Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* -*- linux-c -*- --------------------------------------------------------- *
3 *
4 * linux/fs/devpts/inode.c
5 *
6 * Copyright 1998-2004 H. Peter Anvin -- All Rights Reserved
7 *
8 * ------------------------------------------------------------------------- */
9
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/module.h>
13#include <linux/init.h>
14#include <linux/fs.h>
15#include <linux/sched.h>
16#include <linux/namei.h>
17#include <linux/slab.h>
18#include <linux/mount.h>
19#include <linux/tty.h>
20#include <linux/mutex.h>
21#include <linux/magic.h>
22#include <linux/idr.h>
23#include <linux/devpts_fs.h>
24#include <linux/parser.h>
25#include <linux/fsnotify.h>
26#include <linux/seq_file.h>
27
28#define DEVPTS_DEFAULT_MODE 0600
29/*
30 * ptmx is a new node in /dev/pts and will be unused in legacy (single-
31 * instance) mode. To prevent surprises in user space, set permissions of
32 * ptmx to 0. Use 'chmod' or remount with '-o ptmxmode' to set meaningful
33 * permissions.
34 */
35#define DEVPTS_DEFAULT_PTMX_MODE 0000
36#define PTMX_MINOR 2
37
38/*
39 * sysctl support for setting limits on the number of Unix98 ptys allocated.
40 * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
41 */
42static int pty_limit = NR_UNIX98_PTY_DEFAULT;
43static int pty_reserve = NR_UNIX98_PTY_RESERVE;
44static int pty_limit_min;
45static int pty_limit_max = INT_MAX;
46static atomic_t pty_count = ATOMIC_INIT(0);
47
48static struct ctl_table pty_table[] = {
49 {
50 .procname = "max",
51 .maxlen = sizeof(int),
52 .mode = 0644,
53 .data = &pty_limit,
54 .proc_handler = proc_dointvec_minmax,
55 .extra1 = &pty_limit_min,
56 .extra2 = &pty_limit_max,
57 }, {
58 .procname = "reserve",
59 .maxlen = sizeof(int),
60 .mode = 0644,
61 .data = &pty_reserve,
62 .proc_handler = proc_dointvec_minmax,
63 .extra1 = &pty_limit_min,
64 .extra2 = &pty_limit_max,
65 }, {
66 .procname = "nr",
67 .maxlen = sizeof(int),
68 .mode = 0444,
69 .data = &pty_count,
70 .proc_handler = proc_dointvec,
71 },
72 {}
73};
74
75static struct ctl_table pty_kern_table[] = {
76 {
77 .procname = "pty",
78 .mode = 0555,
79 .child = pty_table,
80 },
81 {}
82};
83
84static struct ctl_table pty_root_table[] = {
85 {
86 .procname = "kernel",
87 .mode = 0555,
88 .child = pty_kern_table,
89 },
90 {}
91};
92
93struct pts_mount_opts {
94 int setuid;
95 int setgid;
96 kuid_t uid;
97 kgid_t gid;
98 umode_t mode;
99 umode_t ptmxmode;
100 int reserve;
101 int max;
102};
103
104enum {
105 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
106 Opt_err
107};
108
109static const match_table_t tokens = {
110 {Opt_uid, "uid=%u"},
111 {Opt_gid, "gid=%u"},
112 {Opt_mode, "mode=%o"},
113 {Opt_ptmxmode, "ptmxmode=%o"},
114 {Opt_newinstance, "newinstance"},
115 {Opt_max, "max=%d"},
116 {Opt_err, NULL}
117};
118
119struct pts_fs_info {
120 struct ida allocated_ptys;
121 struct pts_mount_opts mount_opts;
122 struct super_block *sb;
123 struct dentry *ptmx_dentry;
124};
125
126static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
127{
128 return sb->s_fs_info;
129}
130
131static int devpts_ptmx_path(struct path *path)
132{
133 struct super_block *sb;
134 int err;
135
136 /* Is a devpts filesystem at "pts" in the same directory? */
137 err = path_pts(path);
138 if (err)
139 return err;
140
141 /* Is the path the root of a devpts filesystem? */
142 sb = path->mnt->mnt_sb;
143 if ((sb->s_magic != DEVPTS_SUPER_MAGIC) ||
144 (path->mnt->mnt_root != sb->s_root))
145 return -ENODEV;
146
147 return 0;
148}
149
150/*
151 * Try to find a suitable devpts filesystem. We support the following
152 * scenarios:
153 * - The ptmx device node is located in the same directory as the devpts
154 * mount where the pts device nodes are located.
155 * This is e.g. the case when calling open on the /dev/pts/ptmx device
156 * node when the devpts filesystem is mounted at /dev/pts.
157 * - The ptmx device node is located outside the devpts filesystem mount
158 * where the pts device nodes are located. For example, the ptmx device
159 * is a symlink, separate device node, or bind-mount.
160 * A supported scenario is bind-mounting /dev/pts/ptmx to /dev/ptmx and
161 * then calling open on /dev/ptmx. In this case a suitable pts
162 * subdirectory can be found in the common parent directory /dev of the
163 * devpts mount and the ptmx bind-mount, after resolving the /dev/ptmx
164 * bind-mount.
165 * If no suitable pts subdirectory can be found this function will fail.
166 * This is e.g. the case when bind-mounting /dev/pts/ptmx to /ptmx.
167 */
168struct vfsmount *devpts_mntget(struct file *filp, struct pts_fs_info *fsi)
169{
170 struct path path;
171 int err = 0;
172
173 path = filp->f_path;
174 path_get(&path);
175
176 /* Walk upward while the start point is a bind mount of
177 * a single file.
178 */
179 while (path.mnt->mnt_root == path.dentry)
180 if (follow_up(&path) == 0)
181 break;
182
183 /* devpts_ptmx_path() finds a devpts fs or returns an error. */
184 if ((path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) ||
185 (DEVPTS_SB(path.mnt->mnt_sb) != fsi))
186 err = devpts_ptmx_path(&path);
187 dput(path.dentry);
188 if (!err) {
189 if (DEVPTS_SB(path.mnt->mnt_sb) == fsi)
190 return path.mnt;
191
192 err = -ENODEV;
193 }
194
195 mntput(path.mnt);
196 return ERR_PTR(err);
197}
198
199struct pts_fs_info *devpts_acquire(struct file *filp)
200{
201 struct pts_fs_info *result;
202 struct path path;
203 struct super_block *sb;
204
205 path = filp->f_path;
206 path_get(&path);
207
208 /* Has the devpts filesystem already been found? */
209 if (path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) {
210 int err;
211
212 err = devpts_ptmx_path(&path);
213 if (err) {
214 result = ERR_PTR(err);
215 goto out;
216 }
217 }
218
219 /*
220 * pty code needs to hold extra references in case of last /dev/tty close
221 */
222 sb = path.mnt->mnt_sb;
223 atomic_inc(&sb->s_active);
224 result = DEVPTS_SB(sb);
225
226out:
227 path_put(&path);
228 return result;
229}
230
231void devpts_release(struct pts_fs_info *fsi)
232{
233 deactivate_super(fsi->sb);
234}
235
236#define PARSE_MOUNT 0
237#define PARSE_REMOUNT 1
238
239/*
240 * parse_mount_options():
241 * Set @opts to mount options specified in @data. If an option is not
242 * specified in @data, set it to its default value.
243 *
244 * Note: @data may be NULL (in which case all options are set to default).
245 */
246static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
247{
248 char *p;
249 kuid_t uid;
250 kgid_t gid;
251
252 opts->setuid = 0;
253 opts->setgid = 0;
254 opts->uid = GLOBAL_ROOT_UID;
255 opts->gid = GLOBAL_ROOT_GID;
256 opts->mode = DEVPTS_DEFAULT_MODE;
257 opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
258 opts->max = NR_UNIX98_PTY_MAX;
259
260 /* Only allow instances mounted from the initial mount
261 * namespace to tap the reserve pool of ptys.
262 */
263 if (op == PARSE_MOUNT)
264 opts->reserve =
265 (current->nsproxy->mnt_ns == init_task.nsproxy->mnt_ns);
266
267 while ((p = strsep(&data, ",")) != NULL) {
268 substring_t args[MAX_OPT_ARGS];
269 int token;
270 int option;
271
272 if (!*p)
273 continue;
274
275 token = match_token(p, tokens, args);
276 switch (token) {
277 case Opt_uid:
278 if (match_int(&args[0], &option))
279 return -EINVAL;
280 uid = make_kuid(current_user_ns(), option);
281 if (!uid_valid(uid))
282 return -EINVAL;
283 opts->uid = uid;
284 opts->setuid = 1;
285 break;
286 case Opt_gid:
287 if (match_int(&args[0], &option))
288 return -EINVAL;
289 gid = make_kgid(current_user_ns(), option);
290 if (!gid_valid(gid))
291 return -EINVAL;
292 opts->gid = gid;
293 opts->setgid = 1;
294 break;
295 case Opt_mode:
296 if (match_octal(&args[0], &option))
297 return -EINVAL;
298 opts->mode = option & S_IALLUGO;
299 break;
300 case Opt_ptmxmode:
301 if (match_octal(&args[0], &option))
302 return -EINVAL;
303 opts->ptmxmode = option & S_IALLUGO;
304 break;
305 case Opt_newinstance:
306 break;
307 case Opt_max:
308 if (match_int(&args[0], &option) ||
309 option < 0 || option > NR_UNIX98_PTY_MAX)
310 return -EINVAL;
311 opts->max = option;
312 break;
313 default:
314 pr_err("called with bogus options\n");
315 return -EINVAL;
316 }
317 }
318
319 return 0;
320}
321
322static int mknod_ptmx(struct super_block *sb)
323{
324 int mode;
325 int rc = -ENOMEM;
326 struct dentry *dentry;
327 struct inode *inode;
328 struct dentry *root = sb->s_root;
329 struct pts_fs_info *fsi = DEVPTS_SB(sb);
330 struct pts_mount_opts *opts = &fsi->mount_opts;
331 kuid_t ptmx_uid = current_fsuid();
332 kgid_t ptmx_gid = current_fsgid();
333
334 inode_lock(d_inode(root));
335
336 /* If we have already created ptmx node, return */
337 if (fsi->ptmx_dentry) {
338 rc = 0;
339 goto out;
340 }
341
342 dentry = d_alloc_name(root, "ptmx");
343 if (!dentry) {
344 pr_err("Unable to alloc dentry for ptmx node\n");
345 goto out;
346 }
347
348 /*
349 * Create a new 'ptmx' node in this mount of devpts.
350 */
351 inode = new_inode(sb);
352 if (!inode) {
353 pr_err("Unable to alloc inode for ptmx node\n");
354 dput(dentry);
355 goto out;
356 }
357
358 inode->i_ino = 2;
359 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
360
361 mode = S_IFCHR|opts->ptmxmode;
362 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
363 inode->i_uid = ptmx_uid;
364 inode->i_gid = ptmx_gid;
365
366 d_add(dentry, inode);
367
368 fsi->ptmx_dentry = dentry;
369 rc = 0;
370out:
371 inode_unlock(d_inode(root));
372 return rc;
373}
374
375static void update_ptmx_mode(struct pts_fs_info *fsi)
376{
377 struct inode *inode;
378 if (fsi->ptmx_dentry) {
379 inode = d_inode(fsi->ptmx_dentry);
380 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
381 }
382}
383
384static int devpts_remount(struct super_block *sb, int *flags, char *data)
385{
386 int err;
387 struct pts_fs_info *fsi = DEVPTS_SB(sb);
388 struct pts_mount_opts *opts = &fsi->mount_opts;
389
390 err = parse_mount_options(data, PARSE_REMOUNT, opts);
391
392 /*
393 * parse_mount_options() restores options to default values
394 * before parsing and may have changed ptmxmode. So, update the
395 * mode in the inode too. Bogus options don't fail the remount,
396 * so do this even on error return.
397 */
398 update_ptmx_mode(fsi);
399
400 return err;
401}
402
403static int devpts_show_options(struct seq_file *seq, struct dentry *root)
404{
405 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
406 struct pts_mount_opts *opts = &fsi->mount_opts;
407
408 if (opts->setuid)
409 seq_printf(seq, ",uid=%u",
410 from_kuid_munged(&init_user_ns, opts->uid));
411 if (opts->setgid)
412 seq_printf(seq, ",gid=%u",
413 from_kgid_munged(&init_user_ns, opts->gid));
414 seq_printf(seq, ",mode=%03o", opts->mode);
415 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
416 if (opts->max < NR_UNIX98_PTY_MAX)
417 seq_printf(seq, ",max=%d", opts->max);
418
419 return 0;
420}
421
422static const struct super_operations devpts_sops = {
423 .statfs = simple_statfs,
424 .remount_fs = devpts_remount,
425 .show_options = devpts_show_options,
426};
427
428static void *new_pts_fs_info(struct super_block *sb)
429{
430 struct pts_fs_info *fsi;
431
432 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
433 if (!fsi)
434 return NULL;
435
436 ida_init(&fsi->allocated_ptys);
437 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
438 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
439 fsi->sb = sb;
440
441 return fsi;
442}
443
444static int
445devpts_fill_super(struct super_block *s, void *data, int silent)
446{
447 struct inode *inode;
448 int error;
449
450 s->s_iflags &= ~SB_I_NODEV;
451 s->s_blocksize = 1024;
452 s->s_blocksize_bits = 10;
453 s->s_magic = DEVPTS_SUPER_MAGIC;
454 s->s_op = &devpts_sops;
455 s->s_d_op = &simple_dentry_operations;
456 s->s_time_gran = 1;
457
458 error = -ENOMEM;
459 s->s_fs_info = new_pts_fs_info(s);
460 if (!s->s_fs_info)
461 goto fail;
462
463 error = parse_mount_options(data, PARSE_MOUNT, &DEVPTS_SB(s)->mount_opts);
464 if (error)
465 goto fail;
466
467 error = -ENOMEM;
468 inode = new_inode(s);
469 if (!inode)
470 goto fail;
471 inode->i_ino = 1;
472 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
473 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
474 inode->i_op = &simple_dir_inode_operations;
475 inode->i_fop = &simple_dir_operations;
476 set_nlink(inode, 2);
477
478 s->s_root = d_make_root(inode);
479 if (!s->s_root) {
480 pr_err("get root dentry failed\n");
481 goto fail;
482 }
483
484 error = mknod_ptmx(s);
485 if (error)
486 goto fail_dput;
487
488 return 0;
489fail_dput:
490 dput(s->s_root);
491 s->s_root = NULL;
492fail:
493 return error;
494}
495
496/*
497 * devpts_mount()
498 *
499 * Mount a new (private) instance of devpts. PTYs created in this
500 * instance are independent of the PTYs in other devpts instances.
501 */
502static struct dentry *devpts_mount(struct file_system_type *fs_type,
503 int flags, const char *dev_name, void *data)
504{
505 return mount_nodev(fs_type, flags, data, devpts_fill_super);
506}
507
508static void devpts_kill_sb(struct super_block *sb)
509{
510 struct pts_fs_info *fsi = DEVPTS_SB(sb);
511
512 if (fsi)
513 ida_destroy(&fsi->allocated_ptys);
514 kfree(fsi);
515 kill_litter_super(sb);
516}
517
518static struct file_system_type devpts_fs_type = {
519 .name = "devpts",
520 .mount = devpts_mount,
521 .kill_sb = devpts_kill_sb,
522 .fs_flags = FS_USERNS_MOUNT,
523};
524
525/*
526 * The normal naming convention is simply /dev/pts/<number>; this conforms
527 * to the System V naming convention
528 */
529
530int devpts_new_index(struct pts_fs_info *fsi)
531{
532 int index = -ENOSPC;
533
534 if (atomic_inc_return(&pty_count) >= (pty_limit -
535 (fsi->mount_opts.reserve ? 0 : pty_reserve)))
536 goto out;
537
538 index = ida_alloc_max(&fsi->allocated_ptys, fsi->mount_opts.max - 1,
539 GFP_KERNEL);
540
541out:
542 if (index < 0)
543 atomic_dec(&pty_count);
544 return index;
545}
546
547void devpts_kill_index(struct pts_fs_info *fsi, int idx)
548{
549 ida_free(&fsi->allocated_ptys, idx);
550 atomic_dec(&pty_count);
551}
552
553/**
554 * devpts_pty_new -- create a new inode in /dev/pts/
555 * @ptmx_inode: inode of the master
556 * @device: major+minor of the node to be created
557 * @index: used as a name of the node
558 * @priv: what's given back by devpts_get_priv
559 *
560 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
561 */
562struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
563{
564 struct dentry *dentry;
565 struct super_block *sb = fsi->sb;
566 struct inode *inode;
567 struct dentry *root;
568 struct pts_mount_opts *opts;
569 char s[12];
570
571 root = sb->s_root;
572 opts = &fsi->mount_opts;
573
574 inode = new_inode(sb);
575 if (!inode)
576 return ERR_PTR(-ENOMEM);
577
578 inode->i_ino = index + 3;
579 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
580 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
581 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
582 init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
583
584 sprintf(s, "%d", index);
585
586 dentry = d_alloc_name(root, s);
587 if (dentry) {
588 dentry->d_fsdata = priv;
589 d_add(dentry, inode);
590 fsnotify_create(d_inode(root), dentry);
591 } else {
592 iput(inode);
593 dentry = ERR_PTR(-ENOMEM);
594 }
595
596 return dentry;
597}
598
599/**
600 * devpts_get_priv -- get private data for a slave
601 * @pts_inode: inode of the slave
602 *
603 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
604 */
605void *devpts_get_priv(struct dentry *dentry)
606{
607 if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
608 return NULL;
609 return dentry->d_fsdata;
610}
611
612/**
613 * devpts_pty_kill -- remove inode form /dev/pts/
614 * @inode: inode of the slave to be removed
615 *
616 * This is an inverse operation of devpts_pty_new.
617 */
618void devpts_pty_kill(struct dentry *dentry)
619{
620 WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
621
622 dentry->d_fsdata = NULL;
623 drop_nlink(dentry->d_inode);
624 fsnotify_unlink(d_inode(dentry->d_parent), dentry);
625 d_drop(dentry);
626 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
627}
628
629static int __init init_devpts_fs(void)
630{
631 int err = register_filesystem(&devpts_fs_type);
632 if (!err) {
633 register_sysctl_table(pty_root_table);
634 }
635 return err;
636}
637module_init(init_devpts_fs)
1/* -*- linux-c -*- --------------------------------------------------------- *
2 *
3 * linux/fs/devpts/inode.c
4 *
5 * Copyright 1998-2004 H. Peter Anvin -- All Rights Reserved
6 *
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.
10 *
11 * ------------------------------------------------------------------------- */
12
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/fs.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>
28
29#define DEVPTS_DEFAULT_MODE 0600
30/*
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
34 * permissions.
35 */
36#define DEVPTS_DEFAULT_PTMX_MODE 0000
37#define PTMX_MINOR 2
38
39/*
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.
42 */
43static int pty_limit = NR_UNIX98_PTY_DEFAULT;
44static int pty_reserve = NR_UNIX98_PTY_RESERVE;
45static int pty_limit_min;
46static int pty_limit_max = INT_MAX;
47static int pty_count;
48
49static struct ctl_table pty_table[] = {
50 {
51 .procname = "max",
52 .maxlen = sizeof(int),
53 .mode = 0644,
54 .data = &pty_limit,
55 .proc_handler = proc_dointvec_minmax,
56 .extra1 = &pty_limit_min,
57 .extra2 = &pty_limit_max,
58 }, {
59 .procname = "reserve",
60 .maxlen = sizeof(int),
61 .mode = 0644,
62 .data = &pty_reserve,
63 .proc_handler = proc_dointvec_minmax,
64 .extra1 = &pty_limit_min,
65 .extra2 = &pty_limit_max,
66 }, {
67 .procname = "nr",
68 .maxlen = sizeof(int),
69 .mode = 0444,
70 .data = &pty_count,
71 .proc_handler = proc_dointvec,
72 },
73 {}
74};
75
76static struct ctl_table pty_kern_table[] = {
77 {
78 .procname = "pty",
79 .mode = 0555,
80 .child = pty_table,
81 },
82 {}
83};
84
85static struct ctl_table pty_root_table[] = {
86 {
87 .procname = "kernel",
88 .mode = 0555,
89 .child = pty_kern_table,
90 },
91 {}
92};
93
94static DEFINE_MUTEX(allocated_ptys_lock);
95
96static struct vfsmount *devpts_mnt;
97
98struct pts_mount_opts {
99 int setuid;
100 int setgid;
101 kuid_t uid;
102 kgid_t gid;
103 umode_t mode;
104 umode_t ptmxmode;
105 int newinstance;
106 int max;
107};
108
109enum {
110 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
111 Opt_err
112};
113
114static const match_table_t tokens = {
115 {Opt_uid, "uid=%u"},
116 {Opt_gid, "gid=%u"},
117 {Opt_mode, "mode=%o"},
118#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
119 {Opt_ptmxmode, "ptmxmode=%o"},
120 {Opt_newinstance, "newinstance"},
121 {Opt_max, "max=%d"},
122#endif
123 {Opt_err, NULL}
124};
125
126struct pts_fs_info {
127 struct ida allocated_ptys;
128 struct pts_mount_opts mount_opts;
129 struct dentry *ptmx_dentry;
130};
131
132static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
133{
134 return sb->s_fs_info;
135}
136
137static inline struct super_block *pts_sb_from_inode(struct inode *inode)
138{
139#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
140 if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
141 return inode->i_sb;
142#endif
143 return devpts_mnt->mnt_sb;
144}
145
146#define PARSE_MOUNT 0
147#define PARSE_REMOUNT 1
148
149/*
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).
155 *
156 * Note: @data may be NULL (in which case all options are set to default).
157 */
158static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
159{
160 char *p;
161 kuid_t uid;
162 kgid_t gid;
163
164 opts->setuid = 0;
165 opts->setgid = 0;
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;
171
172 /* newinstance makes sense only on initial mount */
173 if (op == PARSE_MOUNT)
174 opts->newinstance = 0;
175
176 while ((p = strsep(&data, ",")) != NULL) {
177 substring_t args[MAX_OPT_ARGS];
178 int token;
179 int option;
180
181 if (!*p)
182 continue;
183
184 token = match_token(p, tokens, args);
185 switch (token) {
186 case Opt_uid:
187 if (match_int(&args[0], &option))
188 return -EINVAL;
189 uid = make_kuid(current_user_ns(), option);
190 if (!uid_valid(uid))
191 return -EINVAL;
192 opts->uid = uid;
193 opts->setuid = 1;
194 break;
195 case Opt_gid:
196 if (match_int(&args[0], &option))
197 return -EINVAL;
198 gid = make_kgid(current_user_ns(), option);
199 if (!gid_valid(gid))
200 return -EINVAL;
201 opts->gid = gid;
202 opts->setgid = 1;
203 break;
204 case Opt_mode:
205 if (match_octal(&args[0], &option))
206 return -EINVAL;
207 opts->mode = option & S_IALLUGO;
208 break;
209#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
210 case Opt_ptmxmode:
211 if (match_octal(&args[0], &option))
212 return -EINVAL;
213 opts->ptmxmode = option & S_IALLUGO;
214 break;
215 case Opt_newinstance:
216 /* newinstance makes sense only on initial mount */
217 if (op == PARSE_MOUNT)
218 opts->newinstance = 1;
219 break;
220 case Opt_max:
221 if (match_int(&args[0], &option) ||
222 option < 0 || option > NR_UNIX98_PTY_MAX)
223 return -EINVAL;
224 opts->max = option;
225 break;
226#endif
227 default:
228 printk(KERN_ERR "devpts: called with bogus options\n");
229 return -EINVAL;
230 }
231 }
232
233 return 0;
234}
235
236#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
237static int mknod_ptmx(struct super_block *sb)
238{
239 int mode;
240 int rc = -ENOMEM;
241 struct dentry *dentry;
242 struct inode *inode;
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;
246 kuid_t root_uid;
247 kgid_t root_gid;
248
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))
252 return -EINVAL;
253
254 mutex_lock(&root->d_inode->i_mutex);
255
256 /* If we have already created ptmx node, return */
257 if (fsi->ptmx_dentry) {
258 rc = 0;
259 goto out;
260 }
261
262 dentry = d_alloc_name(root, "ptmx");
263 if (!dentry) {
264 printk(KERN_NOTICE "Unable to alloc dentry for ptmx node\n");
265 goto out;
266 }
267
268 /*
269 * Create a new 'ptmx' node in this mount of devpts.
270 */
271 inode = new_inode(sb);
272 if (!inode) {
273 printk(KERN_ERR "Unable to alloc inode for ptmx node\n");
274 dput(dentry);
275 goto out;
276 }
277
278 inode->i_ino = 2;
279 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
280
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;
285
286 d_add(dentry, inode);
287
288 fsi->ptmx_dentry = dentry;
289 rc = 0;
290out:
291 mutex_unlock(&root->d_inode->i_mutex);
292 return rc;
293}
294
295static void update_ptmx_mode(struct pts_fs_info *fsi)
296{
297 struct inode *inode;
298 if (fsi->ptmx_dentry) {
299 inode = fsi->ptmx_dentry->d_inode;
300 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
301 }
302}
303#else
304static inline void update_ptmx_mode(struct pts_fs_info *fsi)
305{
306 return;
307}
308#endif
309
310static int devpts_remount(struct super_block *sb, int *flags, char *data)
311{
312 int err;
313 struct pts_fs_info *fsi = DEVPTS_SB(sb);
314 struct pts_mount_opts *opts = &fsi->mount_opts;
315
316 sync_filesystem(sb);
317 err = parse_mount_options(data, PARSE_REMOUNT, opts);
318
319 /*
320 * parse_mount_options() restores options to default values
321 * before parsing and may have changed ptmxmode. So, update the
322 * mode in the inode too. Bogus options don't fail the remount,
323 * so do this even on error return.
324 */
325 update_ptmx_mode(fsi);
326
327 return err;
328}
329
330static int devpts_show_options(struct seq_file *seq, struct dentry *root)
331{
332 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
333 struct pts_mount_opts *opts = &fsi->mount_opts;
334
335 if (opts->setuid)
336 seq_printf(seq, ",uid=%u", from_kuid_munged(&init_user_ns, opts->uid));
337 if (opts->setgid)
338 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, opts->gid));
339 seq_printf(seq, ",mode=%03o", opts->mode);
340#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
341 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
342 if (opts->max < NR_UNIX98_PTY_MAX)
343 seq_printf(seq, ",max=%d", opts->max);
344#endif
345
346 return 0;
347}
348
349static const struct super_operations devpts_sops = {
350 .statfs = simple_statfs,
351 .remount_fs = devpts_remount,
352 .show_options = devpts_show_options,
353};
354
355static void *new_pts_fs_info(void)
356{
357 struct pts_fs_info *fsi;
358
359 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
360 if (!fsi)
361 return NULL;
362
363 ida_init(&fsi->allocated_ptys);
364 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
365 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
366
367 return fsi;
368}
369
370static int
371devpts_fill_super(struct super_block *s, void *data, int silent)
372{
373 struct inode *inode;
374
375 s->s_blocksize = 1024;
376 s->s_blocksize_bits = 10;
377 s->s_magic = DEVPTS_SUPER_MAGIC;
378 s->s_op = &devpts_sops;
379 s->s_time_gran = 1;
380
381 s->s_fs_info = new_pts_fs_info();
382 if (!s->s_fs_info)
383 goto fail;
384
385 inode = new_inode(s);
386 if (!inode)
387 goto fail;
388 inode->i_ino = 1;
389 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
390 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
391 inode->i_op = &simple_dir_inode_operations;
392 inode->i_fop = &simple_dir_operations;
393 set_nlink(inode, 2);
394
395 s->s_root = d_make_root(inode);
396 if (s->s_root)
397 return 0;
398
399 printk(KERN_ERR "devpts: get root dentry failed\n");
400
401fail:
402 return -ENOMEM;
403}
404
405#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
406static int compare_init_pts_sb(struct super_block *s, void *p)
407{
408 if (devpts_mnt)
409 return devpts_mnt->mnt_sb == s;
410 return 0;
411}
412
413/*
414 * devpts_mount()
415 *
416 * If the '-o newinstance' mount option was specified, mount a new
417 * (private) instance of devpts. PTYs created in this instance are
418 * independent of the PTYs in other devpts instances.
419 *
420 * If the '-o newinstance' option was not specified, mount/remount the
421 * initial kernel mount of devpts. This type of mount gives the
422 * legacy, single-instance semantics.
423 *
424 * The 'newinstance' option is needed to support multiple namespace
425 * semantics in devpts while preserving backward compatibility of the
426 * current 'single-namespace' semantics. i.e all mounts of devpts
427 * without the 'newinstance' mount option should bind to the initial
428 * kernel mount, like mount_single().
429 *
430 * Mounts with 'newinstance' option create a new, private namespace.
431 *
432 * NOTE:
433 *
434 * For single-mount semantics, devpts cannot use mount_single(),
435 * because mount_single()/sget() find and use the super-block from
436 * the most recent mount of devpts. But that recent mount may be a
437 * 'newinstance' mount and mount_single() would pick the newinstance
438 * super-block instead of the initial super-block.
439 */
440static struct dentry *devpts_mount(struct file_system_type *fs_type,
441 int flags, const char *dev_name, void *data)
442{
443 int error;
444 struct pts_mount_opts opts;
445 struct super_block *s;
446
447 error = parse_mount_options(data, PARSE_MOUNT, &opts);
448 if (error)
449 return ERR_PTR(error);
450
451 /* Require newinstance for all user namespace mounts to ensure
452 * the mount options are not changed.
453 */
454 if ((current_user_ns() != &init_user_ns) && !opts.newinstance)
455 return ERR_PTR(-EINVAL);
456
457 if (opts.newinstance)
458 s = sget(fs_type, NULL, set_anon_super, flags, NULL);
459 else
460 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags,
461 NULL);
462
463 if (IS_ERR(s))
464 return ERR_CAST(s);
465
466 if (!s->s_root) {
467 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
468 if (error)
469 goto out_undo_sget;
470 s->s_flags |= MS_ACTIVE;
471 }
472
473 memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
474
475 error = mknod_ptmx(s);
476 if (error)
477 goto out_undo_sget;
478
479 return dget(s->s_root);
480
481out_undo_sget:
482 deactivate_locked_super(s);
483 return ERR_PTR(error);
484}
485
486#else
487/*
488 * This supports only the legacy single-instance semantics (no
489 * multiple-instance semantics)
490 */
491static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
492 const char *dev_name, void *data)
493{
494 return mount_single(fs_type, flags, data, devpts_fill_super);
495}
496#endif
497
498static void devpts_kill_sb(struct super_block *sb)
499{
500 struct pts_fs_info *fsi = DEVPTS_SB(sb);
501
502 ida_destroy(&fsi->allocated_ptys);
503 kfree(fsi);
504 kill_litter_super(sb);
505}
506
507static struct file_system_type devpts_fs_type = {
508 .name = "devpts",
509 .mount = devpts_mount,
510 .kill_sb = devpts_kill_sb,
511#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
512 .fs_flags = FS_USERNS_MOUNT | FS_USERNS_DEV_MOUNT,
513#endif
514};
515
516/*
517 * The normal naming convention is simply /dev/pts/<number>; this conforms
518 * to the System V naming convention
519 */
520
521int devpts_new_index(struct inode *ptmx_inode)
522{
523 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
524 struct pts_fs_info *fsi = DEVPTS_SB(sb);
525 int index;
526 int ida_ret;
527
528retry:
529 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
530 return -ENOMEM;
531
532 mutex_lock(&allocated_ptys_lock);
533 if (pty_count >= pty_limit -
534 (fsi->mount_opts.newinstance ? pty_reserve : 0)) {
535 mutex_unlock(&allocated_ptys_lock);
536 return -ENOSPC;
537 }
538
539 ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
540 if (ida_ret < 0) {
541 mutex_unlock(&allocated_ptys_lock);
542 if (ida_ret == -EAGAIN)
543 goto retry;
544 return -EIO;
545 }
546
547 if (index >= fsi->mount_opts.max) {
548 ida_remove(&fsi->allocated_ptys, index);
549 mutex_unlock(&allocated_ptys_lock);
550 return -ENOSPC;
551 }
552 pty_count++;
553 mutex_unlock(&allocated_ptys_lock);
554 return index;
555}
556
557void devpts_kill_index(struct inode *ptmx_inode, int idx)
558{
559 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
560 struct pts_fs_info *fsi = DEVPTS_SB(sb);
561
562 mutex_lock(&allocated_ptys_lock);
563 ida_remove(&fsi->allocated_ptys, idx);
564 pty_count--;
565 mutex_unlock(&allocated_ptys_lock);
566}
567
568/**
569 * devpts_pty_new -- create a new inode in /dev/pts/
570 * @ptmx_inode: inode of the master
571 * @device: major+minor of the node to be created
572 * @index: used as a name of the node
573 * @priv: what's given back by devpts_get_priv
574 *
575 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
576 */
577struct inode *devpts_pty_new(struct inode *ptmx_inode, dev_t device, int index,
578 void *priv)
579{
580 struct dentry *dentry;
581 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
582 struct inode *inode;
583 struct dentry *root = sb->s_root;
584 struct pts_fs_info *fsi = DEVPTS_SB(sb);
585 struct pts_mount_opts *opts = &fsi->mount_opts;
586 char s[12];
587
588 inode = new_inode(sb);
589 if (!inode)
590 return ERR_PTR(-ENOMEM);
591
592 inode->i_ino = index + 3;
593 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
594 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
595 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
596 init_special_inode(inode, S_IFCHR|opts->mode, device);
597 inode->i_private = priv;
598
599 sprintf(s, "%d", index);
600
601 mutex_lock(&root->d_inode->i_mutex);
602
603 dentry = d_alloc_name(root, s);
604 if (dentry) {
605 d_add(dentry, inode);
606 fsnotify_create(root->d_inode, dentry);
607 } else {
608 iput(inode);
609 inode = ERR_PTR(-ENOMEM);
610 }
611
612 mutex_unlock(&root->d_inode->i_mutex);
613
614 return inode;
615}
616
617/**
618 * devpts_get_priv -- get private data for a slave
619 * @pts_inode: inode of the slave
620 *
621 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
622 */
623void *devpts_get_priv(struct inode *pts_inode)
624{
625 struct dentry *dentry;
626 void *priv = NULL;
627
628 BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
629
630 /* Ensure dentry has not been deleted by devpts_pty_kill() */
631 dentry = d_find_alias(pts_inode);
632 if (!dentry)
633 return NULL;
634
635 if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
636 priv = pts_inode->i_private;
637
638 dput(dentry);
639
640 return priv;
641}
642
643/**
644 * devpts_pty_kill -- remove inode form /dev/pts/
645 * @inode: inode of the slave to be removed
646 *
647 * This is an inverse operation of devpts_pty_new.
648 */
649void devpts_pty_kill(struct inode *inode)
650{
651 struct super_block *sb = pts_sb_from_inode(inode);
652 struct dentry *root = sb->s_root;
653 struct dentry *dentry;
654
655 BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
656
657 mutex_lock(&root->d_inode->i_mutex);
658
659 dentry = d_find_alias(inode);
660
661 drop_nlink(inode);
662 d_delete(dentry);
663 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
664 dput(dentry); /* d_find_alias above */
665
666 mutex_unlock(&root->d_inode->i_mutex);
667}
668
669static int __init init_devpts_fs(void)
670{
671 int err = register_filesystem(&devpts_fs_type);
672 struct ctl_table_header *table;
673
674 if (!err) {
675 table = register_sysctl_table(pty_root_table);
676 devpts_mnt = kern_mount(&devpts_fs_type);
677 if (IS_ERR(devpts_mnt)) {
678 err = PTR_ERR(devpts_mnt);
679 unregister_filesystem(&devpts_fs_type);
680 unregister_sysctl_table(table);
681 }
682 }
683 return err;
684}
685module_init(init_devpts_fs)