xref: /linux/fs/proc/base.c (revision 27258e448eb301cf89e351df87aa8cb916653bf2)
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49 
50 #include <asm/uaccess.h>
51 
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/rcupdate.h>
67 #include <linux/kallsyms.h>
68 #include <linux/stacktrace.h>
69 #include <linux/resource.h>
70 #include <linux/module.h>
71 #include <linux/mount.h>
72 #include <linux/security.h>
73 #include <linux/ptrace.h>
74 #include <linux/tracehook.h>
75 #include <linux/cgroup.h>
76 #include <linux/cpuset.h>
77 #include <linux/audit.h>
78 #include <linux/poll.h>
79 #include <linux/nsproxy.h>
80 #include <linux/oom.h>
81 #include <linux/elf.h>
82 #include <linux/pid_namespace.h>
83 #include <linux/fs_struct.h>
84 #include "internal.h"
85 
86 /* NOTE:
87  *	Implementing inode permission operations in /proc is almost
88  *	certainly an error.  Permission checks need to happen during
89  *	each system call not at open time.  The reason is that most of
90  *	what we wish to check for permissions in /proc varies at runtime.
91  *
92  *	The classic example of a problem is opening file descriptors
93  *	in /proc for a task before it execs a suid executable.
94  */
95 
96 struct pid_entry {
97 	char *name;
98 	int len;
99 	mode_t mode;
100 	const struct inode_operations *iop;
101 	const struct file_operations *fop;
102 	union proc_op op;
103 };
104 
105 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
106 	.name = (NAME),					\
107 	.len  = sizeof(NAME) - 1,			\
108 	.mode = MODE,					\
109 	.iop  = IOP,					\
110 	.fop  = FOP,					\
111 	.op   = OP,					\
112 }
113 
114 #define DIR(NAME, MODE, iops, fops)	\
115 	NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
116 #define LNK(NAME, get_link)					\
117 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
118 		&proc_pid_link_inode_operations, NULL,		\
119 		{ .proc_get_link = get_link } )
120 #define REG(NAME, MODE, fops)				\
121 	NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
122 #define INF(NAME, MODE, read)				\
123 	NOD(NAME, (S_IFREG|(MODE)), 			\
124 		NULL, &proc_info_file_operations,	\
125 		{ .proc_read = read } )
126 #define ONE(NAME, MODE, show)				\
127 	NOD(NAME, (S_IFREG|(MODE)), 			\
128 		NULL, &proc_single_file_operations,	\
129 		{ .proc_show = show } )
130 
131 /*
132  * Count the number of hardlinks for the pid_entry table, excluding the .
133  * and .. links.
134  */
135 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
136 	unsigned int n)
137 {
138 	unsigned int i;
139 	unsigned int count;
140 
141 	count = 0;
142 	for (i = 0; i < n; ++i) {
143 		if (S_ISDIR(entries[i].mode))
144 			++count;
145 	}
146 
147 	return count;
148 }
149 
150 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
151 {
152 	struct fs_struct *fs;
153 	int result = -ENOENT;
154 
155 	task_lock(task);
156 	fs = task->fs;
157 	if (fs) {
158 		read_lock(&fs->lock);
159 		*path = root ? fs->root : fs->pwd;
160 		path_get(path);
161 		read_unlock(&fs->lock);
162 		result = 0;
163 	}
164 	task_unlock(task);
165 	return result;
166 }
167 
168 static int get_nr_threads(struct task_struct *tsk)
169 {
170 	unsigned long flags;
171 	int count = 0;
172 
173 	if (lock_task_sighand(tsk, &flags)) {
174 		count = atomic_read(&tsk->signal->count);
175 		unlock_task_sighand(tsk, &flags);
176 	}
177 	return count;
178 }
179 
180 static int proc_cwd_link(struct inode *inode, struct path *path)
181 {
182 	struct task_struct *task = get_proc_task(inode);
183 	int result = -ENOENT;
184 
185 	if (task) {
186 		result = get_fs_path(task, path, 0);
187 		put_task_struct(task);
188 	}
189 	return result;
190 }
191 
192 static int proc_root_link(struct inode *inode, struct path *path)
193 {
194 	struct task_struct *task = get_proc_task(inode);
195 	int result = -ENOENT;
196 
197 	if (task) {
198 		result = get_fs_path(task, path, 1);
199 		put_task_struct(task);
200 	}
201 	return result;
202 }
203 
204 /*
205  * Return zero if current may access user memory in @task, -error if not.
206  */
207 static int check_mem_permission(struct task_struct *task)
208 {
209 	/*
210 	 * A task can always look at itself, in case it chooses
211 	 * to use system calls instead of load instructions.
212 	 */
213 	if (task == current)
214 		return 0;
215 
216 	/*
217 	 * If current is actively ptrace'ing, and would also be
218 	 * permitted to freshly attach with ptrace now, permit it.
219 	 */
220 	if (task_is_stopped_or_traced(task)) {
221 		int match;
222 		rcu_read_lock();
223 		match = (tracehook_tracer_task(task) == current);
224 		rcu_read_unlock();
225 		if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
226 			return 0;
227 	}
228 
229 	/*
230 	 * Noone else is allowed.
231 	 */
232 	return -EPERM;
233 }
234 
235 struct mm_struct *mm_for_maps(struct task_struct *task)
236 {
237 	struct mm_struct *mm;
238 
239 	if (mutex_lock_killable(&task->cred_guard_mutex))
240 		return NULL;
241 
242 	mm = get_task_mm(task);
243 	if (mm && mm != current->mm &&
244 			!ptrace_may_access(task, PTRACE_MODE_READ)) {
245 		mmput(mm);
246 		mm = NULL;
247 	}
248 	mutex_unlock(&task->cred_guard_mutex);
249 
250 	return mm;
251 }
252 
253 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
254 {
255 	int res = 0;
256 	unsigned int len;
257 	struct mm_struct *mm = get_task_mm(task);
258 	if (!mm)
259 		goto out;
260 	if (!mm->arg_end)
261 		goto out_mm;	/* Shh! No looking before we're done */
262 
263  	len = mm->arg_end - mm->arg_start;
264 
265 	if (len > PAGE_SIZE)
266 		len = PAGE_SIZE;
267 
268 	res = access_process_vm(task, mm->arg_start, buffer, len, 0);
269 
270 	// If the nul at the end of args has been overwritten, then
271 	// assume application is using setproctitle(3).
272 	if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
273 		len = strnlen(buffer, res);
274 		if (len < res) {
275 		    res = len;
276 		} else {
277 			len = mm->env_end - mm->env_start;
278 			if (len > PAGE_SIZE - res)
279 				len = PAGE_SIZE - res;
280 			res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
281 			res = strnlen(buffer, res);
282 		}
283 	}
284 out_mm:
285 	mmput(mm);
286 out:
287 	return res;
288 }
289 
290 static int proc_pid_auxv(struct task_struct *task, char *buffer)
291 {
292 	int res = 0;
293 	struct mm_struct *mm = get_task_mm(task);
294 	if (mm) {
295 		unsigned int nwords = 0;
296 		do {
297 			nwords += 2;
298 		} while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
299 		res = nwords * sizeof(mm->saved_auxv[0]);
300 		if (res > PAGE_SIZE)
301 			res = PAGE_SIZE;
302 		memcpy(buffer, mm->saved_auxv, res);
303 		mmput(mm);
304 	}
305 	return res;
306 }
307 
308 
309 #ifdef CONFIG_KALLSYMS
310 /*
311  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
312  * Returns the resolved symbol.  If that fails, simply return the address.
313  */
314 static int proc_pid_wchan(struct task_struct *task, char *buffer)
315 {
316 	unsigned long wchan;
317 	char symname[KSYM_NAME_LEN];
318 
319 	wchan = get_wchan(task);
320 
321 	if (lookup_symbol_name(wchan, symname) < 0)
322 		if (!ptrace_may_access(task, PTRACE_MODE_READ))
323 			return 0;
324 		else
325 			return sprintf(buffer, "%lu", wchan);
326 	else
327 		return sprintf(buffer, "%s", symname);
328 }
329 #endif /* CONFIG_KALLSYMS */
330 
331 #ifdef CONFIG_STACKTRACE
332 
333 #define MAX_STACK_TRACE_DEPTH	64
334 
335 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
336 			  struct pid *pid, struct task_struct *task)
337 {
338 	struct stack_trace trace;
339 	unsigned long *entries;
340 	int i;
341 
342 	entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
343 	if (!entries)
344 		return -ENOMEM;
345 
346 	trace.nr_entries	= 0;
347 	trace.max_entries	= MAX_STACK_TRACE_DEPTH;
348 	trace.entries		= entries;
349 	trace.skip		= 0;
350 	save_stack_trace_tsk(task, &trace);
351 
352 	for (i = 0; i < trace.nr_entries; i++) {
353 		seq_printf(m, "[<%p>] %pS\n",
354 			   (void *)entries[i], (void *)entries[i]);
355 	}
356 	kfree(entries);
357 
358 	return 0;
359 }
360 #endif
361 
362 #ifdef CONFIG_SCHEDSTATS
363 /*
364  * Provides /proc/PID/schedstat
365  */
366 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
367 {
368 	return sprintf(buffer, "%llu %llu %lu\n",
369 			(unsigned long long)task->se.sum_exec_runtime,
370 			(unsigned long long)task->sched_info.run_delay,
371 			task->sched_info.pcount);
372 }
373 #endif
374 
375 #ifdef CONFIG_LATENCYTOP
376 static int lstats_show_proc(struct seq_file *m, void *v)
377 {
378 	int i;
379 	struct inode *inode = m->private;
380 	struct task_struct *task = get_proc_task(inode);
381 
382 	if (!task)
383 		return -ESRCH;
384 	seq_puts(m, "Latency Top version : v0.1\n");
385 	for (i = 0; i < 32; i++) {
386 		if (task->latency_record[i].backtrace[0]) {
387 			int q;
388 			seq_printf(m, "%i %li %li ",
389 				task->latency_record[i].count,
390 				task->latency_record[i].time,
391 				task->latency_record[i].max);
392 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
393 				char sym[KSYM_SYMBOL_LEN];
394 				char *c;
395 				if (!task->latency_record[i].backtrace[q])
396 					break;
397 				if (task->latency_record[i].backtrace[q] == ULONG_MAX)
398 					break;
399 				sprint_symbol(sym, task->latency_record[i].backtrace[q]);
400 				c = strchr(sym, '+');
401 				if (c)
402 					*c = 0;
403 				seq_printf(m, "%s ", sym);
404 			}
405 			seq_printf(m, "\n");
406 		}
407 
408 	}
409 	put_task_struct(task);
410 	return 0;
411 }
412 
413 static int lstats_open(struct inode *inode, struct file *file)
414 {
415 	return single_open(file, lstats_show_proc, inode);
416 }
417 
418 static ssize_t lstats_write(struct file *file, const char __user *buf,
419 			    size_t count, loff_t *offs)
420 {
421 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
422 
423 	if (!task)
424 		return -ESRCH;
425 	clear_all_latency_tracing(task);
426 	put_task_struct(task);
427 
428 	return count;
429 }
430 
431 static const struct file_operations proc_lstats_operations = {
432 	.open		= lstats_open,
433 	.read		= seq_read,
434 	.write		= lstats_write,
435 	.llseek		= seq_lseek,
436 	.release	= single_release,
437 };
438 
439 #endif
440 
441 /* The badness from the OOM killer */
442 unsigned long badness(struct task_struct *p, unsigned long uptime);
443 static int proc_oom_score(struct task_struct *task, char *buffer)
444 {
445 	unsigned long points;
446 	struct timespec uptime;
447 
448 	do_posix_clock_monotonic_gettime(&uptime);
449 	read_lock(&tasklist_lock);
450 	points = badness(task->group_leader, uptime.tv_sec);
451 	read_unlock(&tasklist_lock);
452 	return sprintf(buffer, "%lu\n", points);
453 }
454 
455 struct limit_names {
456 	char *name;
457 	char *unit;
458 };
459 
460 static const struct limit_names lnames[RLIM_NLIMITS] = {
461 	[RLIMIT_CPU] = {"Max cpu time", "ms"},
462 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
463 	[RLIMIT_DATA] = {"Max data size", "bytes"},
464 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
465 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
466 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
467 	[RLIMIT_NPROC] = {"Max processes", "processes"},
468 	[RLIMIT_NOFILE] = {"Max open files", "files"},
469 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
470 	[RLIMIT_AS] = {"Max address space", "bytes"},
471 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
472 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
473 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
474 	[RLIMIT_NICE] = {"Max nice priority", NULL},
475 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
476 	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
477 };
478 
479 /* Display limits for a process */
480 static int proc_pid_limits(struct task_struct *task, char *buffer)
481 {
482 	unsigned int i;
483 	int count = 0;
484 	unsigned long flags;
485 	char *bufptr = buffer;
486 
487 	struct rlimit rlim[RLIM_NLIMITS];
488 
489 	if (!lock_task_sighand(task, &flags))
490 		return 0;
491 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
492 	unlock_task_sighand(task, &flags);
493 
494 	/*
495 	 * print the file header
496 	 */
497 	count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
498 			"Limit", "Soft Limit", "Hard Limit", "Units");
499 
500 	for (i = 0; i < RLIM_NLIMITS; i++) {
501 		if (rlim[i].rlim_cur == RLIM_INFINITY)
502 			count += sprintf(&bufptr[count], "%-25s %-20s ",
503 					 lnames[i].name, "unlimited");
504 		else
505 			count += sprintf(&bufptr[count], "%-25s %-20lu ",
506 					 lnames[i].name, rlim[i].rlim_cur);
507 
508 		if (rlim[i].rlim_max == RLIM_INFINITY)
509 			count += sprintf(&bufptr[count], "%-20s ", "unlimited");
510 		else
511 			count += sprintf(&bufptr[count], "%-20lu ",
512 					 rlim[i].rlim_max);
513 
514 		if (lnames[i].unit)
515 			count += sprintf(&bufptr[count], "%-10s\n",
516 					 lnames[i].unit);
517 		else
518 			count += sprintf(&bufptr[count], "\n");
519 	}
520 
521 	return count;
522 }
523 
524 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
525 static int proc_pid_syscall(struct task_struct *task, char *buffer)
526 {
527 	long nr;
528 	unsigned long args[6], sp, pc;
529 
530 	if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
531 		return sprintf(buffer, "running\n");
532 
533 	if (nr < 0)
534 		return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
535 
536 	return sprintf(buffer,
537 		       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
538 		       nr,
539 		       args[0], args[1], args[2], args[3], args[4], args[5],
540 		       sp, pc);
541 }
542 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
543 
544 /************************************************************************/
545 /*                       Here the fs part begins                        */
546 /************************************************************************/
547 
548 /* permission checks */
549 static int proc_fd_access_allowed(struct inode *inode)
550 {
551 	struct task_struct *task;
552 	int allowed = 0;
553 	/* Allow access to a task's file descriptors if it is us or we
554 	 * may use ptrace attach to the process and find out that
555 	 * information.
556 	 */
557 	task = get_proc_task(inode);
558 	if (task) {
559 		allowed = ptrace_may_access(task, PTRACE_MODE_READ);
560 		put_task_struct(task);
561 	}
562 	return allowed;
563 }
564 
565 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
566 {
567 	int error;
568 	struct inode *inode = dentry->d_inode;
569 
570 	if (attr->ia_valid & ATTR_MODE)
571 		return -EPERM;
572 
573 	error = inode_change_ok(inode, attr);
574 	if (!error)
575 		error = inode_setattr(inode, attr);
576 	return error;
577 }
578 
579 static const struct inode_operations proc_def_inode_operations = {
580 	.setattr	= proc_setattr,
581 };
582 
583 static int mounts_open_common(struct inode *inode, struct file *file,
584 			      const struct seq_operations *op)
585 {
586 	struct task_struct *task = get_proc_task(inode);
587 	struct nsproxy *nsp;
588 	struct mnt_namespace *ns = NULL;
589 	struct path root;
590 	struct proc_mounts *p;
591 	int ret = -EINVAL;
592 
593 	if (task) {
594 		rcu_read_lock();
595 		nsp = task_nsproxy(task);
596 		if (nsp) {
597 			ns = nsp->mnt_ns;
598 			if (ns)
599 				get_mnt_ns(ns);
600 		}
601 		rcu_read_unlock();
602 		if (ns && get_fs_path(task, &root, 1) == 0)
603 			ret = 0;
604 		put_task_struct(task);
605 	}
606 
607 	if (!ns)
608 		goto err;
609 	if (ret)
610 		goto err_put_ns;
611 
612 	ret = -ENOMEM;
613 	p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
614 	if (!p)
615 		goto err_put_path;
616 
617 	file->private_data = &p->m;
618 	ret = seq_open(file, op);
619 	if (ret)
620 		goto err_free;
621 
622 	p->m.private = p;
623 	p->ns = ns;
624 	p->root = root;
625 	p->event = ns->event;
626 
627 	return 0;
628 
629  err_free:
630 	kfree(p);
631  err_put_path:
632 	path_put(&root);
633  err_put_ns:
634 	put_mnt_ns(ns);
635  err:
636 	return ret;
637 }
638 
639 static int mounts_release(struct inode *inode, struct file *file)
640 {
641 	struct proc_mounts *p = file->private_data;
642 	path_put(&p->root);
643 	put_mnt_ns(p->ns);
644 	return seq_release(inode, file);
645 }
646 
647 static unsigned mounts_poll(struct file *file, poll_table *wait)
648 {
649 	struct proc_mounts *p = file->private_data;
650 	struct mnt_namespace *ns = p->ns;
651 	unsigned res = POLLIN | POLLRDNORM;
652 
653 	poll_wait(file, &ns->poll, wait);
654 
655 	spin_lock(&vfsmount_lock);
656 	if (p->event != ns->event) {
657 		p->event = ns->event;
658 		res |= POLLERR | POLLPRI;
659 	}
660 	spin_unlock(&vfsmount_lock);
661 
662 	return res;
663 }
664 
665 static int mounts_open(struct inode *inode, struct file *file)
666 {
667 	return mounts_open_common(inode, file, &mounts_op);
668 }
669 
670 static const struct file_operations proc_mounts_operations = {
671 	.open		= mounts_open,
672 	.read		= seq_read,
673 	.llseek		= seq_lseek,
674 	.release	= mounts_release,
675 	.poll		= mounts_poll,
676 };
677 
678 static int mountinfo_open(struct inode *inode, struct file *file)
679 {
680 	return mounts_open_common(inode, file, &mountinfo_op);
681 }
682 
683 static const struct file_operations proc_mountinfo_operations = {
684 	.open		= mountinfo_open,
685 	.read		= seq_read,
686 	.llseek		= seq_lseek,
687 	.release	= mounts_release,
688 	.poll		= mounts_poll,
689 };
690 
691 static int mountstats_open(struct inode *inode, struct file *file)
692 {
693 	return mounts_open_common(inode, file, &mountstats_op);
694 }
695 
696 static const struct file_operations proc_mountstats_operations = {
697 	.open		= mountstats_open,
698 	.read		= seq_read,
699 	.llseek		= seq_lseek,
700 	.release	= mounts_release,
701 };
702 
703 #define PROC_BLOCK_SIZE	(3*1024)		/* 4K page size but our output routines use some slack for overruns */
704 
705 static ssize_t proc_info_read(struct file * file, char __user * buf,
706 			  size_t count, loff_t *ppos)
707 {
708 	struct inode * inode = file->f_path.dentry->d_inode;
709 	unsigned long page;
710 	ssize_t length;
711 	struct task_struct *task = get_proc_task(inode);
712 
713 	length = -ESRCH;
714 	if (!task)
715 		goto out_no_task;
716 
717 	if (count > PROC_BLOCK_SIZE)
718 		count = PROC_BLOCK_SIZE;
719 
720 	length = -ENOMEM;
721 	if (!(page = __get_free_page(GFP_TEMPORARY)))
722 		goto out;
723 
724 	length = PROC_I(inode)->op.proc_read(task, (char*)page);
725 
726 	if (length >= 0)
727 		length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
728 	free_page(page);
729 out:
730 	put_task_struct(task);
731 out_no_task:
732 	return length;
733 }
734 
735 static const struct file_operations proc_info_file_operations = {
736 	.read		= proc_info_read,
737 };
738 
739 static int proc_single_show(struct seq_file *m, void *v)
740 {
741 	struct inode *inode = m->private;
742 	struct pid_namespace *ns;
743 	struct pid *pid;
744 	struct task_struct *task;
745 	int ret;
746 
747 	ns = inode->i_sb->s_fs_info;
748 	pid = proc_pid(inode);
749 	task = get_pid_task(pid, PIDTYPE_PID);
750 	if (!task)
751 		return -ESRCH;
752 
753 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
754 
755 	put_task_struct(task);
756 	return ret;
757 }
758 
759 static int proc_single_open(struct inode *inode, struct file *filp)
760 {
761 	int ret;
762 	ret = single_open(filp, proc_single_show, NULL);
763 	if (!ret) {
764 		struct seq_file *m = filp->private_data;
765 
766 		m->private = inode;
767 	}
768 	return ret;
769 }
770 
771 static const struct file_operations proc_single_file_operations = {
772 	.open		= proc_single_open,
773 	.read		= seq_read,
774 	.llseek		= seq_lseek,
775 	.release	= single_release,
776 };
777 
778 static int mem_open(struct inode* inode, struct file* file)
779 {
780 	file->private_data = (void*)((long)current->self_exec_id);
781 	return 0;
782 }
783 
784 static ssize_t mem_read(struct file * file, char __user * buf,
785 			size_t count, loff_t *ppos)
786 {
787 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
788 	char *page;
789 	unsigned long src = *ppos;
790 	int ret = -ESRCH;
791 	struct mm_struct *mm;
792 
793 	if (!task)
794 		goto out_no_task;
795 
796 	if (check_mem_permission(task))
797 		goto out;
798 
799 	ret = -ENOMEM;
800 	page = (char *)__get_free_page(GFP_TEMPORARY);
801 	if (!page)
802 		goto out;
803 
804 	ret = 0;
805 
806 	mm = get_task_mm(task);
807 	if (!mm)
808 		goto out_free;
809 
810 	ret = -EIO;
811 
812 	if (file->private_data != (void*)((long)current->self_exec_id))
813 		goto out_put;
814 
815 	ret = 0;
816 
817 	while (count > 0) {
818 		int this_len, retval;
819 
820 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
821 		retval = access_process_vm(task, src, page, this_len, 0);
822 		if (!retval || check_mem_permission(task)) {
823 			if (!ret)
824 				ret = -EIO;
825 			break;
826 		}
827 
828 		if (copy_to_user(buf, page, retval)) {
829 			ret = -EFAULT;
830 			break;
831 		}
832 
833 		ret += retval;
834 		src += retval;
835 		buf += retval;
836 		count -= retval;
837 	}
838 	*ppos = src;
839 
840 out_put:
841 	mmput(mm);
842 out_free:
843 	free_page((unsigned long) page);
844 out:
845 	put_task_struct(task);
846 out_no_task:
847 	return ret;
848 }
849 
850 #define mem_write NULL
851 
852 #ifndef mem_write
853 /* This is a security hazard */
854 static ssize_t mem_write(struct file * file, const char __user *buf,
855 			 size_t count, loff_t *ppos)
856 {
857 	int copied;
858 	char *page;
859 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
860 	unsigned long dst = *ppos;
861 
862 	copied = -ESRCH;
863 	if (!task)
864 		goto out_no_task;
865 
866 	if (check_mem_permission(task))
867 		goto out;
868 
869 	copied = -ENOMEM;
870 	page = (char *)__get_free_page(GFP_TEMPORARY);
871 	if (!page)
872 		goto out;
873 
874 	copied = 0;
875 	while (count > 0) {
876 		int this_len, retval;
877 
878 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
879 		if (copy_from_user(page, buf, this_len)) {
880 			copied = -EFAULT;
881 			break;
882 		}
883 		retval = access_process_vm(task, dst, page, this_len, 1);
884 		if (!retval) {
885 			if (!copied)
886 				copied = -EIO;
887 			break;
888 		}
889 		copied += retval;
890 		buf += retval;
891 		dst += retval;
892 		count -= retval;
893 	}
894 	*ppos = dst;
895 	free_page((unsigned long) page);
896 out:
897 	put_task_struct(task);
898 out_no_task:
899 	return copied;
900 }
901 #endif
902 
903 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
904 {
905 	switch (orig) {
906 	case 0:
907 		file->f_pos = offset;
908 		break;
909 	case 1:
910 		file->f_pos += offset;
911 		break;
912 	default:
913 		return -EINVAL;
914 	}
915 	force_successful_syscall_return();
916 	return file->f_pos;
917 }
918 
919 static const struct file_operations proc_mem_operations = {
920 	.llseek		= mem_lseek,
921 	.read		= mem_read,
922 	.write		= mem_write,
923 	.open		= mem_open,
924 };
925 
926 static ssize_t environ_read(struct file *file, char __user *buf,
927 			size_t count, loff_t *ppos)
928 {
929 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
930 	char *page;
931 	unsigned long src = *ppos;
932 	int ret = -ESRCH;
933 	struct mm_struct *mm;
934 
935 	if (!task)
936 		goto out_no_task;
937 
938 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
939 		goto out;
940 
941 	ret = -ENOMEM;
942 	page = (char *)__get_free_page(GFP_TEMPORARY);
943 	if (!page)
944 		goto out;
945 
946 	ret = 0;
947 
948 	mm = get_task_mm(task);
949 	if (!mm)
950 		goto out_free;
951 
952 	while (count > 0) {
953 		int this_len, retval, max_len;
954 
955 		this_len = mm->env_end - (mm->env_start + src);
956 
957 		if (this_len <= 0)
958 			break;
959 
960 		max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
961 		this_len = (this_len > max_len) ? max_len : this_len;
962 
963 		retval = access_process_vm(task, (mm->env_start + src),
964 			page, this_len, 0);
965 
966 		if (retval <= 0) {
967 			ret = retval;
968 			break;
969 		}
970 
971 		if (copy_to_user(buf, page, retval)) {
972 			ret = -EFAULT;
973 			break;
974 		}
975 
976 		ret += retval;
977 		src += retval;
978 		buf += retval;
979 		count -= retval;
980 	}
981 	*ppos = src;
982 
983 	mmput(mm);
984 out_free:
985 	free_page((unsigned long) page);
986 out:
987 	put_task_struct(task);
988 out_no_task:
989 	return ret;
990 }
991 
992 static const struct file_operations proc_environ_operations = {
993 	.read		= environ_read,
994 };
995 
996 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
997 				size_t count, loff_t *ppos)
998 {
999 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1000 	char buffer[PROC_NUMBUF];
1001 	size_t len;
1002 	int oom_adjust = OOM_DISABLE;
1003 	unsigned long flags;
1004 
1005 	if (!task)
1006 		return -ESRCH;
1007 
1008 	if (lock_task_sighand(task, &flags)) {
1009 		oom_adjust = task->signal->oom_adj;
1010 		unlock_task_sighand(task, &flags);
1011 	}
1012 
1013 	put_task_struct(task);
1014 
1015 	len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1016 
1017 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1018 }
1019 
1020 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1021 				size_t count, loff_t *ppos)
1022 {
1023 	struct task_struct *task;
1024 	char buffer[PROC_NUMBUF];
1025 	long oom_adjust;
1026 	unsigned long flags;
1027 	int err;
1028 
1029 	memset(buffer, 0, sizeof(buffer));
1030 	if (count > sizeof(buffer) - 1)
1031 		count = sizeof(buffer) - 1;
1032 	if (copy_from_user(buffer, buf, count))
1033 		return -EFAULT;
1034 
1035 	err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1036 	if (err)
1037 		return -EINVAL;
1038 	if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1039 	     oom_adjust != OOM_DISABLE)
1040 		return -EINVAL;
1041 
1042 	task = get_proc_task(file->f_path.dentry->d_inode);
1043 	if (!task)
1044 		return -ESRCH;
1045 	if (!lock_task_sighand(task, &flags)) {
1046 		put_task_struct(task);
1047 		return -ESRCH;
1048 	}
1049 
1050 	if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1051 		unlock_task_sighand(task, &flags);
1052 		put_task_struct(task);
1053 		return -EACCES;
1054 	}
1055 
1056 	task->signal->oom_adj = oom_adjust;
1057 
1058 	unlock_task_sighand(task, &flags);
1059 	put_task_struct(task);
1060 
1061 	return count;
1062 }
1063 
1064 static const struct file_operations proc_oom_adjust_operations = {
1065 	.read		= oom_adjust_read,
1066 	.write		= oom_adjust_write,
1067 };
1068 
1069 #ifdef CONFIG_AUDITSYSCALL
1070 #define TMPBUFLEN 21
1071 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1072 				  size_t count, loff_t *ppos)
1073 {
1074 	struct inode * inode = file->f_path.dentry->d_inode;
1075 	struct task_struct *task = get_proc_task(inode);
1076 	ssize_t length;
1077 	char tmpbuf[TMPBUFLEN];
1078 
1079 	if (!task)
1080 		return -ESRCH;
1081 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1082 				audit_get_loginuid(task));
1083 	put_task_struct(task);
1084 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1085 }
1086 
1087 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1088 				   size_t count, loff_t *ppos)
1089 {
1090 	struct inode * inode = file->f_path.dentry->d_inode;
1091 	char *page, *tmp;
1092 	ssize_t length;
1093 	uid_t loginuid;
1094 
1095 	if (!capable(CAP_AUDIT_CONTROL))
1096 		return -EPERM;
1097 
1098 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1099 		return -EPERM;
1100 
1101 	if (count >= PAGE_SIZE)
1102 		count = PAGE_SIZE - 1;
1103 
1104 	if (*ppos != 0) {
1105 		/* No partial writes. */
1106 		return -EINVAL;
1107 	}
1108 	page = (char*)__get_free_page(GFP_TEMPORARY);
1109 	if (!page)
1110 		return -ENOMEM;
1111 	length = -EFAULT;
1112 	if (copy_from_user(page, buf, count))
1113 		goto out_free_page;
1114 
1115 	page[count] = '\0';
1116 	loginuid = simple_strtoul(page, &tmp, 10);
1117 	if (tmp == page) {
1118 		length = -EINVAL;
1119 		goto out_free_page;
1120 
1121 	}
1122 	length = audit_set_loginuid(current, loginuid);
1123 	if (likely(length == 0))
1124 		length = count;
1125 
1126 out_free_page:
1127 	free_page((unsigned long) page);
1128 	return length;
1129 }
1130 
1131 static const struct file_operations proc_loginuid_operations = {
1132 	.read		= proc_loginuid_read,
1133 	.write		= proc_loginuid_write,
1134 };
1135 
1136 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1137 				  size_t count, loff_t *ppos)
1138 {
1139 	struct inode * inode = file->f_path.dentry->d_inode;
1140 	struct task_struct *task = get_proc_task(inode);
1141 	ssize_t length;
1142 	char tmpbuf[TMPBUFLEN];
1143 
1144 	if (!task)
1145 		return -ESRCH;
1146 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1147 				audit_get_sessionid(task));
1148 	put_task_struct(task);
1149 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1150 }
1151 
1152 static const struct file_operations proc_sessionid_operations = {
1153 	.read		= proc_sessionid_read,
1154 };
1155 #endif
1156 
1157 #ifdef CONFIG_FAULT_INJECTION
1158 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1159 				      size_t count, loff_t *ppos)
1160 {
1161 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1162 	char buffer[PROC_NUMBUF];
1163 	size_t len;
1164 	int make_it_fail;
1165 
1166 	if (!task)
1167 		return -ESRCH;
1168 	make_it_fail = task->make_it_fail;
1169 	put_task_struct(task);
1170 
1171 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1172 
1173 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1174 }
1175 
1176 static ssize_t proc_fault_inject_write(struct file * file,
1177 			const char __user * buf, size_t count, loff_t *ppos)
1178 {
1179 	struct task_struct *task;
1180 	char buffer[PROC_NUMBUF], *end;
1181 	int make_it_fail;
1182 
1183 	if (!capable(CAP_SYS_RESOURCE))
1184 		return -EPERM;
1185 	memset(buffer, 0, sizeof(buffer));
1186 	if (count > sizeof(buffer) - 1)
1187 		count = sizeof(buffer) - 1;
1188 	if (copy_from_user(buffer, buf, count))
1189 		return -EFAULT;
1190 	make_it_fail = simple_strtol(buffer, &end, 0);
1191 	if (*end == '\n')
1192 		end++;
1193 	task = get_proc_task(file->f_dentry->d_inode);
1194 	if (!task)
1195 		return -ESRCH;
1196 	task->make_it_fail = make_it_fail;
1197 	put_task_struct(task);
1198 	if (end - buffer == 0)
1199 		return -EIO;
1200 	return end - buffer;
1201 }
1202 
1203 static const struct file_operations proc_fault_inject_operations = {
1204 	.read		= proc_fault_inject_read,
1205 	.write		= proc_fault_inject_write,
1206 };
1207 #endif
1208 
1209 
1210 #ifdef CONFIG_SCHED_DEBUG
1211 /*
1212  * Print out various scheduling related per-task fields:
1213  */
1214 static int sched_show(struct seq_file *m, void *v)
1215 {
1216 	struct inode *inode = m->private;
1217 	struct task_struct *p;
1218 
1219 	p = get_proc_task(inode);
1220 	if (!p)
1221 		return -ESRCH;
1222 	proc_sched_show_task(p, m);
1223 
1224 	put_task_struct(p);
1225 
1226 	return 0;
1227 }
1228 
1229 static ssize_t
1230 sched_write(struct file *file, const char __user *buf,
1231 	    size_t count, loff_t *offset)
1232 {
1233 	struct inode *inode = file->f_path.dentry->d_inode;
1234 	struct task_struct *p;
1235 
1236 	p = get_proc_task(inode);
1237 	if (!p)
1238 		return -ESRCH;
1239 	proc_sched_set_task(p);
1240 
1241 	put_task_struct(p);
1242 
1243 	return count;
1244 }
1245 
1246 static int sched_open(struct inode *inode, struct file *filp)
1247 {
1248 	int ret;
1249 
1250 	ret = single_open(filp, sched_show, NULL);
1251 	if (!ret) {
1252 		struct seq_file *m = filp->private_data;
1253 
1254 		m->private = inode;
1255 	}
1256 	return ret;
1257 }
1258 
1259 static const struct file_operations proc_pid_sched_operations = {
1260 	.open		= sched_open,
1261 	.read		= seq_read,
1262 	.write		= sched_write,
1263 	.llseek		= seq_lseek,
1264 	.release	= single_release,
1265 };
1266 
1267 #endif
1268 
1269 /*
1270  * We added or removed a vma mapping the executable. The vmas are only mapped
1271  * during exec and are not mapped with the mmap system call.
1272  * Callers must hold down_write() on the mm's mmap_sem for these
1273  */
1274 void added_exe_file_vma(struct mm_struct *mm)
1275 {
1276 	mm->num_exe_file_vmas++;
1277 }
1278 
1279 void removed_exe_file_vma(struct mm_struct *mm)
1280 {
1281 	mm->num_exe_file_vmas--;
1282 	if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1283 		fput(mm->exe_file);
1284 		mm->exe_file = NULL;
1285 	}
1286 
1287 }
1288 
1289 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1290 {
1291 	if (new_exe_file)
1292 		get_file(new_exe_file);
1293 	if (mm->exe_file)
1294 		fput(mm->exe_file);
1295 	mm->exe_file = new_exe_file;
1296 	mm->num_exe_file_vmas = 0;
1297 }
1298 
1299 struct file *get_mm_exe_file(struct mm_struct *mm)
1300 {
1301 	struct file *exe_file;
1302 
1303 	/* We need mmap_sem to protect against races with removal of
1304 	 * VM_EXECUTABLE vmas */
1305 	down_read(&mm->mmap_sem);
1306 	exe_file = mm->exe_file;
1307 	if (exe_file)
1308 		get_file(exe_file);
1309 	up_read(&mm->mmap_sem);
1310 	return exe_file;
1311 }
1312 
1313 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1314 {
1315 	/* It's safe to write the exe_file pointer without exe_file_lock because
1316 	 * this is called during fork when the task is not yet in /proc */
1317 	newmm->exe_file = get_mm_exe_file(oldmm);
1318 }
1319 
1320 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1321 {
1322 	struct task_struct *task;
1323 	struct mm_struct *mm;
1324 	struct file *exe_file;
1325 
1326 	task = get_proc_task(inode);
1327 	if (!task)
1328 		return -ENOENT;
1329 	mm = get_task_mm(task);
1330 	put_task_struct(task);
1331 	if (!mm)
1332 		return -ENOENT;
1333 	exe_file = get_mm_exe_file(mm);
1334 	mmput(mm);
1335 	if (exe_file) {
1336 		*exe_path = exe_file->f_path;
1337 		path_get(&exe_file->f_path);
1338 		fput(exe_file);
1339 		return 0;
1340 	} else
1341 		return -ENOENT;
1342 }
1343 
1344 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1345 {
1346 	struct inode *inode = dentry->d_inode;
1347 	int error = -EACCES;
1348 
1349 	/* We don't need a base pointer in the /proc filesystem */
1350 	path_put(&nd->path);
1351 
1352 	/* Are we allowed to snoop on the tasks file descriptors? */
1353 	if (!proc_fd_access_allowed(inode))
1354 		goto out;
1355 
1356 	error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1357 	nd->last_type = LAST_BIND;
1358 out:
1359 	return ERR_PTR(error);
1360 }
1361 
1362 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1363 {
1364 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1365 	char *pathname;
1366 	int len;
1367 
1368 	if (!tmp)
1369 		return -ENOMEM;
1370 
1371 	pathname = d_path(path, tmp, PAGE_SIZE);
1372 	len = PTR_ERR(pathname);
1373 	if (IS_ERR(pathname))
1374 		goto out;
1375 	len = tmp + PAGE_SIZE - 1 - pathname;
1376 
1377 	if (len > buflen)
1378 		len = buflen;
1379 	if (copy_to_user(buffer, pathname, len))
1380 		len = -EFAULT;
1381  out:
1382 	free_page((unsigned long)tmp);
1383 	return len;
1384 }
1385 
1386 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1387 {
1388 	int error = -EACCES;
1389 	struct inode *inode = dentry->d_inode;
1390 	struct path path;
1391 
1392 	/* Are we allowed to snoop on the tasks file descriptors? */
1393 	if (!proc_fd_access_allowed(inode))
1394 		goto out;
1395 
1396 	error = PROC_I(inode)->op.proc_get_link(inode, &path);
1397 	if (error)
1398 		goto out;
1399 
1400 	error = do_proc_readlink(&path, buffer, buflen);
1401 	path_put(&path);
1402 out:
1403 	return error;
1404 }
1405 
1406 static const struct inode_operations proc_pid_link_inode_operations = {
1407 	.readlink	= proc_pid_readlink,
1408 	.follow_link	= proc_pid_follow_link,
1409 	.setattr	= proc_setattr,
1410 };
1411 
1412 
1413 /* building an inode */
1414 
1415 static int task_dumpable(struct task_struct *task)
1416 {
1417 	int dumpable = 0;
1418 	struct mm_struct *mm;
1419 
1420 	task_lock(task);
1421 	mm = task->mm;
1422 	if (mm)
1423 		dumpable = get_dumpable(mm);
1424 	task_unlock(task);
1425 	if(dumpable == 1)
1426 		return 1;
1427 	return 0;
1428 }
1429 
1430 
1431 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1432 {
1433 	struct inode * inode;
1434 	struct proc_inode *ei;
1435 	const struct cred *cred;
1436 
1437 	/* We need a new inode */
1438 
1439 	inode = new_inode(sb);
1440 	if (!inode)
1441 		goto out;
1442 
1443 	/* Common stuff */
1444 	ei = PROC_I(inode);
1445 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1446 	inode->i_op = &proc_def_inode_operations;
1447 
1448 	/*
1449 	 * grab the reference to task.
1450 	 */
1451 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1452 	if (!ei->pid)
1453 		goto out_unlock;
1454 
1455 	if (task_dumpable(task)) {
1456 		rcu_read_lock();
1457 		cred = __task_cred(task);
1458 		inode->i_uid = cred->euid;
1459 		inode->i_gid = cred->egid;
1460 		rcu_read_unlock();
1461 	}
1462 	security_task_to_inode(task, inode);
1463 
1464 out:
1465 	return inode;
1466 
1467 out_unlock:
1468 	iput(inode);
1469 	return NULL;
1470 }
1471 
1472 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1473 {
1474 	struct inode *inode = dentry->d_inode;
1475 	struct task_struct *task;
1476 	const struct cred *cred;
1477 
1478 	generic_fillattr(inode, stat);
1479 
1480 	rcu_read_lock();
1481 	stat->uid = 0;
1482 	stat->gid = 0;
1483 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1484 	if (task) {
1485 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1486 		    task_dumpable(task)) {
1487 			cred = __task_cred(task);
1488 			stat->uid = cred->euid;
1489 			stat->gid = cred->egid;
1490 		}
1491 	}
1492 	rcu_read_unlock();
1493 	return 0;
1494 }
1495 
1496 /* dentry stuff */
1497 
1498 /*
1499  *	Exceptional case: normally we are not allowed to unhash a busy
1500  * directory. In this case, however, we can do it - no aliasing problems
1501  * due to the way we treat inodes.
1502  *
1503  * Rewrite the inode's ownerships here because the owning task may have
1504  * performed a setuid(), etc.
1505  *
1506  * Before the /proc/pid/status file was created the only way to read
1507  * the effective uid of a /process was to stat /proc/pid.  Reading
1508  * /proc/pid/status is slow enough that procps and other packages
1509  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1510  * made this apply to all per process world readable and executable
1511  * directories.
1512  */
1513 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1514 {
1515 	struct inode *inode = dentry->d_inode;
1516 	struct task_struct *task = get_proc_task(inode);
1517 	const struct cred *cred;
1518 
1519 	if (task) {
1520 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1521 		    task_dumpable(task)) {
1522 			rcu_read_lock();
1523 			cred = __task_cred(task);
1524 			inode->i_uid = cred->euid;
1525 			inode->i_gid = cred->egid;
1526 			rcu_read_unlock();
1527 		} else {
1528 			inode->i_uid = 0;
1529 			inode->i_gid = 0;
1530 		}
1531 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1532 		security_task_to_inode(task, inode);
1533 		put_task_struct(task);
1534 		return 1;
1535 	}
1536 	d_drop(dentry);
1537 	return 0;
1538 }
1539 
1540 static int pid_delete_dentry(struct dentry * dentry)
1541 {
1542 	/* Is the task we represent dead?
1543 	 * If so, then don't put the dentry on the lru list,
1544 	 * kill it immediately.
1545 	 */
1546 	return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1547 }
1548 
1549 static const struct dentry_operations pid_dentry_operations =
1550 {
1551 	.d_revalidate	= pid_revalidate,
1552 	.d_delete	= pid_delete_dentry,
1553 };
1554 
1555 /* Lookups */
1556 
1557 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1558 				struct task_struct *, const void *);
1559 
1560 /*
1561  * Fill a directory entry.
1562  *
1563  * If possible create the dcache entry and derive our inode number and
1564  * file type from dcache entry.
1565  *
1566  * Since all of the proc inode numbers are dynamically generated, the inode
1567  * numbers do not exist until the inode is cache.  This means creating the
1568  * the dcache entry in readdir is necessary to keep the inode numbers
1569  * reported by readdir in sync with the inode numbers reported
1570  * by stat.
1571  */
1572 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1573 	char *name, int len,
1574 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1575 {
1576 	struct dentry *child, *dir = filp->f_path.dentry;
1577 	struct inode *inode;
1578 	struct qstr qname;
1579 	ino_t ino = 0;
1580 	unsigned type = DT_UNKNOWN;
1581 
1582 	qname.name = name;
1583 	qname.len  = len;
1584 	qname.hash = full_name_hash(name, len);
1585 
1586 	child = d_lookup(dir, &qname);
1587 	if (!child) {
1588 		struct dentry *new;
1589 		new = d_alloc(dir, &qname);
1590 		if (new) {
1591 			child = instantiate(dir->d_inode, new, task, ptr);
1592 			if (child)
1593 				dput(new);
1594 			else
1595 				child = new;
1596 		}
1597 	}
1598 	if (!child || IS_ERR(child) || !child->d_inode)
1599 		goto end_instantiate;
1600 	inode = child->d_inode;
1601 	if (inode) {
1602 		ino = inode->i_ino;
1603 		type = inode->i_mode >> 12;
1604 	}
1605 	dput(child);
1606 end_instantiate:
1607 	if (!ino)
1608 		ino = find_inode_number(dir, &qname);
1609 	if (!ino)
1610 		ino = 1;
1611 	return filldir(dirent, name, len, filp->f_pos, ino, type);
1612 }
1613 
1614 static unsigned name_to_int(struct dentry *dentry)
1615 {
1616 	const char *name = dentry->d_name.name;
1617 	int len = dentry->d_name.len;
1618 	unsigned n = 0;
1619 
1620 	if (len > 1 && *name == '0')
1621 		goto out;
1622 	while (len-- > 0) {
1623 		unsigned c = *name++ - '0';
1624 		if (c > 9)
1625 			goto out;
1626 		if (n >= (~0U-9)/10)
1627 			goto out;
1628 		n *= 10;
1629 		n += c;
1630 	}
1631 	return n;
1632 out:
1633 	return ~0U;
1634 }
1635 
1636 #define PROC_FDINFO_MAX 64
1637 
1638 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1639 {
1640 	struct task_struct *task = get_proc_task(inode);
1641 	struct files_struct *files = NULL;
1642 	struct file *file;
1643 	int fd = proc_fd(inode);
1644 
1645 	if (task) {
1646 		files = get_files_struct(task);
1647 		put_task_struct(task);
1648 	}
1649 	if (files) {
1650 		/*
1651 		 * We are not taking a ref to the file structure, so we must
1652 		 * hold ->file_lock.
1653 		 */
1654 		spin_lock(&files->file_lock);
1655 		file = fcheck_files(files, fd);
1656 		if (file) {
1657 			if (path) {
1658 				*path = file->f_path;
1659 				path_get(&file->f_path);
1660 			}
1661 			if (info)
1662 				snprintf(info, PROC_FDINFO_MAX,
1663 					 "pos:\t%lli\n"
1664 					 "flags:\t0%o\n",
1665 					 (long long) file->f_pos,
1666 					 file->f_flags);
1667 			spin_unlock(&files->file_lock);
1668 			put_files_struct(files);
1669 			return 0;
1670 		}
1671 		spin_unlock(&files->file_lock);
1672 		put_files_struct(files);
1673 	}
1674 	return -ENOENT;
1675 }
1676 
1677 static int proc_fd_link(struct inode *inode, struct path *path)
1678 {
1679 	return proc_fd_info(inode, path, NULL);
1680 }
1681 
1682 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1683 {
1684 	struct inode *inode = dentry->d_inode;
1685 	struct task_struct *task = get_proc_task(inode);
1686 	int fd = proc_fd(inode);
1687 	struct files_struct *files;
1688 	const struct cred *cred;
1689 
1690 	if (task) {
1691 		files = get_files_struct(task);
1692 		if (files) {
1693 			rcu_read_lock();
1694 			if (fcheck_files(files, fd)) {
1695 				rcu_read_unlock();
1696 				put_files_struct(files);
1697 				if (task_dumpable(task)) {
1698 					rcu_read_lock();
1699 					cred = __task_cred(task);
1700 					inode->i_uid = cred->euid;
1701 					inode->i_gid = cred->egid;
1702 					rcu_read_unlock();
1703 				} else {
1704 					inode->i_uid = 0;
1705 					inode->i_gid = 0;
1706 				}
1707 				inode->i_mode &= ~(S_ISUID | S_ISGID);
1708 				security_task_to_inode(task, inode);
1709 				put_task_struct(task);
1710 				return 1;
1711 			}
1712 			rcu_read_unlock();
1713 			put_files_struct(files);
1714 		}
1715 		put_task_struct(task);
1716 	}
1717 	d_drop(dentry);
1718 	return 0;
1719 }
1720 
1721 static const struct dentry_operations tid_fd_dentry_operations =
1722 {
1723 	.d_revalidate	= tid_fd_revalidate,
1724 	.d_delete	= pid_delete_dentry,
1725 };
1726 
1727 static struct dentry *proc_fd_instantiate(struct inode *dir,
1728 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1729 {
1730 	unsigned fd = *(const unsigned *)ptr;
1731 	struct file *file;
1732 	struct files_struct *files;
1733  	struct inode *inode;
1734  	struct proc_inode *ei;
1735 	struct dentry *error = ERR_PTR(-ENOENT);
1736 
1737 	inode = proc_pid_make_inode(dir->i_sb, task);
1738 	if (!inode)
1739 		goto out;
1740 	ei = PROC_I(inode);
1741 	ei->fd = fd;
1742 	files = get_files_struct(task);
1743 	if (!files)
1744 		goto out_iput;
1745 	inode->i_mode = S_IFLNK;
1746 
1747 	/*
1748 	 * We are not taking a ref to the file structure, so we must
1749 	 * hold ->file_lock.
1750 	 */
1751 	spin_lock(&files->file_lock);
1752 	file = fcheck_files(files, fd);
1753 	if (!file)
1754 		goto out_unlock;
1755 	if (file->f_mode & FMODE_READ)
1756 		inode->i_mode |= S_IRUSR | S_IXUSR;
1757 	if (file->f_mode & FMODE_WRITE)
1758 		inode->i_mode |= S_IWUSR | S_IXUSR;
1759 	spin_unlock(&files->file_lock);
1760 	put_files_struct(files);
1761 
1762 	inode->i_op = &proc_pid_link_inode_operations;
1763 	inode->i_size = 64;
1764 	ei->op.proc_get_link = proc_fd_link;
1765 	dentry->d_op = &tid_fd_dentry_operations;
1766 	d_add(dentry, inode);
1767 	/* Close the race of the process dying before we return the dentry */
1768 	if (tid_fd_revalidate(dentry, NULL))
1769 		error = NULL;
1770 
1771  out:
1772 	return error;
1773 out_unlock:
1774 	spin_unlock(&files->file_lock);
1775 	put_files_struct(files);
1776 out_iput:
1777 	iput(inode);
1778 	goto out;
1779 }
1780 
1781 static struct dentry *proc_lookupfd_common(struct inode *dir,
1782 					   struct dentry *dentry,
1783 					   instantiate_t instantiate)
1784 {
1785 	struct task_struct *task = get_proc_task(dir);
1786 	unsigned fd = name_to_int(dentry);
1787 	struct dentry *result = ERR_PTR(-ENOENT);
1788 
1789 	if (!task)
1790 		goto out_no_task;
1791 	if (fd == ~0U)
1792 		goto out;
1793 
1794 	result = instantiate(dir, dentry, task, &fd);
1795 out:
1796 	put_task_struct(task);
1797 out_no_task:
1798 	return result;
1799 }
1800 
1801 static int proc_readfd_common(struct file * filp, void * dirent,
1802 			      filldir_t filldir, instantiate_t instantiate)
1803 {
1804 	struct dentry *dentry = filp->f_path.dentry;
1805 	struct inode *inode = dentry->d_inode;
1806 	struct task_struct *p = get_proc_task(inode);
1807 	unsigned int fd, ino;
1808 	int retval;
1809 	struct files_struct * files;
1810 
1811 	retval = -ENOENT;
1812 	if (!p)
1813 		goto out_no_task;
1814 	retval = 0;
1815 
1816 	fd = filp->f_pos;
1817 	switch (fd) {
1818 		case 0:
1819 			if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1820 				goto out;
1821 			filp->f_pos++;
1822 		case 1:
1823 			ino = parent_ino(dentry);
1824 			if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1825 				goto out;
1826 			filp->f_pos++;
1827 		default:
1828 			files = get_files_struct(p);
1829 			if (!files)
1830 				goto out;
1831 			rcu_read_lock();
1832 			for (fd = filp->f_pos-2;
1833 			     fd < files_fdtable(files)->max_fds;
1834 			     fd++, filp->f_pos++) {
1835 				char name[PROC_NUMBUF];
1836 				int len;
1837 
1838 				if (!fcheck_files(files, fd))
1839 					continue;
1840 				rcu_read_unlock();
1841 
1842 				len = snprintf(name, sizeof(name), "%d", fd);
1843 				if (proc_fill_cache(filp, dirent, filldir,
1844 						    name, len, instantiate,
1845 						    p, &fd) < 0) {
1846 					rcu_read_lock();
1847 					break;
1848 				}
1849 				rcu_read_lock();
1850 			}
1851 			rcu_read_unlock();
1852 			put_files_struct(files);
1853 	}
1854 out:
1855 	put_task_struct(p);
1856 out_no_task:
1857 	return retval;
1858 }
1859 
1860 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1861 				    struct nameidata *nd)
1862 {
1863 	return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1864 }
1865 
1866 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1867 {
1868 	return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1869 }
1870 
1871 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1872 				      size_t len, loff_t *ppos)
1873 {
1874 	char tmp[PROC_FDINFO_MAX];
1875 	int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1876 	if (!err)
1877 		err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1878 	return err;
1879 }
1880 
1881 static const struct file_operations proc_fdinfo_file_operations = {
1882 	.open		= nonseekable_open,
1883 	.read		= proc_fdinfo_read,
1884 };
1885 
1886 static const struct file_operations proc_fd_operations = {
1887 	.read		= generic_read_dir,
1888 	.readdir	= proc_readfd,
1889 };
1890 
1891 /*
1892  * /proc/pid/fd needs a special permission handler so that a process can still
1893  * access /proc/self/fd after it has executed a setuid().
1894  */
1895 static int proc_fd_permission(struct inode *inode, int mask)
1896 {
1897 	int rv;
1898 
1899 	rv = generic_permission(inode, mask, NULL);
1900 	if (rv == 0)
1901 		return 0;
1902 	if (task_pid(current) == proc_pid(inode))
1903 		rv = 0;
1904 	return rv;
1905 }
1906 
1907 /*
1908  * proc directories can do almost nothing..
1909  */
1910 static const struct inode_operations proc_fd_inode_operations = {
1911 	.lookup		= proc_lookupfd,
1912 	.permission	= proc_fd_permission,
1913 	.setattr	= proc_setattr,
1914 };
1915 
1916 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1917 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1918 {
1919 	unsigned fd = *(unsigned *)ptr;
1920  	struct inode *inode;
1921  	struct proc_inode *ei;
1922 	struct dentry *error = ERR_PTR(-ENOENT);
1923 
1924 	inode = proc_pid_make_inode(dir->i_sb, task);
1925 	if (!inode)
1926 		goto out;
1927 	ei = PROC_I(inode);
1928 	ei->fd = fd;
1929 	inode->i_mode = S_IFREG | S_IRUSR;
1930 	inode->i_fop = &proc_fdinfo_file_operations;
1931 	dentry->d_op = &tid_fd_dentry_operations;
1932 	d_add(dentry, inode);
1933 	/* Close the race of the process dying before we return the dentry */
1934 	if (tid_fd_revalidate(dentry, NULL))
1935 		error = NULL;
1936 
1937  out:
1938 	return error;
1939 }
1940 
1941 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1942 					struct dentry *dentry,
1943 					struct nameidata *nd)
1944 {
1945 	return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1946 }
1947 
1948 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1949 {
1950 	return proc_readfd_common(filp, dirent, filldir,
1951 				  proc_fdinfo_instantiate);
1952 }
1953 
1954 static const struct file_operations proc_fdinfo_operations = {
1955 	.read		= generic_read_dir,
1956 	.readdir	= proc_readfdinfo,
1957 };
1958 
1959 /*
1960  * proc directories can do almost nothing..
1961  */
1962 static const struct inode_operations proc_fdinfo_inode_operations = {
1963 	.lookup		= proc_lookupfdinfo,
1964 	.setattr	= proc_setattr,
1965 };
1966 
1967 
1968 static struct dentry *proc_pident_instantiate(struct inode *dir,
1969 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1970 {
1971 	const struct pid_entry *p = ptr;
1972 	struct inode *inode;
1973 	struct proc_inode *ei;
1974 	struct dentry *error = ERR_PTR(-ENOENT);
1975 
1976 	inode = proc_pid_make_inode(dir->i_sb, task);
1977 	if (!inode)
1978 		goto out;
1979 
1980 	ei = PROC_I(inode);
1981 	inode->i_mode = p->mode;
1982 	if (S_ISDIR(inode->i_mode))
1983 		inode->i_nlink = 2;	/* Use getattr to fix if necessary */
1984 	if (p->iop)
1985 		inode->i_op = p->iop;
1986 	if (p->fop)
1987 		inode->i_fop = p->fop;
1988 	ei->op = p->op;
1989 	dentry->d_op = &pid_dentry_operations;
1990 	d_add(dentry, inode);
1991 	/* Close the race of the process dying before we return the dentry */
1992 	if (pid_revalidate(dentry, NULL))
1993 		error = NULL;
1994 out:
1995 	return error;
1996 }
1997 
1998 static struct dentry *proc_pident_lookup(struct inode *dir,
1999 					 struct dentry *dentry,
2000 					 const struct pid_entry *ents,
2001 					 unsigned int nents)
2002 {
2003 	struct dentry *error;
2004 	struct task_struct *task = get_proc_task(dir);
2005 	const struct pid_entry *p, *last;
2006 
2007 	error = ERR_PTR(-ENOENT);
2008 
2009 	if (!task)
2010 		goto out_no_task;
2011 
2012 	/*
2013 	 * Yes, it does not scale. And it should not. Don't add
2014 	 * new entries into /proc/<tgid>/ without very good reasons.
2015 	 */
2016 	last = &ents[nents - 1];
2017 	for (p = ents; p <= last; p++) {
2018 		if (p->len != dentry->d_name.len)
2019 			continue;
2020 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2021 			break;
2022 	}
2023 	if (p > last)
2024 		goto out;
2025 
2026 	error = proc_pident_instantiate(dir, dentry, task, p);
2027 out:
2028 	put_task_struct(task);
2029 out_no_task:
2030 	return error;
2031 }
2032 
2033 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2034 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2035 {
2036 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2037 				proc_pident_instantiate, task, p);
2038 }
2039 
2040 static int proc_pident_readdir(struct file *filp,
2041 		void *dirent, filldir_t filldir,
2042 		const struct pid_entry *ents, unsigned int nents)
2043 {
2044 	int i;
2045 	struct dentry *dentry = filp->f_path.dentry;
2046 	struct inode *inode = dentry->d_inode;
2047 	struct task_struct *task = get_proc_task(inode);
2048 	const struct pid_entry *p, *last;
2049 	ino_t ino;
2050 	int ret;
2051 
2052 	ret = -ENOENT;
2053 	if (!task)
2054 		goto out_no_task;
2055 
2056 	ret = 0;
2057 	i = filp->f_pos;
2058 	switch (i) {
2059 	case 0:
2060 		ino = inode->i_ino;
2061 		if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2062 			goto out;
2063 		i++;
2064 		filp->f_pos++;
2065 		/* fall through */
2066 	case 1:
2067 		ino = parent_ino(dentry);
2068 		if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2069 			goto out;
2070 		i++;
2071 		filp->f_pos++;
2072 		/* fall through */
2073 	default:
2074 		i -= 2;
2075 		if (i >= nents) {
2076 			ret = 1;
2077 			goto out;
2078 		}
2079 		p = ents + i;
2080 		last = &ents[nents - 1];
2081 		while (p <= last) {
2082 			if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2083 				goto out;
2084 			filp->f_pos++;
2085 			p++;
2086 		}
2087 	}
2088 
2089 	ret = 1;
2090 out:
2091 	put_task_struct(task);
2092 out_no_task:
2093 	return ret;
2094 }
2095 
2096 #ifdef CONFIG_SECURITY
2097 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2098 				  size_t count, loff_t *ppos)
2099 {
2100 	struct inode * inode = file->f_path.dentry->d_inode;
2101 	char *p = NULL;
2102 	ssize_t length;
2103 	struct task_struct *task = get_proc_task(inode);
2104 
2105 	if (!task)
2106 		return -ESRCH;
2107 
2108 	length = security_getprocattr(task,
2109 				      (char*)file->f_path.dentry->d_name.name,
2110 				      &p);
2111 	put_task_struct(task);
2112 	if (length > 0)
2113 		length = simple_read_from_buffer(buf, count, ppos, p, length);
2114 	kfree(p);
2115 	return length;
2116 }
2117 
2118 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2119 				   size_t count, loff_t *ppos)
2120 {
2121 	struct inode * inode = file->f_path.dentry->d_inode;
2122 	char *page;
2123 	ssize_t length;
2124 	struct task_struct *task = get_proc_task(inode);
2125 
2126 	length = -ESRCH;
2127 	if (!task)
2128 		goto out_no_task;
2129 	if (count > PAGE_SIZE)
2130 		count = PAGE_SIZE;
2131 
2132 	/* No partial writes. */
2133 	length = -EINVAL;
2134 	if (*ppos != 0)
2135 		goto out;
2136 
2137 	length = -ENOMEM;
2138 	page = (char*)__get_free_page(GFP_TEMPORARY);
2139 	if (!page)
2140 		goto out;
2141 
2142 	length = -EFAULT;
2143 	if (copy_from_user(page, buf, count))
2144 		goto out_free;
2145 
2146 	/* Guard against adverse ptrace interaction */
2147 	length = mutex_lock_interruptible(&task->cred_guard_mutex);
2148 	if (length < 0)
2149 		goto out_free;
2150 
2151 	length = security_setprocattr(task,
2152 				      (char*)file->f_path.dentry->d_name.name,
2153 				      (void*)page, count);
2154 	mutex_unlock(&task->cred_guard_mutex);
2155 out_free:
2156 	free_page((unsigned long) page);
2157 out:
2158 	put_task_struct(task);
2159 out_no_task:
2160 	return length;
2161 }
2162 
2163 static const struct file_operations proc_pid_attr_operations = {
2164 	.read		= proc_pid_attr_read,
2165 	.write		= proc_pid_attr_write,
2166 };
2167 
2168 static const struct pid_entry attr_dir_stuff[] = {
2169 	REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2170 	REG("prev",       S_IRUGO,	   proc_pid_attr_operations),
2171 	REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2172 	REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2173 	REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2174 	REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2175 };
2176 
2177 static int proc_attr_dir_readdir(struct file * filp,
2178 			     void * dirent, filldir_t filldir)
2179 {
2180 	return proc_pident_readdir(filp,dirent,filldir,
2181 				   attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2182 }
2183 
2184 static const struct file_operations proc_attr_dir_operations = {
2185 	.read		= generic_read_dir,
2186 	.readdir	= proc_attr_dir_readdir,
2187 };
2188 
2189 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2190 				struct dentry *dentry, struct nameidata *nd)
2191 {
2192 	return proc_pident_lookup(dir, dentry,
2193 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2194 }
2195 
2196 static const struct inode_operations proc_attr_dir_inode_operations = {
2197 	.lookup		= proc_attr_dir_lookup,
2198 	.getattr	= pid_getattr,
2199 	.setattr	= proc_setattr,
2200 };
2201 
2202 #endif
2203 
2204 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2205 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2206 					 size_t count, loff_t *ppos)
2207 {
2208 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2209 	struct mm_struct *mm;
2210 	char buffer[PROC_NUMBUF];
2211 	size_t len;
2212 	int ret;
2213 
2214 	if (!task)
2215 		return -ESRCH;
2216 
2217 	ret = 0;
2218 	mm = get_task_mm(task);
2219 	if (mm) {
2220 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2221 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2222 				MMF_DUMP_FILTER_SHIFT));
2223 		mmput(mm);
2224 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2225 	}
2226 
2227 	put_task_struct(task);
2228 
2229 	return ret;
2230 }
2231 
2232 static ssize_t proc_coredump_filter_write(struct file *file,
2233 					  const char __user *buf,
2234 					  size_t count,
2235 					  loff_t *ppos)
2236 {
2237 	struct task_struct *task;
2238 	struct mm_struct *mm;
2239 	char buffer[PROC_NUMBUF], *end;
2240 	unsigned int val;
2241 	int ret;
2242 	int i;
2243 	unsigned long mask;
2244 
2245 	ret = -EFAULT;
2246 	memset(buffer, 0, sizeof(buffer));
2247 	if (count > sizeof(buffer) - 1)
2248 		count = sizeof(buffer) - 1;
2249 	if (copy_from_user(buffer, buf, count))
2250 		goto out_no_task;
2251 
2252 	ret = -EINVAL;
2253 	val = (unsigned int)simple_strtoul(buffer, &end, 0);
2254 	if (*end == '\n')
2255 		end++;
2256 	if (end - buffer == 0)
2257 		goto out_no_task;
2258 
2259 	ret = -ESRCH;
2260 	task = get_proc_task(file->f_dentry->d_inode);
2261 	if (!task)
2262 		goto out_no_task;
2263 
2264 	ret = end - buffer;
2265 	mm = get_task_mm(task);
2266 	if (!mm)
2267 		goto out_no_mm;
2268 
2269 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2270 		if (val & mask)
2271 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2272 		else
2273 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2274 	}
2275 
2276 	mmput(mm);
2277  out_no_mm:
2278 	put_task_struct(task);
2279  out_no_task:
2280 	return ret;
2281 }
2282 
2283 static const struct file_operations proc_coredump_filter_operations = {
2284 	.read		= proc_coredump_filter_read,
2285 	.write		= proc_coredump_filter_write,
2286 };
2287 #endif
2288 
2289 /*
2290  * /proc/self:
2291  */
2292 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2293 			      int buflen)
2294 {
2295 	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2296 	pid_t tgid = task_tgid_nr_ns(current, ns);
2297 	char tmp[PROC_NUMBUF];
2298 	if (!tgid)
2299 		return -ENOENT;
2300 	sprintf(tmp, "%d", tgid);
2301 	return vfs_readlink(dentry,buffer,buflen,tmp);
2302 }
2303 
2304 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2305 {
2306 	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2307 	pid_t tgid = task_tgid_nr_ns(current, ns);
2308 	char tmp[PROC_NUMBUF];
2309 	if (!tgid)
2310 		return ERR_PTR(-ENOENT);
2311 	sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2312 	return ERR_PTR(vfs_follow_link(nd,tmp));
2313 }
2314 
2315 static const struct inode_operations proc_self_inode_operations = {
2316 	.readlink	= proc_self_readlink,
2317 	.follow_link	= proc_self_follow_link,
2318 };
2319 
2320 /*
2321  * proc base
2322  *
2323  * These are the directory entries in the root directory of /proc
2324  * that properly belong to the /proc filesystem, as they describe
2325  * describe something that is process related.
2326  */
2327 static const struct pid_entry proc_base_stuff[] = {
2328 	NOD("self", S_IFLNK|S_IRWXUGO,
2329 		&proc_self_inode_operations, NULL, {}),
2330 };
2331 
2332 /*
2333  *	Exceptional case: normally we are not allowed to unhash a busy
2334  * directory. In this case, however, we can do it - no aliasing problems
2335  * due to the way we treat inodes.
2336  */
2337 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2338 {
2339 	struct inode *inode = dentry->d_inode;
2340 	struct task_struct *task = get_proc_task(inode);
2341 	if (task) {
2342 		put_task_struct(task);
2343 		return 1;
2344 	}
2345 	d_drop(dentry);
2346 	return 0;
2347 }
2348 
2349 static const struct dentry_operations proc_base_dentry_operations =
2350 {
2351 	.d_revalidate	= proc_base_revalidate,
2352 	.d_delete	= pid_delete_dentry,
2353 };
2354 
2355 static struct dentry *proc_base_instantiate(struct inode *dir,
2356 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2357 {
2358 	const struct pid_entry *p = ptr;
2359 	struct inode *inode;
2360 	struct proc_inode *ei;
2361 	struct dentry *error = ERR_PTR(-EINVAL);
2362 
2363 	/* Allocate the inode */
2364 	error = ERR_PTR(-ENOMEM);
2365 	inode = new_inode(dir->i_sb);
2366 	if (!inode)
2367 		goto out;
2368 
2369 	/* Initialize the inode */
2370 	ei = PROC_I(inode);
2371 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2372 
2373 	/*
2374 	 * grab the reference to the task.
2375 	 */
2376 	ei->pid = get_task_pid(task, PIDTYPE_PID);
2377 	if (!ei->pid)
2378 		goto out_iput;
2379 
2380 	inode->i_mode = p->mode;
2381 	if (S_ISDIR(inode->i_mode))
2382 		inode->i_nlink = 2;
2383 	if (S_ISLNK(inode->i_mode))
2384 		inode->i_size = 64;
2385 	if (p->iop)
2386 		inode->i_op = p->iop;
2387 	if (p->fop)
2388 		inode->i_fop = p->fop;
2389 	ei->op = p->op;
2390 	dentry->d_op = &proc_base_dentry_operations;
2391 	d_add(dentry, inode);
2392 	error = NULL;
2393 out:
2394 	return error;
2395 out_iput:
2396 	iput(inode);
2397 	goto out;
2398 }
2399 
2400 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2401 {
2402 	struct dentry *error;
2403 	struct task_struct *task = get_proc_task(dir);
2404 	const struct pid_entry *p, *last;
2405 
2406 	error = ERR_PTR(-ENOENT);
2407 
2408 	if (!task)
2409 		goto out_no_task;
2410 
2411 	/* Lookup the directory entry */
2412 	last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2413 	for (p = proc_base_stuff; p <= last; p++) {
2414 		if (p->len != dentry->d_name.len)
2415 			continue;
2416 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2417 			break;
2418 	}
2419 	if (p > last)
2420 		goto out;
2421 
2422 	error = proc_base_instantiate(dir, dentry, task, p);
2423 
2424 out:
2425 	put_task_struct(task);
2426 out_no_task:
2427 	return error;
2428 }
2429 
2430 static int proc_base_fill_cache(struct file *filp, void *dirent,
2431 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2432 {
2433 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2434 				proc_base_instantiate, task, p);
2435 }
2436 
2437 #ifdef CONFIG_TASK_IO_ACCOUNTING
2438 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2439 {
2440 	struct task_io_accounting acct = task->ioac;
2441 	unsigned long flags;
2442 
2443 	if (whole && lock_task_sighand(task, &flags)) {
2444 		struct task_struct *t = task;
2445 
2446 		task_io_accounting_add(&acct, &task->signal->ioac);
2447 		while_each_thread(task, t)
2448 			task_io_accounting_add(&acct, &t->ioac);
2449 
2450 		unlock_task_sighand(task, &flags);
2451 	}
2452 	return sprintf(buffer,
2453 			"rchar: %llu\n"
2454 			"wchar: %llu\n"
2455 			"syscr: %llu\n"
2456 			"syscw: %llu\n"
2457 			"read_bytes: %llu\n"
2458 			"write_bytes: %llu\n"
2459 			"cancelled_write_bytes: %llu\n",
2460 			(unsigned long long)acct.rchar,
2461 			(unsigned long long)acct.wchar,
2462 			(unsigned long long)acct.syscr,
2463 			(unsigned long long)acct.syscw,
2464 			(unsigned long long)acct.read_bytes,
2465 			(unsigned long long)acct.write_bytes,
2466 			(unsigned long long)acct.cancelled_write_bytes);
2467 }
2468 
2469 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2470 {
2471 	return do_io_accounting(task, buffer, 0);
2472 }
2473 
2474 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2475 {
2476 	return do_io_accounting(task, buffer, 1);
2477 }
2478 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2479 
2480 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2481 				struct pid *pid, struct task_struct *task)
2482 {
2483 	seq_printf(m, "%08x\n", task->personality);
2484 	return 0;
2485 }
2486 
2487 /*
2488  * Thread groups
2489  */
2490 static const struct file_operations proc_task_operations;
2491 static const struct inode_operations proc_task_inode_operations;
2492 
2493 static const struct pid_entry tgid_base_stuff[] = {
2494 	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2495 	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2496 	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2497 #ifdef CONFIG_NET
2498 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2499 #endif
2500 	REG("environ",    S_IRUSR, proc_environ_operations),
2501 	INF("auxv",       S_IRUSR, proc_pid_auxv),
2502 	ONE("status",     S_IRUGO, proc_pid_status),
2503 	ONE("personality", S_IRUSR, proc_pid_personality),
2504 	INF("limits",	  S_IRUSR, proc_pid_limits),
2505 #ifdef CONFIG_SCHED_DEBUG
2506 	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2507 #endif
2508 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2509 	INF("syscall",    S_IRUSR, proc_pid_syscall),
2510 #endif
2511 	INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2512 	ONE("stat",       S_IRUGO, proc_tgid_stat),
2513 	ONE("statm",      S_IRUGO, proc_pid_statm),
2514 	REG("maps",       S_IRUGO, proc_maps_operations),
2515 #ifdef CONFIG_NUMA
2516 	REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2517 #endif
2518 	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2519 	LNK("cwd",        proc_cwd_link),
2520 	LNK("root",       proc_root_link),
2521 	LNK("exe",        proc_exe_link),
2522 	REG("mounts",     S_IRUGO, proc_mounts_operations),
2523 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2524 	REG("mountstats", S_IRUSR, proc_mountstats_operations),
2525 #ifdef CONFIG_PROC_PAGE_MONITOR
2526 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2527 	REG("smaps",      S_IRUGO, proc_smaps_operations),
2528 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2529 #endif
2530 #ifdef CONFIG_SECURITY
2531 	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2532 #endif
2533 #ifdef CONFIG_KALLSYMS
2534 	INF("wchan",      S_IRUGO, proc_pid_wchan),
2535 #endif
2536 #ifdef CONFIG_STACKTRACE
2537 	ONE("stack",      S_IRUSR, proc_pid_stack),
2538 #endif
2539 #ifdef CONFIG_SCHEDSTATS
2540 	INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2541 #endif
2542 #ifdef CONFIG_LATENCYTOP
2543 	REG("latency",  S_IRUGO, proc_lstats_operations),
2544 #endif
2545 #ifdef CONFIG_PROC_PID_CPUSET
2546 	REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2547 #endif
2548 #ifdef CONFIG_CGROUPS
2549 	REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2550 #endif
2551 	INF("oom_score",  S_IRUGO, proc_oom_score),
2552 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2553 #ifdef CONFIG_AUDITSYSCALL
2554 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2555 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2556 #endif
2557 #ifdef CONFIG_FAULT_INJECTION
2558 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2559 #endif
2560 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2561 	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2562 #endif
2563 #ifdef CONFIG_TASK_IO_ACCOUNTING
2564 	INF("io",	S_IRUGO, proc_tgid_io_accounting),
2565 #endif
2566 };
2567 
2568 static int proc_tgid_base_readdir(struct file * filp,
2569 			     void * dirent, filldir_t filldir)
2570 {
2571 	return proc_pident_readdir(filp,dirent,filldir,
2572 				   tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2573 }
2574 
2575 static const struct file_operations proc_tgid_base_operations = {
2576 	.read		= generic_read_dir,
2577 	.readdir	= proc_tgid_base_readdir,
2578 };
2579 
2580 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2581 	return proc_pident_lookup(dir, dentry,
2582 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2583 }
2584 
2585 static const struct inode_operations proc_tgid_base_inode_operations = {
2586 	.lookup		= proc_tgid_base_lookup,
2587 	.getattr	= pid_getattr,
2588 	.setattr	= proc_setattr,
2589 };
2590 
2591 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2592 {
2593 	struct dentry *dentry, *leader, *dir;
2594 	char buf[PROC_NUMBUF];
2595 	struct qstr name;
2596 
2597 	name.name = buf;
2598 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2599 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2600 	if (dentry) {
2601 		if (!(current->flags & PF_EXITING))
2602 			shrink_dcache_parent(dentry);
2603 		d_drop(dentry);
2604 		dput(dentry);
2605 	}
2606 
2607 	if (tgid == 0)
2608 		goto out;
2609 
2610 	name.name = buf;
2611 	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2612 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2613 	if (!leader)
2614 		goto out;
2615 
2616 	name.name = "task";
2617 	name.len = strlen(name.name);
2618 	dir = d_hash_and_lookup(leader, &name);
2619 	if (!dir)
2620 		goto out_put_leader;
2621 
2622 	name.name = buf;
2623 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2624 	dentry = d_hash_and_lookup(dir, &name);
2625 	if (dentry) {
2626 		shrink_dcache_parent(dentry);
2627 		d_drop(dentry);
2628 		dput(dentry);
2629 	}
2630 
2631 	dput(dir);
2632 out_put_leader:
2633 	dput(leader);
2634 out:
2635 	return;
2636 }
2637 
2638 /**
2639  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2640  * @task: task that should be flushed.
2641  *
2642  * When flushing dentries from proc, one needs to flush them from global
2643  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2644  * in. This call is supposed to do all of this job.
2645  *
2646  * Looks in the dcache for
2647  * /proc/@pid
2648  * /proc/@tgid/task/@pid
2649  * if either directory is present flushes it and all of it'ts children
2650  * from the dcache.
2651  *
2652  * It is safe and reasonable to cache /proc entries for a task until
2653  * that task exits.  After that they just clog up the dcache with
2654  * useless entries, possibly causing useful dcache entries to be
2655  * flushed instead.  This routine is proved to flush those useless
2656  * dcache entries at process exit time.
2657  *
2658  * NOTE: This routine is just an optimization so it does not guarantee
2659  *       that no dcache entries will exist at process exit time it
2660  *       just makes it very unlikely that any will persist.
2661  */
2662 
2663 void proc_flush_task(struct task_struct *task)
2664 {
2665 	int i;
2666 	struct pid *pid, *tgid = NULL;
2667 	struct upid *upid;
2668 
2669 	pid = task_pid(task);
2670 	if (thread_group_leader(task))
2671 		tgid = task_tgid(task);
2672 
2673 	for (i = 0; i <= pid->level; i++) {
2674 		upid = &pid->numbers[i];
2675 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2676 			tgid ? tgid->numbers[i].nr : 0);
2677 	}
2678 
2679 	upid = &pid->numbers[pid->level];
2680 	if (upid->nr == 1)
2681 		pid_ns_release_proc(upid->ns);
2682 }
2683 
2684 static struct dentry *proc_pid_instantiate(struct inode *dir,
2685 					   struct dentry * dentry,
2686 					   struct task_struct *task, const void *ptr)
2687 {
2688 	struct dentry *error = ERR_PTR(-ENOENT);
2689 	struct inode *inode;
2690 
2691 	inode = proc_pid_make_inode(dir->i_sb, task);
2692 	if (!inode)
2693 		goto out;
2694 
2695 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2696 	inode->i_op = &proc_tgid_base_inode_operations;
2697 	inode->i_fop = &proc_tgid_base_operations;
2698 	inode->i_flags|=S_IMMUTABLE;
2699 
2700 	inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2701 		ARRAY_SIZE(tgid_base_stuff));
2702 
2703 	dentry->d_op = &pid_dentry_operations;
2704 
2705 	d_add(dentry, inode);
2706 	/* Close the race of the process dying before we return the dentry */
2707 	if (pid_revalidate(dentry, NULL))
2708 		error = NULL;
2709 out:
2710 	return error;
2711 }
2712 
2713 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2714 {
2715 	struct dentry *result = ERR_PTR(-ENOENT);
2716 	struct task_struct *task;
2717 	unsigned tgid;
2718 	struct pid_namespace *ns;
2719 
2720 	result = proc_base_lookup(dir, dentry);
2721 	if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2722 		goto out;
2723 
2724 	tgid = name_to_int(dentry);
2725 	if (tgid == ~0U)
2726 		goto out;
2727 
2728 	ns = dentry->d_sb->s_fs_info;
2729 	rcu_read_lock();
2730 	task = find_task_by_pid_ns(tgid, ns);
2731 	if (task)
2732 		get_task_struct(task);
2733 	rcu_read_unlock();
2734 	if (!task)
2735 		goto out;
2736 
2737 	result = proc_pid_instantiate(dir, dentry, task, NULL);
2738 	put_task_struct(task);
2739 out:
2740 	return result;
2741 }
2742 
2743 /*
2744  * Find the first task with tgid >= tgid
2745  *
2746  */
2747 struct tgid_iter {
2748 	unsigned int tgid;
2749 	struct task_struct *task;
2750 };
2751 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2752 {
2753 	struct pid *pid;
2754 
2755 	if (iter.task)
2756 		put_task_struct(iter.task);
2757 	rcu_read_lock();
2758 retry:
2759 	iter.task = NULL;
2760 	pid = find_ge_pid(iter.tgid, ns);
2761 	if (pid) {
2762 		iter.tgid = pid_nr_ns(pid, ns);
2763 		iter.task = pid_task(pid, PIDTYPE_PID);
2764 		/* What we to know is if the pid we have find is the
2765 		 * pid of a thread_group_leader.  Testing for task
2766 		 * being a thread_group_leader is the obvious thing
2767 		 * todo but there is a window when it fails, due to
2768 		 * the pid transfer logic in de_thread.
2769 		 *
2770 		 * So we perform the straight forward test of seeing
2771 		 * if the pid we have found is the pid of a thread
2772 		 * group leader, and don't worry if the task we have
2773 		 * found doesn't happen to be a thread group leader.
2774 		 * As we don't care in the case of readdir.
2775 		 */
2776 		if (!iter.task || !has_group_leader_pid(iter.task)) {
2777 			iter.tgid += 1;
2778 			goto retry;
2779 		}
2780 		get_task_struct(iter.task);
2781 	}
2782 	rcu_read_unlock();
2783 	return iter;
2784 }
2785 
2786 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2787 
2788 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2789 	struct tgid_iter iter)
2790 {
2791 	char name[PROC_NUMBUF];
2792 	int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2793 	return proc_fill_cache(filp, dirent, filldir, name, len,
2794 				proc_pid_instantiate, iter.task, NULL);
2795 }
2796 
2797 /* for the /proc/ directory itself, after non-process stuff has been done */
2798 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2799 {
2800 	unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2801 	struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2802 	struct tgid_iter iter;
2803 	struct pid_namespace *ns;
2804 
2805 	if (!reaper)
2806 		goto out_no_task;
2807 
2808 	for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2809 		const struct pid_entry *p = &proc_base_stuff[nr];
2810 		if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2811 			goto out;
2812 	}
2813 
2814 	ns = filp->f_dentry->d_sb->s_fs_info;
2815 	iter.task = NULL;
2816 	iter.tgid = filp->f_pos - TGID_OFFSET;
2817 	for (iter = next_tgid(ns, iter);
2818 	     iter.task;
2819 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
2820 		filp->f_pos = iter.tgid + TGID_OFFSET;
2821 		if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2822 			put_task_struct(iter.task);
2823 			goto out;
2824 		}
2825 	}
2826 	filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2827 out:
2828 	put_task_struct(reaper);
2829 out_no_task:
2830 	return 0;
2831 }
2832 
2833 /*
2834  * Tasks
2835  */
2836 static const struct pid_entry tid_base_stuff[] = {
2837 	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2838 	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2839 	REG("environ",   S_IRUSR, proc_environ_operations),
2840 	INF("auxv",      S_IRUSR, proc_pid_auxv),
2841 	ONE("status",    S_IRUGO, proc_pid_status),
2842 	ONE("personality", S_IRUSR, proc_pid_personality),
2843 	INF("limits",	 S_IRUSR, proc_pid_limits),
2844 #ifdef CONFIG_SCHED_DEBUG
2845 	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2846 #endif
2847 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2848 	INF("syscall",   S_IRUSR, proc_pid_syscall),
2849 #endif
2850 	INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2851 	ONE("stat",      S_IRUGO, proc_tid_stat),
2852 	ONE("statm",     S_IRUGO, proc_pid_statm),
2853 	REG("maps",      S_IRUGO, proc_maps_operations),
2854 #ifdef CONFIG_NUMA
2855 	REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2856 #endif
2857 	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2858 	LNK("cwd",       proc_cwd_link),
2859 	LNK("root",      proc_root_link),
2860 	LNK("exe",       proc_exe_link),
2861 	REG("mounts",    S_IRUGO, proc_mounts_operations),
2862 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2863 #ifdef CONFIG_PROC_PAGE_MONITOR
2864 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2865 	REG("smaps",     S_IRUGO, proc_smaps_operations),
2866 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2867 #endif
2868 #ifdef CONFIG_SECURITY
2869 	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2870 #endif
2871 #ifdef CONFIG_KALLSYMS
2872 	INF("wchan",     S_IRUGO, proc_pid_wchan),
2873 #endif
2874 #ifdef CONFIG_STACKTRACE
2875 	ONE("stack",      S_IRUSR, proc_pid_stack),
2876 #endif
2877 #ifdef CONFIG_SCHEDSTATS
2878 	INF("schedstat", S_IRUGO, proc_pid_schedstat),
2879 #endif
2880 #ifdef CONFIG_LATENCYTOP
2881 	REG("latency",  S_IRUGO, proc_lstats_operations),
2882 #endif
2883 #ifdef CONFIG_PROC_PID_CPUSET
2884 	REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2885 #endif
2886 #ifdef CONFIG_CGROUPS
2887 	REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2888 #endif
2889 	INF("oom_score", S_IRUGO, proc_oom_score),
2890 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2891 #ifdef CONFIG_AUDITSYSCALL
2892 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2893 	REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2894 #endif
2895 #ifdef CONFIG_FAULT_INJECTION
2896 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2897 #endif
2898 #ifdef CONFIG_TASK_IO_ACCOUNTING
2899 	INF("io",	S_IRUGO, proc_tid_io_accounting),
2900 #endif
2901 };
2902 
2903 static int proc_tid_base_readdir(struct file * filp,
2904 			     void * dirent, filldir_t filldir)
2905 {
2906 	return proc_pident_readdir(filp,dirent,filldir,
2907 				   tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2908 }
2909 
2910 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2911 	return proc_pident_lookup(dir, dentry,
2912 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2913 }
2914 
2915 static const struct file_operations proc_tid_base_operations = {
2916 	.read		= generic_read_dir,
2917 	.readdir	= proc_tid_base_readdir,
2918 };
2919 
2920 static const struct inode_operations proc_tid_base_inode_operations = {
2921 	.lookup		= proc_tid_base_lookup,
2922 	.getattr	= pid_getattr,
2923 	.setattr	= proc_setattr,
2924 };
2925 
2926 static struct dentry *proc_task_instantiate(struct inode *dir,
2927 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2928 {
2929 	struct dentry *error = ERR_PTR(-ENOENT);
2930 	struct inode *inode;
2931 	inode = proc_pid_make_inode(dir->i_sb, task);
2932 
2933 	if (!inode)
2934 		goto out;
2935 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2936 	inode->i_op = &proc_tid_base_inode_operations;
2937 	inode->i_fop = &proc_tid_base_operations;
2938 	inode->i_flags|=S_IMMUTABLE;
2939 
2940 	inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2941 		ARRAY_SIZE(tid_base_stuff));
2942 
2943 	dentry->d_op = &pid_dentry_operations;
2944 
2945 	d_add(dentry, inode);
2946 	/* Close the race of the process dying before we return the dentry */
2947 	if (pid_revalidate(dentry, NULL))
2948 		error = NULL;
2949 out:
2950 	return error;
2951 }
2952 
2953 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2954 {
2955 	struct dentry *result = ERR_PTR(-ENOENT);
2956 	struct task_struct *task;
2957 	struct task_struct *leader = get_proc_task(dir);
2958 	unsigned tid;
2959 	struct pid_namespace *ns;
2960 
2961 	if (!leader)
2962 		goto out_no_task;
2963 
2964 	tid = name_to_int(dentry);
2965 	if (tid == ~0U)
2966 		goto out;
2967 
2968 	ns = dentry->d_sb->s_fs_info;
2969 	rcu_read_lock();
2970 	task = find_task_by_pid_ns(tid, ns);
2971 	if (task)
2972 		get_task_struct(task);
2973 	rcu_read_unlock();
2974 	if (!task)
2975 		goto out;
2976 	if (!same_thread_group(leader, task))
2977 		goto out_drop_task;
2978 
2979 	result = proc_task_instantiate(dir, dentry, task, NULL);
2980 out_drop_task:
2981 	put_task_struct(task);
2982 out:
2983 	put_task_struct(leader);
2984 out_no_task:
2985 	return result;
2986 }
2987 
2988 /*
2989  * Find the first tid of a thread group to return to user space.
2990  *
2991  * Usually this is just the thread group leader, but if the users
2992  * buffer was too small or there was a seek into the middle of the
2993  * directory we have more work todo.
2994  *
2995  * In the case of a short read we start with find_task_by_pid.
2996  *
2997  * In the case of a seek we start with the leader and walk nr
2998  * threads past it.
2999  */
3000 static struct task_struct *first_tid(struct task_struct *leader,
3001 		int tid, int nr, struct pid_namespace *ns)
3002 {
3003 	struct task_struct *pos;
3004 
3005 	rcu_read_lock();
3006 	/* Attempt to start with the pid of a thread */
3007 	if (tid && (nr > 0)) {
3008 		pos = find_task_by_pid_ns(tid, ns);
3009 		if (pos && (pos->group_leader == leader))
3010 			goto found;
3011 	}
3012 
3013 	/* If nr exceeds the number of threads there is nothing todo */
3014 	pos = NULL;
3015 	if (nr && nr >= get_nr_threads(leader))
3016 		goto out;
3017 
3018 	/* If we haven't found our starting place yet start
3019 	 * with the leader and walk nr threads forward.
3020 	 */
3021 	for (pos = leader; nr > 0; --nr) {
3022 		pos = next_thread(pos);
3023 		if (pos == leader) {
3024 			pos = NULL;
3025 			goto out;
3026 		}
3027 	}
3028 found:
3029 	get_task_struct(pos);
3030 out:
3031 	rcu_read_unlock();
3032 	return pos;
3033 }
3034 
3035 /*
3036  * Find the next thread in the thread list.
3037  * Return NULL if there is an error or no next thread.
3038  *
3039  * The reference to the input task_struct is released.
3040  */
3041 static struct task_struct *next_tid(struct task_struct *start)
3042 {
3043 	struct task_struct *pos = NULL;
3044 	rcu_read_lock();
3045 	if (pid_alive(start)) {
3046 		pos = next_thread(start);
3047 		if (thread_group_leader(pos))
3048 			pos = NULL;
3049 		else
3050 			get_task_struct(pos);
3051 	}
3052 	rcu_read_unlock();
3053 	put_task_struct(start);
3054 	return pos;
3055 }
3056 
3057 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3058 	struct task_struct *task, int tid)
3059 {
3060 	char name[PROC_NUMBUF];
3061 	int len = snprintf(name, sizeof(name), "%d", tid);
3062 	return proc_fill_cache(filp, dirent, filldir, name, len,
3063 				proc_task_instantiate, task, NULL);
3064 }
3065 
3066 /* for the /proc/TGID/task/ directories */
3067 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3068 {
3069 	struct dentry *dentry = filp->f_path.dentry;
3070 	struct inode *inode = dentry->d_inode;
3071 	struct task_struct *leader = NULL;
3072 	struct task_struct *task;
3073 	int retval = -ENOENT;
3074 	ino_t ino;
3075 	int tid;
3076 	struct pid_namespace *ns;
3077 
3078 	task = get_proc_task(inode);
3079 	if (!task)
3080 		goto out_no_task;
3081 	rcu_read_lock();
3082 	if (pid_alive(task)) {
3083 		leader = task->group_leader;
3084 		get_task_struct(leader);
3085 	}
3086 	rcu_read_unlock();
3087 	put_task_struct(task);
3088 	if (!leader)
3089 		goto out_no_task;
3090 	retval = 0;
3091 
3092 	switch ((unsigned long)filp->f_pos) {
3093 	case 0:
3094 		ino = inode->i_ino;
3095 		if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3096 			goto out;
3097 		filp->f_pos++;
3098 		/* fall through */
3099 	case 1:
3100 		ino = parent_ino(dentry);
3101 		if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3102 			goto out;
3103 		filp->f_pos++;
3104 		/* fall through */
3105 	}
3106 
3107 	/* f_version caches the tgid value that the last readdir call couldn't
3108 	 * return. lseek aka telldir automagically resets f_version to 0.
3109 	 */
3110 	ns = filp->f_dentry->d_sb->s_fs_info;
3111 	tid = (int)filp->f_version;
3112 	filp->f_version = 0;
3113 	for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3114 	     task;
3115 	     task = next_tid(task), filp->f_pos++) {
3116 		tid = task_pid_nr_ns(task, ns);
3117 		if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3118 			/* returning this tgid failed, save it as the first
3119 			 * pid for the next readir call */
3120 			filp->f_version = (u64)tid;
3121 			put_task_struct(task);
3122 			break;
3123 		}
3124 	}
3125 out:
3126 	put_task_struct(leader);
3127 out_no_task:
3128 	return retval;
3129 }
3130 
3131 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3132 {
3133 	struct inode *inode = dentry->d_inode;
3134 	struct task_struct *p = get_proc_task(inode);
3135 	generic_fillattr(inode, stat);
3136 
3137 	if (p) {
3138 		stat->nlink += get_nr_threads(p);
3139 		put_task_struct(p);
3140 	}
3141 
3142 	return 0;
3143 }
3144 
3145 static const struct inode_operations proc_task_inode_operations = {
3146 	.lookup		= proc_task_lookup,
3147 	.getattr	= proc_task_getattr,
3148 	.setattr	= proc_setattr,
3149 };
3150 
3151 static const struct file_operations proc_task_operations = {
3152 	.read		= generic_read_dir,
3153 	.readdir	= proc_task_readdir,
3154 };
3155