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 out:
 202         mutex_unlock(&root->d_inode->i_mutex);
 203         return rc;
 204 }
 205
 206 static void update_ptmx_mode(struct pts_fs_info *fsi)
 207 {
 208         struct inode *inode;
 209         if (fsi->ptmx_dentry) {
 210                 inode = fsi->ptmx_dentry->d_inode;
 211                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 212         }
 213 }
 214 #else
 215 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
 216 {
 217        return;
 218 }
 219 #endif
 220
 221 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 222 {
 223         int err;
 224         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 225         struct pts_mount_opts *opts = &fsi->mount_opts;
 226
 227         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 228
 229         /*
 230          * parse_mount_options() restores options to default values
 231          * before parsing and may have changed ptmxmode. So, update the
 232          * mode in the inode too. Bogus options don't fail the remount,
 233          * so do this even on error return.
 234          */
 235         update_ptmx_mode(fsi);
 236
 237         return err;
 238 }
 239
 240 static int devpts_show_options(struct seq_file *seq, struct vfsmount *vfs)
 241 {
 242         struct pts_fs_info *fsi = DEVPTS_SB(vfs->mnt_sb);
 243         struct pts_mount_opts *opts = &fsi->mount_opts;
 244
 245         if (opts->setuid)
 246                 seq_printf(seq, ",uid=%u", opts->uid);
 247         if (opts->setgid)
 248                 seq_printf(seq, ",gid=%u", opts->gid);
 249         seq_printf(seq, ",mode=%03o", opts->mode);
 250 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 251         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 252 #endif
 253
 254         return 0;
 255 }
 256
 257 static const struct super_operations devpts_sops = {
 258         .statfs         = simple_statfs,
 259         .remount_fs     = devpts_remount,
 260         .show_options   = devpts_show_options,
 261 };
 262
 263 static void *new_pts_fs_info(void)
 264 {
 265         struct pts_fs_info *fsi;
 266
 267         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 268         if (!fsi)
 269                 return NULL;
 270
 271         ida_init(&fsi->allocated_ptys);
 272         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 273         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 274
 275         return fsi;
 276 }
 277
 278 static int
 279 devpts_fill_super(struct super_block *s, void *data, int silent)
 280 {
 281         struct inode *inode;
 282
 283         s->s_blocksize = 1024;
 284         s->s_blocksize_bits = 10;
 285         s->s_magic = DEVPTS_SUPER_MAGIC;
 286         s->s_op = &devpts_sops;
 287         s->s_time_gran = 1;
 288
 289         s->s_fs_info = new_pts_fs_info();
 290         if (!s->s_fs_info)
 291                 goto fail;
 292
 293         inode = new_inode(s);
 294         if (!inode)
 295                 goto free_fsi;
 296         inode->i_ino = 1;
 297         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 298         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 299         inode->i_op = &simple_dir_inode_operations;
 300         inode->i_fop = &simple_dir_operations;
 301         inode->i_nlink = 2;
 302
 303         s->s_root = d_alloc_root(inode);
 304         if (s->s_root)
 305                 return 0;
 306
 307         printk(KERN_ERR "devpts: get root dentry failed\n");
 308         iput(inode);
 309
 310 free_fsi:
 311         kfree(s->s_fs_info);
 312 fail:
 313         return -ENOMEM;
 314 }
 315
 316 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 317 static int compare_init_pts_sb(struct super_block *s, void *p)
 318 {
 319         if (devpts_mnt)
 320                 return devpts_mnt->mnt_sb == s;
 321         return 0;
 322 }
 323
 324 /*
 325  * Safely parse the mount options in @data and update @opts.
 326  *
 327  * devpts ends up parsing options two times during mount, due to the
 328  * two modes of operation it supports. The first parse occurs in
 329  * devpts_get_sb() when determining the mode (single-instance or
 330  * multi-instance mode). The second parse happens in devpts_remount()
 331  * or new_pts_mount() depending on the mode.
 332  *
 333  * Parsing of options modifies the @data making subsequent parsing
 334  * incorrect. So make a local copy of @data and parse it.
 335  *
 336  * Return: 0 On success, -errno on error
 337  */
 338 static int safe_parse_mount_options(void *data, struct pts_mount_opts *opts)
 339 {
 340         int rc;
 341         void *datacp;
 342
 343         if (!data)
 344                 return 0;
 345
 346         /* Use kstrdup() ?  */
 347         datacp = kmalloc(PAGE_SIZE, GFP_KERNEL);
 348         if (!datacp)
 349                 return -ENOMEM;
 350
 351         memcpy(datacp, data, PAGE_SIZE);
 352         rc = parse_mount_options((char *)datacp, PARSE_MOUNT, opts);
 353         kfree(datacp);
 354
 355         return rc;
 356 }
 357
 358 /*
 359  * Mount a new (private) instance of devpts.  PTYs created in this
 360  * instance are independent of the PTYs in other devpts instances.
 361  */
 362 static int new_pts_mount(struct file_system_type *fs_type, int flags,
 363                 void *data, struct vfsmount *mnt)
 364 {
 365         int err;
 366         struct pts_fs_info *fsi;
 367         struct pts_mount_opts *opts;
 368
 369         err = get_sb_nodev(fs_type, flags, data, devpts_fill_super, mnt);
 370         if (err)
 371                 return err;
 372
 373         fsi = DEVPTS_SB(mnt->mnt_sb);
 374         opts = &fsi->mount_opts;
 375
 376         err = parse_mount_options(data, PARSE_MOUNT, opts);
 377         if (err)
 378                 goto fail;
 379
 380         err = mknod_ptmx(mnt->mnt_sb);
 381         if (err)
 382                 goto fail;
 383
 384         return 0;
 385
 386 fail:
 387         dput(mnt->mnt_sb->s_root);
 388         up_write(&mnt->mnt_sb->s_umount);
 389         deactivate_super(mnt->mnt_sb);
 390         return err;
 391 }
 392
 393 /*
 394  * Check if 'newinstance' mount option was specified in @data.
 395  *
 396  * Return: -errno       on error (eg: invalid mount options specified)
 397  *       : 1            if 'newinstance' mount option was specified
 398  *       : 0            if 'newinstance' mount option was NOT specified
 399  */
 400 static int is_new_instance_mount(void *data)
 401 {
 402         int rc;
 403         struct pts_mount_opts opts;
 404
 405         if (!data)
 406                 return 0;
 407
 408         rc = safe_parse_mount_options(data, &opts);
 409         if (!rc)
 410                 rc = opts.newinstance;
 411
 412         return rc;
 413 }
 414
 415 /*
 416  * get_init_pts_sb()
 417  *
 418  *     This interface is needed to support multiple namespace semantics in
 419  *     devpts while preserving backward compatibility of the current 'single-
 420  *     namespace' semantics. i.e all mounts of devpts without the 'newinstance'
 421  *     mount option should bind to the initial kernel mount, like
 422  *     get_sb_single().
 423  *
 424  *     Mounts with 'newinstance' option create a new private namespace.
 425  *
 426  *     But for single-mount semantics, devpts cannot use get_sb_single(),
 427  *     because get_sb_single()/sget() find and use the super-block from
 428  *     the most recent mount of devpts. But that recent mount may be a
 429  *     'newinstance' mount and get_sb_single() would pick the newinstance
 430  *     super-block instead of the initial super-block.
 431  *
 432  *     This interface is identical to get_sb_single() except that it
 433  *     consistently selects the 'single-namespace' superblock even in the
 434  *     presence of the private namespace (i.e 'newinstance') super-blocks.
 435  */
 436 static int get_init_pts_sb(struct file_system_type *fs_type, int flags,
 437                 void *data, struct vfsmount *mnt)
 438 {
 439         struct super_block *s;
 440         struct pts_mount_opts *opts;
 441         struct pts_fs_info *fsi;
 442         int error;
 443
 444         s = sget(fs_type, compare_init_pts_sb, set_anon_super, NULL);
 445         if (IS_ERR(s))
 446                 return PTR_ERR(s);
 447
 448         if (!s->s_root) {
 449                 s->s_flags = flags;
 450                 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 451                 if (error) {
 452                         up_write(&s->s_umount);
 453                         deactivate_super(s);
 454                         return error;
 455                 }
 456                 s->s_flags |= MS_ACTIVE;
 457         }
 458         fsi = DEVPTS_SB(s);
 459         opts = &fsi->mount_opts;
 460         parse_mount_options(data, PARSE_REMOUNT, opts);
 461
 462         simple_set_mnt(mnt, s);
 463         return 0;
 464 }
 465
 466 /*
 467  * Mount or remount the initial kernel mount of devpts. This type of
 468  * mount maintains the legacy, single-instance semantics, while the
 469  * kernel still allows multiple-instances.
 470  */
 471 static int init_pts_mount(struct file_system_type *fs_type, int flags,
 472                 void *data, struct vfsmount *mnt)
 473 {
 474         int err;
 475
 476         err = get_init_pts_sb(fs_type, flags, data, mnt);
 477         if (err)
 478                 return err;
 479
 480         err = mknod_ptmx(mnt->mnt_sb);
 481         if (err) {
 482                 dput(mnt->mnt_sb->s_root);
 483                 up_write(&mnt->mnt_sb->s_umount);
 484                 deactivate_super(mnt->mnt_sb);
 485         }
 486
 487         return err;
 488 }
 489
 490 static int devpts_get_sb(struct file_system_type *fs_type,
 491         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 492 {
 493         int new;
 494
 495         new = is_new_instance_mount(data);
 496         if (new < 0)
 497                 return new;
 498
 499         if (new)
 500                 return new_pts_mount(fs_type, flags, data, mnt);
 501
 502         return init_pts_mount(fs_type, flags, data, mnt);
 503 }
 504 #else
 505 /*
 506  * This supports only the legacy single-instance semantics (no
 507  * multiple-instance semantics)
 508  */
 509 static int devpts_get_sb(struct file_system_type *fs_type, int flags,
 510                 const char *dev_name, void *data, struct vfsmount *mnt)
 511 {
 512         return get_sb_single(fs_type, flags, data, devpts_fill_super, mnt);
 513 }
 514 #endif
 515
 516 static void devpts_kill_sb(struct super_block *sb)
 517 {
 518         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 519
 520         kfree(fsi);
 521         kill_litter_super(sb);
 522 }
 523
 524 static struct file_system_type devpts_fs_type = {
 525         .owner          = THIS_MODULE,
 526         .name           = "devpts",
 527         .get_sb         = devpts_get_sb,
 528         .kill_sb        = devpts_kill_sb,
 529 };
 530
 531 /*
 532  * The normal naming convention is simply /dev/pts/<number>; this conforms
 533  * to the System V naming convention
 534  */
 535
 536 int devpts_new_index(struct inode *ptmx_inode)
 537 {
 538         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 539         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 540         int index;
 541         int ida_ret;
 542
 543 retry:
 544         if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
 545                 return -ENOMEM;
 546
 547         mutex_lock(&allocated_ptys_lock);
 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 >= pty_limit) {
 557                 ida_remove(&fsi->allocated_ptys, index);
 558                 mutex_unlock(&allocated_ptys_lock);
 559                 return -EIO;
 560         }
 561         mutex_unlock(&allocated_ptys_lock);
 562         return index;
 563 }
 564
 565 void devpts_kill_index(struct inode *ptmx_inode, int idx)
 566 {
 567         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 568         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 569
 570         mutex_lock(&allocated_ptys_lock);
 571         ida_remove(&fsi->allocated_ptys, idx);
 572         mutex_unlock(&allocated_ptys_lock);
 573 }
 574
 575 int devpts_pty_new(struct inode *ptmx_inode, struct tty_struct *tty)
 576 {
 577         /* tty layer puts index from devpts_new_index() in here */
 578         int number = tty->index;
 579         struct tty_driver *driver = tty->driver;
 580         dev_t device = MKDEV(driver->major, driver->minor_start+number);
 581         struct dentry *dentry;
 582         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 583         struct inode *inode = new_inode(sb);
 584         struct dentry *root = sb->s_root;
 585         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 586         struct pts_mount_opts *opts = &fsi->mount_opts;
 587         char s[12];
 588
 589         /* We're supposed to be given the slave end of a pty */
 590         BUG_ON(driver->type != TTY_DRIVER_TYPE_PTY);
 591         BUG_ON(driver->subtype != PTY_TYPE_SLAVE);
 592
 593         if (!inode)
 594                 return -ENOMEM;
 595
 596         inode->i_ino = number + 3;
 597         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 598         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 599         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 600         init_special_inode(inode, S_IFCHR|opts->mode, device);
 601         inode->i_private = tty;
 602         tty->driver_data = inode;
 603
 604         sprintf(s, "%d", number);
 605
 606         mutex_lock(&root->d_inode->i_mutex);
 607
 608         dentry = d_alloc_name(root, s);
 609         if (!IS_ERR(dentry)) {
 610                 d_add(dentry, inode);
 611                 fsnotify_create(root->d_inode, dentry);
 612         }
 613
 614         mutex_unlock(&root->d_inode->i_mutex);
 615
 616         return 0;
 617 }
 618
 619 struct tty_struct *devpts_get_tty(struct inode *pts_inode, int number)
 620 {
 621         BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 622
 623         if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
 624                 return (struct tty_struct *)pts_inode->i_private;
 625         return NULL;
 626 }
 627
 628 void devpts_pty_kill(struct tty_struct *tty)
 629 {
 630         struct inode *inode = tty->driver_data;
 631         struct super_block *sb = pts_sb_from_inode(inode);
 632         struct dentry *root = sb->s_root;
 633         struct dentry *dentry;
 634
 635         BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 636
 637         mutex_lock(&root->d_inode->i_mutex);
 638
 639         dentry = d_find_alias(inode);
 640         if (IS_ERR(dentry))
 641                 goto out;
 642
 643         if (dentry) {
 644                 inode->i_nlink--;
 645                 d_delete(dentry);
 646                 dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 647         }
 648
 649         dput(dentry);           /* d_find_alias above */
 650 out:
 651         mutex_unlock(&root->d_inode->i_mutex);
 652 }
 653
 654 static int __init init_devpts_fs(void)
 655 {
 656         int err = register_filesystem(&devpts_fs_type);
 657         if (!err) {
 658                 devpts_mnt = kern_mount(&devpts_fs_type);
 659                 if (IS_ERR(devpts_mnt))
 660                         err = PTR_ERR(devpts_mnt);
 661         }
 662         return err;
 663 }
 664
 665 static void __exit exit_devpts_fs(void)
 666 {
 667         unregister_filesystem(&devpts_fs_type);
 668         mntput(devpts_mnt);
 669 }
 670
 671 module_init(init_devpts_fs)
 672 module_exit(exit_devpts_fs)
 673 MODULE_LICENSE("GPL");