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 */
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;
49 static int pty_count;
50
51 static 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
78 static struct ctl_table pty_kern_table[] = {
79 {
80 .procname = "pty",
81 .mode = 0555,
82 .child = pty_table,
83 },
84 {}
85 };
86
87 static struct ctl_table pty_root_table[] = {
88 {
89 .procname = "kernel",
90 .mode = 0555,
91 .child = pty_kern_table,
92 },
93 {}
94 };
95
96 static DEFINE_MUTEX(allocated_ptys_lock);
97
98 struct 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 reserve;
106 int max;
107 };
108
109 enum {
110 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
111 Opt_err
112 };
113
114 static const match_table_t tokens = {
115 {Opt_uid, "uid=%u"},
116 {Opt_gid, "gid=%u"},
117 {Opt_mode, "mode=%o"},
118 {Opt_ptmxmode, "ptmxmode=%o"},
119 {Opt_newinstance, "newinstance"},
120 {Opt_max, "max=%d"},
121 {Opt_err, NULL}
122 };
123
124 struct pts_fs_info {
125 struct ida allocated_ptys;
126 struct pts_mount_opts mount_opts;
127 struct super_block *sb;
128 struct dentry *ptmx_dentry;
129 };
130
DEVPTS_SB(struct super_block * sb)131 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
132 {
133 return sb->s_fs_info;
134 }
135
devpts_ptmx_path(struct path * path)136 static int devpts_ptmx_path(struct path *path)
137 {
138 struct super_block *sb;
139 int err;
140
141 /* Is a devpts filesystem at "pts" in the same directory? */
142 err = path_pts(path);
143 if (err)
144 return err;
145
146 /* Is the path the root of a devpts filesystem? */
147 sb = path->mnt->mnt_sb;
148 if ((sb->s_magic != DEVPTS_SUPER_MAGIC) ||
149 (path->mnt->mnt_root != sb->s_root))
150 return -ENODEV;
151
152 return 0;
153 }
154
devpts_mntget(struct file * filp,struct pts_fs_info * fsi)155 struct vfsmount *devpts_mntget(struct file *filp, struct pts_fs_info *fsi)
156 {
157 struct path path;
158 int err = 0;
159
160 path = filp->f_path;
161 path_get(&path);
162
163 /* Walk upward while the start point is a bind mount of
164 * a single file.
165 */
166 while (path.mnt->mnt_root == path.dentry)
167 if (follow_up(&path) == 0)
168 break;
169
170 /* devpts_ptmx_path() finds a devpts fs or returns an error. */
171 if ((path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) ||
172 (DEVPTS_SB(path.mnt->mnt_sb) != fsi))
173 err = devpts_ptmx_path(&path);
174 dput(path.dentry);
175 if (!err) {
176 if (DEVPTS_SB(path.mnt->mnt_sb) == fsi)
177 return path.mnt;
178
179 err = -ENODEV;
180 }
181
182 mntput(path.mnt);
183 return ERR_PTR(err);
184 }
185
devpts_acquire(struct file * filp)186 struct pts_fs_info *devpts_acquire(struct file *filp)
187 {
188 struct pts_fs_info *result;
189 struct path path;
190 struct super_block *sb;
191
192 path = filp->f_path;
193 path_get(&path);
194
195 /* Has the devpts filesystem already been found? */
196 if (path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) {
197 int err;
198
199 err = devpts_ptmx_path(&path);
200 if (err) {
201 result = ERR_PTR(err);
202 goto out;
203 }
204 }
205
206 /*
207 * pty code needs to hold extra references in case of last /dev/tty close
208 */
209 sb = path.mnt->mnt_sb;
210 atomic_inc(&sb->s_active);
211 result = DEVPTS_SB(sb);
212
213 out:
214 path_put(&path);
215 return result;
216 }
217
devpts_release(struct pts_fs_info * fsi)218 void devpts_release(struct pts_fs_info *fsi)
219 {
220 deactivate_super(fsi->sb);
221 }
222
223 #define PARSE_MOUNT 0
224 #define PARSE_REMOUNT 1
225
226 /*
227 * parse_mount_options():
228 * Set @opts to mount options specified in @data. If an option is not
229 * specified in @data, set it to its default value.
230 *
231 * Note: @data may be NULL (in which case all options are set to default).
232 */
parse_mount_options(char * data,int op,struct pts_mount_opts * opts)233 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
234 {
235 char *p;
236 kuid_t uid;
237 kgid_t gid;
238
239 opts->setuid = 0;
240 opts->setgid = 0;
241 opts->uid = GLOBAL_ROOT_UID;
242 opts->gid = GLOBAL_ROOT_GID;
243 opts->mode = DEVPTS_DEFAULT_MODE;
244 opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
245 opts->max = NR_UNIX98_PTY_MAX;
246
247 /* Only allow instances mounted from the initial mount
248 * namespace to tap the reserve pool of ptys.
249 */
250 if (op == PARSE_MOUNT)
251 opts->reserve =
252 (current->nsproxy->mnt_ns == init_task.nsproxy->mnt_ns);
253
254 while ((p = strsep(&data, ",")) != NULL) {
255 substring_t args[MAX_OPT_ARGS];
256 int token;
257 int option;
258
259 if (!*p)
260 continue;
261
262 token = match_token(p, tokens, args);
263 switch (token) {
264 case Opt_uid:
265 if (match_int(&args[0], &option))
266 return -EINVAL;
267 uid = make_kuid(current_user_ns(), option);
268 if (!uid_valid(uid))
269 return -EINVAL;
270 opts->uid = uid;
271 opts->setuid = 1;
272 break;
273 case Opt_gid:
274 if (match_int(&args[0], &option))
275 return -EINVAL;
276 gid = make_kgid(current_user_ns(), option);
277 if (!gid_valid(gid))
278 return -EINVAL;
279 opts->gid = gid;
280 opts->setgid = 1;
281 break;
282 case Opt_mode:
283 if (match_octal(&args[0], &option))
284 return -EINVAL;
285 opts->mode = option & S_IALLUGO;
286 break;
287 case Opt_ptmxmode:
288 if (match_octal(&args[0], &option))
289 return -EINVAL;
290 opts->ptmxmode = option & S_IALLUGO;
291 break;
292 case Opt_newinstance:
293 break;
294 case Opt_max:
295 if (match_int(&args[0], &option) ||
296 option < 0 || option > NR_UNIX98_PTY_MAX)
297 return -EINVAL;
298 opts->max = option;
299 break;
300 default:
301 pr_err("called with bogus options\n");
302 return -EINVAL;
303 }
304 }
305
306 return 0;
307 }
308
mknod_ptmx(struct super_block * sb)309 static int mknod_ptmx(struct super_block *sb)
310 {
311 int mode;
312 int rc = -ENOMEM;
313 struct dentry *dentry;
314 struct inode *inode;
315 struct dentry *root = sb->s_root;
316 struct pts_fs_info *fsi = DEVPTS_SB(sb);
317 struct pts_mount_opts *opts = &fsi->mount_opts;
318 kuid_t ptmx_uid = current_fsuid();
319 kgid_t ptmx_gid = current_fsgid();
320
321 inode_lock(d_inode(root));
322
323 /* If we have already created ptmx node, return */
324 if (fsi->ptmx_dentry) {
325 rc = 0;
326 goto out;
327 }
328
329 dentry = d_alloc_name(root, "ptmx");
330 if (!dentry) {
331 pr_err("Unable to alloc dentry for ptmx node\n");
332 goto out;
333 }
334
335 /*
336 * Create a new 'ptmx' node in this mount of devpts.
337 */
338 inode = new_inode(sb);
339 if (!inode) {
340 pr_err("Unable to alloc inode for ptmx node\n");
341 dput(dentry);
342 goto out;
343 }
344
345 inode->i_ino = 2;
346 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
347
348 mode = S_IFCHR|opts->ptmxmode;
349 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
350 inode->i_uid = ptmx_uid;
351 inode->i_gid = ptmx_gid;
352
353 d_add(dentry, inode);
354
355 fsi->ptmx_dentry = dentry;
356 rc = 0;
357 out:
358 inode_unlock(d_inode(root));
359 return rc;
360 }
361
update_ptmx_mode(struct pts_fs_info * fsi)362 static void update_ptmx_mode(struct pts_fs_info *fsi)
363 {
364 struct inode *inode;
365 if (fsi->ptmx_dentry) {
366 inode = d_inode(fsi->ptmx_dentry);
367 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
368 }
369 }
370
devpts_remount(struct super_block * sb,int * flags,char * data)371 static int devpts_remount(struct super_block *sb, int *flags, char *data)
372 {
373 int err;
374 struct pts_fs_info *fsi = DEVPTS_SB(sb);
375 struct pts_mount_opts *opts = &fsi->mount_opts;
376
377 err = parse_mount_options(data, PARSE_REMOUNT, opts);
378
379 /*
380 * parse_mount_options() restores options to default values
381 * before parsing and may have changed ptmxmode. So, update the
382 * mode in the inode too. Bogus options don't fail the remount,
383 * so do this even on error return.
384 */
385 update_ptmx_mode(fsi);
386
387 return err;
388 }
389
devpts_show_options(struct seq_file * seq,struct dentry * root)390 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
391 {
392 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
393 struct pts_mount_opts *opts = &fsi->mount_opts;
394
395 if (opts->setuid)
396 seq_printf(seq, ",uid=%u",
397 from_kuid_munged(&init_user_ns, opts->uid));
398 if (opts->setgid)
399 seq_printf(seq, ",gid=%u",
400 from_kgid_munged(&init_user_ns, opts->gid));
401 seq_printf(seq, ",mode=%03o", opts->mode);
402 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
403 if (opts->max < NR_UNIX98_PTY_MAX)
404 seq_printf(seq, ",max=%d", opts->max);
405
406 return 0;
407 }
408
409 static const struct super_operations devpts_sops = {
410 .statfs = simple_statfs,
411 .remount_fs = devpts_remount,
412 .show_options = devpts_show_options,
413 };
414
new_pts_fs_info(struct super_block * sb)415 static void *new_pts_fs_info(struct super_block *sb)
416 {
417 struct pts_fs_info *fsi;
418
419 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
420 if (!fsi)
421 return NULL;
422
423 ida_init(&fsi->allocated_ptys);
424 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
425 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
426 fsi->sb = sb;
427
428 return fsi;
429 }
430
431 static int
devpts_fill_super(struct super_block * s,void * data,int silent)432 devpts_fill_super(struct super_block *s, void *data, int silent)
433 {
434 struct inode *inode;
435 int error;
436
437 s->s_iflags &= ~SB_I_NODEV;
438 s->s_blocksize = 1024;
439 s->s_blocksize_bits = 10;
440 s->s_magic = DEVPTS_SUPER_MAGIC;
441 s->s_op = &devpts_sops;
442 s->s_d_op = &simple_dentry_operations;
443 s->s_time_gran = 1;
444
445 error = -ENOMEM;
446 s->s_fs_info = new_pts_fs_info(s);
447 if (!s->s_fs_info)
448 goto fail;
449
450 error = parse_mount_options(data, PARSE_MOUNT, &DEVPTS_SB(s)->mount_opts);
451 if (error)
452 goto fail;
453
454 error = -ENOMEM;
455 inode = new_inode(s);
456 if (!inode)
457 goto fail;
458 inode->i_ino = 1;
459 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
460 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
461 inode->i_op = &simple_dir_inode_operations;
462 inode->i_fop = &simple_dir_operations;
463 set_nlink(inode, 2);
464
465 s->s_root = d_make_root(inode);
466 if (!s->s_root) {
467 pr_err("get root dentry failed\n");
468 goto fail;
469 }
470
471 error = mknod_ptmx(s);
472 if (error)
473 goto fail_dput;
474
475 return 0;
476 fail_dput:
477 dput(s->s_root);
478 s->s_root = NULL;
479 fail:
480 return error;
481 }
482
483 /*
484 * devpts_mount()
485 *
486 * Mount a new (private) instance of devpts. PTYs created in this
487 * instance are independent of the PTYs in other devpts instances.
488 */
devpts_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)489 static struct dentry *devpts_mount(struct file_system_type *fs_type,
490 int flags, const char *dev_name, void *data)
491 {
492 return mount_nodev(fs_type, flags, data, devpts_fill_super);
493 }
494
devpts_kill_sb(struct super_block * sb)495 static void devpts_kill_sb(struct super_block *sb)
496 {
497 struct pts_fs_info *fsi = DEVPTS_SB(sb);
498
499 if (fsi)
500 ida_destroy(&fsi->allocated_ptys);
501 kfree(fsi);
502 kill_litter_super(sb);
503 }
504
505 static struct file_system_type devpts_fs_type = {
506 .name = "devpts",
507 .mount = devpts_mount,
508 .kill_sb = devpts_kill_sb,
509 .fs_flags = FS_USERNS_MOUNT,
510 };
511
512 /*
513 * The normal naming convention is simply /dev/pts/<number>; this conforms
514 * to the System V naming convention
515 */
516
devpts_new_index(struct pts_fs_info * fsi)517 int devpts_new_index(struct pts_fs_info *fsi)
518 {
519 int index;
520 int ida_ret;
521
522 retry:
523 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
524 return -ENOMEM;
525
526 mutex_lock(&allocated_ptys_lock);
527 if (pty_count >= (pty_limit -
528 (fsi->mount_opts.reserve ? 0 : pty_reserve))) {
529 mutex_unlock(&allocated_ptys_lock);
530 return -ENOSPC;
531 }
532
533 ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
534 if (ida_ret < 0) {
535 mutex_unlock(&allocated_ptys_lock);
536 if (ida_ret == -EAGAIN)
537 goto retry;
538 return -EIO;
539 }
540
541 if (index >= fsi->mount_opts.max) {
542 ida_remove(&fsi->allocated_ptys, index);
543 mutex_unlock(&allocated_ptys_lock);
544 return -ENOSPC;
545 }
546 pty_count++;
547 mutex_unlock(&allocated_ptys_lock);
548 return index;
549 }
550
devpts_kill_index(struct pts_fs_info * fsi,int idx)551 void devpts_kill_index(struct pts_fs_info *fsi, int idx)
552 {
553 mutex_lock(&allocated_ptys_lock);
554 ida_remove(&fsi->allocated_ptys, idx);
555 pty_count--;
556 mutex_unlock(&allocated_ptys_lock);
557 }
558
559 /**
560 * devpts_pty_new -- create a new inode in /dev/pts/
561 * @ptmx_inode: inode of the master
562 * @device: major+minor of the node to be created
563 * @index: used as a name of the node
564 * @priv: what's given back by devpts_get_priv
565 *
566 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
567 */
devpts_pty_new(struct pts_fs_info * fsi,int index,void * priv)568 struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
569 {
570 struct dentry *dentry;
571 struct super_block *sb = fsi->sb;
572 struct inode *inode;
573 struct dentry *root;
574 struct pts_mount_opts *opts;
575 char s[12];
576
577 root = sb->s_root;
578 opts = &fsi->mount_opts;
579
580 inode = new_inode(sb);
581 if (!inode)
582 return ERR_PTR(-ENOMEM);
583
584 inode->i_ino = index + 3;
585 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
586 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
587 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
588 init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
589
590 sprintf(s, "%d", index);
591
592 dentry = d_alloc_name(root, s);
593 if (dentry) {
594 dentry->d_fsdata = priv;
595 d_add(dentry, inode);
596 fsnotify_create(d_inode(root), dentry);
597 } else {
598 iput(inode);
599 dentry = ERR_PTR(-ENOMEM);
600 }
601
602 return dentry;
603 }
604
605 /**
606 * devpts_get_priv -- get private data for a slave
607 * @pts_inode: inode of the slave
608 *
609 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
610 */
devpts_get_priv(struct dentry * dentry)611 void *devpts_get_priv(struct dentry *dentry)
612 {
613 if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
614 return NULL;
615 return dentry->d_fsdata;
616 }
617
618 /**
619 * devpts_pty_kill -- remove inode form /dev/pts/
620 * @inode: inode of the slave to be removed
621 *
622 * This is an inverse operation of devpts_pty_new.
623 */
devpts_pty_kill(struct dentry * dentry)624 void devpts_pty_kill(struct dentry *dentry)
625 {
626 WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
627
628 dentry->d_fsdata = NULL;
629 drop_nlink(dentry->d_inode);
630 d_delete(dentry);
631 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
632 }
633
init_devpts_fs(void)634 static int __init init_devpts_fs(void)
635 {
636 int err = register_filesystem(&devpts_fs_type);
637 if (!err) {
638 register_sysctl_table(pty_root_table);
639 }
640 return err;
641 }
642 module_init(init_devpts_fs)
643