Loading...
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/module.h>
16#include <linux/init.h>
17#include <linux/fs.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>
30
31#define DEVPTS_DEFAULT_MODE 0600
32/*
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
36 * permissions.
37 */
38#define DEVPTS_DEFAULT_PTMX_MODE 0000
39#define PTMX_MINOR 2
40
41/*
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.
44 */
45static int pty_limit = NR_UNIX98_PTY_DEFAULT;
46static int pty_reserve = NR_UNIX98_PTY_RESERVE;
47static int pty_limit_min;
48static int pty_limit_max = INT_MAX;
49static int pty_count;
50
51static struct ctl_table pty_table[] = {
52 {
53 .procname = "max",
54 .maxlen = sizeof(int),
55 .mode = 0644,
56 .data = &pty_limit,
57 .proc_handler = proc_dointvec_minmax,
58 .extra1 = &pty_limit_min,
59 .extra2 = &pty_limit_max,
60 }, {
61 .procname = "reserve",
62 .maxlen = sizeof(int),
63 .mode = 0644,
64 .data = &pty_reserve,
65 .proc_handler = proc_dointvec_minmax,
66 .extra1 = &pty_limit_min,
67 .extra2 = &pty_limit_max,
68 }, {
69 .procname = "nr",
70 .maxlen = sizeof(int),
71 .mode = 0444,
72 .data = &pty_count,
73 .proc_handler = proc_dointvec,
74 },
75 {}
76};
77
78static struct ctl_table pty_kern_table[] = {
79 {
80 .procname = "pty",
81 .mode = 0555,
82 .child = pty_table,
83 },
84 {}
85};
86
87static struct ctl_table pty_root_table[] = {
88 {
89 .procname = "kernel",
90 .mode = 0555,
91 .child = pty_kern_table,
92 },
93 {}
94};
95
96static DEFINE_MUTEX(allocated_ptys_lock);
97
98static struct vfsmount *devpts_mnt;
99
100struct pts_mount_opts {
101 int setuid;
102 int setgid;
103 kuid_t uid;
104 kgid_t gid;
105 umode_t mode;
106 umode_t ptmxmode;
107 int newinstance;
108 int max;
109};
110
111enum {
112 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
113 Opt_err
114};
115
116static const match_table_t tokens = {
117 {Opt_uid, "uid=%u"},
118 {Opt_gid, "gid=%u"},
119 {Opt_mode, "mode=%o"},
120#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
121 {Opt_ptmxmode, "ptmxmode=%o"},
122 {Opt_newinstance, "newinstance"},
123 {Opt_max, "max=%d"},
124#endif
125 {Opt_err, NULL}
126};
127
128struct pts_fs_info {
129 struct ida allocated_ptys;
130 struct pts_mount_opts mount_opts;
131 struct super_block *sb;
132 struct dentry *ptmx_dentry;
133};
134
135static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
136{
137 return sb->s_fs_info;
138}
139
140static inline struct super_block *pts_sb_from_inode(struct inode *inode)
141{
142#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
143 if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
144 return inode->i_sb;
145#endif
146 if (!devpts_mnt)
147 return NULL;
148 return devpts_mnt->mnt_sb;
149}
150
151#define PARSE_MOUNT 0
152#define PARSE_REMOUNT 1
153
154/*
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).
160 *
161 * Note: @data may be NULL (in which case all options are set to default).
162 */
163static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
164{
165 char *p;
166 kuid_t uid;
167 kgid_t gid;
168
169 opts->setuid = 0;
170 opts->setgid = 0;
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;
176
177 /* newinstance makes sense only on initial mount */
178 if (op == PARSE_MOUNT)
179 opts->newinstance = 0;
180
181 while ((p = strsep(&data, ",")) != NULL) {
182 substring_t args[MAX_OPT_ARGS];
183 int token;
184 int option;
185
186 if (!*p)
187 continue;
188
189 token = match_token(p, tokens, args);
190 switch (token) {
191 case Opt_uid:
192 if (match_int(&args[0], &option))
193 return -EINVAL;
194 uid = make_kuid(current_user_ns(), option);
195 if (!uid_valid(uid))
196 return -EINVAL;
197 opts->uid = uid;
198 opts->setuid = 1;
199 break;
200 case Opt_gid:
201 if (match_int(&args[0], &option))
202 return -EINVAL;
203 gid = make_kgid(current_user_ns(), option);
204 if (!gid_valid(gid))
205 return -EINVAL;
206 opts->gid = gid;
207 opts->setgid = 1;
208 break;
209 case Opt_mode:
210 if (match_octal(&args[0], &option))
211 return -EINVAL;
212 opts->mode = option & S_IALLUGO;
213 break;
214#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
215 case Opt_ptmxmode:
216 if (match_octal(&args[0], &option))
217 return -EINVAL;
218 opts->ptmxmode = option & S_IALLUGO;
219 break;
220 case Opt_newinstance:
221 /* newinstance makes sense only on initial mount */
222 if (op == PARSE_MOUNT)
223 opts->newinstance = 1;
224 break;
225 case Opt_max:
226 if (match_int(&args[0], &option) ||
227 option < 0 || option > NR_UNIX98_PTY_MAX)
228 return -EINVAL;
229 opts->max = option;
230 break;
231#endif
232 default:
233 pr_err("called with bogus options\n");
234 return -EINVAL;
235 }
236 }
237
238 return 0;
239}
240
241#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
242static int mknod_ptmx(struct super_block *sb)
243{
244 int mode;
245 int rc = -ENOMEM;
246 struct dentry *dentry;
247 struct inode *inode;
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;
251 kuid_t root_uid;
252 kgid_t root_gid;
253
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))
257 return -EINVAL;
258
259 inode_lock(d_inode(root));
260
261 /* If we have already created ptmx node, return */
262 if (fsi->ptmx_dentry) {
263 rc = 0;
264 goto out;
265 }
266
267 dentry = d_alloc_name(root, "ptmx");
268 if (!dentry) {
269 pr_err("Unable to alloc dentry for ptmx node\n");
270 goto out;
271 }
272
273 /*
274 * Create a new 'ptmx' node in this mount of devpts.
275 */
276 inode = new_inode(sb);
277 if (!inode) {
278 pr_err("Unable to alloc inode for ptmx node\n");
279 dput(dentry);
280 goto out;
281 }
282
283 inode->i_ino = 2;
284 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
285
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;
290
291 d_add(dentry, inode);
292
293 fsi->ptmx_dentry = dentry;
294 rc = 0;
295out:
296 inode_unlock(d_inode(root));
297 return rc;
298}
299
300static void update_ptmx_mode(struct pts_fs_info *fsi)
301{
302 struct inode *inode;
303 if (fsi->ptmx_dentry) {
304 inode = d_inode(fsi->ptmx_dentry);
305 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
306 }
307}
308#else
309static inline void update_ptmx_mode(struct pts_fs_info *fsi)
310{
311 return;
312}
313#endif
314
315static int devpts_remount(struct super_block *sb, int *flags, char *data)
316{
317 int err;
318 struct pts_fs_info *fsi = DEVPTS_SB(sb);
319 struct pts_mount_opts *opts = &fsi->mount_opts;
320
321 sync_filesystem(sb);
322 err = parse_mount_options(data, PARSE_REMOUNT, opts);
323
324 /*
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.
329 */
330 update_ptmx_mode(fsi);
331
332 return err;
333}
334
335static int devpts_show_options(struct seq_file *seq, struct dentry *root)
336{
337 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
338 struct pts_mount_opts *opts = &fsi->mount_opts;
339
340 if (opts->setuid)
341 seq_printf(seq, ",uid=%u",
342 from_kuid_munged(&init_user_ns, opts->uid));
343 if (opts->setgid)
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);
351#endif
352
353 return 0;
354}
355
356static const struct super_operations devpts_sops = {
357 .statfs = simple_statfs,
358 .remount_fs = devpts_remount,
359 .show_options = devpts_show_options,
360};
361
362static void *new_pts_fs_info(struct super_block *sb)
363{
364 struct pts_fs_info *fsi;
365
366 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
367 if (!fsi)
368 return NULL;
369
370 ida_init(&fsi->allocated_ptys);
371 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
372 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
373 fsi->sb = sb;
374
375 return fsi;
376}
377
378static int
379devpts_fill_super(struct super_block *s, void *data, int silent)
380{
381 struct inode *inode;
382
383 s->s_blocksize = 1024;
384 s->s_blocksize_bits = 10;
385 s->s_magic = DEVPTS_SUPER_MAGIC;
386 s->s_op = &devpts_sops;
387 s->s_time_gran = 1;
388
389 s->s_fs_info = new_pts_fs_info(s);
390 if (!s->s_fs_info)
391 goto fail;
392
393 inode = new_inode(s);
394 if (!inode)
395 goto fail;
396 inode->i_ino = 1;
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;
401 set_nlink(inode, 2);
402
403 s->s_root = d_make_root(inode);
404 if (s->s_root)
405 return 0;
406
407 pr_err("get root dentry failed\n");
408
409fail:
410 return -ENOMEM;
411}
412
413#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
414static int compare_init_pts_sb(struct super_block *s, void *p)
415{
416 if (devpts_mnt)
417 return devpts_mnt->mnt_sb == s;
418 return 0;
419}
420
421/*
422 * devpts_mount()
423 *
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.
427 *
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.
431 *
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().
437 *
438 * Mounts with 'newinstance' option create a new, private namespace.
439 *
440 * NOTE:
441 *
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.
447 */
448static struct dentry *devpts_mount(struct file_system_type *fs_type,
449 int flags, const char *dev_name, void *data)
450{
451 int error;
452 struct pts_mount_opts opts;
453 struct super_block *s;
454
455 error = parse_mount_options(data, PARSE_MOUNT, &opts);
456 if (error)
457 return ERR_PTR(error);
458
459 /* Require newinstance for all user namespace mounts to ensure
460 * the mount options are not changed.
461 */
462 if ((current_user_ns() != &init_user_ns) && !opts.newinstance)
463 return ERR_PTR(-EINVAL);
464
465 if (opts.newinstance)
466 s = sget(fs_type, NULL, set_anon_super, flags, NULL);
467 else
468 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags,
469 NULL);
470
471 if (IS_ERR(s))
472 return ERR_CAST(s);
473
474 if (!s->s_root) {
475 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
476 if (error)
477 goto out_undo_sget;
478 s->s_flags |= MS_ACTIVE;
479 }
480
481 memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
482
483 error = mknod_ptmx(s);
484 if (error)
485 goto out_undo_sget;
486
487 return dget(s->s_root);
488
489out_undo_sget:
490 deactivate_locked_super(s);
491 return ERR_PTR(error);
492}
493
494#else
495/*
496 * This supports only the legacy single-instance semantics (no
497 * multiple-instance semantics)
498 */
499static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
500 const char *dev_name, void *data)
501{
502 return mount_single(fs_type, flags, data, devpts_fill_super);
503}
504#endif
505
506static void devpts_kill_sb(struct super_block *sb)
507{
508 struct pts_fs_info *fsi = DEVPTS_SB(sb);
509
510 ida_destroy(&fsi->allocated_ptys);
511 kfree(fsi);
512 kill_litter_super(sb);
513}
514
515static struct file_system_type devpts_fs_type = {
516 .name = "devpts",
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,
521#endif
522};
523
524/*
525 * The normal naming convention is simply /dev/pts/<number>; this conforms
526 * to the System V naming convention
527 */
528
529int devpts_new_index(struct pts_fs_info *fsi)
530{
531 int index;
532 int ida_ret;
533
534 if (!fsi)
535 return -ENODEV;
536
537retry:
538 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
539 return -ENOMEM;
540
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);
545 return -ENOSPC;
546 }
547
548 ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
549 if (ida_ret < 0) {
550 mutex_unlock(&allocated_ptys_lock);
551 if (ida_ret == -EAGAIN)
552 goto retry;
553 return -EIO;
554 }
555
556 if (index >= fsi->mount_opts.max) {
557 ida_remove(&fsi->allocated_ptys, index);
558 mutex_unlock(&allocated_ptys_lock);
559 return -ENOSPC;
560 }
561 pty_count++;
562 mutex_unlock(&allocated_ptys_lock);
563 return index;
564}
565
566void devpts_kill_index(struct pts_fs_info *fsi, int idx)
567{
568 mutex_lock(&allocated_ptys_lock);
569 ida_remove(&fsi->allocated_ptys, idx);
570 pty_count--;
571 mutex_unlock(&allocated_ptys_lock);
572}
573
574/*
575 * pty code needs to hold extra references in case of last /dev/tty close
576 */
577struct pts_fs_info *devpts_get_ref(struct inode *ptmx_inode, struct file *file)
578{
579 struct super_block *sb;
580 struct pts_fs_info *fsi;
581
582 sb = pts_sb_from_inode(ptmx_inode);
583 if (!sb)
584 return NULL;
585 fsi = DEVPTS_SB(sb);
586 if (!fsi)
587 return NULL;
588
589 atomic_inc(&sb->s_active);
590 return fsi;
591}
592
593void devpts_put_ref(struct pts_fs_info *fsi)
594{
595 deactivate_super(fsi->sb);
596}
597
598/**
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
604 *
605 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
606 */
607struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
608{
609 struct dentry *dentry;
610 struct super_block *sb;
611 struct inode *inode;
612 struct dentry *root;
613 struct pts_mount_opts *opts;
614 char s[12];
615
616 if (!fsi)
617 return ERR_PTR(-ENODEV);
618
619 sb = fsi->sb;
620 root = sb->s_root;
621 opts = &fsi->mount_opts;
622
623 inode = new_inode(sb);
624 if (!inode)
625 return ERR_PTR(-ENOMEM);
626
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));
632
633 sprintf(s, "%d", index);
634
635 dentry = d_alloc_name(root, s);
636 if (dentry) {
637 dentry->d_fsdata = priv;
638 d_add(dentry, inode);
639 fsnotify_create(d_inode(root), dentry);
640 } else {
641 iput(inode);
642 dentry = ERR_PTR(-ENOMEM);
643 }
644
645 return dentry;
646}
647
648/**
649 * devpts_get_priv -- get private data for a slave
650 * @pts_inode: inode of the slave
651 *
652 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
653 */
654void *devpts_get_priv(struct dentry *dentry)
655{
656 WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
657 return dentry->d_fsdata;
658}
659
660/**
661 * devpts_pty_kill -- remove inode form /dev/pts/
662 * @inode: inode of the slave to be removed
663 *
664 * This is an inverse operation of devpts_pty_new.
665 */
666void devpts_pty_kill(struct dentry *dentry)
667{
668 WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
669
670 dentry->d_fsdata = NULL;
671 drop_nlink(dentry->d_inode);
672 d_delete(dentry);
673 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
674}
675
676static int __init init_devpts_fs(void)
677{
678 int err = register_filesystem(&devpts_fs_type);
679 struct ctl_table_header *table;
680
681 if (!err) {
682 struct vfsmount *mnt;
683
684 table = register_sysctl_table(pty_root_table);
685 mnt = kern_mount(&devpts_fs_type);
686 if (IS_ERR(mnt)) {
687 err = PTR_ERR(mnt);
688 unregister_filesystem(&devpts_fs_type);
689 unregister_sysctl_table(table);
690 } else {
691 devpts_mnt = mnt;
692 }
693 }
694 return err;
695}
696module_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)