1/*
2 * proc/fs/generic.c --- generic routines for the proc-fs
3 *
4 * This file contains generic proc-fs routines for handling
5 * directories and files.
6 *
7 * Copyright (C) 1991, 1992 Linus Torvalds.
8 * Copyright (C) 1997 Theodore Ts'o
9 */
10
11#include <linux/errno.h>
12#include <linux/time.h>
13#include <linux/proc_fs.h>
14#include <linux/stat.h>
15#include <linux/mm.h>
16#include <linux/module.h>
17#include <linux/slab.h>
18#include <linux/printk.h>
19#include <linux/mount.h>
20#include <linux/init.h>
21#include <linux/idr.h>
22#include <linux/bitops.h>
23#include <linux/spinlock.h>
24#include <linux/completion.h>
25#include <asm/uaccess.h>
26
27#include "internal.h"
28
29static DEFINE_SPINLOCK(proc_subdir_lock);
30
31static int proc_match(unsigned int len, const char *name, struct proc_dir_entry *de)
32{
33	if (len < de->namelen)
34		return -1;
35	if (len > de->namelen)
36		return 1;
37
38	return memcmp(name, de->name, len);
39}
40
41static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir)
42{
43	return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry,
44			     subdir_node);
45}
46
47static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir)
48{
49	return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry,
50			     subdir_node);
51}
52
53static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir,
54					      const char *name,
55					      unsigned int len)
56{
57	struct rb_node *node = dir->subdir.rb_node;
58
59	while (node) {
60		struct proc_dir_entry *de = container_of(node,
61							 struct proc_dir_entry,
62							 subdir_node);
63		int result = proc_match(len, name, de);
64
65		if (result < 0)
66			node = node->rb_left;
67		else if (result > 0)
68			node = node->rb_right;
69		else
70			return de;
71	}
72	return NULL;
73}
74
75static bool pde_subdir_insert(struct proc_dir_entry *dir,
76			      struct proc_dir_entry *de)
77{
78	struct rb_root *root = &dir->subdir;
79	struct rb_node **new = &root->rb_node, *parent = NULL;
80
81	/* Figure out where to put new node */
82	while (*new) {
83		struct proc_dir_entry *this =
84			container_of(*new, struct proc_dir_entry, subdir_node);
85		int result = proc_match(de->namelen, de->name, this);
86
87		parent = *new;
88		if (result < 0)
89			new = &(*new)->rb_left;
90		else if (result > 0)
91			new = &(*new)->rb_right;
92		else
93			return false;
94	}
95
96	/* Add new node and rebalance tree. */
97	rb_link_node(&de->subdir_node, parent, new);
98	rb_insert_color(&de->subdir_node, root);
99	return true;
100}
101
102static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
103{
104	struct inode *inode = d_inode(dentry);
105	struct proc_dir_entry *de = PDE(inode);
106	int error;
107
108	error = inode_change_ok(inode, iattr);
109	if (error)
110		return error;
111
112	setattr_copy(inode, iattr);
113	mark_inode_dirty(inode);
114
115	proc_set_user(de, inode->i_uid, inode->i_gid);
116	de->mode = inode->i_mode;
117	return 0;
118}
119
120static int proc_getattr(struct vfsmount *mnt, struct dentry *dentry,
121			struct kstat *stat)
122{
123	struct inode *inode = d_inode(dentry);
124	struct proc_dir_entry *de = PDE(inode);
125	if (de && de->nlink)
126		set_nlink(inode, de->nlink);
127
128	generic_fillattr(inode, stat);
129	return 0;
130}
131
132static const struct inode_operations proc_file_inode_operations = {
133	.setattr	= proc_notify_change,
134};
135
136/*
137 * This function parses a name such as "tty/driver/serial", and
138 * returns the struct proc_dir_entry for "/proc/tty/driver", and
139 * returns "serial" in residual.
140 */
141static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
142			     const char **residual)
143{
144	const char     		*cp = name, *next;
145	struct proc_dir_entry	*de;
146	unsigned int		len;
147
148	de = *ret;
149	if (!de)
150		de = &proc_root;
151
152	while (1) {
153		next = strchr(cp, '/');
154		if (!next)
155			break;
156
157		len = next - cp;
158		de = pde_subdir_find(de, cp, len);
159		if (!de) {
160			WARN(1, "name '%s'\n", name);
161			return -ENOENT;
162		}
163		cp += len + 1;
164	}
165	*residual = cp;
166	*ret = de;
167	return 0;
168}
169
170static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
171			   const char **residual)
172{
173	int rv;
174
175	spin_lock(&proc_subdir_lock);
176	rv = __xlate_proc_name(name, ret, residual);
177	spin_unlock(&proc_subdir_lock);
178	return rv;
179}
180
181static DEFINE_IDA(proc_inum_ida);
182static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */
183
184#define PROC_DYNAMIC_FIRST 0xF0000000U
185
186/*
187 * Return an inode number between PROC_DYNAMIC_FIRST and
188 * 0xffffffff, or zero on failure.
189 */
190int proc_alloc_inum(unsigned int *inum)
191{
192	unsigned int i;
193	int error;
194
195retry:
196	if (!ida_pre_get(&proc_inum_ida, GFP_KERNEL))
197		return -ENOMEM;
198
199	spin_lock_irq(&proc_inum_lock);
200	error = ida_get_new(&proc_inum_ida, &i);
201	spin_unlock_irq(&proc_inum_lock);
202	if (error == -EAGAIN)
203		goto retry;
204	else if (error)
205		return error;
206
207	if (i > UINT_MAX - PROC_DYNAMIC_FIRST) {
208		spin_lock_irq(&proc_inum_lock);
209		ida_remove(&proc_inum_ida, i);
210		spin_unlock_irq(&proc_inum_lock);
211		return -ENOSPC;
212	}
213	*inum = PROC_DYNAMIC_FIRST + i;
214	return 0;
215}
216
217void proc_free_inum(unsigned int inum)
218{
219	unsigned long flags;
220	spin_lock_irqsave(&proc_inum_lock, flags);
221	ida_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
222	spin_unlock_irqrestore(&proc_inum_lock, flags);
223}
224
225/*
226 * Don't create negative dentries here, return -ENOENT by hand
227 * instead.
228 */
229struct dentry *proc_lookup_de(struct proc_dir_entry *de, struct inode *dir,
230		struct dentry *dentry)
231{
232	struct inode *inode;
233
234	spin_lock(&proc_subdir_lock);
235	de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len);
236	if (de) {
237		pde_get(de);
238		spin_unlock(&proc_subdir_lock);
239		inode = proc_get_inode(dir->i_sb, de);
240		if (!inode)
241			return ERR_PTR(-ENOMEM);
242		d_set_d_op(dentry, &simple_dentry_operations);
243		d_add(dentry, inode);
244		return NULL;
245	}
246	spin_unlock(&proc_subdir_lock);
247	return ERR_PTR(-ENOENT);
248}
249
250struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
251		unsigned int flags)
252{
253	return proc_lookup_de(PDE(dir), dir, dentry);
254}
255
256/*
257 * This returns non-zero if at EOF, so that the /proc
258 * root directory can use this and check if it should
259 * continue with the <pid> entries..
260 *
261 * Note that the VFS-layer doesn't care about the return
262 * value of the readdir() call, as long as it's non-negative
263 * for success..
264 */
265int proc_readdir_de(struct proc_dir_entry *de, struct file *file,
266		    struct dir_context *ctx)
267{
268	int i;
269
270	if (!dir_emit_dots(file, ctx))
271		return 0;
272
273	spin_lock(&proc_subdir_lock);
274	de = pde_subdir_first(de);
275	i = ctx->pos - 2;
276	for (;;) {
277		if (!de) {
278			spin_unlock(&proc_subdir_lock);
279			return 0;
280		}
281		if (!i)
282			break;
283		de = pde_subdir_next(de);
284		i--;
285	}
286
287	do {
288		struct proc_dir_entry *next;
289		pde_get(de);
290		spin_unlock(&proc_subdir_lock);
291		if (!dir_emit(ctx, de->name, de->namelen,
292			    de->low_ino, de->mode >> 12)) {
293			pde_put(de);
294			return 0;
295		}
296		spin_lock(&proc_subdir_lock);
297		ctx->pos++;
298		next = pde_subdir_next(de);
299		pde_put(de);
300		de = next;
301	} while (de);
302	spin_unlock(&proc_subdir_lock);
303	return 1;
304}
305
306int proc_readdir(struct file *file, struct dir_context *ctx)
307{
308	struct inode *inode = file_inode(file);
309
310	return proc_readdir_de(PDE(inode), file, ctx);
311}
312
313/*
314 * These are the generic /proc directory operations. They
315 * use the in-memory "struct proc_dir_entry" tree to parse
316 * the /proc directory.
317 */
318static const struct file_operations proc_dir_operations = {
319	.llseek			= generic_file_llseek,
320	.read			= generic_read_dir,
321	.iterate		= proc_readdir,
322};
323
324/*
325 * proc directories can do almost nothing..
326 */
327static const struct inode_operations proc_dir_inode_operations = {
328	.lookup		= proc_lookup,
329	.getattr	= proc_getattr,
330	.setattr	= proc_notify_change,
331};
332
333static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
334{
335	int ret;
336
337	ret = proc_alloc_inum(&dp->low_ino);
338	if (ret)
339		return ret;
340
341	spin_lock(&proc_subdir_lock);
342	dp->parent = dir;
343	if (pde_subdir_insert(dir, dp) == false) {
344		WARN(1, "proc_dir_entry '%s/%s' already registered\n",
345		     dir->name, dp->name);
346		spin_unlock(&proc_subdir_lock);
347		proc_free_inum(dp->low_ino);
348		return -EEXIST;
349	}
350	spin_unlock(&proc_subdir_lock);
351
352	return 0;
353}
354
355static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
356					  const char *name,
357					  umode_t mode,
358					  nlink_t nlink)
359{
360	struct proc_dir_entry *ent = NULL;
361	const char *fn;
362	struct qstr qstr;
363
364	if (xlate_proc_name(name, parent, &fn) != 0)
365		goto out;
366	qstr.name = fn;
367	qstr.len = strlen(fn);
368	if (qstr.len == 0 || qstr.len >= 256) {
369		WARN(1, "name len %u\n", qstr.len);
370		return NULL;
371	}
372	if (*parent == &proc_root && name_to_int(&qstr) != ~0U) {
373		WARN(1, "create '/proc/%s' by hand\n", qstr.name);
374		return NULL;
375	}
376	if (is_empty_pde(*parent)) {
377		WARN(1, "attempt to add to permanently empty directory");
378		return NULL;
379	}
380
381	ent = kzalloc(sizeof(struct proc_dir_entry) + qstr.len + 1, GFP_KERNEL);
382	if (!ent)
383		goto out;
384
385	memcpy(ent->name, fn, qstr.len + 1);
386	ent->namelen = qstr.len;
387	ent->mode = mode;
388	ent->nlink = nlink;
389	ent->subdir = RB_ROOT;
390	atomic_set(&ent->count, 1);
391	spin_lock_init(&ent->pde_unload_lock);
392	INIT_LIST_HEAD(&ent->pde_openers);
393out:
394	return ent;
395}
396
397struct proc_dir_entry *proc_symlink(const char *name,
398		struct proc_dir_entry *parent, const char *dest)
399{
400	struct proc_dir_entry *ent;
401
402	ent = __proc_create(&parent, name,
403			  (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
404
405	if (ent) {
406		ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
407		if (ent->data) {
408			strcpy((char*)ent->data,dest);
409			ent->proc_iops = &proc_link_inode_operations;
410			if (proc_register(parent, ent) < 0) {
411				kfree(ent->data);
412				kfree(ent);
413				ent = NULL;
414			}
415		} else {
416			kfree(ent);
417			ent = NULL;
418		}
419	}
420	return ent;
421}
422EXPORT_SYMBOL(proc_symlink);
423
424struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode,
425		struct proc_dir_entry *parent, void *data)
426{
427	struct proc_dir_entry *ent;
428
429	if (mode == 0)
430		mode = S_IRUGO | S_IXUGO;
431
432	ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
433	if (ent) {
434		ent->data = data;
435		ent->proc_fops = &proc_dir_operations;
436		ent->proc_iops = &proc_dir_inode_operations;
437		parent->nlink++;
438		if (proc_register(parent, ent) < 0) {
439			kfree(ent);
440			parent->nlink--;
441			ent = NULL;
442		}
443	}
444	return ent;
445}
446EXPORT_SYMBOL_GPL(proc_mkdir_data);
447
448struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode,
449				       struct proc_dir_entry *parent)
450{
451	return proc_mkdir_data(name, mode, parent, NULL);
452}
453EXPORT_SYMBOL(proc_mkdir_mode);
454
455struct proc_dir_entry *proc_mkdir(const char *name,
456		struct proc_dir_entry *parent)
457{
458	return proc_mkdir_data(name, 0, parent, NULL);
459}
460EXPORT_SYMBOL(proc_mkdir);
461
462struct proc_dir_entry *proc_create_mount_point(const char *name)
463{
464	umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO;
465	struct proc_dir_entry *ent, *parent = NULL;
466
467	ent = __proc_create(&parent, name, mode, 2);
468	if (ent) {
469		ent->data = NULL;
470		ent->proc_fops = NULL;
471		ent->proc_iops = NULL;
472		if (proc_register(parent, ent) < 0) {
473			kfree(ent);
474			parent->nlink--;
475			ent = NULL;
476		}
477	}
478	return ent;
479}
480
481struct proc_dir_entry *proc_create_data(const char *name, umode_t mode,
482					struct proc_dir_entry *parent,
483					const struct file_operations *proc_fops,
484					void *data)
485{
486	struct proc_dir_entry *pde;
487	if ((mode & S_IFMT) == 0)
488		mode |= S_IFREG;
489
490	if (!S_ISREG(mode)) {
491		WARN_ON(1);	/* use proc_mkdir() */
492		return NULL;
493	}
494
495	BUG_ON(proc_fops == NULL);
496
497	if ((mode & S_IALLUGO) == 0)
498		mode |= S_IRUGO;
499	pde = __proc_create(&parent, name, mode, 1);
500	if (!pde)
501		goto out;
502	pde->proc_fops = proc_fops;
503	pde->data = data;
504	pde->proc_iops = &proc_file_inode_operations;
505	if (proc_register(parent, pde) < 0)
506		goto out_free;
507	return pde;
508out_free:
509	kfree(pde);
510out:
511	return NULL;
512}
513EXPORT_SYMBOL(proc_create_data);
514
515void proc_set_size(struct proc_dir_entry *de, loff_t size)
516{
517	de->size = size;
518}
519EXPORT_SYMBOL(proc_set_size);
520
521void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid)
522{
523	de->uid = uid;
524	de->gid = gid;
525}
526EXPORT_SYMBOL(proc_set_user);
527
528static void free_proc_entry(struct proc_dir_entry *de)
529{
530	proc_free_inum(de->low_ino);
531
532	if (S_ISLNK(de->mode))
533		kfree(de->data);
534	kfree(de);
535}
536
537void pde_put(struct proc_dir_entry *pde)
538{
539	if (atomic_dec_and_test(&pde->count))
540		free_proc_entry(pde);
541}
542
543/*
544 * Remove a /proc entry and free it if it's not currently in use.
545 */
546void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
547{
548	struct proc_dir_entry *de = NULL;
549	const char *fn = name;
550	unsigned int len;
551
552	spin_lock(&proc_subdir_lock);
553	if (__xlate_proc_name(name, &parent, &fn) != 0) {
554		spin_unlock(&proc_subdir_lock);
555		return;
556	}
557	len = strlen(fn);
558
559	de = pde_subdir_find(parent, fn, len);
560	if (de)
561		rb_erase(&de->subdir_node, &parent->subdir);
562	spin_unlock(&proc_subdir_lock);
563	if (!de) {
564		WARN(1, "name '%s'\n", name);
565		return;
566	}
567
568	proc_entry_rundown(de);
569
570	if (S_ISDIR(de->mode))
571		parent->nlink--;
572	de->nlink = 0;
573	WARN(pde_subdir_first(de),
574	     "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n",
575	     __func__, de->parent->name, de->name, pde_subdir_first(de)->name);
576	pde_put(de);
577}
578EXPORT_SYMBOL(remove_proc_entry);
579
580int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
581{
582	struct proc_dir_entry *root = NULL, *de, *next;
583	const char *fn = name;
584	unsigned int len;
585
586	spin_lock(&proc_subdir_lock);
587	if (__xlate_proc_name(name, &parent, &fn) != 0) {
588		spin_unlock(&proc_subdir_lock);
589		return -ENOENT;
590	}
591	len = strlen(fn);
592
593	root = pde_subdir_find(parent, fn, len);
594	if (!root) {
595		spin_unlock(&proc_subdir_lock);
596		return -ENOENT;
597	}
598	rb_erase(&root->subdir_node, &parent->subdir);
599
600	de = root;
601	while (1) {
602		next = pde_subdir_first(de);
603		if (next) {
604			rb_erase(&next->subdir_node, &de->subdir);
605			de = next;
606			continue;
607		}
608		spin_unlock(&proc_subdir_lock);
609
610		proc_entry_rundown(de);
611		next = de->parent;
612		if (S_ISDIR(de->mode))
613			next->nlink--;
614		de->nlink = 0;
615		if (de == root)
616			break;
617		pde_put(de);
618
619		spin_lock(&proc_subdir_lock);
620		de = next;
621	}
622	pde_put(root);
623	return 0;
624}
625EXPORT_SYMBOL(remove_proc_subtree);
626
627void *proc_get_parent_data(const struct inode *inode)
628{
629	struct proc_dir_entry *de = PDE(inode);
630	return de->parent->data;
631}
632EXPORT_SYMBOL_GPL(proc_get_parent_data);
633
634void proc_remove(struct proc_dir_entry *de)
635{
636	if (de)
637		remove_proc_subtree(de->name, de->parent);
638}
639EXPORT_SYMBOL(proc_remove);
640
641void *PDE_DATA(const struct inode *inode)
642{
643	return __PDE_DATA(inode);
644}
645EXPORT_SYMBOL(PDE_DATA);
646