xref: /openbmc/linux/fs/proc/base.c (revision 2ec220e27f5040aec1e88901c1b6ea3d135787ad)
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 "internal.h"
84 
85 /* NOTE:
86  *	Implementing inode permission operations in /proc is almost
87  *	certainly an error.  Permission checks need to happen during
88  *	each system call not at open time.  The reason is that most of
89  *	what we wish to check for permissions in /proc varies at runtime.
90  *
91  *	The classic example of a problem is opening file descriptors
92  *	in /proc for a task before it execs a suid executable.
93  */
94 
95 struct pid_entry {
96 	char *name;
97 	int len;
98 	mode_t mode;
99 	const struct inode_operations *iop;
100 	const struct file_operations *fop;
101 	union proc_op op;
102 };
103 
104 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
105 	.name = (NAME),					\
106 	.len  = sizeof(NAME) - 1,			\
107 	.mode = MODE,					\
108 	.iop  = IOP,					\
109 	.fop  = FOP,					\
110 	.op   = OP,					\
111 }
112 
113 #define DIR(NAME, MODE, iops, fops)	\
114 	NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
115 #define LNK(NAME, get_link)					\
116 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
117 		&proc_pid_link_inode_operations, NULL,		\
118 		{ .proc_get_link = get_link } )
119 #define REG(NAME, MODE, fops)				\
120 	NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
121 #define INF(NAME, MODE, read)				\
122 	NOD(NAME, (S_IFREG|(MODE)), 			\
123 		NULL, &proc_info_file_operations,	\
124 		{ .proc_read = read } )
125 #define ONE(NAME, MODE, show)				\
126 	NOD(NAME, (S_IFREG|(MODE)), 			\
127 		NULL, &proc_single_file_operations,	\
128 		{ .proc_show = show } )
129 
130 /*
131  * Count the number of hardlinks for the pid_entry table, excluding the .
132  * and .. links.
133  */
134 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
135 	unsigned int n)
136 {
137 	unsigned int i;
138 	unsigned int count;
139 
140 	count = 0;
141 	for (i = 0; i < n; ++i) {
142 		if (S_ISDIR(entries[i].mode))
143 			++count;
144 	}
145 
146 	return count;
147 }
148 
149 static struct fs_struct *get_fs_struct(struct task_struct *task)
150 {
151 	struct fs_struct *fs;
152 	task_lock(task);
153 	fs = task->fs;
154 	if(fs)
155 		atomic_inc(&fs->count);
156 	task_unlock(task);
157 	return fs;
158 }
159 
160 static int get_nr_threads(struct task_struct *tsk)
161 {
162 	unsigned long flags;
163 	int count = 0;
164 
165 	if (lock_task_sighand(tsk, &flags)) {
166 		count = atomic_read(&tsk->signal->count);
167 		unlock_task_sighand(tsk, &flags);
168 	}
169 	return count;
170 }
171 
172 static int proc_cwd_link(struct inode *inode, struct path *path)
173 {
174 	struct task_struct *task = get_proc_task(inode);
175 	struct fs_struct *fs = NULL;
176 	int result = -ENOENT;
177 
178 	if (task) {
179 		fs = get_fs_struct(task);
180 		put_task_struct(task);
181 	}
182 	if (fs) {
183 		read_lock(&fs->lock);
184 		*path = fs->pwd;
185 		path_get(&fs->pwd);
186 		read_unlock(&fs->lock);
187 		result = 0;
188 		put_fs_struct(fs);
189 	}
190 	return result;
191 }
192 
193 static int proc_root_link(struct inode *inode, struct path *path)
194 {
195 	struct task_struct *task = get_proc_task(inode);
196 	struct fs_struct *fs = NULL;
197 	int result = -ENOENT;
198 
199 	if (task) {
200 		fs = get_fs_struct(task);
201 		put_task_struct(task);
202 	}
203 	if (fs) {
204 		read_lock(&fs->lock);
205 		*path = fs->root;
206 		path_get(&fs->root);
207 		read_unlock(&fs->lock);
208 		result = 0;
209 		put_fs_struct(fs);
210 	}
211 	return result;
212 }
213 
214 /*
215  * Return zero if current may access user memory in @task, -error if not.
216  */
217 static int check_mem_permission(struct task_struct *task)
218 {
219 	/*
220 	 * A task can always look at itself, in case it chooses
221 	 * to use system calls instead of load instructions.
222 	 */
223 	if (task == current)
224 		return 0;
225 
226 	/*
227 	 * If current is actively ptrace'ing, and would also be
228 	 * permitted to freshly attach with ptrace now, permit it.
229 	 */
230 	if (task_is_stopped_or_traced(task)) {
231 		int match;
232 		rcu_read_lock();
233 		match = (tracehook_tracer_task(task) == current);
234 		rcu_read_unlock();
235 		if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
236 			return 0;
237 	}
238 
239 	/*
240 	 * Noone else is allowed.
241 	 */
242 	return -EPERM;
243 }
244 
245 struct mm_struct *mm_for_maps(struct task_struct *task)
246 {
247 	struct mm_struct *mm = get_task_mm(task);
248 	if (!mm)
249 		return NULL;
250 	down_read(&mm->mmap_sem);
251 	task_lock(task);
252 	if (task->mm != mm)
253 		goto out;
254 	if (task->mm != current->mm &&
255 	    __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
256 		goto out;
257 	task_unlock(task);
258 	return mm;
259 out:
260 	task_unlock(task);
261 	up_read(&mm->mmap_sem);
262 	mmput(mm);
263 	return NULL;
264 }
265 
266 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
267 {
268 	int res = 0;
269 	unsigned int len;
270 	struct mm_struct *mm = get_task_mm(task);
271 	if (!mm)
272 		goto out;
273 	if (!mm->arg_end)
274 		goto out_mm;	/* Shh! No looking before we're done */
275 
276  	len = mm->arg_end - mm->arg_start;
277 
278 	if (len > PAGE_SIZE)
279 		len = PAGE_SIZE;
280 
281 	res = access_process_vm(task, mm->arg_start, buffer, len, 0);
282 
283 	// If the nul at the end of args has been overwritten, then
284 	// assume application is using setproctitle(3).
285 	if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
286 		len = strnlen(buffer, res);
287 		if (len < res) {
288 		    res = len;
289 		} else {
290 			len = mm->env_end - mm->env_start;
291 			if (len > PAGE_SIZE - res)
292 				len = PAGE_SIZE - res;
293 			res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
294 			res = strnlen(buffer, res);
295 		}
296 	}
297 out_mm:
298 	mmput(mm);
299 out:
300 	return res;
301 }
302 
303 static int proc_pid_auxv(struct task_struct *task, char *buffer)
304 {
305 	int res = 0;
306 	struct mm_struct *mm = get_task_mm(task);
307 	if (mm) {
308 		unsigned int nwords = 0;
309 		do
310 			nwords += 2;
311 		while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
312 		res = nwords * sizeof(mm->saved_auxv[0]);
313 		if (res > PAGE_SIZE)
314 			res = PAGE_SIZE;
315 		memcpy(buffer, mm->saved_auxv, res);
316 		mmput(mm);
317 	}
318 	return res;
319 }
320 
321 
322 #ifdef CONFIG_KALLSYMS
323 /*
324  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
325  * Returns the resolved symbol.  If that fails, simply return the address.
326  */
327 static int proc_pid_wchan(struct task_struct *task, char *buffer)
328 {
329 	unsigned long wchan;
330 	char symname[KSYM_NAME_LEN];
331 
332 	wchan = get_wchan(task);
333 
334 	if (lookup_symbol_name(wchan, symname) < 0)
335 		return sprintf(buffer, "%lu", wchan);
336 	else
337 		return sprintf(buffer, "%s", symname);
338 }
339 #endif /* CONFIG_KALLSYMS */
340 
341 #ifdef CONFIG_STACKTRACE
342 
343 #define MAX_STACK_TRACE_DEPTH	64
344 
345 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
346 			  struct pid *pid, struct task_struct *task)
347 {
348 	struct stack_trace trace;
349 	unsigned long *entries;
350 	int i;
351 
352 	entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
353 	if (!entries)
354 		return -ENOMEM;
355 
356 	trace.nr_entries	= 0;
357 	trace.max_entries	= MAX_STACK_TRACE_DEPTH;
358 	trace.entries		= entries;
359 	trace.skip		= 0;
360 	save_stack_trace_tsk(task, &trace);
361 
362 	for (i = 0; i < trace.nr_entries; i++) {
363 		seq_printf(m, "[<%p>] %pS\n",
364 			   (void *)entries[i], (void *)entries[i]);
365 	}
366 	kfree(entries);
367 
368 	return 0;
369 }
370 #endif
371 
372 #ifdef CONFIG_SCHEDSTATS
373 /*
374  * Provides /proc/PID/schedstat
375  */
376 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
377 {
378 	return sprintf(buffer, "%llu %llu %lu\n",
379 			(unsigned long long)task->se.sum_exec_runtime,
380 			(unsigned long long)task->sched_info.run_delay,
381 			task->sched_info.pcount);
382 }
383 #endif
384 
385 #ifdef CONFIG_LATENCYTOP
386 static int lstats_show_proc(struct seq_file *m, void *v)
387 {
388 	int i;
389 	struct inode *inode = m->private;
390 	struct task_struct *task = get_proc_task(inode);
391 
392 	if (!task)
393 		return -ESRCH;
394 	seq_puts(m, "Latency Top version : v0.1\n");
395 	for (i = 0; i < 32; i++) {
396 		if (task->latency_record[i].backtrace[0]) {
397 			int q;
398 			seq_printf(m, "%i %li %li ",
399 				task->latency_record[i].count,
400 				task->latency_record[i].time,
401 				task->latency_record[i].max);
402 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
403 				char sym[KSYM_SYMBOL_LEN];
404 				char *c;
405 				if (!task->latency_record[i].backtrace[q])
406 					break;
407 				if (task->latency_record[i].backtrace[q] == ULONG_MAX)
408 					break;
409 				sprint_symbol(sym, task->latency_record[i].backtrace[q]);
410 				c = strchr(sym, '+');
411 				if (c)
412 					*c = 0;
413 				seq_printf(m, "%s ", sym);
414 			}
415 			seq_printf(m, "\n");
416 		}
417 
418 	}
419 	put_task_struct(task);
420 	return 0;
421 }
422 
423 static int lstats_open(struct inode *inode, struct file *file)
424 {
425 	return single_open(file, lstats_show_proc, inode);
426 }
427 
428 static ssize_t lstats_write(struct file *file, const char __user *buf,
429 			    size_t count, loff_t *offs)
430 {
431 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
432 
433 	if (!task)
434 		return -ESRCH;
435 	clear_all_latency_tracing(task);
436 	put_task_struct(task);
437 
438 	return count;
439 }
440 
441 static const struct file_operations proc_lstats_operations = {
442 	.open		= lstats_open,
443 	.read		= seq_read,
444 	.write		= lstats_write,
445 	.llseek		= seq_lseek,
446 	.release	= single_release,
447 };
448 
449 #endif
450 
451 /* The badness from the OOM killer */
452 unsigned long badness(struct task_struct *p, unsigned long uptime);
453 static int proc_oom_score(struct task_struct *task, char *buffer)
454 {
455 	unsigned long points;
456 	struct timespec uptime;
457 
458 	do_posix_clock_monotonic_gettime(&uptime);
459 	read_lock(&tasklist_lock);
460 	points = badness(task, uptime.tv_sec);
461 	read_unlock(&tasklist_lock);
462 	return sprintf(buffer, "%lu\n", points);
463 }
464 
465 struct limit_names {
466 	char *name;
467 	char *unit;
468 };
469 
470 static const struct limit_names lnames[RLIM_NLIMITS] = {
471 	[RLIMIT_CPU] = {"Max cpu time", "ms"},
472 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
473 	[RLIMIT_DATA] = {"Max data size", "bytes"},
474 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
475 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
476 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
477 	[RLIMIT_NPROC] = {"Max processes", "processes"},
478 	[RLIMIT_NOFILE] = {"Max open files", "files"},
479 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
480 	[RLIMIT_AS] = {"Max address space", "bytes"},
481 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
482 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
483 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
484 	[RLIMIT_NICE] = {"Max nice priority", NULL},
485 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
486 	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
487 };
488 
489 /* Display limits for a process */
490 static int proc_pid_limits(struct task_struct *task, char *buffer)
491 {
492 	unsigned int i;
493 	int count = 0;
494 	unsigned long flags;
495 	char *bufptr = buffer;
496 
497 	struct rlimit rlim[RLIM_NLIMITS];
498 
499 	if (!lock_task_sighand(task, &flags))
500 		return 0;
501 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
502 	unlock_task_sighand(task, &flags);
503 
504 	/*
505 	 * print the file header
506 	 */
507 	count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
508 			"Limit", "Soft Limit", "Hard Limit", "Units");
509 
510 	for (i = 0; i < RLIM_NLIMITS; i++) {
511 		if (rlim[i].rlim_cur == RLIM_INFINITY)
512 			count += sprintf(&bufptr[count], "%-25s %-20s ",
513 					 lnames[i].name, "unlimited");
514 		else
515 			count += sprintf(&bufptr[count], "%-25s %-20lu ",
516 					 lnames[i].name, rlim[i].rlim_cur);
517 
518 		if (rlim[i].rlim_max == RLIM_INFINITY)
519 			count += sprintf(&bufptr[count], "%-20s ", "unlimited");
520 		else
521 			count += sprintf(&bufptr[count], "%-20lu ",
522 					 rlim[i].rlim_max);
523 
524 		if (lnames[i].unit)
525 			count += sprintf(&bufptr[count], "%-10s\n",
526 					 lnames[i].unit);
527 		else
528 			count += sprintf(&bufptr[count], "\n");
529 	}
530 
531 	return count;
532 }
533 
534 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
535 static int proc_pid_syscall(struct task_struct *task, char *buffer)
536 {
537 	long nr;
538 	unsigned long args[6], sp, pc;
539 
540 	if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
541 		return sprintf(buffer, "running\n");
542 
543 	if (nr < 0)
544 		return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
545 
546 	return sprintf(buffer,
547 		       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
548 		       nr,
549 		       args[0], args[1], args[2], args[3], args[4], args[5],
550 		       sp, pc);
551 }
552 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
553 
554 /************************************************************************/
555 /*                       Here the fs part begins                        */
556 /************************************************************************/
557 
558 /* permission checks */
559 static int proc_fd_access_allowed(struct inode *inode)
560 {
561 	struct task_struct *task;
562 	int allowed = 0;
563 	/* Allow access to a task's file descriptors if it is us or we
564 	 * may use ptrace attach to the process and find out that
565 	 * information.
566 	 */
567 	task = get_proc_task(inode);
568 	if (task) {
569 		allowed = ptrace_may_access(task, PTRACE_MODE_READ);
570 		put_task_struct(task);
571 	}
572 	return allowed;
573 }
574 
575 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
576 {
577 	int error;
578 	struct inode *inode = dentry->d_inode;
579 
580 	if (attr->ia_valid & ATTR_MODE)
581 		return -EPERM;
582 
583 	error = inode_change_ok(inode, attr);
584 	if (!error)
585 		error = inode_setattr(inode, attr);
586 	return error;
587 }
588 
589 static const struct inode_operations proc_def_inode_operations = {
590 	.setattr	= proc_setattr,
591 };
592 
593 static int mounts_open_common(struct inode *inode, struct file *file,
594 			      const struct seq_operations *op)
595 {
596 	struct task_struct *task = get_proc_task(inode);
597 	struct nsproxy *nsp;
598 	struct mnt_namespace *ns = NULL;
599 	struct fs_struct *fs = NULL;
600 	struct path root;
601 	struct proc_mounts *p;
602 	int ret = -EINVAL;
603 
604 	if (task) {
605 		rcu_read_lock();
606 		nsp = task_nsproxy(task);
607 		if (nsp) {
608 			ns = nsp->mnt_ns;
609 			if (ns)
610 				get_mnt_ns(ns);
611 		}
612 		rcu_read_unlock();
613 		if (ns)
614 			fs = get_fs_struct(task);
615 		put_task_struct(task);
616 	}
617 
618 	if (!ns)
619 		goto err;
620 	if (!fs)
621 		goto err_put_ns;
622 
623 	read_lock(&fs->lock);
624 	root = fs->root;
625 	path_get(&root);
626 	read_unlock(&fs->lock);
627 	put_fs_struct(fs);
628 
629 	ret = -ENOMEM;
630 	p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
631 	if (!p)
632 		goto err_put_path;
633 
634 	file->private_data = &p->m;
635 	ret = seq_open(file, op);
636 	if (ret)
637 		goto err_free;
638 
639 	p->m.private = p;
640 	p->ns = ns;
641 	p->root = root;
642 	p->event = ns->event;
643 
644 	return 0;
645 
646  err_free:
647 	kfree(p);
648  err_put_path:
649 	path_put(&root);
650  err_put_ns:
651 	put_mnt_ns(ns);
652  err:
653 	return ret;
654 }
655 
656 static int mounts_release(struct inode *inode, struct file *file)
657 {
658 	struct proc_mounts *p = file->private_data;
659 	path_put(&p->root);
660 	put_mnt_ns(p->ns);
661 	return seq_release(inode, file);
662 }
663 
664 static unsigned mounts_poll(struct file *file, poll_table *wait)
665 {
666 	struct proc_mounts *p = file->private_data;
667 	struct mnt_namespace *ns = p->ns;
668 	unsigned res = 0;
669 
670 	poll_wait(file, &ns->poll, wait);
671 
672 	spin_lock(&vfsmount_lock);
673 	if (p->event != ns->event) {
674 		p->event = ns->event;
675 		res = POLLERR;
676 	}
677 	spin_unlock(&vfsmount_lock);
678 
679 	return res;
680 }
681 
682 static int mounts_open(struct inode *inode, struct file *file)
683 {
684 	return mounts_open_common(inode, file, &mounts_op);
685 }
686 
687 static const struct file_operations proc_mounts_operations = {
688 	.open		= mounts_open,
689 	.read		= seq_read,
690 	.llseek		= seq_lseek,
691 	.release	= mounts_release,
692 	.poll		= mounts_poll,
693 };
694 
695 static int mountinfo_open(struct inode *inode, struct file *file)
696 {
697 	return mounts_open_common(inode, file, &mountinfo_op);
698 }
699 
700 static const struct file_operations proc_mountinfo_operations = {
701 	.open		= mountinfo_open,
702 	.read		= seq_read,
703 	.llseek		= seq_lseek,
704 	.release	= mounts_release,
705 	.poll		= mounts_poll,
706 };
707 
708 static int mountstats_open(struct inode *inode, struct file *file)
709 {
710 	return mounts_open_common(inode, file, &mountstats_op);
711 }
712 
713 static const struct file_operations proc_mountstats_operations = {
714 	.open		= mountstats_open,
715 	.read		= seq_read,
716 	.llseek		= seq_lseek,
717 	.release	= mounts_release,
718 };
719 
720 #define PROC_BLOCK_SIZE	(3*1024)		/* 4K page size but our output routines use some slack for overruns */
721 
722 static ssize_t proc_info_read(struct file * file, char __user * buf,
723 			  size_t count, loff_t *ppos)
724 {
725 	struct inode * inode = file->f_path.dentry->d_inode;
726 	unsigned long page;
727 	ssize_t length;
728 	struct task_struct *task = get_proc_task(inode);
729 
730 	length = -ESRCH;
731 	if (!task)
732 		goto out_no_task;
733 
734 	if (count > PROC_BLOCK_SIZE)
735 		count = PROC_BLOCK_SIZE;
736 
737 	length = -ENOMEM;
738 	if (!(page = __get_free_page(GFP_TEMPORARY)))
739 		goto out;
740 
741 	length = PROC_I(inode)->op.proc_read(task, (char*)page);
742 
743 	if (length >= 0)
744 		length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
745 	free_page(page);
746 out:
747 	put_task_struct(task);
748 out_no_task:
749 	return length;
750 }
751 
752 static const struct file_operations proc_info_file_operations = {
753 	.read		= proc_info_read,
754 };
755 
756 static int proc_single_show(struct seq_file *m, void *v)
757 {
758 	struct inode *inode = m->private;
759 	struct pid_namespace *ns;
760 	struct pid *pid;
761 	struct task_struct *task;
762 	int ret;
763 
764 	ns = inode->i_sb->s_fs_info;
765 	pid = proc_pid(inode);
766 	task = get_pid_task(pid, PIDTYPE_PID);
767 	if (!task)
768 		return -ESRCH;
769 
770 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
771 
772 	put_task_struct(task);
773 	return ret;
774 }
775 
776 static int proc_single_open(struct inode *inode, struct file *filp)
777 {
778 	int ret;
779 	ret = single_open(filp, proc_single_show, NULL);
780 	if (!ret) {
781 		struct seq_file *m = filp->private_data;
782 
783 		m->private = inode;
784 	}
785 	return ret;
786 }
787 
788 static const struct file_operations proc_single_file_operations = {
789 	.open		= proc_single_open,
790 	.read		= seq_read,
791 	.llseek		= seq_lseek,
792 	.release	= single_release,
793 };
794 
795 static int mem_open(struct inode* inode, struct file* file)
796 {
797 	file->private_data = (void*)((long)current->self_exec_id);
798 	return 0;
799 }
800 
801 static ssize_t mem_read(struct file * file, char __user * buf,
802 			size_t count, loff_t *ppos)
803 {
804 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
805 	char *page;
806 	unsigned long src = *ppos;
807 	int ret = -ESRCH;
808 	struct mm_struct *mm;
809 
810 	if (!task)
811 		goto out_no_task;
812 
813 	if (check_mem_permission(task))
814 		goto out;
815 
816 	ret = -ENOMEM;
817 	page = (char *)__get_free_page(GFP_TEMPORARY);
818 	if (!page)
819 		goto out;
820 
821 	ret = 0;
822 
823 	mm = get_task_mm(task);
824 	if (!mm)
825 		goto out_free;
826 
827 	ret = -EIO;
828 
829 	if (file->private_data != (void*)((long)current->self_exec_id))
830 		goto out_put;
831 
832 	ret = 0;
833 
834 	while (count > 0) {
835 		int this_len, retval;
836 
837 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
838 		retval = access_process_vm(task, src, page, this_len, 0);
839 		if (!retval || check_mem_permission(task)) {
840 			if (!ret)
841 				ret = -EIO;
842 			break;
843 		}
844 
845 		if (copy_to_user(buf, page, retval)) {
846 			ret = -EFAULT;
847 			break;
848 		}
849 
850 		ret += retval;
851 		src += retval;
852 		buf += retval;
853 		count -= retval;
854 	}
855 	*ppos = src;
856 
857 out_put:
858 	mmput(mm);
859 out_free:
860 	free_page((unsigned long) page);
861 out:
862 	put_task_struct(task);
863 out_no_task:
864 	return ret;
865 }
866 
867 #define mem_write NULL
868 
869 #ifndef mem_write
870 /* This is a security hazard */
871 static ssize_t mem_write(struct file * file, const char __user *buf,
872 			 size_t count, loff_t *ppos)
873 {
874 	int copied;
875 	char *page;
876 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
877 	unsigned long dst = *ppos;
878 
879 	copied = -ESRCH;
880 	if (!task)
881 		goto out_no_task;
882 
883 	if (check_mem_permission(task))
884 		goto out;
885 
886 	copied = -ENOMEM;
887 	page = (char *)__get_free_page(GFP_TEMPORARY);
888 	if (!page)
889 		goto out;
890 
891 	copied = 0;
892 	while (count > 0) {
893 		int this_len, retval;
894 
895 		this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
896 		if (copy_from_user(page, buf, this_len)) {
897 			copied = -EFAULT;
898 			break;
899 		}
900 		retval = access_process_vm(task, dst, page, this_len, 1);
901 		if (!retval) {
902 			if (!copied)
903 				copied = -EIO;
904 			break;
905 		}
906 		copied += retval;
907 		buf += retval;
908 		dst += retval;
909 		count -= retval;
910 	}
911 	*ppos = dst;
912 	free_page((unsigned long) page);
913 out:
914 	put_task_struct(task);
915 out_no_task:
916 	return copied;
917 }
918 #endif
919 
920 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
921 {
922 	switch (orig) {
923 	case 0:
924 		file->f_pos = offset;
925 		break;
926 	case 1:
927 		file->f_pos += offset;
928 		break;
929 	default:
930 		return -EINVAL;
931 	}
932 	force_successful_syscall_return();
933 	return file->f_pos;
934 }
935 
936 static const struct file_operations proc_mem_operations = {
937 	.llseek		= mem_lseek,
938 	.read		= mem_read,
939 	.write		= mem_write,
940 	.open		= mem_open,
941 };
942 
943 static ssize_t environ_read(struct file *file, char __user *buf,
944 			size_t count, loff_t *ppos)
945 {
946 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
947 	char *page;
948 	unsigned long src = *ppos;
949 	int ret = -ESRCH;
950 	struct mm_struct *mm;
951 
952 	if (!task)
953 		goto out_no_task;
954 
955 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
956 		goto out;
957 
958 	ret = -ENOMEM;
959 	page = (char *)__get_free_page(GFP_TEMPORARY);
960 	if (!page)
961 		goto out;
962 
963 	ret = 0;
964 
965 	mm = get_task_mm(task);
966 	if (!mm)
967 		goto out_free;
968 
969 	while (count > 0) {
970 		int this_len, retval, max_len;
971 
972 		this_len = mm->env_end - (mm->env_start + src);
973 
974 		if (this_len <= 0)
975 			break;
976 
977 		max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
978 		this_len = (this_len > max_len) ? max_len : this_len;
979 
980 		retval = access_process_vm(task, (mm->env_start + src),
981 			page, this_len, 0);
982 
983 		if (retval <= 0) {
984 			ret = retval;
985 			break;
986 		}
987 
988 		if (copy_to_user(buf, page, retval)) {
989 			ret = -EFAULT;
990 			break;
991 		}
992 
993 		ret += retval;
994 		src += retval;
995 		buf += retval;
996 		count -= retval;
997 	}
998 	*ppos = src;
999 
1000 	mmput(mm);
1001 out_free:
1002 	free_page((unsigned long) page);
1003 out:
1004 	put_task_struct(task);
1005 out_no_task:
1006 	return ret;
1007 }
1008 
1009 static const struct file_operations proc_environ_operations = {
1010 	.read		= environ_read,
1011 };
1012 
1013 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
1014 				size_t count, loff_t *ppos)
1015 {
1016 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1017 	char buffer[PROC_NUMBUF];
1018 	size_t len;
1019 	int oom_adjust;
1020 
1021 	if (!task)
1022 		return -ESRCH;
1023 	oom_adjust = task->oomkilladj;
1024 	put_task_struct(task);
1025 
1026 	len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1027 
1028 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1029 }
1030 
1031 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1032 				size_t count, loff_t *ppos)
1033 {
1034 	struct task_struct *task;
1035 	char buffer[PROC_NUMBUF], *end;
1036 	int oom_adjust;
1037 
1038 	memset(buffer, 0, sizeof(buffer));
1039 	if (count > sizeof(buffer) - 1)
1040 		count = sizeof(buffer) - 1;
1041 	if (copy_from_user(buffer, buf, count))
1042 		return -EFAULT;
1043 	oom_adjust = simple_strtol(buffer, &end, 0);
1044 	if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1045 	     oom_adjust != OOM_DISABLE)
1046 		return -EINVAL;
1047 	if (*end == '\n')
1048 		end++;
1049 	task = get_proc_task(file->f_path.dentry->d_inode);
1050 	if (!task)
1051 		return -ESRCH;
1052 	if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1053 		put_task_struct(task);
1054 		return -EACCES;
1055 	}
1056 	task->oomkilladj = oom_adjust;
1057 	put_task_struct(task);
1058 	if (end - buffer == 0)
1059 		return -EIO;
1060 	return end - buffer;
1061 }
1062 
1063 static const struct file_operations proc_oom_adjust_operations = {
1064 	.read		= oom_adjust_read,
1065 	.write		= oom_adjust_write,
1066 };
1067 
1068 #ifdef CONFIG_AUDITSYSCALL
1069 #define TMPBUFLEN 21
1070 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1071 				  size_t count, loff_t *ppos)
1072 {
1073 	struct inode * inode = file->f_path.dentry->d_inode;
1074 	struct task_struct *task = get_proc_task(inode);
1075 	ssize_t length;
1076 	char tmpbuf[TMPBUFLEN];
1077 
1078 	if (!task)
1079 		return -ESRCH;
1080 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1081 				audit_get_loginuid(task));
1082 	put_task_struct(task);
1083 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1084 }
1085 
1086 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1087 				   size_t count, loff_t *ppos)
1088 {
1089 	struct inode * inode = file->f_path.dentry->d_inode;
1090 	char *page, *tmp;
1091 	ssize_t length;
1092 	uid_t loginuid;
1093 
1094 	if (!capable(CAP_AUDIT_CONTROL))
1095 		return -EPERM;
1096 
1097 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1098 		return -EPERM;
1099 
1100 	if (count >= PAGE_SIZE)
1101 		count = PAGE_SIZE - 1;
1102 
1103 	if (*ppos != 0) {
1104 		/* No partial writes. */
1105 		return -EINVAL;
1106 	}
1107 	page = (char*)__get_free_page(GFP_TEMPORARY);
1108 	if (!page)
1109 		return -ENOMEM;
1110 	length = -EFAULT;
1111 	if (copy_from_user(page, buf, count))
1112 		goto out_free_page;
1113 
1114 	page[count] = '\0';
1115 	loginuid = simple_strtoul(page, &tmp, 10);
1116 	if (tmp == page) {
1117 		length = -EINVAL;
1118 		goto out_free_page;
1119 
1120 	}
1121 	length = audit_set_loginuid(current, loginuid);
1122 	if (likely(length == 0))
1123 		length = count;
1124 
1125 out_free_page:
1126 	free_page((unsigned long) page);
1127 	return length;
1128 }
1129 
1130 static const struct file_operations proc_loginuid_operations = {
1131 	.read		= proc_loginuid_read,
1132 	.write		= proc_loginuid_write,
1133 };
1134 
1135 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1136 				  size_t count, loff_t *ppos)
1137 {
1138 	struct inode * inode = file->f_path.dentry->d_inode;
1139 	struct task_struct *task = get_proc_task(inode);
1140 	ssize_t length;
1141 	char tmpbuf[TMPBUFLEN];
1142 
1143 	if (!task)
1144 		return -ESRCH;
1145 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1146 				audit_get_sessionid(task));
1147 	put_task_struct(task);
1148 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1149 }
1150 
1151 static const struct file_operations proc_sessionid_operations = {
1152 	.read		= proc_sessionid_read,
1153 };
1154 #endif
1155 
1156 #ifdef CONFIG_FAULT_INJECTION
1157 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1158 				      size_t count, loff_t *ppos)
1159 {
1160 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1161 	char buffer[PROC_NUMBUF];
1162 	size_t len;
1163 	int make_it_fail;
1164 
1165 	if (!task)
1166 		return -ESRCH;
1167 	make_it_fail = task->make_it_fail;
1168 	put_task_struct(task);
1169 
1170 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1171 
1172 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1173 }
1174 
1175 static ssize_t proc_fault_inject_write(struct file * file,
1176 			const char __user * buf, size_t count, loff_t *ppos)
1177 {
1178 	struct task_struct *task;
1179 	char buffer[PROC_NUMBUF], *end;
1180 	int make_it_fail;
1181 
1182 	if (!capable(CAP_SYS_RESOURCE))
1183 		return -EPERM;
1184 	memset(buffer, 0, sizeof(buffer));
1185 	if (count > sizeof(buffer) - 1)
1186 		count = sizeof(buffer) - 1;
1187 	if (copy_from_user(buffer, buf, count))
1188 		return -EFAULT;
1189 	make_it_fail = simple_strtol(buffer, &end, 0);
1190 	if (*end == '\n')
1191 		end++;
1192 	task = get_proc_task(file->f_dentry->d_inode);
1193 	if (!task)
1194 		return -ESRCH;
1195 	task->make_it_fail = make_it_fail;
1196 	put_task_struct(task);
1197 	if (end - buffer == 0)
1198 		return -EIO;
1199 	return end - buffer;
1200 }
1201 
1202 static const struct file_operations proc_fault_inject_operations = {
1203 	.read		= proc_fault_inject_read,
1204 	.write		= proc_fault_inject_write,
1205 };
1206 #endif
1207 
1208 
1209 #ifdef CONFIG_SCHED_DEBUG
1210 /*
1211  * Print out various scheduling related per-task fields:
1212  */
1213 static int sched_show(struct seq_file *m, void *v)
1214 {
1215 	struct inode *inode = m->private;
1216 	struct task_struct *p;
1217 
1218 	p = get_proc_task(inode);
1219 	if (!p)
1220 		return -ESRCH;
1221 	proc_sched_show_task(p, m);
1222 
1223 	put_task_struct(p);
1224 
1225 	return 0;
1226 }
1227 
1228 static ssize_t
1229 sched_write(struct file *file, const char __user *buf,
1230 	    size_t count, loff_t *offset)
1231 {
1232 	struct inode *inode = file->f_path.dentry->d_inode;
1233 	struct task_struct *p;
1234 
1235 	p = get_proc_task(inode);
1236 	if (!p)
1237 		return -ESRCH;
1238 	proc_sched_set_task(p);
1239 
1240 	put_task_struct(p);
1241 
1242 	return count;
1243 }
1244 
1245 static int sched_open(struct inode *inode, struct file *filp)
1246 {
1247 	int ret;
1248 
1249 	ret = single_open(filp, sched_show, NULL);
1250 	if (!ret) {
1251 		struct seq_file *m = filp->private_data;
1252 
1253 		m->private = inode;
1254 	}
1255 	return ret;
1256 }
1257 
1258 static const struct file_operations proc_pid_sched_operations = {
1259 	.open		= sched_open,
1260 	.read		= seq_read,
1261 	.write		= sched_write,
1262 	.llseek		= seq_lseek,
1263 	.release	= single_release,
1264 };
1265 
1266 #endif
1267 
1268 /*
1269  * We added or removed a vma mapping the executable. The vmas are only mapped
1270  * during exec and are not mapped with the mmap system call.
1271  * Callers must hold down_write() on the mm's mmap_sem for these
1272  */
1273 void added_exe_file_vma(struct mm_struct *mm)
1274 {
1275 	mm->num_exe_file_vmas++;
1276 }
1277 
1278 void removed_exe_file_vma(struct mm_struct *mm)
1279 {
1280 	mm->num_exe_file_vmas--;
1281 	if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1282 		fput(mm->exe_file);
1283 		mm->exe_file = NULL;
1284 	}
1285 
1286 }
1287 
1288 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1289 {
1290 	if (new_exe_file)
1291 		get_file(new_exe_file);
1292 	if (mm->exe_file)
1293 		fput(mm->exe_file);
1294 	mm->exe_file = new_exe_file;
1295 	mm->num_exe_file_vmas = 0;
1296 }
1297 
1298 struct file *get_mm_exe_file(struct mm_struct *mm)
1299 {
1300 	struct file *exe_file;
1301 
1302 	/* We need mmap_sem to protect against races with removal of
1303 	 * VM_EXECUTABLE vmas */
1304 	down_read(&mm->mmap_sem);
1305 	exe_file = mm->exe_file;
1306 	if (exe_file)
1307 		get_file(exe_file);
1308 	up_read(&mm->mmap_sem);
1309 	return exe_file;
1310 }
1311 
1312 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1313 {
1314 	/* It's safe to write the exe_file pointer without exe_file_lock because
1315 	 * this is called during fork when the task is not yet in /proc */
1316 	newmm->exe_file = get_mm_exe_file(oldmm);
1317 }
1318 
1319 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1320 {
1321 	struct task_struct *task;
1322 	struct mm_struct *mm;
1323 	struct file *exe_file;
1324 
1325 	task = get_proc_task(inode);
1326 	if (!task)
1327 		return -ENOENT;
1328 	mm = get_task_mm(task);
1329 	put_task_struct(task);
1330 	if (!mm)
1331 		return -ENOENT;
1332 	exe_file = get_mm_exe_file(mm);
1333 	mmput(mm);
1334 	if (exe_file) {
1335 		*exe_path = exe_file->f_path;
1336 		path_get(&exe_file->f_path);
1337 		fput(exe_file);
1338 		return 0;
1339 	} else
1340 		return -ENOENT;
1341 }
1342 
1343 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1344 {
1345 	struct inode *inode = dentry->d_inode;
1346 	int error = -EACCES;
1347 
1348 	/* We don't need a base pointer in the /proc filesystem */
1349 	path_put(&nd->path);
1350 
1351 	/* Are we allowed to snoop on the tasks file descriptors? */
1352 	if (!proc_fd_access_allowed(inode))
1353 		goto out;
1354 
1355 	error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1356 	nd->last_type = LAST_BIND;
1357 out:
1358 	return ERR_PTR(error);
1359 }
1360 
1361 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1362 {
1363 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1364 	char *pathname;
1365 	int len;
1366 
1367 	if (!tmp)
1368 		return -ENOMEM;
1369 
1370 	pathname = d_path(path, tmp, PAGE_SIZE);
1371 	len = PTR_ERR(pathname);
1372 	if (IS_ERR(pathname))
1373 		goto out;
1374 	len = tmp + PAGE_SIZE - 1 - pathname;
1375 
1376 	if (len > buflen)
1377 		len = buflen;
1378 	if (copy_to_user(buffer, pathname, len))
1379 		len = -EFAULT;
1380  out:
1381 	free_page((unsigned long)tmp);
1382 	return len;
1383 }
1384 
1385 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1386 {
1387 	int error = -EACCES;
1388 	struct inode *inode = dentry->d_inode;
1389 	struct path path;
1390 
1391 	/* Are we allowed to snoop on the tasks file descriptors? */
1392 	if (!proc_fd_access_allowed(inode))
1393 		goto out;
1394 
1395 	error = PROC_I(inode)->op.proc_get_link(inode, &path);
1396 	if (error)
1397 		goto out;
1398 
1399 	error = do_proc_readlink(&path, buffer, buflen);
1400 	path_put(&path);
1401 out:
1402 	return error;
1403 }
1404 
1405 static const struct inode_operations proc_pid_link_inode_operations = {
1406 	.readlink	= proc_pid_readlink,
1407 	.follow_link	= proc_pid_follow_link,
1408 	.setattr	= proc_setattr,
1409 };
1410 
1411 
1412 /* building an inode */
1413 
1414 static int task_dumpable(struct task_struct *task)
1415 {
1416 	int dumpable = 0;
1417 	struct mm_struct *mm;
1418 
1419 	task_lock(task);
1420 	mm = task->mm;
1421 	if (mm)
1422 		dumpable = get_dumpable(mm);
1423 	task_unlock(task);
1424 	if(dumpable == 1)
1425 		return 1;
1426 	return 0;
1427 }
1428 
1429 
1430 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1431 {
1432 	struct inode * inode;
1433 	struct proc_inode *ei;
1434 	const struct cred *cred;
1435 
1436 	/* We need a new inode */
1437 
1438 	inode = new_inode(sb);
1439 	if (!inode)
1440 		goto out;
1441 
1442 	/* Common stuff */
1443 	ei = PROC_I(inode);
1444 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1445 	inode->i_op = &proc_def_inode_operations;
1446 
1447 	/*
1448 	 * grab the reference to task.
1449 	 */
1450 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1451 	if (!ei->pid)
1452 		goto out_unlock;
1453 
1454 	inode->i_uid = 0;
1455 	inode->i_gid = 0;
1456 	if (task_dumpable(task)) {
1457 		rcu_read_lock();
1458 		cred = __task_cred(task);
1459 		inode->i_uid = cred->euid;
1460 		inode->i_gid = cred->egid;
1461 		rcu_read_unlock();
1462 	}
1463 	security_task_to_inode(task, inode);
1464 
1465 out:
1466 	return inode;
1467 
1468 out_unlock:
1469 	iput(inode);
1470 	return NULL;
1471 }
1472 
1473 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1474 {
1475 	struct inode *inode = dentry->d_inode;
1476 	struct task_struct *task;
1477 	const struct cred *cred;
1478 
1479 	generic_fillattr(inode, stat);
1480 
1481 	rcu_read_lock();
1482 	stat->uid = 0;
1483 	stat->gid = 0;
1484 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1485 	if (task) {
1486 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1487 		    task_dumpable(task)) {
1488 			cred = __task_cred(task);
1489 			stat->uid = cred->euid;
1490 			stat->gid = cred->egid;
1491 		}
1492 	}
1493 	rcu_read_unlock();
1494 	return 0;
1495 }
1496 
1497 /* dentry stuff */
1498 
1499 /*
1500  *	Exceptional case: normally we are not allowed to unhash a busy
1501  * directory. In this case, however, we can do it - no aliasing problems
1502  * due to the way we treat inodes.
1503  *
1504  * Rewrite the inode's ownerships here because the owning task may have
1505  * performed a setuid(), etc.
1506  *
1507  * Before the /proc/pid/status file was created the only way to read
1508  * the effective uid of a /process was to stat /proc/pid.  Reading
1509  * /proc/pid/status is slow enough that procps and other packages
1510  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1511  * made this apply to all per process world readable and executable
1512  * directories.
1513  */
1514 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1515 {
1516 	struct inode *inode = dentry->d_inode;
1517 	struct task_struct *task = get_proc_task(inode);
1518 	const struct cred *cred;
1519 
1520 	if (task) {
1521 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1522 		    task_dumpable(task)) {
1523 			rcu_read_lock();
1524 			cred = __task_cred(task);
1525 			inode->i_uid = cred->euid;
1526 			inode->i_gid = cred->egid;
1527 			rcu_read_unlock();
1528 		} else {
1529 			inode->i_uid = 0;
1530 			inode->i_gid = 0;
1531 		}
1532 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1533 		security_task_to_inode(task, inode);
1534 		put_task_struct(task);
1535 		return 1;
1536 	}
1537 	d_drop(dentry);
1538 	return 0;
1539 }
1540 
1541 static int pid_delete_dentry(struct dentry * dentry)
1542 {
1543 	/* Is the task we represent dead?
1544 	 * If so, then don't put the dentry on the lru list,
1545 	 * kill it immediately.
1546 	 */
1547 	return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1548 }
1549 
1550 static struct dentry_operations pid_dentry_operations =
1551 {
1552 	.d_revalidate	= pid_revalidate,
1553 	.d_delete	= pid_delete_dentry,
1554 };
1555 
1556 /* Lookups */
1557 
1558 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1559 				struct task_struct *, const void *);
1560 
1561 /*
1562  * Fill a directory entry.
1563  *
1564  * If possible create the dcache entry and derive our inode number and
1565  * file type from dcache entry.
1566  *
1567  * Since all of the proc inode numbers are dynamically generated, the inode
1568  * numbers do not exist until the inode is cache.  This means creating the
1569  * the dcache entry in readdir is necessary to keep the inode numbers
1570  * reported by readdir in sync with the inode numbers reported
1571  * by stat.
1572  */
1573 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1574 	char *name, int len,
1575 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1576 {
1577 	struct dentry *child, *dir = filp->f_path.dentry;
1578 	struct inode *inode;
1579 	struct qstr qname;
1580 	ino_t ino = 0;
1581 	unsigned type = DT_UNKNOWN;
1582 
1583 	qname.name = name;
1584 	qname.len  = len;
1585 	qname.hash = full_name_hash(name, len);
1586 
1587 	child = d_lookup(dir, &qname);
1588 	if (!child) {
1589 		struct dentry *new;
1590 		new = d_alloc(dir, &qname);
1591 		if (new) {
1592 			child = instantiate(dir->d_inode, new, task, ptr);
1593 			if (child)
1594 				dput(new);
1595 			else
1596 				child = new;
1597 		}
1598 	}
1599 	if (!child || IS_ERR(child) || !child->d_inode)
1600 		goto end_instantiate;
1601 	inode = child->d_inode;
1602 	if (inode) {
1603 		ino = inode->i_ino;
1604 		type = inode->i_mode >> 12;
1605 	}
1606 	dput(child);
1607 end_instantiate:
1608 	if (!ino)
1609 		ino = find_inode_number(dir, &qname);
1610 	if (!ino)
1611 		ino = 1;
1612 	return filldir(dirent, name, len, filp->f_pos, ino, type);
1613 }
1614 
1615 static unsigned name_to_int(struct dentry *dentry)
1616 {
1617 	const char *name = dentry->d_name.name;
1618 	int len = dentry->d_name.len;
1619 	unsigned n = 0;
1620 
1621 	if (len > 1 && *name == '0')
1622 		goto out;
1623 	while (len-- > 0) {
1624 		unsigned c = *name++ - '0';
1625 		if (c > 9)
1626 			goto out;
1627 		if (n >= (~0U-9)/10)
1628 			goto out;
1629 		n *= 10;
1630 		n += c;
1631 	}
1632 	return n;
1633 out:
1634 	return ~0U;
1635 }
1636 
1637 #define PROC_FDINFO_MAX 64
1638 
1639 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1640 {
1641 	struct task_struct *task = get_proc_task(inode);
1642 	struct files_struct *files = NULL;
1643 	struct file *file;
1644 	int fd = proc_fd(inode);
1645 
1646 	if (task) {
1647 		files = get_files_struct(task);
1648 		put_task_struct(task);
1649 	}
1650 	if (files) {
1651 		/*
1652 		 * We are not taking a ref to the file structure, so we must
1653 		 * hold ->file_lock.
1654 		 */
1655 		spin_lock(&files->file_lock);
1656 		file = fcheck_files(files, fd);
1657 		if (file) {
1658 			if (path) {
1659 				*path = file->f_path;
1660 				path_get(&file->f_path);
1661 			}
1662 			if (info)
1663 				snprintf(info, PROC_FDINFO_MAX,
1664 					 "pos:\t%lli\n"
1665 					 "flags:\t0%o\n",
1666 					 (long long) file->f_pos,
1667 					 file->f_flags);
1668 			spin_unlock(&files->file_lock);
1669 			put_files_struct(files);
1670 			return 0;
1671 		}
1672 		spin_unlock(&files->file_lock);
1673 		put_files_struct(files);
1674 	}
1675 	return -ENOENT;
1676 }
1677 
1678 static int proc_fd_link(struct inode *inode, struct path *path)
1679 {
1680 	return proc_fd_info(inode, path, NULL);
1681 }
1682 
1683 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1684 {
1685 	struct inode *inode = dentry->d_inode;
1686 	struct task_struct *task = get_proc_task(inode);
1687 	int fd = proc_fd(inode);
1688 	struct files_struct *files;
1689 	const struct cred *cred;
1690 
1691 	if (task) {
1692 		files = get_files_struct(task);
1693 		if (files) {
1694 			rcu_read_lock();
1695 			if (fcheck_files(files, fd)) {
1696 				rcu_read_unlock();
1697 				put_files_struct(files);
1698 				if (task_dumpable(task)) {
1699 					rcu_read_lock();
1700 					cred = __task_cred(task);
1701 					inode->i_uid = cred->euid;
1702 					inode->i_gid = cred->egid;
1703 					rcu_read_unlock();
1704 				} else {
1705 					inode->i_uid = 0;
1706 					inode->i_gid = 0;
1707 				}
1708 				inode->i_mode &= ~(S_ISUID | S_ISGID);
1709 				security_task_to_inode(task, inode);
1710 				put_task_struct(task);
1711 				return 1;
1712 			}
1713 			rcu_read_unlock();
1714 			put_files_struct(files);
1715 		}
1716 		put_task_struct(task);
1717 	}
1718 	d_drop(dentry);
1719 	return 0;
1720 }
1721 
1722 static struct dentry_operations tid_fd_dentry_operations =
1723 {
1724 	.d_revalidate	= tid_fd_revalidate,
1725 	.d_delete	= pid_delete_dentry,
1726 };
1727 
1728 static struct dentry *proc_fd_instantiate(struct inode *dir,
1729 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1730 {
1731 	unsigned fd = *(const unsigned *)ptr;
1732 	struct file *file;
1733 	struct files_struct *files;
1734  	struct inode *inode;
1735  	struct proc_inode *ei;
1736 	struct dentry *error = ERR_PTR(-ENOENT);
1737 
1738 	inode = proc_pid_make_inode(dir->i_sb, task);
1739 	if (!inode)
1740 		goto out;
1741 	ei = PROC_I(inode);
1742 	ei->fd = fd;
1743 	files = get_files_struct(task);
1744 	if (!files)
1745 		goto out_iput;
1746 	inode->i_mode = S_IFLNK;
1747 
1748 	/*
1749 	 * We are not taking a ref to the file structure, so we must
1750 	 * hold ->file_lock.
1751 	 */
1752 	spin_lock(&files->file_lock);
1753 	file = fcheck_files(files, fd);
1754 	if (!file)
1755 		goto out_unlock;
1756 	if (file->f_mode & FMODE_READ)
1757 		inode->i_mode |= S_IRUSR | S_IXUSR;
1758 	if (file->f_mode & FMODE_WRITE)
1759 		inode->i_mode |= S_IWUSR | S_IXUSR;
1760 	spin_unlock(&files->file_lock);
1761 	put_files_struct(files);
1762 
1763 	inode->i_op = &proc_pid_link_inode_operations;
1764 	inode->i_size = 64;
1765 	ei->op.proc_get_link = proc_fd_link;
1766 	dentry->d_op = &tid_fd_dentry_operations;
1767 	d_add(dentry, inode);
1768 	/* Close the race of the process dying before we return the dentry */
1769 	if (tid_fd_revalidate(dentry, NULL))
1770 		error = NULL;
1771 
1772  out:
1773 	return error;
1774 out_unlock:
1775 	spin_unlock(&files->file_lock);
1776 	put_files_struct(files);
1777 out_iput:
1778 	iput(inode);
1779 	goto out;
1780 }
1781 
1782 static struct dentry *proc_lookupfd_common(struct inode *dir,
1783 					   struct dentry *dentry,
1784 					   instantiate_t instantiate)
1785 {
1786 	struct task_struct *task = get_proc_task(dir);
1787 	unsigned fd = name_to_int(dentry);
1788 	struct dentry *result = ERR_PTR(-ENOENT);
1789 
1790 	if (!task)
1791 		goto out_no_task;
1792 	if (fd == ~0U)
1793 		goto out;
1794 
1795 	result = instantiate(dir, dentry, task, &fd);
1796 out:
1797 	put_task_struct(task);
1798 out_no_task:
1799 	return result;
1800 }
1801 
1802 static int proc_readfd_common(struct file * filp, void * dirent,
1803 			      filldir_t filldir, instantiate_t instantiate)
1804 {
1805 	struct dentry *dentry = filp->f_path.dentry;
1806 	struct inode *inode = dentry->d_inode;
1807 	struct task_struct *p = get_proc_task(inode);
1808 	unsigned int fd, ino;
1809 	int retval;
1810 	struct files_struct * files;
1811 
1812 	retval = -ENOENT;
1813 	if (!p)
1814 		goto out_no_task;
1815 	retval = 0;
1816 
1817 	fd = filp->f_pos;
1818 	switch (fd) {
1819 		case 0:
1820 			if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1821 				goto out;
1822 			filp->f_pos++;
1823 		case 1:
1824 			ino = parent_ino(dentry);
1825 			if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1826 				goto out;
1827 			filp->f_pos++;
1828 		default:
1829 			files = get_files_struct(p);
1830 			if (!files)
1831 				goto out;
1832 			rcu_read_lock();
1833 			for (fd = filp->f_pos-2;
1834 			     fd < files_fdtable(files)->max_fds;
1835 			     fd++, filp->f_pos++) {
1836 				char name[PROC_NUMBUF];
1837 				int len;
1838 
1839 				if (!fcheck_files(files, fd))
1840 					continue;
1841 				rcu_read_unlock();
1842 
1843 				len = snprintf(name, sizeof(name), "%d", fd);
1844 				if (proc_fill_cache(filp, dirent, filldir,
1845 						    name, len, instantiate,
1846 						    p, &fd) < 0) {
1847 					rcu_read_lock();
1848 					break;
1849 				}
1850 				rcu_read_lock();
1851 			}
1852 			rcu_read_unlock();
1853 			put_files_struct(files);
1854 	}
1855 out:
1856 	put_task_struct(p);
1857 out_no_task:
1858 	return retval;
1859 }
1860 
1861 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1862 				    struct nameidata *nd)
1863 {
1864 	return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1865 }
1866 
1867 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1868 {
1869 	return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1870 }
1871 
1872 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1873 				      size_t len, loff_t *ppos)
1874 {
1875 	char tmp[PROC_FDINFO_MAX];
1876 	int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1877 	if (!err)
1878 		err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1879 	return err;
1880 }
1881 
1882 static const struct file_operations proc_fdinfo_file_operations = {
1883 	.open		= nonseekable_open,
1884 	.read		= proc_fdinfo_read,
1885 };
1886 
1887 static const struct file_operations proc_fd_operations = {
1888 	.read		= generic_read_dir,
1889 	.readdir	= proc_readfd,
1890 };
1891 
1892 /*
1893  * /proc/pid/fd needs a special permission handler so that a process can still
1894  * access /proc/self/fd after it has executed a setuid().
1895  */
1896 static int proc_fd_permission(struct inode *inode, int mask)
1897 {
1898 	int rv;
1899 
1900 	rv = generic_permission(inode, mask, NULL);
1901 	if (rv == 0)
1902 		return 0;
1903 	if (task_pid(current) == proc_pid(inode))
1904 		rv = 0;
1905 	return rv;
1906 }
1907 
1908 /*
1909  * proc directories can do almost nothing..
1910  */
1911 static const struct inode_operations proc_fd_inode_operations = {
1912 	.lookup		= proc_lookupfd,
1913 	.permission	= proc_fd_permission,
1914 	.setattr	= proc_setattr,
1915 };
1916 
1917 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1918 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1919 {
1920 	unsigned fd = *(unsigned *)ptr;
1921  	struct inode *inode;
1922  	struct proc_inode *ei;
1923 	struct dentry *error = ERR_PTR(-ENOENT);
1924 
1925 	inode = proc_pid_make_inode(dir->i_sb, task);
1926 	if (!inode)
1927 		goto out;
1928 	ei = PROC_I(inode);
1929 	ei->fd = fd;
1930 	inode->i_mode = S_IFREG | S_IRUSR;
1931 	inode->i_fop = &proc_fdinfo_file_operations;
1932 	dentry->d_op = &tid_fd_dentry_operations;
1933 	d_add(dentry, inode);
1934 	/* Close the race of the process dying before we return the dentry */
1935 	if (tid_fd_revalidate(dentry, NULL))
1936 		error = NULL;
1937 
1938  out:
1939 	return error;
1940 }
1941 
1942 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1943 					struct dentry *dentry,
1944 					struct nameidata *nd)
1945 {
1946 	return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1947 }
1948 
1949 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1950 {
1951 	return proc_readfd_common(filp, dirent, filldir,
1952 				  proc_fdinfo_instantiate);
1953 }
1954 
1955 static const struct file_operations proc_fdinfo_operations = {
1956 	.read		= generic_read_dir,
1957 	.readdir	= proc_readfdinfo,
1958 };
1959 
1960 /*
1961  * proc directories can do almost nothing..
1962  */
1963 static const struct inode_operations proc_fdinfo_inode_operations = {
1964 	.lookup		= proc_lookupfdinfo,
1965 	.setattr	= proc_setattr,
1966 };
1967 
1968 
1969 static struct dentry *proc_pident_instantiate(struct inode *dir,
1970 	struct dentry *dentry, struct task_struct *task, const void *ptr)
1971 {
1972 	const struct pid_entry *p = ptr;
1973 	struct inode *inode;
1974 	struct proc_inode *ei;
1975 	struct dentry *error = ERR_PTR(-EINVAL);
1976 
1977 	inode = proc_pid_make_inode(dir->i_sb, task);
1978 	if (!inode)
1979 		goto out;
1980 
1981 	ei = PROC_I(inode);
1982 	inode->i_mode = p->mode;
1983 	if (S_ISDIR(inode->i_mode))
1984 		inode->i_nlink = 2;	/* Use getattr to fix if necessary */
1985 	if (p->iop)
1986 		inode->i_op = p->iop;
1987 	if (p->fop)
1988 		inode->i_fop = p->fop;
1989 	ei->op = p->op;
1990 	dentry->d_op = &pid_dentry_operations;
1991 	d_add(dentry, inode);
1992 	/* Close the race of the process dying before we return the dentry */
1993 	if (pid_revalidate(dentry, NULL))
1994 		error = NULL;
1995 out:
1996 	return error;
1997 }
1998 
1999 static struct dentry *proc_pident_lookup(struct inode *dir,
2000 					 struct dentry *dentry,
2001 					 const struct pid_entry *ents,
2002 					 unsigned int nents)
2003 {
2004 	struct inode *inode;
2005 	struct dentry *error;
2006 	struct task_struct *task = get_proc_task(dir);
2007 	const struct pid_entry *p, *last;
2008 
2009 	error = ERR_PTR(-ENOENT);
2010 	inode = NULL;
2011 
2012 	if (!task)
2013 		goto out_no_task;
2014 
2015 	/*
2016 	 * Yes, it does not scale. And it should not. Don't add
2017 	 * new entries into /proc/<tgid>/ without very good reasons.
2018 	 */
2019 	last = &ents[nents - 1];
2020 	for (p = ents; p <= last; p++) {
2021 		if (p->len != dentry->d_name.len)
2022 			continue;
2023 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2024 			break;
2025 	}
2026 	if (p > last)
2027 		goto out;
2028 
2029 	error = proc_pident_instantiate(dir, dentry, task, p);
2030 out:
2031 	put_task_struct(task);
2032 out_no_task:
2033 	return error;
2034 }
2035 
2036 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2037 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2038 {
2039 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2040 				proc_pident_instantiate, task, p);
2041 }
2042 
2043 static int proc_pident_readdir(struct file *filp,
2044 		void *dirent, filldir_t filldir,
2045 		const struct pid_entry *ents, unsigned int nents)
2046 {
2047 	int i;
2048 	struct dentry *dentry = filp->f_path.dentry;
2049 	struct inode *inode = dentry->d_inode;
2050 	struct task_struct *task = get_proc_task(inode);
2051 	const struct pid_entry *p, *last;
2052 	ino_t ino;
2053 	int ret;
2054 
2055 	ret = -ENOENT;
2056 	if (!task)
2057 		goto out_no_task;
2058 
2059 	ret = 0;
2060 	i = filp->f_pos;
2061 	switch (i) {
2062 	case 0:
2063 		ino = inode->i_ino;
2064 		if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2065 			goto out;
2066 		i++;
2067 		filp->f_pos++;
2068 		/* fall through */
2069 	case 1:
2070 		ino = parent_ino(dentry);
2071 		if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2072 			goto out;
2073 		i++;
2074 		filp->f_pos++;
2075 		/* fall through */
2076 	default:
2077 		i -= 2;
2078 		if (i >= nents) {
2079 			ret = 1;
2080 			goto out;
2081 		}
2082 		p = ents + i;
2083 		last = &ents[nents - 1];
2084 		while (p <= last) {
2085 			if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2086 				goto out;
2087 			filp->f_pos++;
2088 			p++;
2089 		}
2090 	}
2091 
2092 	ret = 1;
2093 out:
2094 	put_task_struct(task);
2095 out_no_task:
2096 	return ret;
2097 }
2098 
2099 #ifdef CONFIG_SECURITY
2100 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2101 				  size_t count, loff_t *ppos)
2102 {
2103 	struct inode * inode = file->f_path.dentry->d_inode;
2104 	char *p = NULL;
2105 	ssize_t length;
2106 	struct task_struct *task = get_proc_task(inode);
2107 
2108 	if (!task)
2109 		return -ESRCH;
2110 
2111 	length = security_getprocattr(task,
2112 				      (char*)file->f_path.dentry->d_name.name,
2113 				      &p);
2114 	put_task_struct(task);
2115 	if (length > 0)
2116 		length = simple_read_from_buffer(buf, count, ppos, p, length);
2117 	kfree(p);
2118 	return length;
2119 }
2120 
2121 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2122 				   size_t count, loff_t *ppos)
2123 {
2124 	struct inode * inode = file->f_path.dentry->d_inode;
2125 	char *page;
2126 	ssize_t length;
2127 	struct task_struct *task = get_proc_task(inode);
2128 
2129 	length = -ESRCH;
2130 	if (!task)
2131 		goto out_no_task;
2132 	if (count > PAGE_SIZE)
2133 		count = PAGE_SIZE;
2134 
2135 	/* No partial writes. */
2136 	length = -EINVAL;
2137 	if (*ppos != 0)
2138 		goto out;
2139 
2140 	length = -ENOMEM;
2141 	page = (char*)__get_free_page(GFP_TEMPORARY);
2142 	if (!page)
2143 		goto out;
2144 
2145 	length = -EFAULT;
2146 	if (copy_from_user(page, buf, count))
2147 		goto out_free;
2148 
2149 	length = security_setprocattr(task,
2150 				      (char*)file->f_path.dentry->d_name.name,
2151 				      (void*)page, count);
2152 out_free:
2153 	free_page((unsigned long) page);
2154 out:
2155 	put_task_struct(task);
2156 out_no_task:
2157 	return length;
2158 }
2159 
2160 static const struct file_operations proc_pid_attr_operations = {
2161 	.read		= proc_pid_attr_read,
2162 	.write		= proc_pid_attr_write,
2163 };
2164 
2165 static const struct pid_entry attr_dir_stuff[] = {
2166 	REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2167 	REG("prev",       S_IRUGO,	   proc_pid_attr_operations),
2168 	REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2169 	REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2170 	REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2171 	REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2172 };
2173 
2174 static int proc_attr_dir_readdir(struct file * filp,
2175 			     void * dirent, filldir_t filldir)
2176 {
2177 	return proc_pident_readdir(filp,dirent,filldir,
2178 				   attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2179 }
2180 
2181 static const struct file_operations proc_attr_dir_operations = {
2182 	.read		= generic_read_dir,
2183 	.readdir	= proc_attr_dir_readdir,
2184 };
2185 
2186 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2187 				struct dentry *dentry, struct nameidata *nd)
2188 {
2189 	return proc_pident_lookup(dir, dentry,
2190 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2191 }
2192 
2193 static const struct inode_operations proc_attr_dir_inode_operations = {
2194 	.lookup		= proc_attr_dir_lookup,
2195 	.getattr	= pid_getattr,
2196 	.setattr	= proc_setattr,
2197 };
2198 
2199 #endif
2200 
2201 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2202 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2203 					 size_t count, loff_t *ppos)
2204 {
2205 	struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2206 	struct mm_struct *mm;
2207 	char buffer[PROC_NUMBUF];
2208 	size_t len;
2209 	int ret;
2210 
2211 	if (!task)
2212 		return -ESRCH;
2213 
2214 	ret = 0;
2215 	mm = get_task_mm(task);
2216 	if (mm) {
2217 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2218 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2219 				MMF_DUMP_FILTER_SHIFT));
2220 		mmput(mm);
2221 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2222 	}
2223 
2224 	put_task_struct(task);
2225 
2226 	return ret;
2227 }
2228 
2229 static ssize_t proc_coredump_filter_write(struct file *file,
2230 					  const char __user *buf,
2231 					  size_t count,
2232 					  loff_t *ppos)
2233 {
2234 	struct task_struct *task;
2235 	struct mm_struct *mm;
2236 	char buffer[PROC_NUMBUF], *end;
2237 	unsigned int val;
2238 	int ret;
2239 	int i;
2240 	unsigned long mask;
2241 
2242 	ret = -EFAULT;
2243 	memset(buffer, 0, sizeof(buffer));
2244 	if (count > sizeof(buffer) - 1)
2245 		count = sizeof(buffer) - 1;
2246 	if (copy_from_user(buffer, buf, count))
2247 		goto out_no_task;
2248 
2249 	ret = -EINVAL;
2250 	val = (unsigned int)simple_strtoul(buffer, &end, 0);
2251 	if (*end == '\n')
2252 		end++;
2253 	if (end - buffer == 0)
2254 		goto out_no_task;
2255 
2256 	ret = -ESRCH;
2257 	task = get_proc_task(file->f_dentry->d_inode);
2258 	if (!task)
2259 		goto out_no_task;
2260 
2261 	ret = end - buffer;
2262 	mm = get_task_mm(task);
2263 	if (!mm)
2264 		goto out_no_mm;
2265 
2266 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2267 		if (val & mask)
2268 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2269 		else
2270 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2271 	}
2272 
2273 	mmput(mm);
2274  out_no_mm:
2275 	put_task_struct(task);
2276  out_no_task:
2277 	return ret;
2278 }
2279 
2280 static const struct file_operations proc_coredump_filter_operations = {
2281 	.read		= proc_coredump_filter_read,
2282 	.write		= proc_coredump_filter_write,
2283 };
2284 #endif
2285 
2286 /*
2287  * /proc/self:
2288  */
2289 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2290 			      int buflen)
2291 {
2292 	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2293 	pid_t tgid = task_tgid_nr_ns(current, ns);
2294 	char tmp[PROC_NUMBUF];
2295 	if (!tgid)
2296 		return -ENOENT;
2297 	sprintf(tmp, "%d", tgid);
2298 	return vfs_readlink(dentry,buffer,buflen,tmp);
2299 }
2300 
2301 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2302 {
2303 	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2304 	pid_t tgid = task_tgid_nr_ns(current, ns);
2305 	char tmp[PROC_NUMBUF];
2306 	if (!tgid)
2307 		return ERR_PTR(-ENOENT);
2308 	sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2309 	return ERR_PTR(vfs_follow_link(nd,tmp));
2310 }
2311 
2312 static const struct inode_operations proc_self_inode_operations = {
2313 	.readlink	= proc_self_readlink,
2314 	.follow_link	= proc_self_follow_link,
2315 };
2316 
2317 /*
2318  * proc base
2319  *
2320  * These are the directory entries in the root directory of /proc
2321  * that properly belong to the /proc filesystem, as they describe
2322  * describe something that is process related.
2323  */
2324 static const struct pid_entry proc_base_stuff[] = {
2325 	NOD("self", S_IFLNK|S_IRWXUGO,
2326 		&proc_self_inode_operations, NULL, {}),
2327 };
2328 
2329 /*
2330  *	Exceptional case: normally we are not allowed to unhash a busy
2331  * directory. In this case, however, we can do it - no aliasing problems
2332  * due to the way we treat inodes.
2333  */
2334 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2335 {
2336 	struct inode *inode = dentry->d_inode;
2337 	struct task_struct *task = get_proc_task(inode);
2338 	if (task) {
2339 		put_task_struct(task);
2340 		return 1;
2341 	}
2342 	d_drop(dentry);
2343 	return 0;
2344 }
2345 
2346 static struct dentry_operations proc_base_dentry_operations =
2347 {
2348 	.d_revalidate	= proc_base_revalidate,
2349 	.d_delete	= pid_delete_dentry,
2350 };
2351 
2352 static struct dentry *proc_base_instantiate(struct inode *dir,
2353 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2354 {
2355 	const struct pid_entry *p = ptr;
2356 	struct inode *inode;
2357 	struct proc_inode *ei;
2358 	struct dentry *error = ERR_PTR(-EINVAL);
2359 
2360 	/* Allocate the inode */
2361 	error = ERR_PTR(-ENOMEM);
2362 	inode = new_inode(dir->i_sb);
2363 	if (!inode)
2364 		goto out;
2365 
2366 	/* Initialize the inode */
2367 	ei = PROC_I(inode);
2368 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2369 
2370 	/*
2371 	 * grab the reference to the task.
2372 	 */
2373 	ei->pid = get_task_pid(task, PIDTYPE_PID);
2374 	if (!ei->pid)
2375 		goto out_iput;
2376 
2377 	inode->i_uid = 0;
2378 	inode->i_gid = 0;
2379 	inode->i_mode = p->mode;
2380 	if (S_ISDIR(inode->i_mode))
2381 		inode->i_nlink = 2;
2382 	if (S_ISLNK(inode->i_mode))
2383 		inode->i_size = 64;
2384 	if (p->iop)
2385 		inode->i_op = p->iop;
2386 	if (p->fop)
2387 		inode->i_fop = p->fop;
2388 	ei->op = p->op;
2389 	dentry->d_op = &proc_base_dentry_operations;
2390 	d_add(dentry, inode);
2391 	error = NULL;
2392 out:
2393 	return error;
2394 out_iput:
2395 	iput(inode);
2396 	goto out;
2397 }
2398 
2399 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2400 {
2401 	struct dentry *error;
2402 	struct task_struct *task = get_proc_task(dir);
2403 	const struct pid_entry *p, *last;
2404 
2405 	error = ERR_PTR(-ENOENT);
2406 
2407 	if (!task)
2408 		goto out_no_task;
2409 
2410 	/* Lookup the directory entry */
2411 	last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2412 	for (p = proc_base_stuff; p <= last; p++) {
2413 		if (p->len != dentry->d_name.len)
2414 			continue;
2415 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2416 			break;
2417 	}
2418 	if (p > last)
2419 		goto out;
2420 
2421 	error = proc_base_instantiate(dir, dentry, task, p);
2422 
2423 out:
2424 	put_task_struct(task);
2425 out_no_task:
2426 	return error;
2427 }
2428 
2429 static int proc_base_fill_cache(struct file *filp, void *dirent,
2430 	filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2431 {
2432 	return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2433 				proc_base_instantiate, task, p);
2434 }
2435 
2436 #ifdef CONFIG_TASK_IO_ACCOUNTING
2437 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2438 {
2439 	struct task_io_accounting acct = task->ioac;
2440 	unsigned long flags;
2441 
2442 	if (whole && lock_task_sighand(task, &flags)) {
2443 		struct task_struct *t = task;
2444 
2445 		task_io_accounting_add(&acct, &task->signal->ioac);
2446 		while_each_thread(task, t)
2447 			task_io_accounting_add(&acct, &t->ioac);
2448 
2449 		unlock_task_sighand(task, &flags);
2450 	}
2451 	return sprintf(buffer,
2452 			"rchar: %llu\n"
2453 			"wchar: %llu\n"
2454 			"syscr: %llu\n"
2455 			"syscw: %llu\n"
2456 			"read_bytes: %llu\n"
2457 			"write_bytes: %llu\n"
2458 			"cancelled_write_bytes: %llu\n",
2459 			(unsigned long long)acct.rchar,
2460 			(unsigned long long)acct.wchar,
2461 			(unsigned long long)acct.syscr,
2462 			(unsigned long long)acct.syscw,
2463 			(unsigned long long)acct.read_bytes,
2464 			(unsigned long long)acct.write_bytes,
2465 			(unsigned long long)acct.cancelled_write_bytes);
2466 }
2467 
2468 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2469 {
2470 	return do_io_accounting(task, buffer, 0);
2471 }
2472 
2473 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2474 {
2475 	return do_io_accounting(task, buffer, 1);
2476 }
2477 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2478 
2479 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2480 				struct pid *pid, struct task_struct *task)
2481 {
2482 	seq_printf(m, "%08x\n", task->personality);
2483 	return 0;
2484 }
2485 
2486 /*
2487  * Thread groups
2488  */
2489 static const struct file_operations proc_task_operations;
2490 static const struct inode_operations proc_task_inode_operations;
2491 
2492 static const struct pid_entry tgid_base_stuff[] = {
2493 	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2494 	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2495 	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2496 #ifdef CONFIG_NET
2497 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2498 #endif
2499 	REG("environ",    S_IRUSR, proc_environ_operations),
2500 	INF("auxv",       S_IRUSR, proc_pid_auxv),
2501 	ONE("status",     S_IRUGO, proc_pid_status),
2502 	ONE("personality", S_IRUSR, proc_pid_personality),
2503 	INF("limits",	  S_IRUSR, proc_pid_limits),
2504 #ifdef CONFIG_SCHED_DEBUG
2505 	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2506 #endif
2507 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2508 	INF("syscall",    S_IRUSR, proc_pid_syscall),
2509 #endif
2510 	INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2511 	ONE("stat",       S_IRUGO, proc_tgid_stat),
2512 	ONE("statm",      S_IRUGO, proc_pid_statm),
2513 	REG("maps",       S_IRUGO, proc_maps_operations),
2514 #ifdef CONFIG_NUMA
2515 	REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2516 #endif
2517 	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2518 	LNK("cwd",        proc_cwd_link),
2519 	LNK("root",       proc_root_link),
2520 	LNK("exe",        proc_exe_link),
2521 	REG("mounts",     S_IRUGO, proc_mounts_operations),
2522 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2523 	REG("mountstats", S_IRUSR, proc_mountstats_operations),
2524 #ifdef CONFIG_PROC_PAGE_MONITOR
2525 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2526 	REG("smaps",      S_IRUGO, proc_smaps_operations),
2527 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2528 #endif
2529 #ifdef CONFIG_SECURITY
2530 	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2531 #endif
2532 #ifdef CONFIG_KALLSYMS
2533 	INF("wchan",      S_IRUGO, proc_pid_wchan),
2534 #endif
2535 #ifdef CONFIG_STACKTRACE
2536 	ONE("stack",      S_IRUSR, proc_pid_stack),
2537 #endif
2538 #ifdef CONFIG_SCHEDSTATS
2539 	INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2540 #endif
2541 #ifdef CONFIG_LATENCYTOP
2542 	REG("latency",  S_IRUGO, proc_lstats_operations),
2543 #endif
2544 #ifdef CONFIG_PROC_PID_CPUSET
2545 	REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2546 #endif
2547 #ifdef CONFIG_CGROUPS
2548 	REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2549 #endif
2550 	INF("oom_score",  S_IRUGO, proc_oom_score),
2551 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2552 #ifdef CONFIG_AUDITSYSCALL
2553 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2554 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2555 #endif
2556 #ifdef CONFIG_FAULT_INJECTION
2557 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2558 #endif
2559 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2560 	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2561 #endif
2562 #ifdef CONFIG_TASK_IO_ACCOUNTING
2563 	INF("io",	S_IRUGO, proc_tgid_io_accounting),
2564 #endif
2565 };
2566 
2567 static int proc_tgid_base_readdir(struct file * filp,
2568 			     void * dirent, filldir_t filldir)
2569 {
2570 	return proc_pident_readdir(filp,dirent,filldir,
2571 				   tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2572 }
2573 
2574 static const struct file_operations proc_tgid_base_operations = {
2575 	.read		= generic_read_dir,
2576 	.readdir	= proc_tgid_base_readdir,
2577 };
2578 
2579 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2580 	return proc_pident_lookup(dir, dentry,
2581 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2582 }
2583 
2584 static const struct inode_operations proc_tgid_base_inode_operations = {
2585 	.lookup		= proc_tgid_base_lookup,
2586 	.getattr	= pid_getattr,
2587 	.setattr	= proc_setattr,
2588 };
2589 
2590 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2591 {
2592 	struct dentry *dentry, *leader, *dir;
2593 	char buf[PROC_NUMBUF];
2594 	struct qstr name;
2595 
2596 	name.name = buf;
2597 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2598 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2599 	if (dentry) {
2600 		if (!(current->flags & PF_EXITING))
2601 			shrink_dcache_parent(dentry);
2602 		d_drop(dentry);
2603 		dput(dentry);
2604 	}
2605 
2606 	if (tgid == 0)
2607 		goto out;
2608 
2609 	name.name = buf;
2610 	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2611 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2612 	if (!leader)
2613 		goto out;
2614 
2615 	name.name = "task";
2616 	name.len = strlen(name.name);
2617 	dir = d_hash_and_lookup(leader, &name);
2618 	if (!dir)
2619 		goto out_put_leader;
2620 
2621 	name.name = buf;
2622 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2623 	dentry = d_hash_and_lookup(dir, &name);
2624 	if (dentry) {
2625 		shrink_dcache_parent(dentry);
2626 		d_drop(dentry);
2627 		dput(dentry);
2628 	}
2629 
2630 	dput(dir);
2631 out_put_leader:
2632 	dput(leader);
2633 out:
2634 	return;
2635 }
2636 
2637 /**
2638  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2639  * @task: task that should be flushed.
2640  *
2641  * When flushing dentries from proc, one needs to flush them from global
2642  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2643  * in. This call is supposed to do all of this job.
2644  *
2645  * Looks in the dcache for
2646  * /proc/@pid
2647  * /proc/@tgid/task/@pid
2648  * if either directory is present flushes it and all of it'ts children
2649  * from the dcache.
2650  *
2651  * It is safe and reasonable to cache /proc entries for a task until
2652  * that task exits.  After that they just clog up the dcache with
2653  * useless entries, possibly causing useful dcache entries to be
2654  * flushed instead.  This routine is proved to flush those useless
2655  * dcache entries at process exit time.
2656  *
2657  * NOTE: This routine is just an optimization so it does not guarantee
2658  *       that no dcache entries will exist at process exit time it
2659  *       just makes it very unlikely that any will persist.
2660  */
2661 
2662 void proc_flush_task(struct task_struct *task)
2663 {
2664 	int i;
2665 	struct pid *pid, *tgid = NULL;
2666 	struct upid *upid;
2667 
2668 	pid = task_pid(task);
2669 	if (thread_group_leader(task))
2670 		tgid = task_tgid(task);
2671 
2672 	for (i = 0; i <= pid->level; i++) {
2673 		upid = &pid->numbers[i];
2674 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2675 			tgid ? tgid->numbers[i].nr : 0);
2676 	}
2677 
2678 	upid = &pid->numbers[pid->level];
2679 	if (upid->nr == 1)
2680 		pid_ns_release_proc(upid->ns);
2681 }
2682 
2683 static struct dentry *proc_pid_instantiate(struct inode *dir,
2684 					   struct dentry * dentry,
2685 					   struct task_struct *task, const void *ptr)
2686 {
2687 	struct dentry *error = ERR_PTR(-ENOENT);
2688 	struct inode *inode;
2689 
2690 	inode = proc_pid_make_inode(dir->i_sb, task);
2691 	if (!inode)
2692 		goto out;
2693 
2694 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2695 	inode->i_op = &proc_tgid_base_inode_operations;
2696 	inode->i_fop = &proc_tgid_base_operations;
2697 	inode->i_flags|=S_IMMUTABLE;
2698 
2699 	inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2700 		ARRAY_SIZE(tgid_base_stuff));
2701 
2702 	dentry->d_op = &pid_dentry_operations;
2703 
2704 	d_add(dentry, inode);
2705 	/* Close the race of the process dying before we return the dentry */
2706 	if (pid_revalidate(dentry, NULL))
2707 		error = NULL;
2708 out:
2709 	return error;
2710 }
2711 
2712 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2713 {
2714 	struct dentry *result = ERR_PTR(-ENOENT);
2715 	struct task_struct *task;
2716 	unsigned tgid;
2717 	struct pid_namespace *ns;
2718 
2719 	result = proc_base_lookup(dir, dentry);
2720 	if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2721 		goto out;
2722 
2723 	tgid = name_to_int(dentry);
2724 	if (tgid == ~0U)
2725 		goto out;
2726 
2727 	ns = dentry->d_sb->s_fs_info;
2728 	rcu_read_lock();
2729 	task = find_task_by_pid_ns(tgid, ns);
2730 	if (task)
2731 		get_task_struct(task);
2732 	rcu_read_unlock();
2733 	if (!task)
2734 		goto out;
2735 
2736 	result = proc_pid_instantiate(dir, dentry, task, NULL);
2737 	put_task_struct(task);
2738 out:
2739 	return result;
2740 }
2741 
2742 /*
2743  * Find the first task with tgid >= tgid
2744  *
2745  */
2746 struct tgid_iter {
2747 	unsigned int tgid;
2748 	struct task_struct *task;
2749 };
2750 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2751 {
2752 	struct pid *pid;
2753 
2754 	if (iter.task)
2755 		put_task_struct(iter.task);
2756 	rcu_read_lock();
2757 retry:
2758 	iter.task = NULL;
2759 	pid = find_ge_pid(iter.tgid, ns);
2760 	if (pid) {
2761 		iter.tgid = pid_nr_ns(pid, ns);
2762 		iter.task = pid_task(pid, PIDTYPE_PID);
2763 		/* What we to know is if the pid we have find is the
2764 		 * pid of a thread_group_leader.  Testing for task
2765 		 * being a thread_group_leader is the obvious thing
2766 		 * todo but there is a window when it fails, due to
2767 		 * the pid transfer logic in de_thread.
2768 		 *
2769 		 * So we perform the straight forward test of seeing
2770 		 * if the pid we have found is the pid of a thread
2771 		 * group leader, and don't worry if the task we have
2772 		 * found doesn't happen to be a thread group leader.
2773 		 * As we don't care in the case of readdir.
2774 		 */
2775 		if (!iter.task || !has_group_leader_pid(iter.task)) {
2776 			iter.tgid += 1;
2777 			goto retry;
2778 		}
2779 		get_task_struct(iter.task);
2780 	}
2781 	rcu_read_unlock();
2782 	return iter;
2783 }
2784 
2785 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2786 
2787 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2788 	struct tgid_iter iter)
2789 {
2790 	char name[PROC_NUMBUF];
2791 	int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2792 	return proc_fill_cache(filp, dirent, filldir, name, len,
2793 				proc_pid_instantiate, iter.task, NULL);
2794 }
2795 
2796 /* for the /proc/ directory itself, after non-process stuff has been done */
2797 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2798 {
2799 	unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2800 	struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2801 	struct tgid_iter iter;
2802 	struct pid_namespace *ns;
2803 
2804 	if (!reaper)
2805 		goto out_no_task;
2806 
2807 	for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2808 		const struct pid_entry *p = &proc_base_stuff[nr];
2809 		if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2810 			goto out;
2811 	}
2812 
2813 	ns = filp->f_dentry->d_sb->s_fs_info;
2814 	iter.task = NULL;
2815 	iter.tgid = filp->f_pos - TGID_OFFSET;
2816 	for (iter = next_tgid(ns, iter);
2817 	     iter.task;
2818 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
2819 		filp->f_pos = iter.tgid + TGID_OFFSET;
2820 		if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2821 			put_task_struct(iter.task);
2822 			goto out;
2823 		}
2824 	}
2825 	filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2826 out:
2827 	put_task_struct(reaper);
2828 out_no_task:
2829 	return 0;
2830 }
2831 
2832 /*
2833  * Tasks
2834  */
2835 static const struct pid_entry tid_base_stuff[] = {
2836 	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2837 	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2838 	REG("environ",   S_IRUSR, proc_environ_operations),
2839 	INF("auxv",      S_IRUSR, proc_pid_auxv),
2840 	ONE("status",    S_IRUGO, proc_pid_status),
2841 	ONE("personality", S_IRUSR, proc_pid_personality),
2842 	INF("limits",	 S_IRUSR, proc_pid_limits),
2843 #ifdef CONFIG_SCHED_DEBUG
2844 	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2845 #endif
2846 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2847 	INF("syscall",   S_IRUSR, proc_pid_syscall),
2848 #endif
2849 	INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2850 	ONE("stat",      S_IRUGO, proc_tid_stat),
2851 	ONE("statm",     S_IRUGO, proc_pid_statm),
2852 	REG("maps",      S_IRUGO, proc_maps_operations),
2853 #ifdef CONFIG_NUMA
2854 	REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2855 #endif
2856 	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2857 	LNK("cwd",       proc_cwd_link),
2858 	LNK("root",      proc_root_link),
2859 	LNK("exe",       proc_exe_link),
2860 	REG("mounts",    S_IRUGO, proc_mounts_operations),
2861 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2862 #ifdef CONFIG_PROC_PAGE_MONITOR
2863 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2864 	REG("smaps",     S_IRUGO, proc_smaps_operations),
2865 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2866 #endif
2867 #ifdef CONFIG_SECURITY
2868 	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2869 #endif
2870 #ifdef CONFIG_KALLSYMS
2871 	INF("wchan",     S_IRUGO, proc_pid_wchan),
2872 #endif
2873 #ifdef CONFIG_STACKTRACE
2874 	ONE("stack",      S_IRUSR, proc_pid_stack),
2875 #endif
2876 #ifdef CONFIG_SCHEDSTATS
2877 	INF("schedstat", S_IRUGO, proc_pid_schedstat),
2878 #endif
2879 #ifdef CONFIG_LATENCYTOP
2880 	REG("latency",  S_IRUGO, proc_lstats_operations),
2881 #endif
2882 #ifdef CONFIG_PROC_PID_CPUSET
2883 	REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2884 #endif
2885 #ifdef CONFIG_CGROUPS
2886 	REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2887 #endif
2888 	INF("oom_score", S_IRUGO, proc_oom_score),
2889 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2890 #ifdef CONFIG_AUDITSYSCALL
2891 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2892 	REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2893 #endif
2894 #ifdef CONFIG_FAULT_INJECTION
2895 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2896 #endif
2897 #ifdef CONFIG_TASK_IO_ACCOUNTING
2898 	INF("io",	S_IRUGO, proc_tid_io_accounting),
2899 #endif
2900 };
2901 
2902 static int proc_tid_base_readdir(struct file * filp,
2903 			     void * dirent, filldir_t filldir)
2904 {
2905 	return proc_pident_readdir(filp,dirent,filldir,
2906 				   tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2907 }
2908 
2909 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2910 	return proc_pident_lookup(dir, dentry,
2911 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2912 }
2913 
2914 static const struct file_operations proc_tid_base_operations = {
2915 	.read		= generic_read_dir,
2916 	.readdir	= proc_tid_base_readdir,
2917 };
2918 
2919 static const struct inode_operations proc_tid_base_inode_operations = {
2920 	.lookup		= proc_tid_base_lookup,
2921 	.getattr	= pid_getattr,
2922 	.setattr	= proc_setattr,
2923 };
2924 
2925 static struct dentry *proc_task_instantiate(struct inode *dir,
2926 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2927 {
2928 	struct dentry *error = ERR_PTR(-ENOENT);
2929 	struct inode *inode;
2930 	inode = proc_pid_make_inode(dir->i_sb, task);
2931 
2932 	if (!inode)
2933 		goto out;
2934 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2935 	inode->i_op = &proc_tid_base_inode_operations;
2936 	inode->i_fop = &proc_tid_base_operations;
2937 	inode->i_flags|=S_IMMUTABLE;
2938 
2939 	inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2940 		ARRAY_SIZE(tid_base_stuff));
2941 
2942 	dentry->d_op = &pid_dentry_operations;
2943 
2944 	d_add(dentry, inode);
2945 	/* Close the race of the process dying before we return the dentry */
2946 	if (pid_revalidate(dentry, NULL))
2947 		error = NULL;
2948 out:
2949 	return error;
2950 }
2951 
2952 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2953 {
2954 	struct dentry *result = ERR_PTR(-ENOENT);
2955 	struct task_struct *task;
2956 	struct task_struct *leader = get_proc_task(dir);
2957 	unsigned tid;
2958 	struct pid_namespace *ns;
2959 
2960 	if (!leader)
2961 		goto out_no_task;
2962 
2963 	tid = name_to_int(dentry);
2964 	if (tid == ~0U)
2965 		goto out;
2966 
2967 	ns = dentry->d_sb->s_fs_info;
2968 	rcu_read_lock();
2969 	task = find_task_by_pid_ns(tid, ns);
2970 	if (task)
2971 		get_task_struct(task);
2972 	rcu_read_unlock();
2973 	if (!task)
2974 		goto out;
2975 	if (!same_thread_group(leader, task))
2976 		goto out_drop_task;
2977 
2978 	result = proc_task_instantiate(dir, dentry, task, NULL);
2979 out_drop_task:
2980 	put_task_struct(task);
2981 out:
2982 	put_task_struct(leader);
2983 out_no_task:
2984 	return result;
2985 }
2986 
2987 /*
2988  * Find the first tid of a thread group to return to user space.
2989  *
2990  * Usually this is just the thread group leader, but if the users
2991  * buffer was too small or there was a seek into the middle of the
2992  * directory we have more work todo.
2993  *
2994  * In the case of a short read we start with find_task_by_pid.
2995  *
2996  * In the case of a seek we start with the leader and walk nr
2997  * threads past it.
2998  */
2999 static struct task_struct *first_tid(struct task_struct *leader,
3000 		int tid, int nr, struct pid_namespace *ns)
3001 {
3002 	struct task_struct *pos;
3003 
3004 	rcu_read_lock();
3005 	/* Attempt to start with the pid of a thread */
3006 	if (tid && (nr > 0)) {
3007 		pos = find_task_by_pid_ns(tid, ns);
3008 		if (pos && (pos->group_leader == leader))
3009 			goto found;
3010 	}
3011 
3012 	/* If nr exceeds the number of threads there is nothing todo */
3013 	pos = NULL;
3014 	if (nr && nr >= get_nr_threads(leader))
3015 		goto out;
3016 
3017 	/* If we haven't found our starting place yet start
3018 	 * with the leader and walk nr threads forward.
3019 	 */
3020 	for (pos = leader; nr > 0; --nr) {
3021 		pos = next_thread(pos);
3022 		if (pos == leader) {
3023 			pos = NULL;
3024 			goto out;
3025 		}
3026 	}
3027 found:
3028 	get_task_struct(pos);
3029 out:
3030 	rcu_read_unlock();
3031 	return pos;
3032 }
3033 
3034 /*
3035  * Find the next thread in the thread list.
3036  * Return NULL if there is an error or no next thread.
3037  *
3038  * The reference to the input task_struct is released.
3039  */
3040 static struct task_struct *next_tid(struct task_struct *start)
3041 {
3042 	struct task_struct *pos = NULL;
3043 	rcu_read_lock();
3044 	if (pid_alive(start)) {
3045 		pos = next_thread(start);
3046 		if (thread_group_leader(pos))
3047 			pos = NULL;
3048 		else
3049 			get_task_struct(pos);
3050 	}
3051 	rcu_read_unlock();
3052 	put_task_struct(start);
3053 	return pos;
3054 }
3055 
3056 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3057 	struct task_struct *task, int tid)
3058 {
3059 	char name[PROC_NUMBUF];
3060 	int len = snprintf(name, sizeof(name), "%d", tid);
3061 	return proc_fill_cache(filp, dirent, filldir, name, len,
3062 				proc_task_instantiate, task, NULL);
3063 }
3064 
3065 /* for the /proc/TGID/task/ directories */
3066 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3067 {
3068 	struct dentry *dentry = filp->f_path.dentry;
3069 	struct inode *inode = dentry->d_inode;
3070 	struct task_struct *leader = NULL;
3071 	struct task_struct *task;
3072 	int retval = -ENOENT;
3073 	ino_t ino;
3074 	int tid;
3075 	unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
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 (pos) {
3093 	case 0:
3094 		ino = inode->i_ino;
3095 		if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
3096 			goto out;
3097 		pos++;
3098 		/* fall through */
3099 	case 1:
3100 		ino = parent_ino(dentry);
3101 		if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
3102 			goto out;
3103 		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, pos - 2, ns);
3114 	     task;
3115 	     task = next_tid(task), 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 	filp->f_pos = pos;
3127 	put_task_struct(leader);
3128 out_no_task:
3129 	return retval;
3130 }
3131 
3132 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3133 {
3134 	struct inode *inode = dentry->d_inode;
3135 	struct task_struct *p = get_proc_task(inode);
3136 	generic_fillattr(inode, stat);
3137 
3138 	if (p) {
3139 		stat->nlink += get_nr_threads(p);
3140 		put_task_struct(p);
3141 	}
3142 
3143 	return 0;
3144 }
3145 
3146 static const struct inode_operations proc_task_inode_operations = {
3147 	.lookup		= proc_task_lookup,
3148 	.getattr	= proc_task_getattr,
3149 	.setattr	= proc_setattr,
3150 };
3151 
3152 static const struct file_operations proc_task_operations = {
3153 	.read		= generic_read_dir,
3154 	.readdir	= proc_task_readdir,
3155 };
3156