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/slab.h>
17 #include <linux/mount.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_read(struct file *file, char __user *buf, size_t nbytes,
190 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
193 spin_lock(&pde->pde_unload_lock);
194 if (!pde->proc_fops) {
195 spin_unlock(&pde->pde_unload_lock);
199 spin_unlock(&pde->pde_unload_lock);
201 rv = __proc_file_read(file, buf, nbytes, ppos);
208 proc_file_write(struct file *file, const char __user *buffer,
209 size_t count, loff_t *ppos)
211 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
214 if (pde->write_proc) {
215 spin_lock(&pde->pde_unload_lock);
216 if (!pde->proc_fops) {
217 spin_unlock(&pde->pde_unload_lock);
221 spin_unlock(&pde->pde_unload_lock);
223 /* FIXME: does this routine need ppos? probably... */
224 rv = pde->write_proc(file, buffer, count, pde->data);
232 proc_file_lseek(struct file *file, loff_t offset, int orig)
234 loff_t retval = -EINVAL;
237 offset += file->f_pos;
240 if (offset < 0 || offset > MAX_NON_LFS)
242 file->f_pos = retval = offset;
247 static const struct file_operations proc_file_operations = {
248 .llseek = proc_file_lseek,
249 .read = proc_file_read,
250 .write = proc_file_write,
253 static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
255 struct inode *inode = dentry->d_inode;
256 struct proc_dir_entry *de = PDE(inode);
259 error = inode_change_ok(inode, iattr);
263 error = inode_setattr(inode, iattr);
267 de->uid = inode->i_uid;
268 de->gid = inode->i_gid;
269 de->mode = inode->i_mode;
274 static int proc_getattr(struct vfsmount *mnt, struct dentry *dentry,
277 struct inode *inode = dentry->d_inode;
278 struct proc_dir_entry *de = PROC_I(inode)->pde;
280 inode->i_nlink = de->nlink;
282 generic_fillattr(inode, stat);
286 static const struct inode_operations proc_file_inode_operations = {
287 .setattr = proc_notify_change,
291 * This function parses a name such as "tty/driver/serial", and
292 * returns the struct proc_dir_entry for "/proc/tty/driver", and
293 * returns "serial" in residual.
295 static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
296 const char **residual)
298 const char *cp = name, *next;
299 struct proc_dir_entry *de;
307 next = strchr(cp, '/');
312 for (de = de->subdir; de ; de = de->next) {
313 if (proc_match(len, cp, de))
317 WARN(1, "name '%s'\n", name);
327 static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
328 const char **residual)
332 spin_lock(&proc_subdir_lock);
333 rv = __xlate_proc_name(name, ret, residual);
334 spin_unlock(&proc_subdir_lock);
338 static DEFINE_IDA(proc_inum_ida);
339 static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */
341 #define PROC_DYNAMIC_FIRST 0xF0000000U
344 * Return an inode number between PROC_DYNAMIC_FIRST and
345 * 0xffffffff, or zero on failure.
347 static unsigned int get_inode_number(void)
353 if (ida_pre_get(&proc_inum_ida, GFP_KERNEL) == 0)
356 spin_lock(&proc_inum_lock);
357 error = ida_get_new(&proc_inum_ida, &i);
358 spin_unlock(&proc_inum_lock);
359 if (error == -EAGAIN)
364 if (i > UINT_MAX - PROC_DYNAMIC_FIRST) {
365 spin_lock(&proc_inum_lock);
366 ida_remove(&proc_inum_ida, i);
367 spin_unlock(&proc_inum_lock);
370 return PROC_DYNAMIC_FIRST + i;
373 static void release_inode_number(unsigned int inum)
375 spin_lock(&proc_inum_lock);
376 ida_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
377 spin_unlock(&proc_inum_lock);
380 static void *proc_follow_link(struct dentry *dentry, struct nameidata *nd)
382 nd_set_link(nd, PDE(dentry->d_inode)->data);
386 static const struct inode_operations proc_link_inode_operations = {
387 .readlink = generic_readlink,
388 .follow_link = proc_follow_link,
392 * As some entries in /proc are volatile, we want to
393 * get rid of unused dentries. This could be made
394 * smarter: we could keep a "volatile" flag in the
395 * inode to indicate which ones to keep.
397 static int proc_delete_dentry(struct dentry * dentry)
402 static const struct dentry_operations proc_dentry_operations =
404 .d_delete = proc_delete_dentry,
408 * Don't create negative dentries here, return -ENOENT by hand
411 struct dentry *proc_lookup_de(struct proc_dir_entry *de, struct inode *dir,
412 struct dentry *dentry)
414 struct inode *inode = NULL;
417 spin_lock(&proc_subdir_lock);
418 for (de = de->subdir; de ; de = de->next) {
419 if (de->namelen != dentry->d_name.len)
421 if (!memcmp(dentry->d_name.name, de->name, de->namelen)) {
426 spin_unlock(&proc_subdir_lock);
428 inode = proc_get_inode(dir->i_sb, ino, de);
432 spin_unlock(&proc_subdir_lock);
436 dentry->d_op = &proc_dentry_operations;
437 d_add(dentry, inode);
442 return ERR_PTR(error);
445 struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
446 struct nameidata *nd)
448 return proc_lookup_de(PDE(dir), dir, dentry);
452 * This returns non-zero if at EOF, so that the /proc
453 * root directory can use this and check if it should
454 * continue with the <pid> entries..
456 * Note that the VFS-layer doesn't care about the return
457 * value of the readdir() call, as long as it's non-negative
460 int proc_readdir_de(struct proc_dir_entry *de, struct file *filp, void *dirent,
465 struct inode *inode = filp->f_path.dentry->d_inode;
472 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
478 if (filldir(dirent, "..", 2, i,
479 parent_ino(filp->f_path.dentry),
486 spin_lock(&proc_subdir_lock);
492 spin_unlock(&proc_subdir_lock);
502 struct proc_dir_entry *next;
504 /* filldir passes info to user space */
506 spin_unlock(&proc_subdir_lock);
507 if (filldir(dirent, de->name, de->namelen, filp->f_pos,
508 de->low_ino, de->mode >> 12) < 0) {
512 spin_lock(&proc_subdir_lock);
518 spin_unlock(&proc_subdir_lock);
525 int proc_readdir(struct file *filp, void *dirent, filldir_t filldir)
527 struct inode *inode = filp->f_path.dentry->d_inode;
529 return proc_readdir_de(PDE(inode), filp, dirent, filldir);
533 * These are the generic /proc directory operations. They
534 * use the in-memory "struct proc_dir_entry" tree to parse
535 * the /proc directory.
537 static const struct file_operations proc_dir_operations = {
538 .llseek = generic_file_llseek,
539 .read = generic_read_dir,
540 .readdir = proc_readdir,
544 * proc directories can do almost nothing..
546 static const struct inode_operations proc_dir_inode_operations = {
547 .lookup = proc_lookup,
548 .getattr = proc_getattr,
549 .setattr = proc_notify_change,
552 static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
555 struct proc_dir_entry *tmp;
557 i = get_inode_number();
562 if (S_ISDIR(dp->mode)) {
563 if (dp->proc_iops == NULL) {
564 dp->proc_fops = &proc_dir_operations;
565 dp->proc_iops = &proc_dir_inode_operations;
568 } else if (S_ISLNK(dp->mode)) {
569 if (dp->proc_iops == NULL)
570 dp->proc_iops = &proc_link_inode_operations;
571 } else if (S_ISREG(dp->mode)) {
572 if (dp->proc_fops == NULL)
573 dp->proc_fops = &proc_file_operations;
574 if (dp->proc_iops == NULL)
575 dp->proc_iops = &proc_file_inode_operations;
578 spin_lock(&proc_subdir_lock);
580 for (tmp = dir->subdir; tmp; tmp = tmp->next)
581 if (strcmp(tmp->name, dp->name) == 0) {
582 WARN(1, KERN_WARNING "proc_dir_entry '%s/%s' already registered\n",
583 dir->name, dp->name);
587 dp->next = dir->subdir;
590 spin_unlock(&proc_subdir_lock);
595 static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
600 struct proc_dir_entry *ent = NULL;
601 const char *fn = name;
604 /* make sure name is valid */
605 if (!name || !strlen(name)) goto out;
607 if (xlate_proc_name(name, parent, &fn) != 0)
610 /* At this point there must not be any '/' characters beyond *fn */
616 ent = kmalloc(sizeof(struct proc_dir_entry) + len + 1, GFP_KERNEL);
619 memset(ent, 0, sizeof(struct proc_dir_entry));
620 memcpy(((char *) ent) + sizeof(struct proc_dir_entry), fn, len + 1);
621 ent->name = ((char *) ent) + sizeof(*ent);
625 atomic_set(&ent->count, 1);
627 spin_lock_init(&ent->pde_unload_lock);
628 ent->pde_unload_completion = NULL;
629 INIT_LIST_HEAD(&ent->pde_openers);
634 struct proc_dir_entry *proc_symlink(const char *name,
635 struct proc_dir_entry *parent, const char *dest)
637 struct proc_dir_entry *ent;
639 ent = __proc_create(&parent, name,
640 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
643 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
645 strcpy((char*)ent->data,dest);
646 if (proc_register(parent, ent) < 0) {
658 EXPORT_SYMBOL(proc_symlink);
660 struct proc_dir_entry *proc_mkdir_mode(const char *name, mode_t mode,
661 struct proc_dir_entry *parent)
663 struct proc_dir_entry *ent;
665 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
667 if (proc_register(parent, ent) < 0) {
675 struct proc_dir_entry *proc_net_mkdir(struct net *net, const char *name,
676 struct proc_dir_entry *parent)
678 struct proc_dir_entry *ent;
680 ent = __proc_create(&parent, name, S_IFDIR | S_IRUGO | S_IXUGO, 2);
683 if (proc_register(parent, ent) < 0) {
690 EXPORT_SYMBOL_GPL(proc_net_mkdir);
692 struct proc_dir_entry *proc_mkdir(const char *name,
693 struct proc_dir_entry *parent)
695 return proc_mkdir_mode(name, S_IRUGO | S_IXUGO, parent);
697 EXPORT_SYMBOL(proc_mkdir);
699 struct proc_dir_entry *create_proc_entry(const char *name, mode_t mode,
700 struct proc_dir_entry *parent)
702 struct proc_dir_entry *ent;
706 if ((mode & S_IALLUGO) == 0)
707 mode |= S_IRUGO | S_IXUGO;
710 if ((mode & S_IFMT) == 0)
712 if ((mode & S_IALLUGO) == 0)
717 ent = __proc_create(&parent, name, mode, nlink);
719 if (proc_register(parent, ent) < 0) {
726 EXPORT_SYMBOL(create_proc_entry);
728 struct proc_dir_entry *proc_create_data(const char *name, mode_t mode,
729 struct proc_dir_entry *parent,
730 const struct file_operations *proc_fops,
733 struct proc_dir_entry *pde;
737 if ((mode & S_IALLUGO) == 0)
738 mode |= S_IRUGO | S_IXUGO;
741 if ((mode & S_IFMT) == 0)
743 if ((mode & S_IALLUGO) == 0)
748 pde = __proc_create(&parent, name, mode, nlink);
751 pde->proc_fops = proc_fops;
753 if (proc_register(parent, pde) < 0)
761 EXPORT_SYMBOL(proc_create_data);
763 static void free_proc_entry(struct proc_dir_entry *de)
765 unsigned int ino = de->low_ino;
767 if (ino < PROC_DYNAMIC_FIRST)
770 release_inode_number(ino);
772 if (S_ISLNK(de->mode))
777 void pde_put(struct proc_dir_entry *pde)
779 if (atomic_dec_and_test(&pde->count))
780 free_proc_entry(pde);
784 * Remove a /proc entry and free it if it's not currently in use.
786 void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
788 struct proc_dir_entry **p;
789 struct proc_dir_entry *de = NULL;
790 const char *fn = name;
793 spin_lock(&proc_subdir_lock);
794 if (__xlate_proc_name(name, &parent, &fn) != 0) {
795 spin_unlock(&proc_subdir_lock);
800 for (p = &parent->subdir; *p; p=&(*p)->next ) {
801 if (proc_match(len, fn, *p)) {
808 spin_unlock(&proc_subdir_lock);
810 WARN(1, "name '%s'\n", name);
814 spin_lock(&de->pde_unload_lock);
816 * Stop accepting new callers into module. If you're
817 * dynamically allocating ->proc_fops, save a pointer somewhere.
819 de->proc_fops = NULL;
820 /* Wait until all existing callers into module are done. */
821 if (de->pde_users > 0) {
822 DECLARE_COMPLETION_ONSTACK(c);
824 if (!de->pde_unload_completion)
825 de->pde_unload_completion = &c;
827 spin_unlock(&de->pde_unload_lock);
829 wait_for_completion(de->pde_unload_completion);
831 goto continue_removing;
833 spin_unlock(&de->pde_unload_lock);
836 spin_lock(&de->pde_unload_lock);
837 while (!list_empty(&de->pde_openers)) {
838 struct pde_opener *pdeo;
840 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
842 spin_unlock(&de->pde_unload_lock);
843 pdeo->release(pdeo->inode, pdeo->file);
845 spin_lock(&de->pde_unload_lock);
847 spin_unlock(&de->pde_unload_lock);
849 if (S_ISDIR(de->mode))
852 WARN(de->subdir, KERN_WARNING "%s: removing non-empty directory "
853 "'%s/%s', leaking at least '%s'\n", __func__,
854 de->parent->name, de->name, de->subdir->name);
857 EXPORT_SYMBOL(remove_proc_entry);