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 DEFINE_SPINLOCK(proc_subdir_lock);
30 static int proc_match(int len, const char *name, struct proc_dir_entry *de)
32 if (de->namelen != len)
34 return !memcmp(name, de->name, len);
37 /* buffer size is one page but our output routines use some slack for overruns */
38 #define PROC_BLOCK_SIZE (PAGE_SIZE - 1024)
41 proc_file_read(struct file *file, char __user *buf, size_t nbytes,
44 struct inode * inode = file->f_path.dentry->d_inode;
50 struct proc_dir_entry * dp;
51 unsigned long long pos;
54 * Gaah, please just use "seq_file" instead. The legacy /proc
55 * interfaces cut loff_t down to off_t for reads, and ignore
56 * the offset entirely for writes..
59 if (pos > MAX_NON_LFS)
61 if (nbytes > MAX_NON_LFS - pos)
62 nbytes = MAX_NON_LFS - pos;
65 if (!(page = (char*) __get_free_page(GFP_TEMPORARY)))
68 while ((nbytes > 0) && !eof) {
69 count = min_t(size_t, PROC_BLOCK_SIZE, nbytes);
74 * How to be a proc read function
75 * ------------------------------
77 * int f(char *buffer, char **start, off_t offset,
78 * int count, int *peof, void *dat)
80 * Assume that the buffer is "count" bytes in size.
82 * If you know you have supplied all the data you
85 * You have three ways to return data:
86 * 0) Leave *start = NULL. (This is the default.)
87 * Put the data of the requested offset at that
88 * offset within the buffer. Return the number (n)
89 * of bytes there are from the beginning of the
90 * buffer up to the last byte of data. If the
91 * number of supplied bytes (= n - offset) is
92 * greater than zero and you didn't signal eof
93 * and the reader is prepared to take more data
94 * you will be called again with the requested
95 * offset advanced by the number of bytes
96 * absorbed. This interface is useful for files
97 * no larger than the buffer.
98 * 1) Set *start = an unsigned long value less than
99 * the buffer address but greater than zero.
100 * Put the data of the requested offset at the
101 * beginning of the buffer. Return the number of
102 * bytes of data placed there. If this number is
103 * greater than zero and you didn't signal eof
104 * and the reader is prepared to take more data
105 * you will be called again with the requested
106 * offset advanced by *start. This interface is
107 * useful when you have a large file consisting
108 * of a series of blocks which you want to count
109 * and return as wholes.
110 * (Hack by Paul.Russell@rustcorp.com.au)
111 * 2) Set *start = an address within the buffer.
112 * Put the data of the requested offset at *start.
113 * Return the number of bytes of data placed there.
114 * If this number is greater than zero and you
115 * didn't signal eof and the reader is prepared to
116 * take more data you will be called again with the
117 * requested offset advanced by the number of bytes
120 n = dp->read_proc(page, &start, *ppos,
121 count, &eof, dp->data);
125 if (n == 0) /* end of file */
127 if (n < 0) { /* error */
136 "proc_file_read: Apparent buffer overflow!\n");
144 start = page + *ppos;
145 } else if (start < page) {
148 "proc_file_read: Apparent buffer overflow!\n");
153 * Don't reduce n because doing so might
154 * cut off part of a data block.
157 "proc_file_read: Read count exceeded\n");
159 } else /* start >= page */ {
160 unsigned long startoff = (unsigned long)(start - page);
161 if (n > (PAGE_SIZE - startoff)) {
163 "proc_file_read: Apparent buffer overflow!\n");
164 n = PAGE_SIZE - startoff;
170 n -= copy_to_user(buf, start < page ? page : start, n);
177 *ppos += start < page ? (unsigned long)start : n;
182 free_page((unsigned long) page);
187 proc_file_write(struct file *file, const char __user *buffer,
188 size_t count, loff_t *ppos)
190 struct inode *inode = file->f_path.dentry->d_inode;
191 struct proc_dir_entry * dp;
198 /* FIXME: does this routine need ppos? probably... */
199 return dp->write_proc(file, buffer, count, dp->data);
204 proc_file_lseek(struct file *file, loff_t offset, int orig)
206 loff_t retval = -EINVAL;
209 offset += file->f_pos;
212 if (offset < 0 || offset > MAX_NON_LFS)
214 file->f_pos = retval = offset;
219 static const struct file_operations proc_file_operations = {
220 .llseek = proc_file_lseek,
221 .read = proc_file_read,
222 .write = proc_file_write,
225 static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
227 struct inode *inode = dentry->d_inode;
228 struct proc_dir_entry *de = PDE(inode);
231 error = inode_change_ok(inode, iattr);
235 error = inode_setattr(inode, iattr);
239 de->uid = inode->i_uid;
240 de->gid = inode->i_gid;
241 de->mode = inode->i_mode;
246 static int proc_getattr(struct vfsmount *mnt, struct dentry *dentry,
249 struct inode *inode = dentry->d_inode;
250 struct proc_dir_entry *de = PROC_I(inode)->pde;
252 inode->i_nlink = de->nlink;
254 generic_fillattr(inode, stat);
258 static const struct inode_operations proc_file_inode_operations = {
259 .setattr = proc_notify_change,
263 * This function parses a name such as "tty/driver/serial", and
264 * returns the struct proc_dir_entry for "/proc/tty/driver", and
265 * returns "serial" in residual.
267 static int xlate_proc_name(const char *name,
268 struct proc_dir_entry **ret, const char **residual)
270 const char *cp = name, *next;
271 struct proc_dir_entry *de;
279 spin_lock(&proc_subdir_lock);
281 next = strchr(cp, '/');
286 for (de = de->subdir; de ; de = de->next) {
287 if (proc_match(len, cp, de))
299 spin_unlock(&proc_subdir_lock);
303 static DEFINE_IDA(proc_inum_ida);
304 static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */
306 #define PROC_DYNAMIC_FIRST 0xF0000000U
309 * Return an inode number between PROC_DYNAMIC_FIRST and
310 * 0xffffffff, or zero on failure.
312 static unsigned int get_inode_number(void)
318 if (ida_pre_get(&proc_inum_ida, GFP_KERNEL) == 0)
321 spin_lock(&proc_inum_lock);
322 error = ida_get_new(&proc_inum_ida, &i);
323 spin_unlock(&proc_inum_lock);
324 if (error == -EAGAIN)
329 if (i > UINT_MAX - PROC_DYNAMIC_FIRST) {
330 spin_lock(&proc_inum_lock);
331 ida_remove(&proc_inum_ida, i);
332 spin_unlock(&proc_inum_lock);
335 return PROC_DYNAMIC_FIRST + i;
338 static void release_inode_number(unsigned int inum)
340 spin_lock(&proc_inum_lock);
341 ida_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
342 spin_unlock(&proc_inum_lock);
345 static void *proc_follow_link(struct dentry *dentry, struct nameidata *nd)
347 nd_set_link(nd, PDE(dentry->d_inode)->data);
351 static const struct inode_operations proc_link_inode_operations = {
352 .readlink = generic_readlink,
353 .follow_link = proc_follow_link,
357 * As some entries in /proc are volatile, we want to
358 * get rid of unused dentries. This could be made
359 * smarter: we could keep a "volatile" flag in the
360 * inode to indicate which ones to keep.
362 static int proc_delete_dentry(struct dentry * dentry)
367 static struct dentry_operations proc_dentry_operations =
369 .d_delete = proc_delete_dentry,
373 * Don't create negative dentries here, return -ENOENT by hand
376 struct dentry *proc_lookup_de(struct proc_dir_entry *de, struct inode *dir,
377 struct dentry *dentry)
379 struct inode *inode = NULL;
383 spin_lock(&proc_subdir_lock);
384 for (de = de->subdir; de ; de = de->next) {
385 if (de->namelen != dentry->d_name.len)
387 if (!memcmp(dentry->d_name.name, de->name, de->namelen)) {
392 spin_unlock(&proc_subdir_lock);
394 inode = proc_get_inode(dir->i_sb, ino, de);
398 spin_unlock(&proc_subdir_lock);
403 dentry->d_op = &proc_dentry_operations;
404 d_add(dentry, inode);
409 return ERR_PTR(error);
412 struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
413 struct nameidata *nd)
415 return proc_lookup_de(PDE(dir), dir, dentry);
419 * This returns non-zero if at EOF, so that the /proc
420 * root directory can use this and check if it should
421 * continue with the <pid> entries..
423 * Note that the VFS-layer doesn't care about the return
424 * value of the readdir() call, as long as it's non-negative
427 int proc_readdir_de(struct proc_dir_entry *de, struct file *filp, void *dirent,
432 struct inode *inode = filp->f_path.dentry->d_inode;
441 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
447 if (filldir(dirent, "..", 2, i,
448 parent_ino(filp->f_path.dentry),
455 spin_lock(&proc_subdir_lock);
461 spin_unlock(&proc_subdir_lock);
471 struct proc_dir_entry *next;
473 /* filldir passes info to user space */
475 spin_unlock(&proc_subdir_lock);
476 if (filldir(dirent, de->name, de->namelen, filp->f_pos,
477 de->low_ino, de->mode >> 12) < 0) {
481 spin_lock(&proc_subdir_lock);
487 spin_unlock(&proc_subdir_lock);
490 out: unlock_kernel();
494 int proc_readdir(struct file *filp, void *dirent, filldir_t filldir)
496 struct inode *inode = filp->f_path.dentry->d_inode;
498 return proc_readdir_de(PDE(inode), filp, dirent, filldir);
502 * These are the generic /proc directory operations. They
503 * use the in-memory "struct proc_dir_entry" tree to parse
504 * the /proc directory.
506 static const struct file_operations proc_dir_operations = {
507 .read = generic_read_dir,
508 .readdir = proc_readdir,
512 * proc directories can do almost nothing..
514 static const struct inode_operations proc_dir_inode_operations = {
515 .lookup = proc_lookup,
516 .getattr = proc_getattr,
517 .setattr = proc_notify_change,
520 static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
523 struct proc_dir_entry *tmp;
525 i = get_inode_number();
530 if (S_ISDIR(dp->mode)) {
531 if (dp->proc_iops == NULL) {
532 dp->proc_fops = &proc_dir_operations;
533 dp->proc_iops = &proc_dir_inode_operations;
536 } else if (S_ISLNK(dp->mode)) {
537 if (dp->proc_iops == NULL)
538 dp->proc_iops = &proc_link_inode_operations;
539 } else if (S_ISREG(dp->mode)) {
540 if (dp->proc_fops == NULL)
541 dp->proc_fops = &proc_file_operations;
542 if (dp->proc_iops == NULL)
543 dp->proc_iops = &proc_file_inode_operations;
546 spin_lock(&proc_subdir_lock);
548 for (tmp = dir->subdir; tmp; tmp = tmp->next)
549 if (strcmp(tmp->name, dp->name) == 0) {
550 WARN(1, KERN_WARNING "proc_dir_entry '%s/%s' already registered\n",
551 dir->name, dp->name);
555 dp->next = dir->subdir;
558 spin_unlock(&proc_subdir_lock);
563 static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
568 struct proc_dir_entry *ent = NULL;
569 const char *fn = name;
572 /* make sure name is valid */
573 if (!name || !strlen(name)) goto out;
575 if (xlate_proc_name(name, parent, &fn) != 0)
578 /* At this point there must not be any '/' characters beyond *fn */
584 ent = kmalloc(sizeof(struct proc_dir_entry) + len + 1, GFP_KERNEL);
587 memset(ent, 0, sizeof(struct proc_dir_entry));
588 memcpy(((char *) ent) + sizeof(struct proc_dir_entry), fn, len + 1);
589 ent->name = ((char *) ent) + sizeof(*ent);
593 atomic_set(&ent->count, 1);
595 spin_lock_init(&ent->pde_unload_lock);
596 ent->pde_unload_completion = NULL;
597 INIT_LIST_HEAD(&ent->pde_openers);
602 struct proc_dir_entry *proc_symlink(const char *name,
603 struct proc_dir_entry *parent, const char *dest)
605 struct proc_dir_entry *ent;
607 ent = __proc_create(&parent, name,
608 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
611 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
613 strcpy((char*)ent->data,dest);
614 if (proc_register(parent, ent) < 0) {
627 struct proc_dir_entry *proc_mkdir_mode(const char *name, mode_t mode,
628 struct proc_dir_entry *parent)
630 struct proc_dir_entry *ent;
632 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
634 if (proc_register(parent, ent) < 0) {
642 struct proc_dir_entry *proc_net_mkdir(struct net *net, const char *name,
643 struct proc_dir_entry *parent)
645 struct proc_dir_entry *ent;
647 ent = __proc_create(&parent, name, S_IFDIR | S_IRUGO | S_IXUGO, 2);
650 if (proc_register(parent, ent) < 0) {
657 EXPORT_SYMBOL_GPL(proc_net_mkdir);
659 struct proc_dir_entry *proc_mkdir(const char *name,
660 struct proc_dir_entry *parent)
662 return proc_mkdir_mode(name, S_IRUGO | S_IXUGO, parent);
665 struct proc_dir_entry *create_proc_entry(const char *name, mode_t mode,
666 struct proc_dir_entry *parent)
668 struct proc_dir_entry *ent;
672 if ((mode & S_IALLUGO) == 0)
673 mode |= S_IRUGO | S_IXUGO;
676 if ((mode & S_IFMT) == 0)
678 if ((mode & S_IALLUGO) == 0)
683 ent = __proc_create(&parent, name, mode, nlink);
685 if (proc_register(parent, ent) < 0) {
693 struct proc_dir_entry *proc_create_data(const char *name, mode_t mode,
694 struct proc_dir_entry *parent,
695 const struct file_operations *proc_fops,
698 struct proc_dir_entry *pde;
702 if ((mode & S_IALLUGO) == 0)
703 mode |= S_IRUGO | S_IXUGO;
706 if ((mode & S_IFMT) == 0)
708 if ((mode & S_IALLUGO) == 0)
713 pde = __proc_create(&parent, name, mode, nlink);
716 pde->proc_fops = proc_fops;
718 if (proc_register(parent, pde) < 0)
727 void free_proc_entry(struct proc_dir_entry *de)
729 unsigned int ino = de->low_ino;
731 if (ino < PROC_DYNAMIC_FIRST)
734 release_inode_number(ino);
736 if (S_ISLNK(de->mode))
742 * Remove a /proc entry and free it if it's not currently in use.
744 void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
746 struct proc_dir_entry **p;
747 struct proc_dir_entry *de = NULL;
748 const char *fn = name;
751 if (xlate_proc_name(name, &parent, &fn) != 0)
755 spin_lock(&proc_subdir_lock);
756 for (p = &parent->subdir; *p; p=&(*p)->next ) {
757 if (proc_match(len, fn, *p)) {
764 spin_unlock(&proc_subdir_lock);
768 spin_lock(&de->pde_unload_lock);
770 * Stop accepting new callers into module. If you're
771 * dynamically allocating ->proc_fops, save a pointer somewhere.
773 de->proc_fops = NULL;
774 /* Wait until all existing callers into module are done. */
775 if (de->pde_users > 0) {
776 DECLARE_COMPLETION_ONSTACK(c);
778 if (!de->pde_unload_completion)
779 de->pde_unload_completion = &c;
781 spin_unlock(&de->pde_unload_lock);
783 wait_for_completion(de->pde_unload_completion);
785 goto continue_removing;
787 spin_unlock(&de->pde_unload_lock);
790 spin_lock(&de->pde_unload_lock);
791 while (!list_empty(&de->pde_openers)) {
792 struct pde_opener *pdeo;
794 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
796 spin_unlock(&de->pde_unload_lock);
797 pdeo->release(pdeo->inode, pdeo->file);
799 spin_lock(&de->pde_unload_lock);
801 spin_unlock(&de->pde_unload_lock);
803 if (S_ISDIR(de->mode))
806 WARN(de->subdir, KERN_WARNING "%s: removing non-empty directory "
807 "'%s/%s', leaking at least '%s'\n", __func__,
808 de->parent->name, de->name, de->subdir->name);
809 if (atomic_dec_and_test(&de->count))