2 * proc/fs/generic.c --- generic routines for the proc-fs
4 * This file contains generic proc-fs routines for handling
5 * directories and files.
7 * Copyright (C) 1991, 1992 Linus Torvalds.
8 * Copyright (C) 1997 Theodore Ts'o
11 #include <linux/errno.h>
12 #include <linux/time.h>
13 #include <linux/proc_fs.h>
14 #include <linux/stat.h>
15 #include <linux/module.h>
16 #include <linux/mount.h>
17 #include <linux/smp_lock.h>
18 #include <linux/init.h>
19 #include <linux/idr.h>
20 #include <linux/namei.h>
21 #include <linux/bitops.h>
22 #include <linux/spinlock.h>
23 #include <linux/completion.h>
24 #include <asm/uaccess.h>
28 static ssize_t
proc_file_read(struct file
*file
, char __user
*buf
,
29 size_t nbytes
, loff_t
*ppos
);
30 static ssize_t
proc_file_write(struct file
*file
, const char __user
*buffer
,
31 size_t count
, loff_t
*ppos
);
32 static loff_t
proc_file_lseek(struct file
*, loff_t
, int);
34 DEFINE_SPINLOCK(proc_subdir_lock
);
36 static int proc_match(int len
, const char *name
, struct proc_dir_entry
*de
)
38 if (de
->namelen
!= len
)
40 return !memcmp(name
, de
->name
, len
);
43 static const struct file_operations proc_file_operations
= {
44 .llseek
= proc_file_lseek
,
45 .read
= proc_file_read
,
46 .write
= proc_file_write
,
49 /* buffer size is one page but our output routines use some slack for overruns */
50 #define PROC_BLOCK_SIZE (PAGE_SIZE - 1024)
53 proc_file_read(struct file
*file
, char __user
*buf
, size_t nbytes
,
56 struct inode
* inode
= file
->f_path
.dentry
->d_inode
;
62 struct proc_dir_entry
* dp
;
63 unsigned long long pos
;
66 * Gaah, please just use "seq_file" instead. The legacy /proc
67 * interfaces cut loff_t down to off_t for reads, and ignore
68 * the offset entirely for writes..
71 if (pos
> MAX_NON_LFS
)
73 if (nbytes
> MAX_NON_LFS
- pos
)
74 nbytes
= MAX_NON_LFS
- pos
;
77 if (!(page
= (char*) __get_free_page(GFP_KERNEL
)))
80 while ((nbytes
> 0) && !eof
) {
81 count
= min_t(size_t, PROC_BLOCK_SIZE
, nbytes
);
85 /* Handle old net routines */
86 n
= dp
->get_info(page
, &start
, *ppos
, count
);
89 } else if (dp
->read_proc
) {
91 * How to be a proc read function
92 * ------------------------------
94 * int f(char *buffer, char **start, off_t offset,
95 * int count, int *peof, void *dat)
97 * Assume that the buffer is "count" bytes in size.
99 * If you know you have supplied all the data you
102 * You have three ways to return data:
103 * 0) Leave *start = NULL. (This is the default.)
104 * Put the data of the requested offset at that
105 * offset within the buffer. Return the number (n)
106 * of bytes there are from the beginning of the
107 * buffer up to the last byte of data. If the
108 * number of supplied bytes (= n - offset) is
109 * greater than zero and you didn't signal eof
110 * and the reader is prepared to take more data
111 * you will be called again with the requested
112 * offset advanced by the number of bytes
113 * absorbed. This interface is useful for files
114 * no larger than the buffer.
115 * 1) Set *start = an unsigned long value less than
116 * the buffer address but greater than zero.
117 * Put the data of the requested offset at the
118 * beginning of the buffer. Return the number of
119 * bytes of data placed there. If this number is
120 * greater than zero and you didn't signal eof
121 * and the reader is prepared to take more data
122 * you will be called again with the requested
123 * offset advanced by *start. This interface is
124 * useful when you have a large file consisting
125 * of a series of blocks which you want to count
126 * and return as wholes.
127 * (Hack by Paul.Russell@rustcorp.com.au)
128 * 2) Set *start = an address within the buffer.
129 * Put the data of the requested offset at *start.
130 * Return the number of bytes of data placed there.
131 * If this number is greater than zero and you
132 * didn't signal eof and the reader is prepared to
133 * take more data you will be called again with the
134 * requested offset advanced by the number of bytes
137 n
= dp
->read_proc(page
, &start
, *ppos
,
138 count
, &eof
, dp
->data
);
142 if (n
== 0) /* end of file */
144 if (n
< 0) { /* error */
153 "proc_file_read: Apparent buffer overflow!\n");
161 start
= page
+ *ppos
;
162 } else if (start
< page
) {
165 "proc_file_read: Apparent buffer overflow!\n");
170 * Don't reduce n because doing so might
171 * cut off part of a data block.
174 "proc_file_read: Read count exceeded\n");
176 } else /* start >= page */ {
177 unsigned long startoff
= (unsigned long)(start
- page
);
178 if (n
> (PAGE_SIZE
- startoff
)) {
180 "proc_file_read: Apparent buffer overflow!\n");
181 n
= PAGE_SIZE
- startoff
;
187 n
-= copy_to_user(buf
, start
< page
? page
: start
, n
);
194 *ppos
+= start
< page
? (unsigned long)start
: n
;
199 free_page((unsigned long) page
);
204 proc_file_write(struct file
*file
, const char __user
*buffer
,
205 size_t count
, loff_t
*ppos
)
207 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
208 struct proc_dir_entry
* dp
;
215 /* FIXME: does this routine need ppos? probably... */
216 return dp
->write_proc(file
, buffer
, count
, dp
->data
);
221 proc_file_lseek(struct file
*file
, loff_t offset
, int orig
)
223 loff_t retval
= -EINVAL
;
226 offset
+= file
->f_pos
;
229 if (offset
< 0 || offset
> MAX_NON_LFS
)
231 file
->f_pos
= retval
= offset
;
236 static int proc_notify_change(struct dentry
*dentry
, struct iattr
*iattr
)
238 struct inode
*inode
= dentry
->d_inode
;
239 struct proc_dir_entry
*de
= PDE(inode
);
242 error
= inode_change_ok(inode
, iattr
);
246 error
= inode_setattr(inode
, iattr
);
250 de
->uid
= inode
->i_uid
;
251 de
->gid
= inode
->i_gid
;
252 de
->mode
= inode
->i_mode
;
257 static int proc_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
260 struct inode
*inode
= dentry
->d_inode
;
261 struct proc_dir_entry
*de
= PROC_I(inode
)->pde
;
263 inode
->i_nlink
= de
->nlink
;
265 generic_fillattr(inode
, stat
);
269 static const struct inode_operations proc_file_inode_operations
= {
270 .setattr
= proc_notify_change
,
274 * This function parses a name such as "tty/driver/serial", and
275 * returns the struct proc_dir_entry for "/proc/tty/driver", and
276 * returns "serial" in residual.
278 static int xlate_proc_name(const char *name
,
279 struct proc_dir_entry
**ret
, const char **residual
)
281 const char *cp
= name
, *next
;
282 struct proc_dir_entry
*de
;
286 spin_lock(&proc_subdir_lock
);
289 next
= strchr(cp
, '/');
294 for (de
= de
->subdir
; de
; de
= de
->next
) {
295 if (proc_match(len
, cp
, de
))
307 spin_unlock(&proc_subdir_lock
);
311 static DEFINE_IDR(proc_inum_idr
);
312 static DEFINE_SPINLOCK(proc_inum_lock
); /* protects the above */
314 #define PROC_DYNAMIC_FIRST 0xF0000000UL
317 * Return an inode number between PROC_DYNAMIC_FIRST and
318 * 0xffffffff, or zero on failure.
320 static unsigned int get_inode_number(void)
326 if (idr_pre_get(&proc_inum_idr
, GFP_KERNEL
) == 0)
329 spin_lock(&proc_inum_lock
);
330 error
= idr_get_new(&proc_inum_idr
, NULL
, &i
);
331 spin_unlock(&proc_inum_lock
);
332 if (error
== -EAGAIN
)
337 inum
= (i
& MAX_ID_MASK
) + PROC_DYNAMIC_FIRST
;
339 /* inum will never be more than 0xf0ffffff, so no check
346 static void release_inode_number(unsigned int inum
)
348 int id
= (inum
- PROC_DYNAMIC_FIRST
) | ~MAX_ID_MASK
;
350 spin_lock(&proc_inum_lock
);
351 idr_remove(&proc_inum_idr
, id
);
352 spin_unlock(&proc_inum_lock
);
355 static void *proc_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
357 nd_set_link(nd
, PDE(dentry
->d_inode
)->data
);
361 static const struct inode_operations proc_link_inode_operations
= {
362 .readlink
= generic_readlink
,
363 .follow_link
= proc_follow_link
,
367 * As some entries in /proc are volatile, we want to
368 * get rid of unused dentries. This could be made
369 * smarter: we could keep a "volatile" flag in the
370 * inode to indicate which ones to keep.
372 static int proc_delete_dentry(struct dentry
* dentry
)
377 static struct dentry_operations proc_dentry_operations
=
379 .d_delete
= proc_delete_dentry
,
383 * Don't create negative dentries here, return -ENOENT by hand
386 struct dentry
*proc_lookup(struct inode
* dir
, struct dentry
*dentry
, struct nameidata
*nd
)
388 struct inode
*inode
= NULL
;
389 struct proc_dir_entry
* de
;
393 spin_lock(&proc_subdir_lock
);
396 for (de
= de
->subdir
; de
; de
= de
->next
) {
397 if (de
->namelen
!= dentry
->d_name
.len
)
399 if (!memcmp(dentry
->d_name
.name
, de
->name
, de
->namelen
)) {
400 unsigned int ino
= de
->low_ino
;
403 spin_unlock(&proc_subdir_lock
);
405 inode
= proc_get_inode(dir
->i_sb
, ino
, de
);
406 spin_lock(&proc_subdir_lock
);
411 spin_unlock(&proc_subdir_lock
);
415 dentry
->d_op
= &proc_dentry_operations
;
416 d_add(dentry
, inode
);
420 return ERR_PTR(error
);
424 * This returns non-zero if at EOF, so that the /proc
425 * root directory can use this and check if it should
426 * continue with the <pid> entries..
428 * Note that the VFS-layer doesn't care about the return
429 * value of the readdir() call, as long as it's non-negative
432 int proc_readdir(struct file
* filp
,
433 void * dirent
, filldir_t filldir
)
435 struct proc_dir_entry
* de
;
438 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
452 if (filldir(dirent
, ".", 1, i
, ino
, DT_DIR
) < 0)
458 if (filldir(dirent
, "..", 2, i
,
459 parent_ino(filp
->f_path
.dentry
),
466 spin_lock(&proc_subdir_lock
);
472 spin_unlock(&proc_subdir_lock
);
482 struct proc_dir_entry
*next
;
484 /* filldir passes info to user space */
486 spin_unlock(&proc_subdir_lock
);
487 if (filldir(dirent
, de
->name
, de
->namelen
, filp
->f_pos
,
488 de
->low_ino
, de
->mode
>> 12) < 0) {
492 spin_lock(&proc_subdir_lock
);
498 spin_unlock(&proc_subdir_lock
);
501 out
: unlock_kernel();
506 * These are the generic /proc directory operations. They
507 * use the in-memory "struct proc_dir_entry" tree to parse
508 * the /proc directory.
510 static const struct file_operations proc_dir_operations
= {
511 .read
= generic_read_dir
,
512 .readdir
= proc_readdir
,
516 * proc directories can do almost nothing..
518 static const struct inode_operations proc_dir_inode_operations
= {
519 .lookup
= proc_lookup
,
520 .getattr
= proc_getattr
,
521 .setattr
= proc_notify_change
,
524 static int proc_register(struct proc_dir_entry
* dir
, struct proc_dir_entry
* dp
)
528 i
= get_inode_number();
533 if (S_ISDIR(dp
->mode
)) {
534 if (dp
->proc_iops
== NULL
) {
535 dp
->proc_fops
= &proc_dir_operations
;
536 dp
->proc_iops
= &proc_dir_inode_operations
;
539 } else if (S_ISLNK(dp
->mode
)) {
540 if (dp
->proc_iops
== NULL
)
541 dp
->proc_iops
= &proc_link_inode_operations
;
542 } else if (S_ISREG(dp
->mode
)) {
543 if (dp
->proc_fops
== NULL
)
544 dp
->proc_fops
= &proc_file_operations
;
545 if (dp
->proc_iops
== NULL
)
546 dp
->proc_iops
= &proc_file_inode_operations
;
549 spin_lock(&proc_subdir_lock
);
550 dp
->next
= dir
->subdir
;
553 spin_unlock(&proc_subdir_lock
);
559 * Kill an inode that got unregistered..
561 static void proc_kill_inodes(struct proc_dir_entry
*de
)
564 struct super_block
*sb
= proc_mnt
->mnt_sb
;
567 * Actually it's a partial revoke().
570 list_for_each(p
, &sb
->s_files
) {
571 struct file
* filp
= list_entry(p
, struct file
, f_u
.fu_list
);
572 struct dentry
* dentry
= filp
->f_path
.dentry
;
573 struct inode
* inode
;
574 const struct file_operations
*fops
;
576 if (dentry
->d_op
!= &proc_dentry_operations
)
578 inode
= dentry
->d_inode
;
579 if (PDE(inode
) != de
)
588 static struct proc_dir_entry
*proc_create(struct proc_dir_entry
**parent
,
593 struct proc_dir_entry
*ent
= NULL
;
594 const char *fn
= name
;
597 /* make sure name is valid */
598 if (!name
|| !strlen(name
)) goto out
;
600 if (!(*parent
) && xlate_proc_name(name
, parent
, &fn
) != 0)
603 /* At this point there must not be any '/' characters beyond *fn */
609 ent
= kmalloc(sizeof(struct proc_dir_entry
) + len
+ 1, GFP_KERNEL
);
612 memset(ent
, 0, sizeof(struct proc_dir_entry
));
613 memcpy(((char *) ent
) + sizeof(struct proc_dir_entry
), fn
, len
+ 1);
614 ent
->name
= ((char *) ent
) + sizeof(*ent
);
619 spin_lock_init(&ent
->pde_unload_lock
);
620 ent
->pde_unload_completion
= NULL
;
625 struct proc_dir_entry
*proc_symlink(const char *name
,
626 struct proc_dir_entry
*parent
, const char *dest
)
628 struct proc_dir_entry
*ent
;
630 ent
= proc_create(&parent
,name
,
631 (S_IFLNK
| S_IRUGO
| S_IWUGO
| S_IXUGO
),1);
634 ent
->data
= kmalloc((ent
->size
=strlen(dest
))+1, GFP_KERNEL
);
636 strcpy((char*)ent
->data
,dest
);
637 if (proc_register(parent
, ent
) < 0) {
650 struct proc_dir_entry
*proc_mkdir_mode(const char *name
, mode_t mode
,
651 struct proc_dir_entry
*parent
)
653 struct proc_dir_entry
*ent
;
655 ent
= proc_create(&parent
, name
, S_IFDIR
| mode
, 2);
657 if (proc_register(parent
, ent
) < 0) {
665 struct proc_dir_entry
*proc_mkdir(const char *name
,
666 struct proc_dir_entry
*parent
)
668 return proc_mkdir_mode(name
, S_IRUGO
| S_IXUGO
, parent
);
671 struct proc_dir_entry
*create_proc_entry(const char *name
, mode_t mode
,
672 struct proc_dir_entry
*parent
)
674 struct proc_dir_entry
*ent
;
678 if ((mode
& S_IALLUGO
) == 0)
679 mode
|= S_IRUGO
| S_IXUGO
;
682 if ((mode
& S_IFMT
) == 0)
684 if ((mode
& S_IALLUGO
) == 0)
689 ent
= proc_create(&parent
,name
,mode
,nlink
);
691 if (proc_register(parent
, ent
) < 0) {
699 void free_proc_entry(struct proc_dir_entry
*de
)
701 unsigned int ino
= de
->low_ino
;
703 if (ino
< PROC_DYNAMIC_FIRST
)
706 release_inode_number(ino
);
708 if (S_ISLNK(de
->mode
) && de
->data
)
714 * Remove a /proc entry and free it if it's not currently in use.
715 * If it is in use, we set the 'deleted' flag.
717 void remove_proc_entry(const char *name
, struct proc_dir_entry
*parent
)
719 struct proc_dir_entry
**p
;
720 struct proc_dir_entry
*de
;
721 const char *fn
= name
;
724 if (!parent
&& xlate_proc_name(name
, &parent
, &fn
) != 0)
728 spin_lock(&proc_subdir_lock
);
729 for (p
= &parent
->subdir
; *p
; p
=&(*p
)->next
) {
730 if (!proc_match(len
, fn
, *p
))
736 spin_lock(&de
->pde_unload_lock
);
738 * Stop accepting new callers into module. If you're
739 * dynamically allocating ->proc_fops, save a pointer somewhere.
741 de
->proc_fops
= NULL
;
742 /* Wait until all existing callers into module are done. */
743 if (de
->pde_users
> 0) {
744 DECLARE_COMPLETION_ONSTACK(c
);
746 if (!de
->pde_unload_completion
)
747 de
->pde_unload_completion
= &c
;
749 spin_unlock(&de
->pde_unload_lock
);
750 spin_unlock(&proc_subdir_lock
);
752 wait_for_completion(de
->pde_unload_completion
);
754 spin_lock(&proc_subdir_lock
);
755 goto continue_removing
;
757 spin_unlock(&de
->pde_unload_lock
);
760 if (S_ISDIR(de
->mode
))
762 if (!S_ISREG(de
->mode
))
763 proc_kill_inodes(de
);
766 if (!atomic_read(&de
->count
))
770 printk("remove_proc_entry: %s/%s busy, count=%d\n",
771 parent
->name
, de
->name
, atomic_read(&de
->count
));
775 spin_unlock(&proc_subdir_lock
);