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/fs_context.h>
16#include <linux/fs_parser.h>
17#include <linux/sched.h>
18#include <linux/namei.h>
19#include <linux/slab.h>
20#include <linux/mount.h>
21#include <linux/tty.h>
22#include <linux/mutex.h>
23#include <linux/magic.h>
24#include <linux/idr.h>
25#include <linux/devpts_fs.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 atomic_t pty_count = ATOMIC_INIT(0);
48
49static const 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
75struct pts_mount_opts {
76 int setuid;
77 int setgid;
78 kuid_t uid;
79 kgid_t gid;
80 umode_t mode;
81 umode_t ptmxmode;
82 int reserve;
83 int max;
84};
85
86enum {
87 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
88 Opt_err
89};
90
91static const struct fs_parameter_spec devpts_param_specs[] = {
92 fsparam_gid ("gid", Opt_gid),
93 fsparam_s32 ("max", Opt_max),
94 fsparam_u32oct ("mode", Opt_mode),
95 fsparam_flag ("newinstance", Opt_newinstance),
96 fsparam_u32oct ("ptmxmode", Opt_ptmxmode),
97 fsparam_uid ("uid", Opt_uid),
98 {}
99};
100
101struct pts_fs_info {
102 struct ida allocated_ptys;
103 struct pts_mount_opts mount_opts;
104 struct super_block *sb;
105 struct dentry *ptmx_dentry;
106};
107
108static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
109{
110 return sb->s_fs_info;
111}
112
113static int devpts_ptmx_path(struct path *path)
114{
115 struct super_block *sb;
116 int err;
117
118 /* Is a devpts filesystem at "pts" in the same directory? */
119 err = path_pts(path);
120 if (err)
121 return err;
122
123 /* Is the path the root of a devpts filesystem? */
124 sb = path->mnt->mnt_sb;
125 if ((sb->s_magic != DEVPTS_SUPER_MAGIC) ||
126 (path->mnt->mnt_root != sb->s_root))
127 return -ENODEV;
128
129 return 0;
130}
131
132/*
133 * Try to find a suitable devpts filesystem. We support the following
134 * scenarios:
135 * - The ptmx device node is located in the same directory as the devpts
136 * mount where the pts device nodes are located.
137 * This is e.g. the case when calling open on the /dev/pts/ptmx device
138 * node when the devpts filesystem is mounted at /dev/pts.
139 * - The ptmx device node is located outside the devpts filesystem mount
140 * where the pts device nodes are located. For example, the ptmx device
141 * is a symlink, separate device node, or bind-mount.
142 * A supported scenario is bind-mounting /dev/pts/ptmx to /dev/ptmx and
143 * then calling open on /dev/ptmx. In this case a suitable pts
144 * subdirectory can be found in the common parent directory /dev of the
145 * devpts mount and the ptmx bind-mount, after resolving the /dev/ptmx
146 * bind-mount.
147 * If no suitable pts subdirectory can be found this function will fail.
148 * This is e.g. the case when bind-mounting /dev/pts/ptmx to /ptmx.
149 */
150struct vfsmount *devpts_mntget(struct file *filp, struct pts_fs_info *fsi)
151{
152 struct path path;
153 int err = 0;
154
155 path = filp->f_path;
156 path_get(&path);
157
158 /* Walk upward while the start point is a bind mount of
159 * a single file.
160 */
161 while (path.mnt->mnt_root == path.dentry)
162 if (follow_up(&path) == 0)
163 break;
164
165 /* devpts_ptmx_path() finds a devpts fs or returns an error. */
166 if ((path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) ||
167 (DEVPTS_SB(sb: path.mnt->mnt_sb) != fsi))
168 err = devpts_ptmx_path(path: &path);
169 dput(path.dentry);
170 if (!err) {
171 if (DEVPTS_SB(sb: path.mnt->mnt_sb) == fsi)
172 return path.mnt;
173
174 err = -ENODEV;
175 }
176
177 mntput(mnt: path.mnt);
178 return ERR_PTR(error: err);
179}
180
181struct pts_fs_info *devpts_acquire(struct file *filp)
182{
183 struct pts_fs_info *result;
184 struct path path;
185 struct super_block *sb;
186
187 path = filp->f_path;
188 path_get(&path);
189
190 /* Has the devpts filesystem already been found? */
191 if (path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) {
192 int err;
193
194 err = devpts_ptmx_path(path: &path);
195 if (err) {
196 result = ERR_PTR(error: err);
197 goto out;
198 }
199 }
200
201 /*
202 * pty code needs to hold extra references in case of last /dev/tty close
203 */
204 sb = path.mnt->mnt_sb;
205 atomic_inc(v: &sb->s_active);
206 result = DEVPTS_SB(sb);
207
208out:
209 path_put(&path);
210 return result;
211}
212
213void devpts_release(struct pts_fs_info *fsi)
214{
215 deactivate_super(sb: fsi->sb);
216}
217
218/*
219 * devpts_parse_param - Parse mount parameters
220 */
221static int devpts_parse_param(struct fs_context *fc, struct fs_parameter *param)
222{
223 struct pts_fs_info *fsi = fc->s_fs_info;
224 struct pts_mount_opts *opts = &fsi->mount_opts;
225 struct fs_parse_result result;
226 int opt;
227
228 opt = fs_parse(fc, desc: devpts_param_specs, param, result: &result);
229 if (opt < 0)
230 return opt;
231
232 switch (opt) {
233 case Opt_uid:
234 opts->uid = result.uid;
235 opts->setuid = 1;
236 break;
237 case Opt_gid:
238 opts->gid = result.gid;
239 opts->setgid = 1;
240 break;
241 case Opt_mode:
242 opts->mode = result.uint_32 & S_IALLUGO;
243 break;
244 case Opt_ptmxmode:
245 opts->ptmxmode = result.uint_32 & S_IALLUGO;
246 break;
247 case Opt_newinstance:
248 break;
249 case Opt_max:
250 if (result.uint_32 > NR_UNIX98_PTY_MAX)
251 return invalf(fc, "max out of range");
252 opts->max = result.uint_32;
253 break;
254 }
255
256 return 0;
257}
258
259static int mknod_ptmx(struct super_block *sb, struct fs_context *fc)
260{
261 int mode;
262 int rc = -ENOMEM;
263 struct dentry *dentry;
264 struct inode *inode;
265 struct dentry *root = sb->s_root;
266 struct pts_fs_info *fsi = DEVPTS_SB(sb);
267 struct pts_mount_opts *opts = &fsi->mount_opts;
268 kuid_t ptmx_uid = current_fsuid();
269 kgid_t ptmx_gid = current_fsgid();
270
271 inode_lock(inode: d_inode(dentry: root));
272
273 /* If we have already created ptmx node, return */
274 if (fsi->ptmx_dentry) {
275 rc = 0;
276 goto out;
277 }
278
279 dentry = d_alloc_name(root, "ptmx");
280 if (!dentry) {
281 pr_err("Unable to alloc dentry for ptmx node\n");
282 goto out;
283 }
284
285 /*
286 * Create a new 'ptmx' node in this mount of devpts.
287 */
288 inode = new_inode(sb);
289 if (!inode) {
290 pr_err("Unable to alloc inode for ptmx node\n");
291 dput(dentry);
292 goto out;
293 }
294
295 inode->i_ino = 2;
296 simple_inode_init_ts(inode);
297
298 mode = S_IFCHR|opts->ptmxmode;
299 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
300 inode->i_uid = ptmx_uid;
301 inode->i_gid = ptmx_gid;
302
303 d_add(dentry, inode);
304
305 fsi->ptmx_dentry = dentry;
306 rc = 0;
307out:
308 inode_unlock(inode: d_inode(dentry: root));
309 return rc;
310}
311
312static void update_ptmx_mode(struct pts_fs_info *fsi)
313{
314 struct inode *inode;
315 if (fsi->ptmx_dentry) {
316 inode = d_inode(dentry: fsi->ptmx_dentry);
317 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
318 }
319}
320
321static int devpts_reconfigure(struct fs_context *fc)
322{
323 struct pts_fs_info *fsi = DEVPTS_SB(sb: fc->root->d_sb);
324 struct pts_fs_info *new = fc->s_fs_info;
325
326 /* Apply the revised options. We don't want to change ->reserve.
327 * Ideally, we'd update each option conditionally on it having been
328 * explicitly changed, but the default is to reset everything so that
329 * would break UAPI...
330 */
331 fsi->mount_opts.setuid = new->mount_opts.setuid;
332 fsi->mount_opts.setgid = new->mount_opts.setgid;
333 fsi->mount_opts.uid = new->mount_opts.uid;
334 fsi->mount_opts.gid = new->mount_opts.gid;
335 fsi->mount_opts.mode = new->mount_opts.mode;
336 fsi->mount_opts.ptmxmode = new->mount_opts.ptmxmode;
337 fsi->mount_opts.max = new->mount_opts.max;
338
339 /*
340 * parse_mount_options() restores options to default values
341 * before parsing and may have changed ptmxmode. So, update the
342 * mode in the inode too. Bogus options don't fail the remount,
343 * so do this even on error return.
344 */
345 update_ptmx_mode(fsi);
346
347 return 0;
348}
349
350static int devpts_show_options(struct seq_file *seq, struct dentry *root)
351{
352 struct pts_fs_info *fsi = DEVPTS_SB(sb: root->d_sb);
353 struct pts_mount_opts *opts = &fsi->mount_opts;
354
355 if (opts->setuid)
356 seq_printf(m: seq, fmt: ",uid=%u",
357 from_kuid_munged(to: &init_user_ns, kuid: opts->uid));
358 if (opts->setgid)
359 seq_printf(m: seq, fmt: ",gid=%u",
360 from_kgid_munged(to: &init_user_ns, kgid: opts->gid));
361 seq_printf(m: seq, fmt: ",mode=%03o", opts->mode);
362 seq_printf(m: seq, fmt: ",ptmxmode=%03o", opts->ptmxmode);
363 if (opts->max < NR_UNIX98_PTY_MAX)
364 seq_printf(m: seq, fmt: ",max=%d", opts->max);
365
366 return 0;
367}
368
369static const struct super_operations devpts_sops = {
370 .statfs = simple_statfs,
371 .show_options = devpts_show_options,
372};
373
374static int devpts_fill_super(struct super_block *s, struct fs_context *fc)
375{
376 struct pts_fs_info *fsi = DEVPTS_SB(sb: s);
377 struct inode *inode;
378
379 s->s_iflags &= ~SB_I_NODEV;
380 s->s_blocksize = 1024;
381 s->s_blocksize_bits = 10;
382 s->s_magic = DEVPTS_SUPER_MAGIC;
383 s->s_op = &devpts_sops;
384 s->s_d_flags = DCACHE_DONTCACHE;
385 s->s_time_gran = 1;
386 fsi->sb = s;
387
388 inode = new_inode(sb: s);
389 if (!inode)
390 return -ENOMEM;
391 inode->i_ino = 1;
392 simple_inode_init_ts(inode);
393 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
394 inode->i_op = &simple_dir_inode_operations;
395 inode->i_fop = &simple_dir_operations;
396 set_nlink(inode, nlink: 2);
397
398 s->s_root = d_make_root(inode);
399 if (!s->s_root) {
400 pr_err("get root dentry failed\n");
401 return -ENOMEM;
402 }
403
404 return mknod_ptmx(sb: s, fc);
405}
406
407/*
408 * devpts_get_tree()
409 *
410 * Mount a new (private) instance of devpts. PTYs created in this
411 * instance are independent of the PTYs in other devpts instances.
412 */
413static int devpts_get_tree(struct fs_context *fc)
414{
415 return get_tree_nodev(fc, fill_super: devpts_fill_super);
416}
417
418static void devpts_free_fc(struct fs_context *fc)
419{
420 kfree(objp: fc->s_fs_info);
421}
422
423static const struct fs_context_operations devpts_context_ops = {
424 .free = devpts_free_fc,
425 .parse_param = devpts_parse_param,
426 .get_tree = devpts_get_tree,
427 .reconfigure = devpts_reconfigure,
428};
429
430/*
431 * Set up the filesystem mount context.
432 */
433static int devpts_init_fs_context(struct fs_context *fc)
434{
435 struct pts_fs_info *fsi;
436
437 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
438 if (!fsi)
439 return -ENOMEM;
440
441 ida_init(ida: &fsi->allocated_ptys);
442 fsi->mount_opts.uid = GLOBAL_ROOT_UID;
443 fsi->mount_opts.gid = GLOBAL_ROOT_GID;
444 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
445 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
446 fsi->mount_opts.max = NR_UNIX98_PTY_MAX;
447
448 if (fc->purpose == FS_CONTEXT_FOR_MOUNT &&
449 current->nsproxy->mnt_ns == init_task.nsproxy->mnt_ns)
450 fsi->mount_opts.reserve = true;
451
452 fc->s_fs_info = fsi;
453 fc->ops = &devpts_context_ops;
454 return 0;
455}
456
457static void devpts_kill_sb(struct super_block *sb)
458{
459 struct pts_fs_info *fsi = DEVPTS_SB(sb);
460
461 if (fsi)
462 ida_destroy(ida: &fsi->allocated_ptys);
463 kfree(objp: fsi);
464 kill_litter_super(sb);
465}
466
467static struct file_system_type devpts_fs_type = {
468 .name = "devpts",
469 .init_fs_context = devpts_init_fs_context,
470 .parameters = devpts_param_specs,
471 .kill_sb = devpts_kill_sb,
472 .fs_flags = FS_USERNS_MOUNT,
473};
474
475/*
476 * The normal naming convention is simply /dev/pts/<number>; this conforms
477 * to the System V naming convention
478 */
479
480int devpts_new_index(struct pts_fs_info *fsi)
481{
482 int index = -ENOSPC;
483
484 if (atomic_inc_return(v: &pty_count) >= (pty_limit -
485 (fsi->mount_opts.reserve ? 0 : pty_reserve)))
486 goto out;
487
488 index = ida_alloc_max(ida: &fsi->allocated_ptys, max: fsi->mount_opts.max - 1,
489 GFP_KERNEL);
490
491out:
492 if (index < 0)
493 atomic_dec(v: &pty_count);
494 return index;
495}
496
497void devpts_kill_index(struct pts_fs_info *fsi, int idx)
498{
499 ida_free(&fsi->allocated_ptys, id: idx);
500 atomic_dec(v: &pty_count);
501}
502
503/**
504 * devpts_pty_new -- create a new inode in /dev/pts/
505 * @fsi: Filesystem info for this instance.
506 * @index: used as a name of the node
507 * @priv: what's given back by devpts_get_priv
508 *
509 * The dentry for the created inode is returned.
510 * Remove it from /dev/pts/ with devpts_pty_kill().
511 */
512struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
513{
514 struct dentry *dentry;
515 struct super_block *sb = fsi->sb;
516 struct inode *inode;
517 struct dentry *root;
518 struct pts_mount_opts *opts;
519 char s[12];
520
521 root = sb->s_root;
522 opts = &fsi->mount_opts;
523
524 inode = new_inode(sb);
525 if (!inode)
526 return ERR_PTR(error: -ENOMEM);
527
528 inode->i_ino = index + 3;
529 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
530 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
531 simple_inode_init_ts(inode);
532 init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
533
534 sprintf(buf: s, fmt: "%d", index);
535
536 dentry = d_alloc_name(root, s);
537 if (dentry) {
538 dentry->d_fsdata = priv;
539 d_add(dentry, inode);
540 fsnotify_create(dir: d_inode(dentry: root), dentry);
541 } else {
542 iput(inode);
543 dentry = ERR_PTR(error: -ENOMEM);
544 }
545
546 return dentry;
547}
548
549/**
550 * devpts_get_priv -- get private data for a slave
551 * @dentry: dentry of the slave
552 *
553 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
554 */
555void *devpts_get_priv(struct dentry *dentry)
556{
557 if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
558 return NULL;
559 return dentry->d_fsdata;
560}
561
562/**
563 * devpts_pty_kill -- remove inode form /dev/pts/
564 * @dentry: dentry of the slave to be removed
565 *
566 * This is an inverse operation of devpts_pty_new.
567 */
568void devpts_pty_kill(struct dentry *dentry)
569{
570 WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
571
572 dentry->d_fsdata = NULL;
573 drop_nlink(inode: dentry->d_inode);
574 d_drop(dentry);
575 fsnotify_unlink(dir: d_inode(dentry: dentry->d_parent), dentry);
576 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
577}
578
579static int __init init_devpts_fs(void)
580{
581 int err = register_filesystem(&devpts_fs_type);
582 if (!err) {
583 register_sysctl("kernel/pty", pty_table);
584 }
585 return err;
586}
587module_init(init_devpts_fs)
588