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 super_block *sb;
132 struct dentry *ptmx_dentry;
135 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
137 return sb->s_fs_info;
140 static inline struct super_block *pts_sb_from_inode(struct inode *inode)
142 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
143 if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
148 return devpts_mnt->mnt_sb;
151 #define PARSE_MOUNT 0
152 #define PARSE_REMOUNT 1
155 * parse_mount_options():
156 * Set @opts to mount options specified in @data. If an option is not
157 * specified in @data, set it to its default value. The exception is
158 * 'newinstance' option which can only be set/cleared on a mount (i.e.
159 * cannot be changed during remount).
161 * Note: @data may be NULL (in which case all options are set to default).
163 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
171 opts->uid = GLOBAL_ROOT_UID;
172 opts->gid = GLOBAL_ROOT_GID;
173 opts->mode = DEVPTS_DEFAULT_MODE;
174 opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
175 opts->max = NR_UNIX98_PTY_MAX;
177 /* newinstance makes sense only on initial mount */
178 if (op == PARSE_MOUNT)
179 opts->newinstance = 0;
181 while ((p = strsep(&data, ",")) != NULL) {
182 substring_t args[MAX_OPT_ARGS];
189 token = match_token(p, tokens, args);
192 if (match_int(&args[0], &option))
194 uid = make_kuid(current_user_ns(), option);
201 if (match_int(&args[0], &option))
203 gid = make_kgid(current_user_ns(), option);
210 if (match_octal(&args[0], &option))
212 opts->mode = option & S_IALLUGO;
214 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
216 if (match_octal(&args[0], &option))
218 opts->ptmxmode = option & S_IALLUGO;
220 case Opt_newinstance:
221 /* newinstance makes sense only on initial mount */
222 if (op == PARSE_MOUNT)
223 opts->newinstance = 1;
226 if (match_int(&args[0], &option) ||
227 option < 0 || option > NR_UNIX98_PTY_MAX)
233 pr_err("called with bogus options\n");
241 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
242 static int mknod_ptmx(struct super_block *sb)
246 struct dentry *dentry;
248 struct dentry *root = sb->s_root;
249 struct pts_fs_info *fsi = DEVPTS_SB(sb);
250 struct pts_mount_opts *opts = &fsi->mount_opts;
254 root_uid = make_kuid(current_user_ns(), 0);
255 root_gid = make_kgid(current_user_ns(), 0);
256 if (!uid_valid(root_uid) || !gid_valid(root_gid))
259 inode_lock(d_inode(root));
261 /* If we have already created ptmx node, return */
262 if (fsi->ptmx_dentry) {
267 dentry = d_alloc_name(root, "ptmx");
269 pr_err("Unable to alloc dentry for ptmx node\n");
274 * Create a new 'ptmx' node in this mount of devpts.
276 inode = new_inode(sb);
278 pr_err("Unable to alloc inode for ptmx node\n");
284 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
286 mode = S_IFCHR|opts->ptmxmode;
287 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
288 inode->i_uid = root_uid;
289 inode->i_gid = root_gid;
291 d_add(dentry, inode);
293 fsi->ptmx_dentry = dentry;
296 inode_unlock(d_inode(root));
300 static void update_ptmx_mode(struct pts_fs_info *fsi)
303 if (fsi->ptmx_dentry) {
304 inode = d_inode(fsi->ptmx_dentry);
305 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
309 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
315 static int devpts_remount(struct super_block *sb, int *flags, char *data)
318 struct pts_fs_info *fsi = DEVPTS_SB(sb);
319 struct pts_mount_opts *opts = &fsi->mount_opts;
322 err = parse_mount_options(data, PARSE_REMOUNT, opts);
325 * parse_mount_options() restores options to default values
326 * before parsing and may have changed ptmxmode. So, update the
327 * mode in the inode too. Bogus options don't fail the remount,
328 * so do this even on error return.
330 update_ptmx_mode(fsi);
335 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
337 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
338 struct pts_mount_opts *opts = &fsi->mount_opts;
341 seq_printf(seq, ",uid=%u",
342 from_kuid_munged(&init_user_ns, opts->uid));
344 seq_printf(seq, ",gid=%u",
345 from_kgid_munged(&init_user_ns, opts->gid));
346 seq_printf(seq, ",mode=%03o", opts->mode);
347 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
348 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
349 if (opts->max < NR_UNIX98_PTY_MAX)
350 seq_printf(seq, ",max=%d", opts->max);
356 static const struct super_operations devpts_sops = {
357 .statfs = simple_statfs,
358 .remount_fs = devpts_remount,
359 .show_options = devpts_show_options,
362 static void *new_pts_fs_info(struct super_block *sb)
364 struct pts_fs_info *fsi;
366 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
370 ida_init(&fsi->allocated_ptys);
371 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
372 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
379 devpts_fill_super(struct super_block *s, void *data, int silent)
383 s->s_blocksize = 1024;
384 s->s_blocksize_bits = 10;
385 s->s_magic = DEVPTS_SUPER_MAGIC;
386 s->s_op = &devpts_sops;
389 s->s_fs_info = new_pts_fs_info(s);
393 inode = new_inode(s);
397 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
398 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
399 inode->i_op = &simple_dir_inode_operations;
400 inode->i_fop = &simple_dir_operations;
403 s->s_root = d_make_root(inode);
407 pr_err("get root dentry failed\n");
413 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
414 static int compare_init_pts_sb(struct super_block *s, void *p)
417 return devpts_mnt->mnt_sb == s;
424 * If the '-o newinstance' mount option was specified, mount a new
425 * (private) instance of devpts. PTYs created in this instance are
426 * independent of the PTYs in other devpts instances.
428 * If the '-o newinstance' option was not specified, mount/remount the
429 * initial kernel mount of devpts. This type of mount gives the
430 * legacy, single-instance semantics.
432 * The 'newinstance' option is needed to support multiple namespace
433 * semantics in devpts while preserving backward compatibility of the
434 * current 'single-namespace' semantics. i.e all mounts of devpts
435 * without the 'newinstance' mount option should bind to the initial
436 * kernel mount, like mount_single().
438 * Mounts with 'newinstance' option create a new, private namespace.
442 * For single-mount semantics, devpts cannot use mount_single(),
443 * because mount_single()/sget() find and use the super-block from
444 * the most recent mount of devpts. But that recent mount may be a
445 * 'newinstance' mount and mount_single() would pick the newinstance
446 * super-block instead of the initial super-block.
448 static struct dentry *devpts_mount(struct file_system_type *fs_type,
449 int flags, const char *dev_name, void *data)
452 struct pts_mount_opts opts;
453 struct super_block *s;
455 error = parse_mount_options(data, PARSE_MOUNT, &opts);
457 return ERR_PTR(error);
459 /* Require newinstance for all user namespace mounts to ensure
460 * the mount options are not changed.
462 if ((current_user_ns() != &init_user_ns) && !opts.newinstance)
463 return ERR_PTR(-EINVAL);
465 if (opts.newinstance)
466 s = sget(fs_type, NULL, set_anon_super, flags, NULL);
468 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags,
475 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
478 s->s_flags |= MS_ACTIVE;
481 memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
483 error = mknod_ptmx(s);
487 return dget(s->s_root);
490 deactivate_locked_super(s);
491 return ERR_PTR(error);
496 * This supports only the legacy single-instance semantics (no
497 * multiple-instance semantics)
499 static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
500 const char *dev_name, void *data)
502 return mount_single(fs_type, flags, data, devpts_fill_super);
506 static void devpts_kill_sb(struct super_block *sb)
508 struct pts_fs_info *fsi = DEVPTS_SB(sb);
510 ida_destroy(&fsi->allocated_ptys);
512 kill_litter_super(sb);
515 static struct file_system_type devpts_fs_type = {
517 .mount = devpts_mount,
518 .kill_sb = devpts_kill_sb,
519 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
520 .fs_flags = FS_USERNS_MOUNT | FS_USERNS_DEV_MOUNT,
525 * The normal naming convention is simply /dev/pts/<number>; this conforms
526 * to the System V naming convention
529 int devpts_new_index(struct pts_fs_info *fsi)
538 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
541 mutex_lock(&allocated_ptys_lock);
542 if (pty_count >= pty_limit -
543 (fsi->mount_opts.newinstance ? pty_reserve : 0)) {
544 mutex_unlock(&allocated_ptys_lock);
548 ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
550 mutex_unlock(&allocated_ptys_lock);
551 if (ida_ret == -EAGAIN)
556 if (index >= fsi->mount_opts.max) {
557 ida_remove(&fsi->allocated_ptys, index);
558 mutex_unlock(&allocated_ptys_lock);
562 mutex_unlock(&allocated_ptys_lock);
566 void devpts_kill_index(struct pts_fs_info *fsi, int idx)
568 mutex_lock(&allocated_ptys_lock);
569 ida_remove(&fsi->allocated_ptys, idx);
571 mutex_unlock(&allocated_ptys_lock);
575 * pty code needs to hold extra references in case of last /dev/tty close
577 struct pts_fs_info *devpts_get_ref(struct inode *ptmx_inode, struct file *file)
579 struct super_block *sb;
580 struct pts_fs_info *fsi;
582 sb = pts_sb_from_inode(ptmx_inode);
589 atomic_inc(&sb->s_active);
593 void devpts_put_ref(struct pts_fs_info *fsi)
595 deactivate_super(fsi->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 dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
609 struct dentry *dentry;
610 struct super_block *sb;
613 struct pts_mount_opts *opts;
617 return ERR_PTR(-ENODEV);
621 opts = &fsi->mount_opts;
623 inode = new_inode(sb);
625 return ERR_PTR(-ENOMEM);
627 inode->i_ino = index + 3;
628 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
629 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
630 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
631 init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
633 sprintf(s, "%d", index);
635 dentry = d_alloc_name(root, s);
637 dentry->d_fsdata = priv;
638 d_add(dentry, inode);
639 fsnotify_create(d_inode(root), dentry);
642 dentry = ERR_PTR(-ENOMEM);
649 * devpts_get_priv -- get private data for a slave
650 * @pts_inode: inode of the slave
652 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
654 void *devpts_get_priv(struct dentry *dentry)
656 WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
657 return dentry->d_fsdata;
661 * devpts_pty_kill -- remove inode form /dev/pts/
662 * @inode: inode of the slave to be removed
664 * This is an inverse operation of devpts_pty_new.
666 void devpts_pty_kill(struct dentry *dentry)
668 WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
670 dentry->d_fsdata = NULL;
671 drop_nlink(dentry->d_inode);
673 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
676 static int __init init_devpts_fs(void)
678 int err = register_filesystem(&devpts_fs_type);
679 struct ctl_table_header *table;
682 struct vfsmount *mnt;
684 table = register_sysctl_table(pty_root_table);
685 mnt = kern_mount(&devpts_fs_type);
688 unregister_filesystem(&devpts_fs_type);
689 unregister_sysctl_table(table);
696 module_init(init_devpts_fs)