1 /* -*- linux-c -*- --------------------------------------------------------- *
3 * linux/fs/devpts/inode.c
5 * Copyright 1998-2004 H. Peter Anvin -- All Rights Reserved
7 * This file is part of the Linux kernel and is made available under
8 * the terms of the GNU General Public License, version 2, or at your
9 * option, any later version, incorporated herein by reference.
11 * ------------------------------------------------------------------------- */
13 #include <linux/module.h>
14 #include <linux/init.h>
16 #include <linux/sched.h>
17 #include <linux/namei.h>
18 #include <linux/slab.h>
19 #include <linux/mount.h>
20 #include <linux/tty.h>
21 #include <linux/mutex.h>
22 #include <linux/magic.h>
23 #include <linux/idr.h>
24 #include <linux/devpts_fs.h>
25 #include <linux/parser.h>
26 #include <linux/fsnotify.h>
27 #include <linux/seq_file.h>
29 #define DEVPTS_DEFAULT_MODE 0600
31 * ptmx is a new node in /dev/pts and will be unused in legacy (single-
32 * instance) mode. To prevent surprises in user space, set permissions of
33 * ptmx to 0. Use 'chmod' or remount with '-o ptmxmode' to set meaningful
36 #define DEVPTS_DEFAULT_PTMX_MODE 0000
40 * sysctl support for setting limits on the number of Unix98 ptys allocated.
41 * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
43 static int pty_limit = NR_UNIX98_PTY_DEFAULT;
44 static int pty_reserve = NR_UNIX98_PTY_RESERVE;
45 static int pty_limit_min;
46 static int pty_limit_max = INT_MAX;
49 static struct ctl_table pty_table[] = {
52 .maxlen = sizeof(int),
55 .proc_handler = proc_dointvec_minmax,
56 .extra1 = &pty_limit_min,
57 .extra2 = &pty_limit_max,
59 .procname = "reserve",
60 .maxlen = sizeof(int),
63 .proc_handler = proc_dointvec_minmax,
64 .extra1 = &pty_limit_min,
65 .extra2 = &pty_limit_max,
68 .maxlen = sizeof(int),
71 .proc_handler = proc_dointvec,
76 static struct ctl_table pty_kern_table[] = {
85 static struct ctl_table pty_root_table[] = {
89 .child = pty_kern_table,
94 static DEFINE_MUTEX(allocated_ptys_lock);
96 static struct vfsmount *devpts_mnt;
98 struct pts_mount_opts {
110 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
114 static const match_table_t tokens = {
117 {Opt_mode, "mode=%o"},
118 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
119 {Opt_ptmxmode, "ptmxmode=%o"},
120 {Opt_newinstance, "newinstance"},
127 struct ida allocated_ptys;
128 struct pts_mount_opts mount_opts;
129 struct dentry *ptmx_dentry;
132 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
134 return sb->s_fs_info;
137 static inline struct super_block *pts_sb_from_inode(struct inode *inode)
139 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
140 if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
143 return devpts_mnt->mnt_sb;
146 #define PARSE_MOUNT 0
147 #define PARSE_REMOUNT 1
150 * parse_mount_options():
151 * Set @opts to mount options specified in @data. If an option is not
152 * specified in @data, set it to its default value. The exception is
153 * 'newinstance' option which can only be set/cleared on a mount (i.e.
154 * cannot be changed during remount).
156 * Note: @data may be NULL (in which case all options are set to default).
158 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
166 opts->uid = GLOBAL_ROOT_UID;
167 opts->gid = GLOBAL_ROOT_GID;
168 opts->mode = DEVPTS_DEFAULT_MODE;
169 opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
170 opts->max = NR_UNIX98_PTY_MAX;
172 /* newinstance makes sense only on initial mount */
173 if (op == PARSE_MOUNT)
174 opts->newinstance = 0;
176 while ((p = strsep(&data, ",")) != NULL) {
177 substring_t args[MAX_OPT_ARGS];
184 token = match_token(p, tokens, args);
187 if (match_int(&args[0], &option))
189 uid = make_kuid(current_user_ns(), option);
196 if (match_int(&args[0], &option))
198 gid = make_kgid(current_user_ns(), option);
205 if (match_octal(&args[0], &option))
207 opts->mode = option & S_IALLUGO;
209 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
211 if (match_octal(&args[0], &option))
213 opts->ptmxmode = option & S_IALLUGO;
215 case Opt_newinstance:
216 /* newinstance makes sense only on initial mount */
217 if (op == PARSE_MOUNT)
218 opts->newinstance = 1;
221 if (match_int(&args[0], &option) ||
222 option < 0 || option > NR_UNIX98_PTY_MAX)
228 printk(KERN_ERR "devpts: called with bogus options\n");
236 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
237 static int mknod_ptmx(struct super_block *sb)
241 struct dentry *dentry;
243 struct dentry *root = sb->s_root;
244 struct pts_fs_info *fsi = DEVPTS_SB(sb);
245 struct pts_mount_opts *opts = &fsi->mount_opts;
249 root_uid = make_kuid(current_user_ns(), 0);
250 root_gid = make_kgid(current_user_ns(), 0);
251 if (!uid_valid(root_uid) || !gid_valid(root_gid))
254 mutex_lock(&root->d_inode->i_mutex);
256 /* If we have already created ptmx node, return */
257 if (fsi->ptmx_dentry) {
262 dentry = d_alloc_name(root, "ptmx");
264 printk(KERN_NOTICE "Unable to alloc dentry for ptmx node\n");
269 * Create a new 'ptmx' node in this mount of devpts.
271 inode = new_inode(sb);
273 printk(KERN_ERR "Unable to alloc inode for ptmx node\n");
279 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
281 mode = S_IFCHR|opts->ptmxmode;
282 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
283 inode->i_uid = root_uid;
284 inode->i_gid = root_gid;
286 d_add(dentry, inode);
288 fsi->ptmx_dentry = dentry;
291 mutex_unlock(&root->d_inode->i_mutex);
295 static void update_ptmx_mode(struct pts_fs_info *fsi)
298 if (fsi->ptmx_dentry) {
299 inode = fsi->ptmx_dentry->d_inode;
300 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
304 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
310 static int devpts_remount(struct super_block *sb, int *flags, char *data)
313 struct pts_fs_info *fsi = DEVPTS_SB(sb);
314 struct pts_mount_opts *opts = &fsi->mount_opts;
316 err = parse_mount_options(data, PARSE_REMOUNT, opts);
319 * parse_mount_options() restores options to default values
320 * before parsing and may have changed ptmxmode. So, update the
321 * mode in the inode too. Bogus options don't fail the remount,
322 * so do this even on error return.
324 update_ptmx_mode(fsi);
329 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
331 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
332 struct pts_mount_opts *opts = &fsi->mount_opts;
335 seq_printf(seq, ",uid=%u", from_kuid_munged(&init_user_ns, opts->uid));
337 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, opts->gid));
338 seq_printf(seq, ",mode=%03o", opts->mode);
339 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
340 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
341 if (opts->max < NR_UNIX98_PTY_MAX)
342 seq_printf(seq, ",max=%d", opts->max);
348 static const struct super_operations devpts_sops = {
349 .statfs = simple_statfs,
350 .remount_fs = devpts_remount,
351 .show_options = devpts_show_options,
354 static void *new_pts_fs_info(void)
356 struct pts_fs_info *fsi;
358 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
362 ida_init(&fsi->allocated_ptys);
363 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
364 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
370 devpts_fill_super(struct super_block *s, void *data, int silent)
374 s->s_blocksize = 1024;
375 s->s_blocksize_bits = 10;
376 s->s_magic = DEVPTS_SUPER_MAGIC;
377 s->s_op = &devpts_sops;
380 s->s_fs_info = new_pts_fs_info();
384 inode = new_inode(s);
388 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
389 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
390 inode->i_op = &simple_dir_inode_operations;
391 inode->i_fop = &simple_dir_operations;
394 s->s_root = d_make_root(inode);
398 printk(KERN_ERR "devpts: get root dentry failed\n");
404 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
405 static int compare_init_pts_sb(struct super_block *s, void *p)
408 return devpts_mnt->mnt_sb == s;
415 * If the '-o newinstance' mount option was specified, mount a new
416 * (private) instance of devpts. PTYs created in this instance are
417 * independent of the PTYs in other devpts instances.
419 * If the '-o newinstance' option was not specified, mount/remount the
420 * initial kernel mount of devpts. This type of mount gives the
421 * legacy, single-instance semantics.
423 * The 'newinstance' option is needed to support multiple namespace
424 * semantics in devpts while preserving backward compatibility of the
425 * current 'single-namespace' semantics. i.e all mounts of devpts
426 * without the 'newinstance' mount option should bind to the initial
427 * kernel mount, like mount_single().
429 * Mounts with 'newinstance' option create a new, private namespace.
433 * For single-mount semantics, devpts cannot use mount_single(),
434 * because mount_single()/sget() find and use the super-block from
435 * the most recent mount of devpts. But that recent mount may be a
436 * 'newinstance' mount and mount_single() would pick the newinstance
437 * super-block instead of the initial super-block.
439 static struct dentry *devpts_mount(struct file_system_type *fs_type,
440 int flags, const char *dev_name, void *data)
443 struct pts_mount_opts opts;
444 struct super_block *s;
446 error = parse_mount_options(data, PARSE_MOUNT, &opts);
448 return ERR_PTR(error);
450 /* Require newinstance for all user namespace mounts to ensure
451 * the mount options are not changed.
453 if ((current_user_ns() != &init_user_ns) && !opts.newinstance)
454 return ERR_PTR(-EINVAL);
456 if (opts.newinstance)
457 s = sget(fs_type, NULL, set_anon_super, flags, NULL);
459 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags,
466 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
469 s->s_flags |= MS_ACTIVE;
472 memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
474 error = mknod_ptmx(s);
478 return dget(s->s_root);
481 deactivate_locked_super(s);
482 return ERR_PTR(error);
487 * This supports only the legacy single-instance semantics (no
488 * multiple-instance semantics)
490 static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
491 const char *dev_name, void *data)
493 return mount_single(fs_type, flags, data, devpts_fill_super);
497 static void devpts_kill_sb(struct super_block *sb)
499 struct pts_fs_info *fsi = DEVPTS_SB(sb);
502 kill_litter_super(sb);
505 static struct file_system_type devpts_fs_type = {
507 .mount = devpts_mount,
508 .kill_sb = devpts_kill_sb,
509 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
510 .fs_flags = FS_USERNS_MOUNT | FS_USERNS_DEV_MOUNT,
515 * The normal naming convention is simply /dev/pts/<number>; this conforms
516 * to the System V naming convention
519 int devpts_new_index(struct inode *ptmx_inode)
521 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
522 struct pts_fs_info *fsi = DEVPTS_SB(sb);
527 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
530 mutex_lock(&allocated_ptys_lock);
531 if (pty_count >= pty_limit -
532 (fsi->mount_opts.newinstance ? pty_reserve : 0)) {
533 mutex_unlock(&allocated_ptys_lock);
537 ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
539 mutex_unlock(&allocated_ptys_lock);
540 if (ida_ret == -EAGAIN)
545 if (index >= fsi->mount_opts.max) {
546 ida_remove(&fsi->allocated_ptys, index);
547 mutex_unlock(&allocated_ptys_lock);
551 mutex_unlock(&allocated_ptys_lock);
555 void devpts_kill_index(struct inode *ptmx_inode, int idx)
557 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
558 struct pts_fs_info *fsi = DEVPTS_SB(sb);
560 mutex_lock(&allocated_ptys_lock);
561 ida_remove(&fsi->allocated_ptys, idx);
563 mutex_unlock(&allocated_ptys_lock);
567 * devpts_pty_new -- create a new inode in /dev/pts/
568 * @ptmx_inode: inode of the master
569 * @device: major+minor of the node to be created
570 * @index: used as a name of the node
571 * @priv: what's given back by devpts_get_priv
573 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
575 struct inode *devpts_pty_new(struct inode *ptmx_inode, dev_t device, int index,
578 struct dentry *dentry;
579 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
581 struct dentry *root = sb->s_root;
582 struct pts_fs_info *fsi = DEVPTS_SB(sb);
583 struct pts_mount_opts *opts = &fsi->mount_opts;
586 inode = new_inode(sb);
588 return ERR_PTR(-ENOMEM);
590 inode->i_ino = index + 3;
591 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
592 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
593 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
594 init_special_inode(inode, S_IFCHR|opts->mode, device);
595 inode->i_private = priv;
597 sprintf(s, "%d", index);
599 mutex_lock(&root->d_inode->i_mutex);
601 dentry = d_alloc_name(root, s);
603 d_add(dentry, inode);
604 fsnotify_create(root->d_inode, dentry);
607 inode = ERR_PTR(-ENOMEM);
610 mutex_unlock(&root->d_inode->i_mutex);
616 * devpts_get_priv -- get private data for a slave
617 * @pts_inode: inode of the slave
619 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
621 void *devpts_get_priv(struct inode *pts_inode)
623 struct dentry *dentry;
626 BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
628 /* Ensure dentry has not been deleted by devpts_pty_kill() */
629 dentry = d_find_alias(pts_inode);
633 if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
634 priv = pts_inode->i_private;
642 * devpts_pty_kill -- remove inode form /dev/pts/
643 * @inode: inode of the slave to be removed
645 * This is an inverse operation of devpts_pty_new.
647 void devpts_pty_kill(struct inode *inode)
649 struct super_block *sb = pts_sb_from_inode(inode);
650 struct dentry *root = sb->s_root;
651 struct dentry *dentry;
653 BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
655 mutex_lock(&root->d_inode->i_mutex);
657 dentry = d_find_alias(inode);
661 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
662 dput(dentry); /* d_find_alias above */
664 mutex_unlock(&root->d_inode->i_mutex);
667 static int __init init_devpts_fs(void)
669 int err = register_filesystem(&devpts_fs_type);
670 struct ctl_table_header *table;
673 table = register_sysctl_table(pty_root_table);
674 devpts_mnt = kern_mount(&devpts_fs_type);
675 if (IS_ERR(devpts_mnt)) {
676 err = PTR_ERR(devpts_mnt);
677 unregister_filesystem(&devpts_fs_type);
678 unregister_sysctl_table(table);
683 module_init(init_devpts_fs)