gro: Allow tunnel stacking in the case of FOU/GUE
[linux/fpc-iii.git] / fs / devpts / inode.c
blob91360444adf5e1c11687c958a6f4b09e89baea31
1 /* -*- linux-c -*- --------------------------------------------------------- *
3 * linux/fs/devpts/inode.c
5 * Copyright 1998-2004 H. Peter Anvin -- All Rights Reserved
7 * This file is part of the Linux kernel and is made available under
8 * the terms of the GNU General Public License, version 2, or at your
9 * option, any later version, incorporated herein by reference.
11 * ------------------------------------------------------------------------- */
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/module.h>
16 #include <linux/init.h>
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>
31 #define DEVPTS_DEFAULT_MODE 0600
33 * ptmx is a new node in /dev/pts and will be unused in legacy (single-
34 * instance) mode. To prevent surprises in user space, set permissions of
35 * ptmx to 0. Use 'chmod' or remount with '-o ptmxmode' to set meaningful
36 * permissions.
38 #define DEVPTS_DEFAULT_PTMX_MODE 0000
39 #define PTMX_MINOR 2
42 * sysctl support for setting limits on the number of Unix98 ptys allocated.
43 * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
45 static int pty_limit = NR_UNIX98_PTY_DEFAULT;
46 static int pty_reserve = NR_UNIX98_PTY_RESERVE;
47 static int pty_limit_min;
48 static int pty_limit_max = INT_MAX;
49 static int pty_count;
51 static struct ctl_table pty_table[] = {
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,
78 static struct ctl_table pty_kern_table[] = {
80 .procname = "pty",
81 .mode = 0555,
82 .child = pty_table,
87 static struct ctl_table pty_root_table[] = {
89 .procname = "kernel",
90 .mode = 0555,
91 .child = pty_kern_table,
96 static DEFINE_MUTEX(allocated_ptys_lock);
98 static struct vfsmount *devpts_mnt;
100 struct 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;
111 enum {
112 Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance, Opt_max,
113 Opt_err
116 static 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}
128 struct pts_fs_info {
129 struct ida allocated_ptys;
130 struct pts_mount_opts mount_opts;
131 struct dentry *ptmx_dentry;
134 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
136 return sb->s_fs_info;
139 static inline struct super_block *pts_sb_from_inode(struct inode *inode)
141 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
142 if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
143 return inode->i_sb;
144 #endif
145 return devpts_mnt->mnt_sb;
148 #define PARSE_MOUNT 0
149 #define PARSE_REMOUNT 1
152 * parse_mount_options():
153 * Set @opts to mount options specified in @data. If an option is not
154 * specified in @data, set it to its default value. The exception is
155 * 'newinstance' option which can only be set/cleared on a mount (i.e.
156 * cannot be changed during remount).
158 * Note: @data may be NULL (in which case all options are set to default).
160 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
162 char *p;
163 kuid_t uid;
164 kgid_t gid;
166 opts->setuid = 0;
167 opts->setgid = 0;
168 opts->uid = GLOBAL_ROOT_UID;
169 opts->gid = GLOBAL_ROOT_GID;
170 opts->mode = DEVPTS_DEFAULT_MODE;
171 opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
172 opts->max = NR_UNIX98_PTY_MAX;
174 /* newinstance makes sense only on initial mount */
175 if (op == PARSE_MOUNT)
176 opts->newinstance = 0;
178 while ((p = strsep(&data, ",")) != NULL) {
179 substring_t args[MAX_OPT_ARGS];
180 int token;
181 int option;
183 if (!*p)
184 continue;
186 token = match_token(p, tokens, args);
187 switch (token) {
188 case Opt_uid:
189 if (match_int(&args[0], &option))
190 return -EINVAL;
191 uid = make_kuid(current_user_ns(), option);
192 if (!uid_valid(uid))
193 return -EINVAL;
194 opts->uid = uid;
195 opts->setuid = 1;
196 break;
197 case Opt_gid:
198 if (match_int(&args[0], &option))
199 return -EINVAL;
200 gid = make_kgid(current_user_ns(), option);
201 if (!gid_valid(gid))
202 return -EINVAL;
203 opts->gid = gid;
204 opts->setgid = 1;
205 break;
206 case Opt_mode:
207 if (match_octal(&args[0], &option))
208 return -EINVAL;
209 opts->mode = option & S_IALLUGO;
210 break;
211 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
212 case Opt_ptmxmode:
213 if (match_octal(&args[0], &option))
214 return -EINVAL;
215 opts->ptmxmode = option & S_IALLUGO;
216 break;
217 case Opt_newinstance:
218 /* newinstance makes sense only on initial mount */
219 if (op == PARSE_MOUNT)
220 opts->newinstance = 1;
221 break;
222 case Opt_max:
223 if (match_int(&args[0], &option) ||
224 option < 0 || option > NR_UNIX98_PTY_MAX)
225 return -EINVAL;
226 opts->max = option;
227 break;
228 #endif
229 default:
230 pr_err("called with bogus options\n");
231 return -EINVAL;
235 return 0;
238 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
239 static int mknod_ptmx(struct super_block *sb)
241 int mode;
242 int rc = -ENOMEM;
243 struct dentry *dentry;
244 struct inode *inode;
245 struct dentry *root = sb->s_root;
246 struct pts_fs_info *fsi = DEVPTS_SB(sb);
247 struct pts_mount_opts *opts = &fsi->mount_opts;
248 kuid_t root_uid;
249 kgid_t root_gid;
251 root_uid = make_kuid(current_user_ns(), 0);
252 root_gid = make_kgid(current_user_ns(), 0);
253 if (!uid_valid(root_uid) || !gid_valid(root_gid))
254 return -EINVAL;
256 mutex_lock(&d_inode(root)->i_mutex);
258 /* If we have already created ptmx node, return */
259 if (fsi->ptmx_dentry) {
260 rc = 0;
261 goto out;
264 dentry = d_alloc_name(root, "ptmx");
265 if (!dentry) {
266 pr_err("Unable to alloc dentry for ptmx node\n");
267 goto out;
271 * Create a new 'ptmx' node in this mount of devpts.
273 inode = new_inode(sb);
274 if (!inode) {
275 pr_err("Unable to alloc inode for ptmx node\n");
276 dput(dentry);
277 goto out;
280 inode->i_ino = 2;
281 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
283 mode = S_IFCHR|opts->ptmxmode;
284 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
285 inode->i_uid = root_uid;
286 inode->i_gid = root_gid;
288 d_add(dentry, inode);
290 fsi->ptmx_dentry = dentry;
291 rc = 0;
292 out:
293 mutex_unlock(&d_inode(root)->i_mutex);
294 return rc;
297 static void update_ptmx_mode(struct pts_fs_info *fsi)
299 struct inode *inode;
300 if (fsi->ptmx_dentry) {
301 inode = d_inode(fsi->ptmx_dentry);
302 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
305 #else
306 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
308 return;
310 #endif
312 static int devpts_remount(struct super_block *sb, int *flags, char *data)
314 int err;
315 struct pts_fs_info *fsi = DEVPTS_SB(sb);
316 struct pts_mount_opts *opts = &fsi->mount_opts;
318 sync_filesystem(sb);
319 err = parse_mount_options(data, PARSE_REMOUNT, opts);
322 * parse_mount_options() restores options to default values
323 * before parsing and may have changed ptmxmode. So, update the
324 * mode in the inode too. Bogus options don't fail the remount,
325 * so do this even on error return.
327 update_ptmx_mode(fsi);
329 return err;
332 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
334 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
335 struct pts_mount_opts *opts = &fsi->mount_opts;
337 if (opts->setuid)
338 seq_printf(seq, ",uid=%u",
339 from_kuid_munged(&init_user_ns, opts->uid));
340 if (opts->setgid)
341 seq_printf(seq, ",gid=%u",
342 from_kgid_munged(&init_user_ns, opts->gid));
343 seq_printf(seq, ",mode=%03o", opts->mode);
344 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
345 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
346 if (opts->max < NR_UNIX98_PTY_MAX)
347 seq_printf(seq, ",max=%d", opts->max);
348 #endif
350 return 0;
353 static const struct super_operations devpts_sops = {
354 .statfs = simple_statfs,
355 .remount_fs = devpts_remount,
356 .show_options = devpts_show_options,
359 static void *new_pts_fs_info(void)
361 struct pts_fs_info *fsi;
363 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
364 if (!fsi)
365 return NULL;
367 ida_init(&fsi->allocated_ptys);
368 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
369 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
371 return fsi;
374 static int
375 devpts_fill_super(struct super_block *s, void *data, int silent)
377 struct inode *inode;
379 s->s_blocksize = 1024;
380 s->s_blocksize_bits = 10;
381 s->s_magic = DEVPTS_SUPER_MAGIC;
382 s->s_op = &devpts_sops;
383 s->s_time_gran = 1;
385 s->s_fs_info = new_pts_fs_info();
386 if (!s->s_fs_info)
387 goto fail;
389 inode = new_inode(s);
390 if (!inode)
391 goto fail;
392 inode->i_ino = 1;
393 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
394 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
395 inode->i_op = &simple_dir_inode_operations;
396 inode->i_fop = &simple_dir_operations;
397 set_nlink(inode, 2);
399 s->s_root = d_make_root(inode);
400 if (s->s_root)
401 return 0;
403 pr_err("get root dentry failed\n");
405 fail:
406 return -ENOMEM;
409 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
410 static int compare_init_pts_sb(struct super_block *s, void *p)
412 if (devpts_mnt)
413 return devpts_mnt->mnt_sb == s;
414 return 0;
418 * devpts_mount()
420 * If the '-o newinstance' mount option was specified, mount a new
421 * (private) instance of devpts. PTYs created in this instance are
422 * independent of the PTYs in other devpts instances.
424 * If the '-o newinstance' option was not specified, mount/remount the
425 * initial kernel mount of devpts. This type of mount gives the
426 * legacy, single-instance semantics.
428 * The 'newinstance' option is needed to support multiple namespace
429 * semantics in devpts while preserving backward compatibility of the
430 * current 'single-namespace' semantics. i.e all mounts of devpts
431 * without the 'newinstance' mount option should bind to the initial
432 * kernel mount, like mount_single().
434 * Mounts with 'newinstance' option create a new, private namespace.
436 * NOTE:
438 * For single-mount semantics, devpts cannot use mount_single(),
439 * because mount_single()/sget() find and use the super-block from
440 * the most recent mount of devpts. But that recent mount may be a
441 * 'newinstance' mount and mount_single() would pick the newinstance
442 * super-block instead of the initial super-block.
444 static struct dentry *devpts_mount(struct file_system_type *fs_type,
445 int flags, const char *dev_name, void *data)
447 int error;
448 struct pts_mount_opts opts;
449 struct super_block *s;
451 error = parse_mount_options(data, PARSE_MOUNT, &opts);
452 if (error)
453 return ERR_PTR(error);
455 /* Require newinstance for all user namespace mounts to ensure
456 * the mount options are not changed.
458 if ((current_user_ns() != &init_user_ns) && !opts.newinstance)
459 return ERR_PTR(-EINVAL);
461 if (opts.newinstance)
462 s = sget(fs_type, NULL, set_anon_super, flags, NULL);
463 else
464 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags,
465 NULL);
467 if (IS_ERR(s))
468 return ERR_CAST(s);
470 if (!s->s_root) {
471 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
472 if (error)
473 goto out_undo_sget;
474 s->s_flags |= MS_ACTIVE;
477 memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
479 error = mknod_ptmx(s);
480 if (error)
481 goto out_undo_sget;
483 return dget(s->s_root);
485 out_undo_sget:
486 deactivate_locked_super(s);
487 return ERR_PTR(error);
490 #else
492 * This supports only the legacy single-instance semantics (no
493 * multiple-instance semantics)
495 static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
496 const char *dev_name, void *data)
498 return mount_single(fs_type, flags, data, devpts_fill_super);
500 #endif
502 static void devpts_kill_sb(struct super_block *sb)
504 struct pts_fs_info *fsi = DEVPTS_SB(sb);
506 ida_destroy(&fsi->allocated_ptys);
507 kfree(fsi);
508 kill_litter_super(sb);
511 static struct file_system_type devpts_fs_type = {
512 .name = "devpts",
513 .mount = devpts_mount,
514 .kill_sb = devpts_kill_sb,
515 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
516 .fs_flags = FS_USERNS_MOUNT | FS_USERNS_DEV_MOUNT,
517 #endif
521 * The normal naming convention is simply /dev/pts/<number>; this conforms
522 * to the System V naming convention
525 int devpts_new_index(struct inode *ptmx_inode)
527 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
528 struct pts_fs_info *fsi = DEVPTS_SB(sb);
529 int index;
530 int ida_ret;
532 retry:
533 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
534 return -ENOMEM;
536 mutex_lock(&allocated_ptys_lock);
537 if (pty_count >= pty_limit -
538 (fsi->mount_opts.newinstance ? pty_reserve : 0)) {
539 mutex_unlock(&allocated_ptys_lock);
540 return -ENOSPC;
543 ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
544 if (ida_ret < 0) {
545 mutex_unlock(&allocated_ptys_lock);
546 if (ida_ret == -EAGAIN)
547 goto retry;
548 return -EIO;
551 if (index >= fsi->mount_opts.max) {
552 ida_remove(&fsi->allocated_ptys, index);
553 mutex_unlock(&allocated_ptys_lock);
554 return -ENOSPC;
556 pty_count++;
557 mutex_unlock(&allocated_ptys_lock);
558 return index;
561 void devpts_kill_index(struct inode *ptmx_inode, int idx)
563 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
564 struct pts_fs_info *fsi = DEVPTS_SB(sb);
566 mutex_lock(&allocated_ptys_lock);
567 ida_remove(&fsi->allocated_ptys, idx);
568 pty_count--;
569 mutex_unlock(&allocated_ptys_lock);
573 * pty code needs to hold extra references in case of last /dev/tty close
576 void devpts_add_ref(struct inode *ptmx_inode)
578 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
580 atomic_inc(&sb->s_active);
581 ihold(ptmx_inode);
584 void devpts_del_ref(struct inode *ptmx_inode)
586 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
588 iput(ptmx_inode);
589 deactivate_super(sb);
593 * devpts_pty_new -- create a new inode in /dev/pts/
594 * @ptmx_inode: inode of the master
595 * @device: major+minor of the node to be created
596 * @index: used as a name of the node
597 * @priv: what's given back by devpts_get_priv
599 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
601 struct inode *devpts_pty_new(struct inode *ptmx_inode, dev_t device, int index,
602 void *priv)
604 struct dentry *dentry;
605 struct super_block *sb = pts_sb_from_inode(ptmx_inode);
606 struct inode *inode;
607 struct dentry *root = sb->s_root;
608 struct pts_fs_info *fsi = DEVPTS_SB(sb);
609 struct pts_mount_opts *opts = &fsi->mount_opts;
610 char s[12];
612 inode = new_inode(sb);
613 if (!inode)
614 return ERR_PTR(-ENOMEM);
616 inode->i_ino = index + 3;
617 inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
618 inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
619 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
620 init_special_inode(inode, S_IFCHR|opts->mode, device);
621 inode->i_private = priv;
623 sprintf(s, "%d", index);
625 mutex_lock(&d_inode(root)->i_mutex);
627 dentry = d_alloc_name(root, s);
628 if (dentry) {
629 d_add(dentry, inode);
630 fsnotify_create(d_inode(root), dentry);
631 } else {
632 iput(inode);
633 inode = ERR_PTR(-ENOMEM);
636 mutex_unlock(&d_inode(root)->i_mutex);
638 return inode;
642 * devpts_get_priv -- get private data for a slave
643 * @pts_inode: inode of the slave
645 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
647 void *devpts_get_priv(struct inode *pts_inode)
649 struct dentry *dentry;
650 void *priv = NULL;
652 BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
654 /* Ensure dentry has not been deleted by devpts_pty_kill() */
655 dentry = d_find_alias(pts_inode);
656 if (!dentry)
657 return NULL;
659 if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
660 priv = pts_inode->i_private;
662 dput(dentry);
664 return priv;
668 * devpts_pty_kill -- remove inode form /dev/pts/
669 * @inode: inode of the slave to be removed
671 * This is an inverse operation of devpts_pty_new.
673 void devpts_pty_kill(struct inode *inode)
675 struct super_block *sb = pts_sb_from_inode(inode);
676 struct dentry *root = sb->s_root;
677 struct dentry *dentry;
679 BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
681 mutex_lock(&d_inode(root)->i_mutex);
683 dentry = d_find_alias(inode);
685 drop_nlink(inode);
686 d_delete(dentry);
687 dput(dentry); /* d_alloc_name() in devpts_pty_new() */
688 dput(dentry); /* d_find_alias above */
690 mutex_unlock(&d_inode(root)->i_mutex);
693 static int __init init_devpts_fs(void)
695 int err = register_filesystem(&devpts_fs_type);
696 struct ctl_table_header *table;
698 if (!err) {
699 table = register_sysctl_table(pty_root_table);
700 devpts_mnt = kern_mount(&devpts_fs_type);
701 if (IS_ERR(devpts_mnt)) {
702 err = PTR_ERR(devpts_mnt);
703 unregister_filesystem(&devpts_fs_type);
704 unregister_sysctl_table(table);
707 return err;
709 module_init(init_devpts_fs)