fs/devpts/inode.c

   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/mount.h>
  19 #include <linux/tty.h>
  20 #include <linux/mutex.h>
  21 #include <linux/idr.h>
  22 #include <linux/devpts_fs.h>
  23 #include <linux/parser.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/seq_file.h>
  26
  27 #define DEVPTS_SUPER_MAGIC 0x1cd1
  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 extern int pty_limit;                   /* Config limit on Unix98 ptys */
  40 static DEFINE_MUTEX(allocated_ptys_lock);
  41
  42 static struct vfsmount *devpts_mnt;
  43
  44 struct pts_mount_opts {
  45         int setuid;
  46         int setgid;
  47         uid_t   uid;
  48         gid_t   gid;
  49         umode_t mode;
  50         umode_t ptmxmode;
  51         int newinstance;
  52 };
  53
  54 enum {
  55         Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,
  56         Opt_err
  57 };
  58
  59 static const match_table_t tokens = {
  60         {Opt_uid, "uid=%u"},
  61         {Opt_gid, "gid=%u"},
  62         {Opt_mode, "mode=%o"},
  63 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
  64         {Opt_ptmxmode, "ptmxmode=%o"},
  65         {Opt_newinstance, "newinstance"},
  66 #endif
  67         {Opt_err, NULL}
  68 };
  69
  70 struct pts_fs_info {
  71         struct ida allocated_ptys;
  72         struct pts_mount_opts mount_opts;
  73         struct dentry *ptmx_dentry;
  74 };
  75
  76 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
  77 {
  78         return sb->s_fs_info;
  79 }
  80
  81 static inline struct super_block *pts_sb_from_inode(struct inode *inode)
  82 {
  83 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
  84         if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
  85                 return inode->i_sb;
  86 #endif
  87         return devpts_mnt->mnt_sb;
  88 }
  89
  90 #define PARSE_MOUNT     0
  91 #define PARSE_REMOUNT   1
  92
  93 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
  94 {
  95         char *p;
  96
  97         opts->setuid  = 0;
  98         opts->setgid  = 0;
  99         opts->uid     = 0;
 100         opts->gid     = 0;
 101         opts->mode    = DEVPTS_DEFAULT_MODE;
 102         opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 103
 104         /* newinstance makes sense only on initial mount */
 105         if (op == PARSE_MOUNT)
 106                 opts->newinstance = 0;
 107
 108         while ((p = strsep(&data, ",")) != NULL) {
 109                 substring_t args[MAX_OPT_ARGS];
 110                 int token;
 111                 int option;
 112
 113                 if (!*p)
 114                         continue;
 115
 116                 token = match_token(p, tokens, args);
 117                 switch (token) {
 118                 case Opt_uid:
 119                         if (match_int(&args[0], &option))
 120                                 return -EINVAL;
 121                         opts->uid = option;
 122                         opts->setuid = 1;
 123                         break;
 124                 case Opt_gid:
 125                         if (match_int(&args[0], &option))
 126                                 return -EINVAL;
 127                         opts->gid = option;
 128                         opts->setgid = 1;
 129                         break;
 130                 case Opt_mode:
 131                         if (match_octal(&args[0], &option))
 132                                 return -EINVAL;
 133                         opts->mode = option & S_IALLUGO;
 134                         break;
 135 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 136                 case Opt_ptmxmode:
 137                         if (match_octal(&args[0], &option))
 138                                 return -EINVAL;
 139                         opts->ptmxmode = option & S_IALLUGO;
 140                         break;
 141                 case Opt_newinstance:
 142                         /* newinstance makes sense only on initial mount */
 143                         if (op == PARSE_MOUNT)
 144                                 opts->newinstance = 1;
 145                         break;
 146 #endif
 147                 default:
 148                         printk(KERN_ERR "devpts: called with bogus options\n");
 149                         return -EINVAL;
 150                 }
 151         }
 152
 153         return 0;
 154 }
 155
 156 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 157 static int mknod_ptmx(struct super_block *sb)
 158 {
 159         int mode;
 160         int rc = -ENOMEM;
 161         struct dentry *dentry;
 162         struct inode *inode;
 163         struct dentry *root = sb->s_root;
 164         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 165         struct pts_mount_opts *opts = &fsi->mount_opts;
 166
 167         mutex_lock(&root->d_inode->i_mutex);
 168
 169         /* If we have already created ptmx node, return */
 170         if (fsi->ptmx_dentry) {
 171                 rc = 0;
 172                 goto out;
 173         }
 174
 175         dentry = d_alloc_name(root, "ptmx");
 176         if (!dentry) {
 177                 printk(KERN_NOTICE "Unable to alloc dentry for ptmx node\n");
 178                 goto out;
 179         }
 180
 181         /*
 182          * Create a new 'ptmx' node in this mount of devpts.
 183          */
 184         inode = new_inode(sb);
 185         if (!inode) {
 186                 printk(KERN_ERR "Unable to alloc inode for ptmx node\n");
 187                 dput(dentry);
 188                 goto out;
 189         }
 190
 191         inode->i_ino = 2;
 192         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 193
 194         mode = S_IFCHR|opts->ptmxmode;
 195         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 196
 197         d_add(dentry, inode);
 198
 199         fsi->ptmx_dentry = dentry;
 200         rc = 0;
 201
 202         printk(KERN_DEBUG "Created ptmx node in devpts ino %lu\n",
 203                         inode->i_ino);
 204 out:
 205         mutex_unlock(&root->d_inode->i_mutex);
 206         return rc;
 207 }
 208
 209 static void update_ptmx_mode(struct pts_fs_info *fsi)
 210 {
 211         struct inode *inode;
 212         if (fsi->ptmx_dentry) {
 213                 inode = fsi->ptmx_dentry->d_inode;
 214                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 215         }
 216 }
 217 #else
 218 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
 219 {
 220        return;
 221 }
 222 #endif
 223
 224 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 225 {
 226         int err;
 227         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 228         struct pts_mount_opts *opts = &fsi->mount_opts;
 229
 230         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 231
 232         /*
 233          * parse_mount_options() restores options to default values
 234          * before parsing and may have changed ptmxmode. So, update the
 235          * mode in the inode too. Bogus options don't fail the remount,
 236          * so do this even on error return.
 237          */
 238         update_ptmx_mode(fsi);
 239
 240         return err;
 241 }
 242
 243 static int devpts_show_options(struct seq_file *seq, struct vfsmount *vfs)
 244 {
 245         struct pts_fs_info *fsi = DEVPTS_SB(vfs->mnt_sb);
 246         struct pts_mount_opts *opts = &fsi->mount_opts;
 247
 248         if (opts->setuid)
 249                 seq_printf(seq, ",uid=%u", opts->uid);
 250         if (opts->setgid)
 251                 seq_printf(seq, ",gid=%u", opts->gid);
 252         seq_printf(seq, ",mode=%03o", opts->mode);
 253 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 254         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 255 #endif
 256
 257         return 0;
 258 }
 259
 260 static const struct super_operations devpts_sops = {
 261         .statfs         = simple_statfs,
 262         .remount_fs     = devpts_remount,
 263         .show_options   = devpts_show_options,
 264 };
 265
 266 static void *new_pts_fs_info(void)
 267 {
 268         struct pts_fs_info *fsi;
 269
 270         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 271         if (!fsi)
 272                 return NULL;
 273
 274         ida_init(&fsi->allocated_ptys);
 275         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 276         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 277
 278         return fsi;
 279 }
 280
 281 static int
 282 devpts_fill_super(struct super_block *s, void *data, int silent)
 283 {
 284         struct inode *inode;
 285
 286         s->s_blocksize = 1024;
 287         s->s_blocksize_bits = 10;
 288         s->s_magic = DEVPTS_SUPER_MAGIC;
 289         s->s_op = &devpts_sops;
 290         s->s_time_gran = 1;
 291
 292         s->s_fs_info = new_pts_fs_info();
 293         if (!s->s_fs_info)
 294                 goto fail;
 295
 296         inode = new_inode(s);
 297         if (!inode)
 298                 goto free_fsi;
 299         inode->i_ino = 1;
 300         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 301         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 302         inode->i_op = &simple_dir_inode_operations;
 303         inode->i_fop = &simple_dir_operations;
 304         inode->i_nlink = 2;
 305
 306         s->s_root = d_alloc_root(inode);
 307         if (s->s_root)
 308                 return 0;
 309
 310         printk(KERN_ERR "devpts: get root dentry failed\n");
 311         iput(inode);
 312
 313 free_fsi:
 314         kfree(s->s_fs_info);
 315 fail:
 316         return -ENOMEM;
 317 }
 318
 319 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 320 static int compare_init_pts_sb(struct super_block *s, void *p)
 321 {
 322         if (devpts_mnt)
 323                 return devpts_mnt->mnt_sb == s;
 324         return 0;
 325 }
 326
 327 /*
 328  * Safely parse the mount options in @data and update @opts.
 329  *
 330  * devpts ends up parsing options two times during mount, due to the
 331  * two modes of operation it supports. The first parse occurs in
 332  * devpts_get_sb() when determining the mode (single-instance or
 333  * multi-instance mode). The second parse happens in devpts_remount()
 334  * or new_pts_mount() depending on the mode.
 335  *
 336  * Parsing of options modifies the @data making subsequent parsing
 337  * incorrect. So make a local copy of @data and parse it.
 338  *
 339  * Return: 0 On success, -errno on error
 340  */
 341 static int safe_parse_mount_options(void *data, struct pts_mount_opts *opts)
 342 {
 343         int rc;
 344         void *datacp;
 345
 346         if (!data)
 347                 return 0;
 348
 349         /* Use kstrdup() ?  */
 350         datacp = kmalloc(PAGE_SIZE, GFP_KERNEL);
 351         if (!datacp)
 352                 return -ENOMEM;
 353
 354         memcpy(datacp, data, PAGE_SIZE);
 355         rc = parse_mount_options((char *)datacp, PARSE_MOUNT, opts);
 356         kfree(datacp);
 357
 358         return rc;
 359 }
 360
 361 /*
 362  * Mount a new (private) instance of devpts.  PTYs created in this
 363  * instance are independent of the PTYs in other devpts instances.
 364  */
 365 static int new_pts_mount(struct file_system_type *fs_type, int flags,
 366                 void *data, struct vfsmount *mnt)
 367 {
 368         int err;
 369         struct pts_fs_info *fsi;
 370         struct pts_mount_opts *opts;
 371
 372         printk(KERN_NOTICE "devpts: newinstance mount\n");
 373
 374         err = get_sb_nodev(fs_type, flags, data, devpts_fill_super, mnt);
 375         if (err)
 376                 return err;
 377
 378         fsi = DEVPTS_SB(mnt->mnt_sb);
 379         opts = &fsi->mount_opts;
 380
 381         err = parse_mount_options(data, PARSE_MOUNT, opts);
 382         if (err)
 383                 goto fail;
 384
 385         err = mknod_ptmx(mnt->mnt_sb);
 386         if (err)
 387                 goto fail;
 388
 389         return 0;
 390
 391 fail:
 392         dput(mnt->mnt_sb->s_root);
 393         deactivate_super(mnt->mnt_sb);
 394         return err;
 395 }
 396
 397 /*
 398  * Check if 'newinstance' mount option was specified in @data.
 399  *
 400  * Return: -errno       on error (eg: invalid mount options specified)
 401  *       : 1            if 'newinstance' mount option was specified
 402  *       : 0            if 'newinstance' mount option was NOT specified
 403  */
 404 static int is_new_instance_mount(void *data)
 405 {
 406         int rc;
 407         struct pts_mount_opts opts;
 408
 409         if (!data)
 410                 return 0;
 411
 412         rc = safe_parse_mount_options(data, &opts);
 413         if (!rc)
 414                 rc = opts.newinstance;
 415
 416         return rc;
 417 }
 418
 419 /*
 420  * get_init_pts_sb()
 421  *
 422  *     This interface is needed to support multiple namespace semantics in
 423  *     devpts while preserving backward compatibility of the current 'single-
 424  *     namespace' semantics. i.e all mounts of devpts without the 'newinstance'
 425  *     mount option should bind to the initial kernel mount, like
 426  *     get_sb_single().
 427  *
 428  *     Mounts with 'newinstance' option create a new private namespace.
 429  *
 430  *     But for single-mount semantics, devpts cannot use get_sb_single(),
 431  *     because get_sb_single()/sget() find and use the super-block from
 432  *     the most recent mount of devpts. But that recent mount may be a
 433  *     'newinstance' mount and get_sb_single() would pick the newinstance
 434  *     super-block instead of the initial super-block.
 435  *
 436  *     This interface is identical to get_sb_single() except that it
 437  *     consistently selects the 'single-namespace' superblock even in the
 438  *     presence of the private namespace (i.e 'newinstance') super-blocks.
 439  */
 440 static int get_init_pts_sb(struct file_system_type *fs_type, int flags,
 441                 void *data, struct vfsmount *mnt)
 442 {
 443         struct super_block *s;
 444         int error;
 445
 446         s = sget(fs_type, compare_init_pts_sb, set_anon_super, NULL);
 447         if (IS_ERR(s))
 448                 return PTR_ERR(s);
 449
 450         if (!s->s_root) {
 451                 s->s_flags = flags;
 452                 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 453                 if (error) {
 454                         up_write(&s->s_umount);
 455                         deactivate_super(s);
 456                         return error;
 457                 }
 458                 s->s_flags |= MS_ACTIVE;
 459         }
 460         do_remount_sb(s, flags, data, 0);
 461         return simple_set_mnt(mnt, s);
 462 }
 463
 464 /*
 465  * Mount or remount the initial kernel mount of devpts. This type of
 466  * mount maintains the legacy, single-instance semantics, while the
 467  * kernel still allows multiple-instances.
 468  */
 469 static int init_pts_mount(struct file_system_type *fs_type, int flags,
 470                 void *data, struct vfsmount *mnt)
 471 {
 472         int err;
 473
 474         err = get_init_pts_sb(fs_type, flags, data, mnt);
 475         if (err)
 476                 return err;
 477
 478         err = mknod_ptmx(mnt->mnt_sb);
 479         if (err) {
 480                 dput(mnt->mnt_sb->s_root);
 481                 deactivate_super(mnt->mnt_sb);
 482         }
 483
 484         return err;
 485 }
 486
 487 static int devpts_get_sb(struct file_system_type *fs_type,
 488         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 489 {
 490         int new;
 491
 492         new = is_new_instance_mount(data);
 493         if (new < 0)
 494                 return new;
 495
 496         if (new)
 497                 return new_pts_mount(fs_type, flags, data, mnt);
 498
 499         return init_pts_mount(fs_type, flags, data, mnt);
 500 }
 501 #else
 502 /*
 503  * This supports only the legacy single-instance semantics (no
 504  * multiple-instance semantics)
 505  */
 506 static int devpts_get_sb(struct file_system_type *fs_type, int flags,
 507                 const char *dev_name, void *data, struct vfsmount *mnt)
 508 {
 509         return get_sb_single(fs_type, flags, data, devpts_fill_super, mnt);
 510 }
 511 #endif
 512
 513 static void devpts_kill_sb(struct super_block *sb)
 514 {
 515         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 516
 517         kfree(fsi);
 518         kill_litter_super(sb);
 519 }
 520
 521 static struct file_system_type devpts_fs_type = {
 522         .owner          = THIS_MODULE,
 523         .name           = "devpts",
 524         .get_sb         = devpts_get_sb,
 525         .kill_sb        = devpts_kill_sb,
 526 };
 527
 528 /*
 529  * The normal naming convention is simply /dev/pts/<number>; this conforms
 530  * to the System V naming convention
 531  */
 532
 533 int devpts_new_index(struct inode *ptmx_inode)
 534 {
 535         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 536         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 537         int index;
 538         int ida_ret;
 539
 540 retry:
 541         if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
 542                 return -ENOMEM;
 543
 544         mutex_lock(&allocated_ptys_lock);
 545         ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 546         if (ida_ret < 0) {
 547                 mutex_unlock(&allocated_ptys_lock);
 548                 if (ida_ret == -EAGAIN)
 549                         goto retry;
 550                 return -EIO;
 551         }
 552
 553         if (index >= pty_limit) {
 554                 ida_remove(&fsi->allocated_ptys, index);
 555                 mutex_unlock(&allocated_ptys_lock);
 556                 return -EIO;
 557         }
 558         mutex_unlock(&allocated_ptys_lock);
 559         return index;
 560 }
 561
 562 void devpts_kill_index(struct inode *ptmx_inode, int idx)
 563 {
 564         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 565         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 566
 567         mutex_lock(&allocated_ptys_lock);
 568         ida_remove(&fsi->allocated_ptys, idx);
 569         mutex_unlock(&allocated_ptys_lock);
 570 }
 571
 572 int devpts_pty_new(struct inode *ptmx_inode, struct tty_struct *tty)
 573 {
 574         /* tty layer puts index from devpts_new_index() in here */
 575         int number = tty->index;
 576         struct tty_driver *driver = tty->driver;
 577         dev_t device = MKDEV(driver->major, driver->minor_start+number);
 578         struct dentry *dentry;
 579         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 580         struct inode *inode = new_inode(sb);
 581         struct dentry *root = sb->s_root;
 582         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 583         struct pts_mount_opts *opts = &fsi->mount_opts;
 584         char s[12];
 585
 586         /* We're supposed to be given the slave end of a pty */
 587         BUG_ON(driver->type != TTY_DRIVER_TYPE_PTY);
 588         BUG_ON(driver->subtype != PTY_TYPE_SLAVE);
 589
 590         if (!inode)
 591                 return -ENOMEM;
 592
 593         inode->i_ino = number + 3;
 594         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 595         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 596         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 597         init_special_inode(inode, S_IFCHR|opts->mode, device);
 598         inode->i_private = tty;
 599         tty->driver_data = inode;
 600
 601         sprintf(s, "%d", number);
 602
 603         mutex_lock(&root->d_inode->i_mutex);
 604
 605         dentry = d_alloc_name(root, s);
 606         if (!IS_ERR(dentry)) {
 607                 d_add(dentry, inode);
 608                 fsnotify_create(root->d_inode, dentry);
 609         }
 610
 611         mutex_unlock(&root->d_inode->i_mutex);
 612
 613         return 0;
 614 }
 615
 616 struct tty_struct *devpts_get_tty(struct inode *pts_inode, int number)
 617 {
 618         BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 619
 620         if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
 621                 return (struct tty_struct *)pts_inode->i_private;
 622         return NULL;
 623 }
 624
 625 void devpts_pty_kill(struct tty_struct *tty)
 626 {
 627         struct inode *inode = tty->driver_data;
 628         struct super_block *sb = pts_sb_from_inode(inode);
 629         struct dentry *root = sb->s_root;
 630         struct dentry *dentry;
 631
 632         BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 633
 634         mutex_lock(&root->d_inode->i_mutex);
 635
 636         dentry = d_find_alias(inode);
 637         if (IS_ERR(dentry))
 638                 goto out;
 639
 640         if (dentry) {
 641                 inode->i_nlink--;
 642                 d_delete(dentry);
 643                 dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 644         }
 645
 646         dput(dentry);           /* d_find_alias above */
 647 out:
 648         mutex_unlock(&root->d_inode->i_mutex);
 649 }
 650
 651 static int __init init_devpts_fs(void)
 652 {
 653         int err = register_filesystem(&devpts_fs_type);
 654         if (!err) {
 655                 devpts_mnt = kern_mount(&devpts_fs_type);
 656                 if (IS_ERR(devpts_mnt))
 657                         err = PTR_ERR(devpts_mnt);
 658         }
 659         return err;
 660 }
 661
 662 static void __exit exit_devpts_fs(void)
 663 {
 664         unregister_filesystem(&devpts_fs_type);
 665         mntput(devpts_mnt);
 666 }
 667
 668 module_init(init_devpts_fs)
 669 module_exit(exit_devpts_fs)
 670 MODULE_LICENSE("GPL");