON(!inode);
901 
902 	p = get_proc_task(inode);
903 	if (!p)
904 		return -ESRCH;
905 	proc_sched_set_task(p);
906 
907 	put_task_struct(p);
908 
909 	return count;
910 }
911 
912 static int sched_open(struct inode *inode, struct file *filp)
913 {
914 	int ret;
915 
916 	ret = single_open(filp, sched_show, NULL);
917 	if (!ret) {
918 		struct seq_file *m = filp->private_data;
919 
920 		m->private = inode;
921 	}
922 	return ret;
923 }
924 
925 static const struct file_operations proc_pid_sched_operations = {
926 	.open		= sched_open,
927 	.read		= seq_read,
928 	.write		= sched_write,
929 	.llseek		= seq_lseek,
930 	.release	= single_release,
931 };
932 
933 #endif
934 
935 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
936 {
937 	struct inode *inode = dentry->d_inode;
938 	int error = -EACCES;
939 
940 	/* We don't need a base pointer in the /proc filesystem */
941 	path_release(nd);
942 
943 	/* Are we allowed to snoop on the tasks file descriptors? */
944 	if (!proc_fd_access_allowed(inode))
945 		goto out;
946 
947 	error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
948 	nd->last_type = LAST_BIND;
949 out:
950 	return ERR_PTR(error);
951 }
952 
953 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
954 			    char __user *buffer, int buflen)
955 {
956 	struct inode * inode;
957 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
958 	char *path;
959 	int len;
960 
961 	if (!tmp)
962 		return -ENOMEM;
963 
964 	inode = dentry->d_inode;
965 	path = d_path(dentry, mnt, tmp, PAGE_SIZE);
966 	len = PTR_ERR(path);
967 	if (IS_ERR(path))
968 		goto out;
969 	len = tmp + PAGE_SIZE - 1 - path;
970 
971 	if (len > buflen)
972 		len = buflen;
973 	if (copy_to_user(buffer, path, len))
974 		len = -EFAULT;
975  out:
976 	free_page((unsigned long)tmp);
977 	return len;
978 }
979 
980 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
981 {
982 	int error = -EACCES;
983 	struct inode *inode = dentry->d_inode;
984 	struct dentry *de;
985 	struct vfsmount *mnt = NULL;
986 
987 	/* Are we allowed to snoop on the tasks file descriptors? */
988 	if (!proc_fd_access_allowed(inode))
989 		goto out;
990 
991 	error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
992 	if (error)
993 		goto out;
994 
995 	error = do_proc_readlink(de, mnt, buffer, buflen);
996 	dput(de);
997 	mntput(mnt);
998 out:
999 	return error;
1000 }
1001 
1002 static const struct inode_operations proc_pid_link_inode_operations = {
1003 	.readlink	= proc_pid_readlink,
1004 	.follow_link	= proc_pid_follow_link,
1005 	.setattr	= proc_setattr,
1006 };
1007 
1008 
1009 /* building an inode */
1010 
1011 static int task_dumpable(struct task_struct *task)
1012 {
1013 	int dumpable = 0;
1014 	struct mm_struct *mm;
1015 
1016 	task_lock(task);
1017 	mm = task->mm;
1018 	if (mm)
1019 		dumpable = get_dumpable(mm);
1020 	task_unlock(task);
1021 	if(dumpable == 1)
1022 		return 1;
1023 	return 0;
1024 }
1025 
1026 
1027 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1028 {
1029 	struct inode * inode;
1030 	struct proc_inode *ei;
1031 
1032 	/* We need a new inode */
1033 
1034 	inode = new_inode(sb);
1035 	if (!inode)
1036 		goto out;
1037 
1038 	/* Common stuff */
1039 	ei = PROC_I(inode);
1040 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1041 	inode->i_op = &proc_def_inode_operations;
1042 
1043 	/*
1044 	 * grab the reference to task.
1045 	 */
1046 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1047 	if (!ei->pid)
1048 		goto out_unlock;
1049 
1050 	inode->i_uid = 0;
1051 	inode->i_gid = 0;
1052 	if (task_dumpable(task)) {
1053 		inode->i_uid = task->euid;
1054 		inode->i_gid = task->egid;
1055 	}
1056 	security_task_to_inode(task, inode);
1057 
1058 out:
1059 	return inode;
1060 
1061 out_unlock:
1062 	iput(inode);
1063 	return NULL;
1064 }
1065 
1066 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1067 {
1068 	struct inode *inode = dentry->d_inode;
1069 	struct task_struct *task;
1070 	generic_fillattr(inode, stat);
1071 
1072 	rcu_read_lock();
1073 	stat->uid = 0;
1074 	stat->gid = 0;
1075 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1076 	if (task) {
1077 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1078 		    task_dumpable(task)) {
1079 			stat->uid = task->euid;
1080 			stat->gid = task->egid;
1081 		}
1082 	}
1083 	rcu_read_unlock();
1084 	return 0;
1085 }
1086 
1087 /* dentry stuff */
1088 
1089 /*
1090  *	Exceptional case: normally we are not allowed to unhash a busy
1091  * directory. In this case, however, we can do it - no aliasing problems
1092  * due to the way we treat inodes.
1093  *
1094  * Rewrite the inode's ownerships here because the owning task may have
1095  * performed a setuid(), etc.
1096  *
1097  * Before the /proc/pid/status file was created the only way to read
1098  * the effective uid of a /process was to stat /proc/pid.  Reading
1099  * /proc/pid/status is slow enough that procps and other packages
1100  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1101  * made this apply to all per process world readable and executable
1102  * directories.
1103  */
1104 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1105 {
1106 	struct inode *inode = dentry->d_inode;
1107 	struct task_struct *task = get_proc_task(inode);
1108 	if (task) {
1109 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1110 		    task_dumpable(task)) {
1111 			inode->i_uid = task->euid;
1112 			inode->i_gid = task->egid;
1113 		} else {
1114 			inode->i_uid = 0;
1115 			inode->i_gid = 0;
1116 		}
1117 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1118 		security_task_to_inode(task, inode);
1119 		put_task_struct(task);
1120 		return 1;
1121 	}
1122 	d_drop(dentry);
1123 	return 0;
1124 }
1125 
1126 static int pid_delete_dentry(struct dentry * dentry)
1127 {
1128 	/* Is the task we represent dead?
1129 	 * If so, then don't put the dentry on the lru list,
1130 	 * kill it immediately.
1131 	 */
1132 	return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1133 }
1134 
1135 static struct dentry_operations pid_dentry_operations =
1136 {
1137 	.d_revalidate	= pid_revalidate,
1138 	.d_delete	= pid_delete_dentry,
1139 };
1140 
1141 /* Lookups */
1142 
1143 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1144 				struct task_struct *, const void *);
1145 
1146 /*
1147  * Fill a directory entry.
1148  *
1149  * If possible create the dcache entry and derive our inode number and
1150  * file type from dcache entry.
1151  *
1152  * Since all of the proc inode numbers are dynamically generated, the inode
1153  * numbers do not exist until the inode is cache.  This means creating the
1154  * the dcache entry in readdir is necessary to keep the inode numbers
1155  * reported by readdir in sync with the inode numbers reported
1156  * by stat.
1157  */
1158 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1159 	char *name, int len,
1160 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1161 {
1162 	struct dentry *child, *dir = filp->f_path.dentry;
1163 	struct inode *inode;
1164 	struct qstr qname;
1165 	ino_t ino = 0;
1166 	unsigned type = DT_UNKNOWN;
1167 
1168 	qname.name = name;
1169 	qname.len  = len;
1170 	qname.hash = full_name_hash(name, len);
1171 
1172 	child = d_lookup(dir, &qname);
1173 	if (!child) {
1174 		struct dentry *new;
1175 		new = d_alloc(dir, &qname);
1176 		if (new) {
1177 			child = instantiate(dir->d_inode, new, task, ptr);
1178 			if (child)
1179 				dput(new);
1180 			else
1181 				child = new;
1182 		}
1183 	}
1184 	if (!child || IS_ERR(child) || !child->d_inode)
1185 		goto end_instantiate;
1186 	inode = child->d_inode;
1187 	if (inode) {
1188 		ino = inode->i_ino;
1189 		type = inode->i_mode >> 12;
1190 	}
1191 	dput(child);
1192 end_instantiate:
1193 	if (!ino)
1194 		ino = find_inode_number(dir, &qname);
1195 	if (!ino)
1196 		ino = 1;
1197 	return filldir(dirent, name, len, filp->f_pos, ino, type);
1198 }
1199 
1200 static unsigned name_to_int(struct dentry *dentry)
1201 {
1202 	const char *name = dentry->d_name.name;
1203 	int len = dentry->d_name.len;
1204 	unsigned n = 0;
1205 
1206 	if (len > 1 && *name == '0')
1207 		goto out;
1208 	while (len-- > 0) {
1209 		unsigned c = *name++ - '0';
1210 		if (c > 9)
1211 			goto out;
1212 		if (n >= (~0U-9)/10)
1213 			goto out;
1214 		n *= 10;
1215 		n += c;
1216 	}
1217 	return n;
1218 out:
1219 	return ~0U;
1220 }
1221 
1222 #define PROC_FDINFO_MAX 64
1223 
1224 static int proc_fd_info(struct inode *inode, struct dentry **dentry,
1225 			struct vfsmount **mnt, char *info)
1226 {
1227 	struct task_struct *task = get_proc_task(inode);
1228 	struct files_struct *files = NULL;
1229 	struct file *file;
1230 	int fd = proc_fd(inode);
1231 
1232 	if (task) {
1233 		files = get_files_struct(task);
1234 		put_task_struct(task);
1235 	}
1236 	if (files) {
1237 		/*
1238 		 * We are not taking a ref to the file structure, so we must
1239 		 * hold ->file_lock.
1240 		 */
1241 		spin_lock(&files->file_lock);
1242 		file = fcheck_files(files, fd);
1243 		if (file) {
1244 			if (mnt)
1245 				*mnt = mntget(file->f_path.mnt);
1246 			if (dentry)
1247 				*dentry = dget(file->f_path.dentry);
1248 			if (info)
1249 				snprintf(info, PROC_FDINFO_MAX,
1250 					 "pos:\t%lli\n"
1251 					 "flags:\t0%o\n",
1252 					 (long long) file->f_pos,
1253 					 file->f_flags);
1254 			spin_unlock(&files->file_lock);
1255 			put_files_struct(files);
1256 			return 0;
1257 		}
1258 		spin_unlock(&files->file_lock);
1259 		put_files_struct(files);
1260 	}
1261 	return -ENOENT;
1262 }
1263 
1264 static int proc_fd_link(struct inode *inode, struct dentry **dentry,
1265 			struct vfsmount **mnt)
1266 {
1267 	return proc_fd_info(inode, dentry, mnt, NULL);
1268 }
1269 
1270 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1271 {
1272 	struct inode *inode = dentry->d_inode;
1273 	struct task_struct *task = get_proc_task(inode);
1274 	int fd = proc_fd(inode);
1275 	struct files_struct *files;
1276 
1277 	if (task) {
1278 		files = get_files_struct(task);
1279 		if (files) {
1280 			rcu_read_lock();
1281 			if (fcheck_files(files, fd)) {
1282 				rcu_read_unlock();
1283 				put_files_struct(files);
1284 				if (task_dumpable(task)) {
1285 					inode->i_uid = task->euid;
1286 					inode->i_gid = task->egid;
1287 				} else {
1288 					inode->i_uid = 0;
1289 					inode->i_gid = 0;
1290 				}
1291 				inode->i_mode &= ~(S_ISUID | S_ISGID);
1292 				security_task_to_inode(task, inode);
1293 				put_task_struct(task);
1294 				return 1;
1295 			}
1296 			rcu_read_unlock();
1297 			put_files_struct(files);
1298 		}
1299 		put_task_struct(task);
1300 	}
1301 	d_drop(dentry);
1302 	return 0;
1303 }
1304 
1305 static struct dentry_operations tid_fd_dentry_operations =
1306 {
1307 	.d_revalidate	= tid_fd_revalidate,
1308 	.d_delete	= pid_delete_dentry,
1309 };
1310 
1311 static struct dentry *proc_fd_instantiate(struct inode *dir,
1312 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1313 {
1314 	unsigned fd = *(const unsigned *)ptr;
1315 	struct file *file;
1316 	struct files_struct *files;
1317  	struct inode *inode;
1318  	struct proc_inode *ei;
1319 	struct dentry *error = ERR_PTR(-ENOENT);
1320 
1321 	inode = proc_pid_make_inode(dir->i_sb, task);
1322 	if (!inode)
1323 		goto out;
1324 	ei = PROC_I(inode);
1325 	ei->fd = fd;
1326 	files = get_files_struct(task);
1327 	if (!files)
1328 		goto out_iput;
1329 	inode->i_mode = S_IFLNK;
1330 
1331 	/*
1332 	 * We are not taking a ref to the file structure, so we must
1333 	 * hold ->file_lock.
1334 	 */
1335 	spin_lock(&files->file_lock);
1336 	file = fcheck_files(files, fd);
1337 	if (!file)
1338 		goto out_unlock;
1339 	if (file->f_mode & 1)
1340 		inode->i_mode |= S_IRUSR | S_IXUSR;
1341 	if (file->f_mode & 2)
1342 		inode->i_mode |= S_IWUSR | S_IXUSR;
1343 	spin_unlock(&files->file_lock);
1344 	put_files_struct(files);
1345 
1346 	inode->i_op = &proc_pid_link_inode_operations;
1347 	inode->i_size = 64;
1348 	ei->op.proc_get_link = proc_fd_link;
1349 	dentry->d_op = &tid_fd_dentry_operations;
1350 	d_add(dentry, inode);
1351 	/* Close the race of the process dying before we return the dentry */
1352 	if (tid_fd_revalidate(dentry, NULL))
1353 		error = NULL;
1354 
1355  out:
1356 	return error;
1357 out_unlock:
1358 	spin_unlock(&files->file_lock);
1359 	put_files_struct(files);
1360 out_iput:
1361 	iput(inode);
1362 	goto out;
1363 }
1364 
1365 static struct dentry *proc_lookupfd_common(struct inode *dir,
1366 					   struct dentry *dentry,
1367 					   instantiate_t instantiate)
1368 {
1369 	struct task_struct *task = get_proc_task(dir);
1370 	unsigned fd = name_to_int(dentry);
1371 	struct dentry *result = ERR_PTR(-ENOENT);
1372 
1373 	if (!task)
1374 		goto out_no_task;
1375 	if (fd == ~0U)
1376 		goto out;
1377 
1378 	result = instantiate(dir, dentry, task, &fd);
1379 out:
1380 	put_task_struct(task);
1381 out_no_task:
1382 	return result;
1383 }
1384 
1385 static int proc_readfd_common(struct file * filp, void * dirent,
1386 			      filldir_t filldir, instantiate_t instantiate)
1387 {
1388 	struct dentry *dentry = filp->f_path.dentry;
1389 	struct inode *inode = dentry->d_inode;
1390 	struct task_struct *p = get_proc_task(inode);
1391 	unsigned int fd, tid, ino;
1392 	int retval;
1393 	struct files_struct * files;
1394 	struct fdtable *fdt;
1395 
1396 	retval = -ENOENT;
1397 	if (!p)
1398 		goto out_no_task;
1399 	retval = 0;
1400 	tid = p->pid;
1401 
1402 	fd = filp->f_pos;
1403 	switch (fd) {
1404 		case 0:
1405 			if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1406 				goto out;
1407 			filp->f_pos++;
1408 		case 1:
1409 			ino = parent_ino(dentry);
1410 			if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1411 				goto out;
1412 			filp->f_pos++;
1413 		default:
1414 			files = get_files_struct(p);
1415 			if (!files)
1416 				goto out;
1417 			rcu_read_lock();
1418 			fdt = files_fdtable(files);
1419 			for (fd = filp->f_pos-2;
1420 			     fd < fdt->max_fds;
1421 			     fd++, filp->f_pos++) {
1422 				char name[PROC_NUMBUF];
1423 				int len;
1424 
1425 				if (!fcheck_files(files, fd))
1426 					continue;
1427 				rcu_read_unlock();
1428 
1429 				len = snprintf(name, sizeof(name), "%d", fd);
1430 				if (proc_fill_cache(filp, dirent, filldir,
1431 						    name, len, instantiate,
1432 						    p, &fd) < 0) {
1433 					rcu_read_lock();
1434 					break;
1435 				}
1436 				rcu_read_lock();
1437 			}
1438 			rcu_read_unlock();
1439 			put_files_struct(files);
1440 	}
1441 out:
1442 	put_task_struct(p);
1443 out_no_task:
1444 	return retval;
1445 }
1446 
1447 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1448 				    struct nameidata *nd)
1449 {
1450 	return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1451 }
1452 
1453 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1454 {
1455 	return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1456 }
1457 
1458 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1459 				      size_t len, loff_t *ppos)
1460 {
1461 	char tmp[PROC_FDINFO_MAX];
1462 	int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, NULL, tmp);
1463 	if (!err)
1464 		err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1465 	return err;
1466 }
1467 
1468 static const struct file_operations proc_fdinfo_file_operations = {
1469 	.open		= nonseekable_open,
1470 	.read		= proc_fdinfo_read,
1471 };
1472 
1473 static const struct file_operations proc_fd_operations = {
1474 	.read		= generic_read_dir,
1475 	.readdir	= proc_readfd,
1476 };
1477 
1478 /*
1479  * /proc/pid/fd needs a special permission handler so that a process can still
1480  * access /proc/self/fd after it has executed a setuid().
1481  */
1482 static int proc_fd_permission(struct inode *inode, int mask,
1483 				struct nameidata *nd)
1484 {
1485 	int rv;
1486 
1487 	rv = generic_permission(inode, mask, NULL);
1488 	if (rv == 0)
1489 		return 0;
1490 	if (task_pid(current) == proc_pid(inode))
1491 		rv = 0;
1492 	return rv;
1493 }
1494 
1495 /*
1496  * proc directories can do almost nothing..
1497  */
1498 static const struct inode_operations proc_fd_inode_operations = {
1499 	.lookup		= proc_lookupfd,
1500 	.permission	= proc_fd_permission,
1501 	.setattr	= proc_setattr,
1502 };
1503 
1504 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1505 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1506 {
1507 	unsigned fd = *(unsigned *)ptr;
1508  	struct inode *inode;
1509  	struct proc_inode *ei;
1510 	struct dentry *error = ERR_PTR(-ENOENT);
1511 
1512 	inode = proc_pid_make_inode(dir->i_sb, task);
1513 	if (!inode)
1514 		goto out;
1515 	ei = PROC_I(inode);
1516 	ei->fd = fd;
1517 	inode->i_mode = S_IFREG | S_IRUSR;
1518 	inode->i_fop = &proc_fdinfo_file_operations;
1519 	dentry->d_op = &tid_fd_dentry_operations;
1520 	d_add(dentry, inode);
1521 	/* Close the race of the process dying before we return the dentry */
1522 	if (tid_fd_revalidate(dentry, NULL))
1523 		error = NULL;
1524 
1525  out:
1526 	return error;
1527 }
1528 
1529 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1530 					struct dentry *dentry,
1531 					struct nameidata *nd)
1532 {
1533 	return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1534 }
1535 
1536 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1537 {
1538 	return proc_readfd_common(filp, dirent, filldir,
1539 				  proc_fdinfo_instantiate);
1540 }
1541 
1542 static const struct file_operations proc_fdinfo_operations = {
1543 	.read		= generic_read_dir,
1544 	.readdir	= proc_readfdinfo,
1545 };
1546 
1547 /*
1548  * proc directories can do almost nothing..
1549  */
1550 static const struct inode_operations proc_fdinfo_inode_operations = {
1551 	.lookup		= proc_lookupfdinfo,
1552 	.setattr	= proc_setattr,
1553 };
1554 
1555 
1556 static struct dentry *proc_pident_instantiate(struct inode *dir,
1557 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1558 {
1559 	const struct pid_entry *p = ptr;
1560 	struct inode *inode;
1561 	struct proc_inode *ei;
1562 	struct dentry *error = ERR_PTR(-EINVAL);
1563 
1564 	inode = proc_pid_make_inode(dir->i_sb, task);
1565 	if (!inode)
1566 		goto out;
1567 
1568 	ei = PROC_I(inode);
1569 	inode->i_mode = p->mode;
1570 	if (S_ISDIR(inode->i_mode))
1571 		inode->i_nlink = 2;	/* Use getattr to fix if necessary */
1572 	if (p->iop)
1573 		inode->i_op = p->iop;
1574 	if (p->fop)
1575 		inode->i_fop = p->fop;
1576 	ei->op = p->op;
1577 	dentry->d_op = &pid_dentry_operations;
1578 	d_add(dentry, inode);
1579 	/* Close the race of the process dying before we return the dentry */
1580 	if (pid_revalidate(dentry, NULL))
1581 		error = NULL;
1582 out:
1583 	return error;
1584 }
1585 
1586 static struct dentry *proc_pident_lookup(struct inode *dir,
1587 					 struct dentry *dentry,
1588 					 const struct pid_entry *ents,
1589 					 unsigned int nents)
1590 {
1591 	struct inode *inode;
1592 	struct dentry *error;
1593 	struct task_struct *task = get_proc_task(dir);
1594 	const struct pid_entry *p, *last;
1595 
1596 	error = ERR_PTR(-ENOENT);
1597 	inode = NULL;
1598 
1599 	if (!task)
1600 		goto out_no_task;
1601 
1602 	/*
1603 	 * Yes, it does not scale. And it should not. Don't add
1604 	 * new entries into /proc/<tgid>/ without very good reasons.
1605 	 */
1606 	last = &ents[nents - 1];
1607 	for (p = ents; p <= last; p++) {
1608 		if (p->len != dentry->d_name.len)
1609 			continue;
1610 		if (!memcmp(dentry->d_name.name, p->name, p->len))
1611 			break;
1612 	}
1613 	if (p > last)
1614 		goto out;
1615 
1616 	error = proc_pident_instantiate(dir, dentry, task, p);
1617 out:
1618 	put_task_struct(task);
1619 out_no_task:
1620 	return error;
1621 }
1622 
1623 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1624 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1625 {
1626 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1627 				proc_pident_instantiate, task, p);
1628 }
1629 
1630 static int proc_pident_readdir(struct file *filp,
1631 		void *dirent, filldir_t filldir,
1632 		const struct pid_entry *ents, unsigned int nents)
1633 {
1634 	int i;
1635 	int pid;
1636 	struct dentry *dentry = filp->f_path.dentry;
1637 	struct inode *inode = dentry->d_inode;
1638 	struct task_struct *task = get_proc_task(inode);
1639 	const struct pid_entry *p, *last;
1640 	ino_t ino;
1641 	int ret;
1642 
1643 	ret = -ENOENT;
1644 	if (!task)
1645 		goto out_no_task;
1646 
1647 	ret = 0;
1648 	pid = task->pid;
1649 	i = filp->f_pos;
1650 	switch (i) {
1651 	case 0:
1652 		ino = inode->i_ino;
1653 		if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1654 			goto out;
1655 		i++;
1656 		filp->f_pos++;
1657 		/* fall through */
1658 	case 1:
1659 		ino = parent_ino(dentry);
1660 		if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1661 			goto out;
1662 		i++;
1663 		filp->f_pos++;
1664 		/* fall through */
1665 	default:
1666 		i -= 2;
1667 		if (i >= nents) {
1668 			ret = 1;
1669 			goto out;
1670 		}
1671 		p = ents + i;
1672 		last = &ents[nents - 1];
1673 		while (p <= last) {
1674 			if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1675 				goto out;
1676 			filp->f_pos++;
1677 			p++;
1678 		}
1679 	}
1680 
1681 	ret = 1;
1682 out:
1683 	put_task_struct(task);
1684 out_no_task:
1685 	return ret;
1686 }
1687 
1688 #ifdef CONFIG_SECURITY
1689 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1690 				  size_t count, loff_t *ppos)
1691 {
1692 	struct inode * inode = file->f_path.dentry->d_inode;
1693 	char *p = NULL;
1694 	ssize_t length;
1695 	struct task_struct *task = get_proc_task(inode);
1696 
1697 	if (!task)
1698 		return -ESRCH;
1699 
1700 	length = security_getprocattr(task,
1701 				      (char*)file->f_path.dentry->d_name.name,
1702 				      &p);
1703 	put_task_struct(task);
1704 	if (length > 0)
1705 		length = simple_read_from_buffer(buf, count, ppos, p, length);
1706 	kfree(p);
1707 	return length;
1708 }
1709 
1710 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1711 				   size_t count, loff_t *ppos)
1712 {
1713 	struct inode * inode = file->f_path.dentry->d_inode;
1714 	char *page;
1715 	ssize_t length;
1716 	struct task_struct *task = get_proc_task(inode);
1717 
1718 	length = -ESRCH;
1719 	if (!task)
1720 		goto out_no_task;
1721 	if (count > PAGE_SIZE)
1722 		count = PAGE_SIZE;
1723 
1724 	/* No partial writes. */
1725 	length = -EINVAL;
1726 	if (*ppos != 0)
1727 		goto out;
1728 
1729 	length = -ENOMEM;
1730 	page = (char*)__get_free_page(GFP_TEMPORARY);
1731 	if (!page)
1732 		goto out;
1733 
1734 	length = -EFAULT;
1735 	if (copy_from_user(page, buf, count))
1736 		goto out_free;
1737 
1738 	length = security_setprocattr(task,
1739 				      (char*)file->f_path.dentry->d_name.name,
1740 				      (void*)page, count);
1741 out_free:
1742 	free_page((unsigned long) page);
1743 out:
1744 	put_task_struct(task);
1745 out_no_task:
1746 	return length;
1747 }
1748 
1749 static const struct file_operations proc_pid_attr_operations = {
1750 	.read		= proc_pid_attr_read,
1751 	.write		= proc_pid_attr_write,
1752 };
1753 
1754 static const struct pid_entry attr_dir_stuff[] = {
1755 	REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1756 	REG("prev",       S_IRUGO,	   pid_attr),
1757 	REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1758 	REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1759 	REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1760 	REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1761 };
1762 
1763 static int proc_attr_dir_readdir(struct file * filp,
1764 			     void * dirent, filldir_t filldir)
1765 {
1766 	return proc_pident_readdir(filp,dirent,filldir,
1767 				   attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1768 }
1769 
1770 static const struct file_operations proc_attr_dir_operations = {
1771 	.read		= generic_read_dir,
1772 	.readdir	= proc_attr_dir_readdir,
1773 };
1774 
1775 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
1776 				struct dentry *dentry, struct nameidata *nd)
1777 {
1778 	return proc_pident_lookup(dir, dentry,
1779 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
1780 }
1781 
1782 static const struct inode_operations proc_attr_dir_inode_operations = {
1783 	.lookup		= proc_attr_dir_lookup,
1784 	.getattr	= pid_getattr,
1785 	.setattr	= proc_setattr,
1786 };
1787 
1788 #endif
1789 
1790 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1791 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
1792 					 size_t count, loff_t *ppos)
1793 {
1794 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1795 	struct mm_struct *mm;
1796 	char buffer[PROC_NUMBUF];
1797 	size_t len;
1798 	int ret;
1799 
1800 	if (!task)
1801 		return -ESRCH;
1802 
1803 	ret = 0;
1804 	mm = get_task_mm(task);
1805 	if (mm) {
1806 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
1807 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
1808 				MMF_DUMP_FILTER_SHIFT));
1809 		mmput(mm);
1810 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
1811 	}
1812 
1813 	put_task_struct(task);
1814 
1815 	return ret;
1816 }
1817 
1818 static ssize_t proc_coredump_filter_write(struct file *file,
1819 					  const char __user *buf,
1820 					  size_t count,
1821 					  loff_t *ppos)
1822 {
1823 	struct task_struct *task;
1824 	struct mm_struct *mm;
1825 	char buffer[PROC_NUMBUF], *end;
1826 	unsigned int val;
1827 	int ret;
1828 	int i;
1829 	unsigned long mask;
1830 
1831 	ret = -EFAULT;
1832 	memset(buffer, 0, sizeof(buffer));
1833 	if (count > sizeof(buffer) - 1)
1834 		count = sizeof(buffer) - 1;
1835 	if (copy_from_user(buffer, buf, count))
1836 		goto out_no_task;
1837 
1838 	ret = -EINVAL;
1839 	val = (unsigned int)simple_strtoul(buffer, &end, 0);
1840 	if (*end == '\n')
1841 		end++;
1842 	if (end - buffer == 0)
1843 		goto out_no_task;
1844 
1845 	ret = -ESRCH;
1846 	task = get_proc_task(file->f_dentry->d_inode);
1847 	if (!task)
1848 		goto out_no_task;
1849 
1850 	ret = end - buffer;
1851 	mm = get_task_mm(task);
1852 	if (!mm)
1853 		goto out_no_mm;
1854 
1855 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
1856 		if (val & mask)
1857 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
1858 		else
1859 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
1860 	}
1861 
1862 	mmput(mm);
1863  out_no_mm:
1864 	put_task_struct(task);
1865  out_no_task:
1866 	return ret;
1867 }
1868 
1869 static const struct file_operations proc_coredump_filter_operations = {
1870 	.read		= proc_coredump_filter_read,
1871 	.write		= proc_coredump_filter_write,
1872 };
1873 #endif
1874 
1875 /*
1876  * /proc/self:
1877  */
1878 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1879 			      int buflen)
1880 {
1881 	char tmp[PROC_NUMBUF];
1882 	sprintf(tmp, "%d", current->tgid);
1883 	return vfs_readlink(dentry,buffer,buflen,tmp);
1884 }
1885 
1886 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1887 {
1888 	char tmp[PROC_NUMBUF];
1889 	sprintf(tmp, "%d", current->tgid);
1890 	return ERR_PTR(vfs_follow_link(nd,tmp));
1891 }
1892 
1893 static const struct inode_operations proc_self_inode_operations = {
1894 	.readlink	= proc_self_readlink,
1895 	.follow_link	= proc_self_follow_link,
1896 };
1897 
1898 /*
1899  * proc base
1900  *
1901  * These are the directory entries in the root directory of /proc
1902  * that properly belong to the /proc filesystem, as they describe
1903  * describe something that is process related.
1904  */
1905 static const struct pid_entry proc_base_stuff[] = {
1906 	NOD("self", S_IFLNK|S_IRWXUGO,
1907 		&proc_self_inode_operations, NULL, {}),
1908 };
1909 
1910 /*
1911  *	Exceptional case: normally we are not allowed to unhash a busy
1912  * directory. In this case, however, we can do it - no aliasing problems
1913  * due to the way we treat inodes.
1914  */
1915 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
1916 {
1917 	struct inode *inode = dentry->d_inode;
1918 	struct task_struct *task = get_proc_task(inode);
1919 	if (task) {
1920 		put_task_struct(task);
1921 		return 1;
1922 	}
1923 	d_drop(dentry);
1924 	return 0;
1925 }
1926 
1927 static struct dentry_operations proc_base_dentry_operations =
1928 {
1929 	.d_revalidate	= proc_base_revalidate,
1930 	.d_delete	= pid_delete_dentry,
1931 };
1932 
1933 static struct dentry *proc_base_instantiate(struct inode *dir,
1934 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1935 {
1936 	const struct pid_entry *p = ptr;
1937 	struct inode *inode;
1938 	struct proc_inode *ei;
1939 	struct dentry *error = ERR_PTR(-EINVAL);
1940 
1941 	/* Allocate the inode */
1942 	error = ERR_PTR(-ENOMEM);
1943 	inode = new_inode(dir->i_sb);
1944 	if (!inode)
1945 		goto out;
1946 
1947 	/* Initialize the inode */
1948 	ei = PROC_I(inode);
1949 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1950 
1951 	/*
1952 	 * grab the reference to the task.
1953 	 */
1954 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1955 	if (!ei->pid)
1956 		goto out_iput;
1957 
1958 	inode->i_uid = 0;
1959 	inode->i_gid = 0;
1960 	inode->i_mode = p->mode;
1961 	if (S_ISDIR(inode->i_mode))
1962 		inode->i_nlink = 2;
1963 	if (S_ISLNK(inode->i_mode))
1964 		inode->i_size = 64;
1965 	if (p->iop)
1966 		inode->i_op = p->iop;
1967 	if (p->fop)
1968 		inode->i_fop = p->fop;
1969 	ei->op = p->op;
1970 	dentry->d_op = &proc_base_dentry_operations;
1971 	d_add(dentry, inode);
1972 	error = NULL;
1973 out:
1974 	return error;
1975 out_iput:
1976 	iput(inode);
1977 	goto out;
1978 }
1979 
1980 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
1981 {
1982 	struct dentry *error;
1983 	struct task_struct *task = get_proc_task(dir);
1984 	const struct pid_entry *p, *last;
1985 
1986 	error = ERR_PTR(-ENOENT);
1987 
1988 	if (!task)
1989 		goto out_no_task;
1990 
1991 	/* Lookup the directory entry */
1992 	last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
1993 	for (p = proc_base_stuff; p <= last; p++) {
1994 		if (p->len != dentry->d_name.len)
1995 			continue;
1996 		if (!memcmp(dentry->d_name.name, p->name, p->len))
1997 			break;
1998 	}
1999 	if (p > last)
2000 		goto out;
2001 
2002 	error = proc_base_instantiate(dir, dentry, task, p);
2003 
2004 out:
2005 	put_task_struct(task);
2006 out_no_task:
2007 	return error;
2008 }
2009 
2010 static int proc_base_fill_cache(struct file *filp, void *dirent,
2011 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2012 {
2013 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2014 				proc_base_instantiate, task, p);
2015 }
2016 
2017 #ifdef CONFIG_TASK_IO_ACCOUNTING
2018 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2019 {
2020 	return sprintf(buffer,
2021 #ifdef CONFIG_TASK_XACCT
2022 			"rchar: %llu\n"
2023 			"wchar: %llu\n"
2024 			"syscr: %llu\n"
2025 			"syscw: %llu\n"
2026 #endif
2027 			"read_bytes: %llu\n"
2028 			"write_bytes: %llu\n"
2029 			"cancelled_write_bytes: %llu\n",
2030 #ifdef CONFIG_TASK_XACCT
2031 			(unsigned long long)task->rchar,
2032 			(unsigned long long)task->wchar,
2033 			(unsigned long long)task->syscr,
2034 			(unsigned long long)task->syscw,
2035 #endif
2036 			(unsigned long long)task->ioac.read_bytes,
2037 			(unsigned long long)task->ioac.write_bytes,
2038 			(unsigned long long)task->ioac.cancelled_write_bytes);
2039 }
2040 #endif
2041 
2042 /*
2043  * Thread groups
2044  */
2045 static const struct file_operations proc_task_operations;
2046 static const struct inode_operations proc_task_inode_operations;
2047 
2048 static const struct pid_entry tgid_base_stuff[] = {
2049 	DIR("task",       S_IRUGO|S_IXUGO, task),
2050 	DIR("fd",         S_IRUSR|S_IXUSR, fd),
2051 	DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2052 	INF("environ",    S_IRUSR, pid_environ),
2053 	INF("auxv",       S_IRUSR, pid_auxv),
2054 	INF("status",     S_IRUGO, pid_status),
2055 #ifdef CONFIG_SCHED_DEBUG
2056 	REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2057 #endif
2058 	INF("cmdline",    S_IRUGO, pid_cmdline),
2059 	INF("stat",       S_IRUGO, tgid_stat),
2060 	INF("statm",      S_IRUGO, pid_statm),
2061 	REG("maps",       S_IRUGO, maps),
2062 #ifdef CONFIG_NUMA
2063 	REG("numa_maps",  S_IRUGO, numa_maps),
2064 #endif
2065 	REG("mem",        S_IRUSR|S_IWUSR, mem),
2066 	LNK("cwd",        cwd),
2067 	LNK("root",       root),
2068 	LNK("exe",        exe),
2069 	REG("mounts",     S_IRUGO, mounts),
2070 	REG("mountstats", S_IRUSR, mountstats),
2071 #ifdef CONFIG_MMU
2072 	REG("clear_refs", S_IWUSR, clear_refs),
2073 	REG("smaps",      S_IRUGO, smaps),
2074 #endif
2075 #ifdef CONFIG_SECURITY
2076 	DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2077 #endif
2078 #ifdef CONFIG_KALLSYMS
2079 	INF("wchan",      S_IRUGO, pid_wchan),
2080 #endif
2081 #ifdef CONFIG_SCHEDSTATS
2082 	INF("schedstat",  S_IRUGO, pid_schedstat),
2083 #endif
2084 #ifdef CONFIG_CPUSETS
2085 	REG("cpuset",     S_IRUGO, cpuset),
2086 #endif
2087 	INF("oom_score",  S_IRUGO, oom_score),
2088 	REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2089 #ifdef CONFIG_AUDITSYSCALL
2090 	REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2091 #endif
2092 #ifdef CONFIG_FAULT_INJECTION
2093 	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2094 #endif
2095 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2096 	REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2097 #endif
2098 #ifdef CONFIG_TASK_IO_ACCOUNTING
2099 	INF("io",	S_IRUGO, pid_io_accounting),
2100 #endif
2101 };
2102 
2103 static int proc_tgid_base_readdir(struct file * filp,
2104 			     void * dirent, filldir_t filldir)
2105 {
2106 	return proc_pident_readdir(filp,dirent,filldir,
2107 				   tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2108 }
2109 
2110 static const struct file_operations proc_tgid_base_operations = {
2111 	.read		= generic_read_dir,
2112 	.readdir	= proc_tgid_base_readdir,
2113 };
2114 
2115 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2116 	return proc_pident_lookup(dir, dentry,
2117 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2118 }
2119 
2120 static const struct inode_operations proc_tgid_base_inode_operations = {
2121 	.lookup		= proc_tgid_base_lookup,
2122 	.getattr	= pid_getattr,
2123 	.setattr	= proc_setattr,
2124 };
2125 
2126 /**
2127  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2128  *
2129  * @task: task that should be flushed.
2130  *
2131  * Looks in the dcache for
2132  * /proc/@pid
2133  * /proc/@tgid/task/@pid
2134  * if either directory is present flushes it and all of it'ts children
2135  * from the dcache.
2136  *
2137  * It is safe and reasonable to cache /proc entries for a task until
2138  * that task exits.  After that they just clog up the dcache with
2139  * useless entries, possibly causing useful dcache entries to be
2140  * flushed instead.  This routine is proved to flush those useless
2141  * dcache entries at process exit time.
2142  *
2143  * NOTE: This routine is just an optimization so it does not guarantee
2144  *       that no dcache entries will exist at process exit time it
2145  *       just makes it very unlikely that any will persist.
2146  */
2147 void proc_flush_task(struct task_struct *task)
2148 {
2149 	struct dentry *dentry, *leader, *dir;
2150 	char buf[PROC_NUMBUF];
2151 	struct qstr name;
2152 
2153 	name.name = buf;
2154 	name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
2155 	dentry = d_hash_and_lookup(proc_mnt->mnt_root, &name);
2156 	if (dentry) {
2157 		shrink_dcache_parent(dentry);
2158 		d_drop(dentry);
2159 		dput(dentry);
2160 	}
2161 
2162 	if (thread_group_leader(task))
2163 		goto out;
2164 
2165 	name.name = buf;
2166 	name.len = snprintf(buf, sizeof(buf), "%d", task->tgid);
2167 	leader = d_hash_and_lookup(proc_mnt->mnt_root, &name);
2168 	if (!leader)
2169 		goto out;
2170 
2171 	name.name = "task";
2172 	name.len = strlen(name.name);
2173 	dir = d_hash_and_lookup(leader, &name);
2174 	if (!dir)
2175 		goto out_put_leader;
2176 
2177 	name.name = buf;
2178 	name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
2179 	dentry = d_hash_and_lookup(dir, &name);
2180 	if (dentry) {
2181 		shrink_dcache_parent(dentry);
2182 		d_drop(dentry);
2183 		dput(dentry);
2184 	}
2185 
2186 	dput(dir);
2187 out_put_leader:
2188 	dput(leader);
2189 out:
2190 	return;
2191 }
2192 
2193 static struct dentry *proc_pid_instantiate(struct inode *dir,
2194 					   struct dentry * dentry,
2195 					   struct task_struct *task, const void *ptr)
2196 {
2197 	struct dentry *error = ERR_PTR(-ENOENT);
2198 	struct inode *inode;
2199 
2200 	inode = proc_pid_make_inode(dir->i_sb, task);
2201 	if (!inode)
2202 		goto out;
2203 
2204 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2205 	inode->i_op = &proc_tgid_base_inode_operations;
2206 	inode->i_fop = &proc_tgid_base_operations;
2207 	inode->i_flags|=S_IMMUTABLE;
2208 	inode->i_nlink = 5;
2209 #ifdef CONFIG_SECURITY
2210 	inode->i_nlink += 1;
2211 #endif
2212 
2213 	dentry->d_op = &pid_dentry_operations;
2214 
2215 	d_add(dentry, inode);
2216 	/* Close the race of the process dying before we return the dentry */
2217 	if (pid_revalidate(dentry, NULL))
2218 		error = NULL;
2219 out:
2220 	return error;
2221 }
2222 
2223 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2224 {
2225 	struct dentry *result = ERR_PTR(-ENOENT);
2226 	struct task_struct *task;
2227 	unsigned tgid;
2228 
2229 	result = proc_base_lookup(dir, dentry);
2230 	if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2231 		goto out;
2232 
2233 	tgid = name_to_int(dentry);
2234 	if (tgid == ~0U)
2235 		goto out;
2236 
2237 	rcu_read_lock();
2238 	task = find_task_by_pid(tgid);
2239 	if (task)
2240 		get_task_struct(task);
2241 	rcu_read_unlock();
2242 	if (!task)
2243 		goto out;
2244 
2245 	result = proc_pid_instantiate(dir, dentry, task, NULL);
2246 	put_task_struct(task);
2247 out:
2248 	return result;
2249 }
2250 
2251 /*
2252  * Find the first task with tgid >= tgid
2253  *
2254  */
2255 static struct task_struct *next_tgid(unsigned int tgid)
2256 {
2257 	struct task_struct *task;
2258 	struct pid *pid;
2259 
2260 	rcu_read_lock();
2261 retry:
2262 	task = NULL;
2263 	pid = find_ge_pid(tgid);
2264 	if (pid) {
2265 		tgid = pid->nr + 1;
2266 		task = pid_task(pid, PIDTYPE_PID);
2267 		/* What we to know is if the pid we have find is the
2268 		 * pid of a thread_group_leader.  Testing for task
2269 		 * being a thread_group_leader is the obvious thing
2270 		 * todo but there is a window when it fails, due to
2271 		 * the pid transfer logic in de_thread.
2272 		 *
2273 		 * So we perform the straight forward test of seeing
2274 		 * if the pid we have found is the pid of a thread
2275 		 * group leader, and don't worry if the task we have
2276 		 * found doesn't happen to be a thread group leader.
2277 		 * As we don't care in the case of readdir.
2278 		 */
2279 		if (!task || !has_group_leader_pid(task))
2280 			goto retry;
2281 		get_task_struct(task);
2282 	}
2283 	rcu_read_unlock();
2284 	return task;
2285 }
2286 
2287 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2288 
2289 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2290 	struct task_struct *task, int tgid)
2291 {
2292 	char name[PROC_NUMBUF];
2293 	int len = snprintf(name, sizeof(name), "%d", tgid);
2294 	return proc_fill_cache(filp, dirent, filldir, name, len,
2295 				proc_pid_instantiate, task, NULL);
2296 }
2297 
2298 /* for the /proc/ directory itself, after non-process stuff has been done */
2299 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2300 {
2301 	unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2302 	struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2303 	struct task_struct *task;
2304 	int tgid;
2305 
2306 	if (!reaper)
2307 		goto out_no_task;
2308 
2309 	for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2310 		const struct pid_entry *p = &proc_base_stuff[nr];
2311 		if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2312 			goto out;
2313 	}
2314 
2315 	tgid = filp->f_pos - TGID_OFFSET;
2316 	for (task = next_tgid(tgid);
2317 	     task;
2318 	     put_task_struct(task), task = next_tgid(tgid + 1)) {
2319 		tgid = task->pid;
2320 		filp->f_pos = tgid + TGID_OFFSET;
2321 		if (proc_pid_fill_cache(filp, dirent, filldir, task, tgid) < 0) {
2322 			put_task_struct(task);
2323 			goto out;
2324 		}
2325 	}
2326 	filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2327 out:
2328 	put_task_struct(reaper);
2329 out_no_task:
2330 	return 0;
2331 }
2332 
2333 /*
2334  * Tasks
2335  */
2336 static const struct pid_entry tid_base_stuff[] = {
2337 	DIR("fd",        S_IRUSR|S_IXUSR, fd),
2338 	DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2339 	INF("environ",   S_IRUSR, pid_environ),
2340 	INF("auxv",      S_IRUSR, pid_auxv),
2341 	INF("status",    S_IRUGO, pid_status),
2342 #ifdef CONFIG_SCHED_DEBUG
2343 	REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2344 #endif
2345 	INF("cmdline",   S_IRUGO, pid_cmdline),
2346 	INF("stat",      S_IRUGO, tid_stat),
2347 	INF("statm",     S_IRUGO, pid_statm),
2348 	REG("maps",      S_IRUGO, maps),
2349 #ifdef CONFIG_NUMA
2350 	REG("numa_maps", S_IRUGO, numa_maps),
2351 #endif
2352 	REG("mem",       S_IRUSR|S_IWUSR, mem),
2353 	LNK("cwd",       cwd),
2354 	LNK("root",      root),
2355 	LNK("exe",       exe),
2356 	REG("mounts",    S_IRUGO, mounts),
2357 #ifdef CONFIG_MMU
2358 	REG("clear_refs", S_IWUSR, clear_refs),
2359 	REG("smaps",     S_IRUGO, smaps),
2360 #endif
2361 #ifdef CONFIG_SECURITY
2362 	DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2363 #endif
2364 #ifdef CONFIG_KALLSYMS
2365 	INF("wchan",     S_IRUGO, pid_wchan),
2366 #endif
2367 #ifdef CONFIG_SCHEDSTATS
2368 	INF("schedstat", S_IRUGO, pid_schedstat),
2369 #endif
2370 #ifdef CONFIG_CPUSETS
2371 	REG("cpuset",    S_IRUGO, cpuset),
2372 #endif
2373 	INF("oom_score", S_IRUGO, oom_score),
2374 	REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2375 #ifdef CONFIG_AUDITSYSCALL
2376 	REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2377 #endif
2378 #ifdef CONFIG_FAULT_INJECTION
2379 	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2380 #endif
2381 };
2382 
2383 static int proc_tid_base_readdir(struct file * filp,
2384 			     void * dirent, filldir_t filldir)
2385 {
2386 	return proc_pident_readdir(filp,dirent,filldir,
2387 				   tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2388 }
2389 
2390 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2391 	return proc_pident_lookup(dir, dentry,
2392 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2393 }
2394 
2395 static const struct file_operations proc_tid_base_operations = {
2396 	.read		= generic_read_dir,
2397 	.readdir	= proc_tid_base_readdir,
2398 };
2399 
2400 static const struct inode_operations proc_tid_base_inode_operations = {
2401 	.lookup		= proc_tid_base_lookup,
2402 	.getattr	= pid_getattr,
2403 	.setattr	= proc_setattr,
2404 };
2405 
2406 static struct dentry *proc_task_instantiate(struct inode *dir,
2407 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2408 {
2409 	struct dentry *error = ERR_PTR(-ENOENT);
2410 	struct inode *inode;
2411 	inode = proc_pid_make_inode(dir->i_sb, task);
2412 
2413 	if (!inode)
2414 		goto out;
2415 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2416 	inode->i_op = &proc_tid_base_inode_operations;
2417 	inode->i_fop = &proc_tid_base_operations;
2418 	inode->i_flags|=S_IMMUTABLE;
2419 	inode->i_nlink = 4;
2420 #ifdef CONFIG_SECURITY
2421 	inode->i_nlink += 1;
2422 #endif
2423 
2424 	dentry->d_op = &pid_dentry_operations;
2425 
2426 	d_add(dentry, inode);
2427 	/* Close the race of the process dying before we return the dentry */
2428 	if (pid_revalidate(dentry, NULL))
2429 		error = NULL;
2430 out:
2431 	return error;
2432 }
2433 
2434 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2435 {
2436 	struct dentry *result = ERR_PTR(-ENOENT);
2437 	struct task_struct *task;
2438 	struct task_struct *leader = get_proc_task(dir);
2439 	unsigned tid;
2440 
2441 	if (!leader)
2442 		goto out_no_task;
2443 
2444 	tid = name_to_int(dentry);
2445 	if (tid == ~0U)
2446 		goto out;
2447 
2448 	rcu_read_lock();
2449 	task = find_task_by_pid(tid);
2450 	if (task)
2451 		get_task_struct(task);
2452 	rcu_read_unlock();
2453 	if (!task)
2454 		goto out;
2455 	if (leader->tgid != task->tgid)
2456 		goto out_drop_task;
2457 
2458 	result = proc_task_instantiate(dir, dentry, task, NULL);
2459 out_drop_task:
2460 	put_task_struct(task);
2461 out:
2462 	put_task_struct(leader);
2463 out_no_task:
2464 	return result;
2465 }
2466 
2467 /*
2468  * Find the first tid of a thread group to return to user space.
2469  *
2470  * Usually this is just the thread group leader, but if the users
2471  * buffer was too small or there was a seek into the middle of the
2472  * directory we have more work todo.
2473  *
2474  * In the case of a short read we start with find_task_by_pid.
2475  *
2476  * In the case of a seek we start with the leader and walk nr
2477  * threads past it.
2478  */
2479 static struct task_struct *first_tid(struct task_struct *leader,
2480 					int tid, int nr)
2481 {
2482 	struct task_struct *pos;
2483 
2484 	rcu_read_lock();
2485 	/* Attempt to start with the pid of a thread */
2486 	if (tid && (nr > 0)) {
2487 		pos = find_task_by_pid(tid);
2488 		if (pos && (pos->group_leader == leader))
2489 			goto found;
2490 	}
2491 
2492 	/* If nr exceeds the number of threads there is nothing todo */
2493 	pos = NULL;
2494 	if (nr && nr >= get_nr_threads(leader))
2495 		goto out;
2496 
2497 	/* If we haven't found our starting place yet start
2498 	 * with the leader and walk nr threads forward.
2499 	 */
2500 	for (pos = leader; nr > 0; --nr) {
2501 		pos = next_thread(pos);
2502 		if (pos == leader) {
2503 			pos = NULL;
2504 			goto out;
2505 		}
2506 	}
2507 found:
2508 	get_task_struct(pos);
2509 out:
2510 	rcu_read_unlock();
2511 	return pos;
2512 }
2513 
2514 /*
2515  * Find the next thread in the thread list.
2516  * Return NULL if there is an error or no next thread.
2517  *
2518  * The reference to the input task_struct is released.
2519  */
2520 static struct task_struct *next_tid(struct task_struct *start)
2521 {
2522 	struct task_struct *pos = NULL;
2523 	rcu_read_lock();
2524 	if (pid_alive(start)) {
2525 		pos = next_thread(start);
2526 		if (thread_group_leader(pos))
2527 			pos = NULL;
2528 		else
2529 			get_task_struct(pos);
2530 	}
2531 	rcu_read_unlock();
2532 	put_task_struct(start);
2533 	return pos;
2534 }
2535 
2536 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2537 	struct task_struct *task, int tid)
2538 {
2539 	char name[PROC_NUMBUF];
2540 	int len = snprintf(name, sizeof(name), "%d", tid);
2541 	return proc_fill_cache(filp, dirent, filldir, name, len,
2542 				proc_task_instantiate, task, NULL);
2543 }
2544 
2545 /* for the /proc/TGID/task/ directories */
2546 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2547 {
2548 	struct dentry *dentry = filp->f_path.dentry;
2549 	struct inode *inode = dentry->d_inode;
2550 	struct task_struct *leader = NULL;
2551 	struct task_struct *task;
2552 	int retval = -ENOENT;
2553 	ino_t ino;
2554 	int tid;
2555 	unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2556 
2557 	task = get_proc_task(inode);
2558 	if (!task)
2559 		goto out_no_task;
2560 	rcu_read_lock();
2561 	if (pid_alive(task)) {
2562 		leader = task->group_leader;
2563 		get_task_struct(leader);
2564 	}
2565 	rcu_read_unlock();
2566 	put_task_struct(task);
2567 	if (!leader)
2568 		goto out_no_task;
2569 	retval = 0;
2570 
2571 	switch (pos) {
2572 	case 0:
2573 		ino = inode->i_ino;
2574 		if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2575 			goto out;
2576 		pos++;
2577 		/* fall through */
2578 	case 1:
2579 		ino = parent_ino(dentry);
2580 		if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2581 			goto out;
2582 		pos++;
2583 		/* fall through */
2584 	}
2585 
2586 	/* f_version caches the tgid value that the last readdir call couldn't
2587 	 * return. lseek aka telldir automagically resets f_version to 0.
2588 	 */
2589 	tid = filp->f_version;
2590 	filp->f_version = 0;
2591 	for (task = first_tid(leader, tid, pos - 2);
2592 	     task;
2593 	     task = next_tid(task), pos++) {
2594 		tid = task->pid;
2595 		if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2596 			/* returning this tgid failed, save it as the first
2597 			 * pid for the next readir call */
2598 			filp->f_version = tid;
2599 			put_task_struct(task);
2600 			break;
2601 		}
2602 	}
2603 out:
2604 	filp->f_pos = pos;
2605 	put_task_struct(leader);
2606 out_no_task:
2607 	return retval;
2608 }
2609 
2610 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2611 {
2612 	struct inode *inode = dentry->d_inode;
2613 	struct task_struct *p = get_proc_task(inode);
2614 	generic_fillattr(inode, stat);
2615 
2616 	if (p) {
2617 		rcu_read_lock();
2618 		stat->nlink += get_nr_threads(p);
2619 		rcu_read_unlock();
2620 		put_task_struct(p);
2621 	}
2622 
2623 	return 0;
2624 }
2625 
2626 static const struct inode_operations proc_task_inode_operations = {
2627 	.lookup		= proc_task_lookup,
2628 	.getattr	= proc_task_getattr,
2629 	.setattr	= proc_setattr,
2630 };
2631 
2632 static const struct file_operations proc_task_operations = {
2633 	.read		= generic_read_dir,
2634 	.readdir	= proc_task_readdir,
2635 };
2636