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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/module.h>
16 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/namei.h>
20 #include <linux/slab.h>
21 #include <linux/mount.h>
22 #include <linux/tty.h>
23 #include <linux/mutex.h>
24 #include <linux/magic.h>
25 #include <linux/idr.h>
26 #include <linux/devpts_fs.h>
27 #include <linux/parser.h>
28 #include <linux/fsnotify.h>
29 #include <linux/seq_file.h>
31 #define DEVPTS_DEFAULT_MODE 0600
33 * ptmx is a new node in /dev/pts and will be unused in legacy (single-
34 * instance) mode. To prevent surprises in user space, set permissions of
35 * ptmx to 0. Use 'chmod' or remount with '-o ptmxmode' to set meaningful
38 #define DEVPTS_DEFAULT_PTMX_MODE 0000
42 * sysctl support for setting limits on the number of Unix98 ptys allocated.
43 * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
45 static int pty_limit = NR_UNIX98_PTY_DEFAULT;
46 static int pty_reserve = NR_UNIX98_PTY_RESERVE;
47 static int pty_limit_min;
48 static int pty_limit_max = INT_MAX;
51 static struct ctl_table pty_table[] = {
54 .maxlen = sizeof(int),
57 .proc_handler = proc_dointvec_minmax,
58 .extra1 = &pty_limit_min,
59 .extra2 = &pty_limit_max,
61 .procname = "reserve",
62 .maxlen = sizeof(int),
65 .proc_handler = proc_dointvec_minmax,
66 .extra1 = &pty_limit_min,
67 .extra2 = &pty_limit_max,
70 .maxlen = sizeof(int),
73 .proc_handler = proc_dointvec,
78 static struct ctl_table pty_kern_table[] = {
87 static struct ctl_table pty_root_table[] = {
91 .child = pty_kern_table,
96 static DEFINE_MUTEX(allocated_ptys_lock);
98 static struct vfsmount *devpts_mnt;
100 struct pts_mount_opts {
112 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
116 static const match_table_t tokens = {
119 {Opt_mode, "mode=%o"},
120 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
121 {Opt_ptmxmode, "ptmxmode=%o"},
122 {Opt_newinstance, "newinstance"},
129 struct ida allocated_ptys;
130 struct pts_mount_opts mount_opts;
131 struct dentry *ptmx_dentry;
134 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
136 return sb->s_fs_info;
139 static inline struct super_block *pts_sb_from_inode(struct inode *inode)
141 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
142 if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
147 return devpts_mnt->mnt_sb;
150 #define PARSE_MOUNT 0
151 #define PARSE_REMOUNT 1
154 * parse_mount_options():
155 * Set @opts to mount options specified in @data. If an option is not
156 * specified in @data, set it to its default value. The exception is
157 * 'newinstance' option which can only be set/cleared on a mount (i.e.
158 * cannot be changed during remount).
160 * Note: @data may be NULL (in which case all options are set to default).
162 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
170 opts->uid = GLOBAL_ROOT_UID;
171 opts->gid = GLOBAL_ROOT_GID;
172 opts->mode = DEVPTS_DEFAULT_MODE;
173 opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
174 opts->max = NR_UNIX98_PTY_MAX;
176 /* newinstance makes sense only on initial mount */
177 if (op == PARSE_MOUNT)
178 opts->newinstance = 0;
180 while ((p = strsep(&data, ",")) != NULL) {
181 substring_t args[MAX_OPT_ARGS];
188 token = match_token(p, tokens, args);
191 if (match_int(&args[0], &option))
193 uid = make_kuid(current_user_ns(), option);
200 if (match_int(&args[0], &option))
202 gid = make_kgid(current_user_ns(), option);
209 if (match_octal(&args[0], &option))
211 opts->mode = option & S_IALLUGO;
213 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
215 if (match_octal(&args[0], &option))
217 opts->ptmxmode = option & S_IALLUGO;
219 case Opt_newinstance:
220 /* newinstance makes sense only on initial mount */
221 if (op == PARSE_MOUNT)
222 opts->newinstance = 1;
225 if (match_int(&args[0], &option) ||
226 option < 0 || option > NR_UNIX98_PTY_MAX)
232 pr_err("called with bogus options\n");
240 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
241 static int mknod_ptmx(struct super_block *sb)
245 struct dentry *dentry;
247 struct dentry *root = sb->s_root;
248 struct pts_fs_info *fsi = DEVPTS_SB(sb);
249 struct pts_mount_opts *opts = &fsi->mount_opts;
253 root_uid = make_kuid(current_user_ns(), 0);
254 root_gid = make_kgid(current_user_ns(), 0);
255 if (!uid_valid(root_uid) || !gid_valid(root_gid))
258 inode_lock(d_inode(root));
260 /* If we have already created ptmx node, return */
261 if (fsi->ptmx_dentry) {
266 dentry = d_alloc_name(root, "ptmx");
268 pr_err("Unable to alloc dentry for ptmx node\n");
273 * Create a new 'ptmx' node in this mount of devpts.
275 inode = new_inode(sb);
277 pr_err("Unable to alloc inode for ptmx node\n");
283 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
285 mode = S_IFCHR|opts->ptmxmode;
286 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
287 inode->i_uid = root_uid;
288 inode->i_gid = root_gid;
290 d_add(dentry, inode);
292 fsi->ptmx_dentry = dentry;
295 inode_unlock(d_inode(root));
299 static void update_ptmx_mode(struct pts_fs_info *fsi)
302 if (fsi->ptmx_dentry) {
303 inode = d_inode(fsi->ptmx_dentry);
304 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
308 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
314 static int devpts_remount(struct super_block *sb, int *flags, char *data)
317 struct pts_fs_info *fsi = DEVPTS_SB(sb);
318 struct pts_mount_opts *opts = &fsi->mount_opts;
321 err = parse_mount_options(data, PARSE_REMOUNT, opts);
324 * parse_mount_options() restores options to default values
325 * before parsing and may have changed ptmxmode. So, update the
326 * mode in the inode too. Bogus options don't fail the remount,
327 * so do this even on error return.
329 update_ptmx_mode(fsi);
334 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
336 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
337 struct pts_mount_opts *opts = &fsi->mount_opts;
340 seq_printf(seq, ",uid=%u",
341 from_kuid_munged(&init_user_ns, opts->uid));
343 seq_printf(seq, ",gid=%u",
344 from_kgid_munged(&init_user_ns, opts->gid));
345 seq_printf(seq, ",mode=%03o", opts->mode);
346 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
347 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
348 if (opts->max < NR_UNIX98_PTY_MAX)
349 seq_printf(seq, ",max=%d", opts->max);
355 static const struct super_operations devpts_sops = {
356 .statfs = simple_statfs,
357 .remount_fs = devpts_remount,
358 .show_options = devpts_show_options,
361 static void *new_pts_fs_info(void)
363 struct pts_fs_info *fsi;
365 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
369 ida_init(&fsi->allocated_ptys);
370 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
371 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
377 devpts_fill_super(struct super_block *s, void *data, int silent)
381 s->s_blocksize = 1024;
382 s->s_blocksize_bits = 10;
383 s->s_magic = DEVPTS_SUPER_MAGIC;
384 s->s_op = &devpts_sops;
387 s->s_fs_info = new_pts_fs_info();
391 inode = new_inode(s);
395 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
396 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
397 inode->i_op = &simple_dir_inode_operations;
398 inode->i_fop = &simple_dir_operations;
401 s->s_root = d_make_root(inode);
405 pr_err("get root dentry failed\n");
411 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
412 static int compare_init_pts_sb(struct super_block *s, void *p)
415 return devpts_mnt->mnt_sb == s;
422 * If the '-o newinstance' mount option was specified, mount a new
423 * (private) instance of devpts. PTYs created in this instance are
424 * independent of the PTYs in other devpts instances.
426 * If the '-o newinstance' option was not specified, mount/remount the
427 * initial kernel mount of devpts. This type of mount gives the
428 * legacy, single-instance semantics.
430 * The 'newinstance' option is needed to support multiple namespace
431 * semantics in devpts while preserving backward compatibility of the
432 * current 'single-namespace' semantics. i.e all mounts of devpts
433 * without the 'newinstance' mount option should bind to the initial
434 * kernel mount, like mount_single().
436 * Mounts with 'newinstance' option create a new, private namespace.
440 * For single-mount semantics, devpts cannot use mount_single(),
441 * because mount_single()/sget() find and use the super-block from
442 * the most recent mount of devpts. But that recent mount may be a
443 * 'newinstance' mount and mount_single() would pick the newinstance
444 * super-block instead of the initial super-block.
446 static struct dentry *devpts_mount(struct file_system_type *fs_type,
447 int flags, const char *dev_name, void *data)
450 struct pts_mount_opts opts;
451 struct super_block *s;
453 error = parse_mount_options(data, PARSE_MOUNT, &opts);
455 return ERR_PTR(error);
457 /* Require newinstance for all user namespace mounts to ensure
458 * the mount options are not changed.
460 if ((current_user_ns() != &init_user_ns) && !opts.newinstance)
461 return ERR_PTR(-EINVAL);
463 if (opts.newinstance)
464 s = sget(fs_type, NULL, set_anon_super, flags, NULL);
466 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags,
473 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
476 s->s_flags |= MS_ACTIVE;
479 memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
481 error = mknod_ptmx(s);
485 return dget(s->s_root);
488 deactivate_locked_super(s);
489 return ERR_PTR(error);
494 * This supports only the legacy single-instance semantics (no
495 * multiple-instance semantics)
497 static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
498 const char *dev_name, void *data)
500 return mount_single(fs_type, flags, data, devpts_fill_super);
504 static void devpts_kill_sb(struct super_block *sb)
506 struct pts_fs_info *fsi = DEVPTS_SB(sb);
508 ida_destroy(&fsi->allocated_ptys);
510 kill_litter_super(sb);
513 static struct file_system_type devpts_fs_type = {
515 .mount = devpts_mount,
516 .kill_sb = devpts_kill_sb,
517 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
518 .fs_flags = FS_USERNS_MOUNT | FS_USERNS_DEV_MOUNT,
523 * The normal naming convention is simply /dev/pts/<number>; this conforms
524 * to the System V naming convention
527 int devpts_new_index(struct inode *ptmx_inode)
529 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
530 struct pts_fs_info *fsi;
539 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
542 mutex_lock(&allocated_ptys_lock);
543 if (pty_count >= pty_limit -
544 (fsi->mount_opts.newinstance ? pty_reserve : 0)) {
545 mutex_unlock(&allocated_ptys_lock);
549 ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
551 mutex_unlock(&allocated_ptys_lock);
552 if (ida_ret == -EAGAIN)
557 if (index >= fsi->mount_opts.max) {
558 ida_remove(&fsi->allocated_ptys, index);
559 mutex_unlock(&allocated_ptys_lock);
563 mutex_unlock(&allocated_ptys_lock);
567 void devpts_kill_index(struct inode *ptmx_inode, int idx)
569 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
570 struct pts_fs_info *fsi = DEVPTS_SB(sb);
572 mutex_lock(&allocated_ptys_lock);
573 ida_remove(&fsi->allocated_ptys, idx);
575 mutex_unlock(&allocated_ptys_lock);
579 * pty code needs to hold extra references in case of last /dev/tty close
582 void devpts_add_ref(struct inode *ptmx_inode)
584 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
586 atomic_inc(&sb->s_active);
590 void devpts_del_ref(struct inode *ptmx_inode)
592 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
595 deactivate_super(sb);
599 * devpts_pty_new -- create a new inode in /dev/pts/
600 * @ptmx_inode: inode of the master
601 * @device: major+minor of the node to be created
602 * @index: used as a name of the node
603 * @priv: what's given back by devpts_get_priv
605 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
607 struct inode *devpts_pty_new(struct inode *ptmx_inode, dev_t device, int index,
610 struct dentry *dentry;
611 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
614 struct pts_fs_info *fsi;
615 struct pts_mount_opts *opts;
619 return ERR_PTR(-ENODEV);
623 opts = &fsi->mount_opts;
625 inode = new_inode(sb);
627 return ERR_PTR(-ENOMEM);
629 inode->i_ino = index + 3;
630 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
631 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
632 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
633 init_special_inode(inode, S_IFCHR|opts->mode, device);
634 inode->i_private = priv;
636 sprintf(s, "%d", index);
638 inode_lock(d_inode(root));
640 dentry = d_alloc_name(root, s);
642 d_add(dentry, inode);
643 fsnotify_create(d_inode(root), dentry);
646 inode = ERR_PTR(-ENOMEM);
649 inode_unlock(d_inode(root));
655 * devpts_get_priv -- get private data for a slave
656 * @pts_inode: inode of the slave
658 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
660 void *devpts_get_priv(struct inode *pts_inode)
662 struct dentry *dentry;
665 BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
667 /* Ensure dentry has not been deleted by devpts_pty_kill() */
668 dentry = d_find_alias(pts_inode);
672 if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
673 priv = pts_inode->i_private;
681 * devpts_pty_kill -- remove inode form /dev/pts/
682 * @inode: inode of the slave to be removed
684 * This is an inverse operation of devpts_pty_new.
686 void devpts_pty_kill(struct inode *inode)
688 struct super_block *sb = pts_sb_from_inode(inode);
689 struct dentry *root = sb->s_root;
690 struct dentry *dentry;
692 BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
694 inode_lock(d_inode(root));
696 dentry = d_find_alias(inode);
700 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
701 dput(dentry); /* d_find_alias above */
703 inode_unlock(d_inode(root));
706 static int __init init_devpts_fs(void)
708 int err = register_filesystem(&devpts_fs_type);
709 struct ctl_table_header *table;
712 struct vfsmount *mnt;
714 table = register_sysctl_table(pty_root_table);
715 mnt = kern_mount(&devpts_fs_type);
718 unregister_filesystem(&devpts_fs_type);
719 unregister_sysctl_table(table);
726 module_init(init_devpts_fs)