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/slab.h>
  19 #include <linux/mount.h>
  20 #include <linux/tty.h>
  21 #include <linux/mutex.h>
  22 #include <linux/magic.h>
  23 #include <linux/idr.h>
  24 #include <linux/devpts_fs.h>
  25 #include <linux/parser.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  */
  43 static int pty_limit = NR_UNIX98_PTY_DEFAULT;
  44 static int pty_reserve = NR_UNIX98_PTY_RESERVE;
  45 static int pty_limit_min;
  46 static int pty_limit_max = INT_MAX;
  47 static int pty_count;
  48
  49 static 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 };
  75
  76 static struct ctl_table pty_kern_table[] = {
  77         {
  78                 .procname       = "pty",
  79                 .mode           = 0555,
  80                 .child          = pty_table,
  81         },
  82         {}
  83 };
  84
  85 static struct ctl_table pty_root_table[] = {
  86         {
  87                 .procname       = "kernel",
  88                 .mode           = 0555,
  89                 .child          = pty_kern_table,
  90         },
  91         {}
  92 };
  93
  94 static DEFINE_MUTEX(allocated_ptys_lock);
  95
  96 static struct vfsmount *devpts_mnt;
  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 newinstance;
 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 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 119         {Opt_ptmxmode, "ptmxmode=%o"},
 120         {Opt_newinstance, "newinstance"},
 121         {Opt_max, "max=%d"},
 122 #endif
 123         {Opt_err, NULL}
 124 };
 125
 126 struct pts_fs_info {
 127         struct ida allocated_ptys;
 128         struct pts_mount_opts mount_opts;
 129         struct dentry *ptmx_dentry;
 130 };
 131
 132 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
 133 {
 134         return sb->s_fs_info;
 135 }
 136
 137 static inline struct super_block *pts_sb_from_inode(struct inode *inode)
 138 {
 139 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 140         if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
 141                 return inode->i_sb;
 142 #endif
 143         return devpts_mnt->mnt_sb;
 144 }
 145
 146 #define PARSE_MOUNT     0
 147 #define PARSE_REMOUNT   1
 148
 149 /*
 150  * parse_mount_options():
 151  *      Set @opts to mount options specified in @data. If an option is not
 152  *      specified in @data, set it to its default value. The exception is
 153  *      'newinstance' option which can only be set/cleared on a mount (i.e.
 154  *      cannot be changed during remount).
 155  *
 156  * Note: @data may be NULL (in which case all options are set to default).
 157  */
 158 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 159 {
 160         char *p;
 161         kuid_t uid;
 162         kgid_t gid;
 163
 164         opts->setuid  = 0;
 165         opts->setgid  = 0;
 166         opts->uid     = GLOBAL_ROOT_UID;
 167         opts->gid     = GLOBAL_ROOT_GID;
 168         opts->mode    = DEVPTS_DEFAULT_MODE;
 169         opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 170         opts->max     = NR_UNIX98_PTY_MAX;
 171
 172         /* newinstance makes sense only on initial mount */
 173         if (op == PARSE_MOUNT)
 174                 opts->newinstance = 0;
 175
 176         while ((p = strsep(&data, ",")) != NULL) {
 177                 substring_t args[MAX_OPT_ARGS];
 178                 int token;
 179                 int option;
 180
 181                 if (!*p)
 182                         continue;
 183
 184                 token = match_token(p, tokens, args);
 185                 switch (token) {
 186                 case Opt_uid:
 187                         if (match_int(&args[0], &option))
 188                                 return -EINVAL;
 189                         uid = make_kuid(current_user_ns(), option);
 190                         if (!uid_valid(uid))
 191                                 return -EINVAL;
 192                         opts->uid = uid;
 193                         opts->setuid = 1;
 194                         break;
 195                 case Opt_gid:
 196                         if (match_int(&args[0], &option))
 197                                 return -EINVAL;
 198                         gid = make_kgid(current_user_ns(), option);
 199                         if (!gid_valid(gid))
 200                                 return -EINVAL;
 201                         opts->gid = gid;
 202                         opts->setgid = 1;
 203                         break;
 204                 case Opt_mode:
 205                         if (match_octal(&args[0], &option))
 206                                 return -EINVAL;
 207                         opts->mode = option & S_IALLUGO;
 208                         break;
 209 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 210                 case Opt_ptmxmode:
 211                         if (match_octal(&args[0], &option))
 212                                 return -EINVAL;
 213                         opts->ptmxmode = option & S_IALLUGO;
 214                         break;
 215                 case Opt_newinstance:
 216                         /* newinstance makes sense only on initial mount */
 217                         if (op == PARSE_MOUNT)
 218                                 opts->newinstance = 1;
 219                         break;
 220                 case Opt_max:
 221                         if (match_int(&args[0], &option) ||
 222                             option < 0 || option > NR_UNIX98_PTY_MAX)
 223                                 return -EINVAL;
 224                         opts->max = option;
 225                         break;
 226 #endif
 227                 default:
 228                         printk(KERN_ERR "devpts: called with bogus options\n");
 229                         return -EINVAL;
 230                 }
 231         }
 232
 233         return 0;
 234 }
 235
 236 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 237 static int mknod_ptmx(struct super_block *sb)
 238 {
 239         int mode;
 240         int rc = -ENOMEM;
 241         struct dentry *dentry;
 242         struct inode *inode;
 243         struct dentry *root = sb->s_root;
 244         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 245         struct pts_mount_opts *opts = &fsi->mount_opts;
 246
 247         mutex_lock(&root->d_inode->i_mutex);
 248
 249         /* If we have already created ptmx node, return */
 250         if (fsi->ptmx_dentry) {
 251                 rc = 0;
 252                 goto out;
 253         }
 254
 255         dentry = d_alloc_name(root, "ptmx");
 256         if (!dentry) {
 257                 printk(KERN_NOTICE "Unable to alloc dentry for ptmx node\n");
 258                 goto out;
 259         }
 260
 261         /*
 262          * Create a new 'ptmx' node in this mount of devpts.
 263          */
 264         inode = new_inode(sb);
 265         if (!inode) {
 266                 printk(KERN_ERR "Unable to alloc inode for ptmx node\n");
 267                 dput(dentry);
 268                 goto out;
 269         }
 270
 271         inode->i_ino = 2;
 272         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 273
 274         mode = S_IFCHR|opts->ptmxmode;
 275         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 276
 277         d_add(dentry, inode);
 278
 279         fsi->ptmx_dentry = dentry;
 280         rc = 0;
 281 out:
 282         mutex_unlock(&root->d_inode->i_mutex);
 283         return rc;
 284 }
 285
 286 static void update_ptmx_mode(struct pts_fs_info *fsi)
 287 {
 288         struct inode *inode;
 289         if (fsi->ptmx_dentry) {
 290                 inode = fsi->ptmx_dentry->d_inode;
 291                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 292         }
 293 }
 294 #else
 295 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
 296 {
 297        return;
 298 }
 299 #endif
 300
 301 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 302 {
 303         int err;
 304         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 305         struct pts_mount_opts *opts = &fsi->mount_opts;
 306
 307         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 308
 309         /*
 310          * parse_mount_options() restores options to default values
 311          * before parsing and may have changed ptmxmode. So, update the
 312          * mode in the inode too. Bogus options don't fail the remount,
 313          * so do this even on error return.
 314          */
 315         update_ptmx_mode(fsi);
 316
 317         return err;
 318 }
 319
 320 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
 321 {
 322         struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
 323         struct pts_mount_opts *opts = &fsi->mount_opts;
 324
 325         if (opts->setuid)
 326                 seq_printf(seq, ",uid=%u", from_kuid_munged(&init_user_ns, opts->uid));
 327         if (opts->setgid)
 328                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, opts->gid));
 329         seq_printf(seq, ",mode=%03o", opts->mode);
 330 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 331         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 332         if (opts->max < NR_UNIX98_PTY_MAX)
 333                 seq_printf(seq, ",max=%d", opts->max);
 334 #endif
 335
 336         return 0;
 337 }
 338
 339 static const struct super_operations devpts_sops = {
 340         .statfs         = simple_statfs,
 341         .remount_fs     = devpts_remount,
 342         .show_options   = devpts_show_options,
 343 };
 344
 345 static void *new_pts_fs_info(void)
 346 {
 347         struct pts_fs_info *fsi;
 348
 349         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 350         if (!fsi)
 351                 return NULL;
 352
 353         ida_init(&fsi->allocated_ptys);
 354         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 355         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 356
 357         return fsi;
 358 }
 359
 360 static int
 361 devpts_fill_super(struct super_block *s, void *data, int silent)
 362 {
 363         struct inode *inode;
 364
 365         s->s_blocksize = 1024;
 366         s->s_blocksize_bits = 10;
 367         s->s_magic = DEVPTS_SUPER_MAGIC;
 368         s->s_op = &devpts_sops;
 369         s->s_time_gran = 1;
 370
 371         s->s_fs_info = new_pts_fs_info();
 372         if (!s->s_fs_info)
 373                 goto fail;
 374
 375         inode = new_inode(s);
 376         if (!inode)
 377                 goto fail;
 378         inode->i_ino = 1;
 379         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 380         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 381         inode->i_op = &simple_dir_inode_operations;
 382         inode->i_fop = &simple_dir_operations;
 383         set_nlink(inode, 2);
 384
 385         s->s_root = d_make_root(inode);
 386         if (s->s_root)
 387                 return 0;
 388
 389         printk(KERN_ERR "devpts: get root dentry failed\n");
 390
 391 fail:
 392         return -ENOMEM;
 393 }
 394
 395 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 396 static int compare_init_pts_sb(struct super_block *s, void *p)
 397 {
 398         if (devpts_mnt)
 399                 return devpts_mnt->mnt_sb == s;
 400         return 0;
 401 }
 402
 403 /*
 404  * devpts_mount()
 405  *
 406  *     If the '-o newinstance' mount option was specified, mount a new
 407  *     (private) instance of devpts.  PTYs created in this instance are
 408  *     independent of the PTYs in other devpts instances.
 409  *
 410  *     If the '-o newinstance' option was not specified, mount/remount the
 411  *     initial kernel mount of devpts.  This type of mount gives the
 412  *     legacy, single-instance semantics.
 413  *
 414  *     The 'newinstance' option is needed to support multiple namespace
 415  *     semantics in devpts while preserving backward compatibility of the
 416  *     current 'single-namespace' semantics. i.e all mounts of devpts
 417  *     without the 'newinstance' mount option should bind to the initial
 418  *     kernel mount, like mount_single().
 419  *
 420  *     Mounts with 'newinstance' option create a new, private namespace.
 421  *
 422  *     NOTE:
 423  *
 424  *     For single-mount semantics, devpts cannot use mount_single(),
 425  *     because mount_single()/sget() find and use the super-block from
 426  *     the most recent mount of devpts. But that recent mount may be a
 427  *     'newinstance' mount and mount_single() would pick the newinstance
 428  *     super-block instead of the initial super-block.
 429  */
 430 static struct dentry *devpts_mount(struct file_system_type *fs_type,
 431         int flags, const char *dev_name, void *data)
 432 {
 433         int error;
 434         struct pts_mount_opts opts;
 435         struct super_block *s;
 436
 437         error = parse_mount_options(data, PARSE_MOUNT, &opts);
 438         if (error)
 439                 return ERR_PTR(error);
 440
 441         if (opts.newinstance)
 442                 s = sget(fs_type, NULL, set_anon_super, flags, NULL);
 443         else
 444                 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags,
 445                          NULL);
 446
 447         if (IS_ERR(s))
 448                 return ERR_CAST(s);
 449
 450         if (!s->s_root) {
 451                 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 452                 if (error)
 453                         goto out_undo_sget;
 454                 s->s_flags |= MS_ACTIVE;
 455         }
 456
 457         memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
 458
 459         error = mknod_ptmx(s);
 460         if (error)
 461                 goto out_undo_sget;
 462
 463         return dget(s->s_root);
 464
 465 out_undo_sget:
 466         deactivate_locked_super(s);
 467         return ERR_PTR(error);
 468 }
 469
 470 #else
 471 /*
 472  * This supports only the legacy single-instance semantics (no
 473  * multiple-instance semantics)
 474  */
 475 static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
 476                 const char *dev_name, void *data)
 477 {
 478         return mount_single(fs_type, flags, data, devpts_fill_super);
 479 }
 480 #endif
 481
 482 static void devpts_kill_sb(struct super_block *sb)
 483 {
 484         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 485
 486         kfree(fsi);
 487         kill_litter_super(sb);
 488 }
 489
 490 static struct file_system_type devpts_fs_type = {
 491         .name           = "devpts",
 492         .mount          = devpts_mount,
 493         .kill_sb        = devpts_kill_sb,
 494 };
 495
 496 /*
 497  * The normal naming convention is simply /dev/pts/<number>; this conforms
 498  * to the System V naming convention
 499  */
 500
 501 int devpts_new_index(struct inode *ptmx_inode)
 502 {
 503         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 504         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 505         int index;
 506         int ida_ret;
 507
 508 retry:
 509         if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
 510                 return -ENOMEM;
 511
 512         mutex_lock(&allocated_ptys_lock);
 513         if (pty_count >= pty_limit -
 514                         (fsi->mount_opts.newinstance ? pty_reserve : 0)) {
 515                 mutex_unlock(&allocated_ptys_lock);
 516                 return -ENOSPC;
 517         }
 518
 519         ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 520         if (ida_ret < 0) {
 521                 mutex_unlock(&allocated_ptys_lock);
 522                 if (ida_ret == -EAGAIN)
 523                         goto retry;
 524                 return -EIO;
 525         }
 526
 527         if (index >= fsi->mount_opts.max) {
 528                 ida_remove(&fsi->allocated_ptys, index);
 529                 mutex_unlock(&allocated_ptys_lock);
 530                 return -ENOSPC;
 531         }
 532         pty_count++;
 533         mutex_unlock(&allocated_ptys_lock);
 534         return index;
 535 }
 536
 537 void devpts_kill_index(struct inode *ptmx_inode, int idx)
 538 {
 539         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 540         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 541
 542         mutex_lock(&allocated_ptys_lock);
 543         ida_remove(&fsi->allocated_ptys, idx);
 544         pty_count--;
 545         mutex_unlock(&allocated_ptys_lock);
 546 }
 547
 548 int devpts_pty_new(struct inode *ptmx_inode, struct tty_struct *tty)
 549 {
 550         /* tty layer puts index from devpts_new_index() in here */
 551         int number = tty->index;
 552         struct tty_driver *driver = tty->driver;
 553         dev_t device = MKDEV(driver->major, driver->minor_start+number);
 554         struct dentry *dentry;
 555         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 556         struct inode *inode = new_inode(sb);
 557         struct dentry *root = sb->s_root;
 558         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 559         struct pts_mount_opts *opts = &fsi->mount_opts;
 560         int ret = 0;
 561         char s[12];
 562
 563         /* We're supposed to be given the slave end of a pty */
 564         BUG_ON(driver->type != TTY_DRIVER_TYPE_PTY);
 565         BUG_ON(driver->subtype != PTY_TYPE_SLAVE);
 566
 567         if (!inode)
 568                 return -ENOMEM;
 569
 570         inode->i_ino = number + 3;
 571         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 572         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 573         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 574         init_special_inode(inode, S_IFCHR|opts->mode, device);
 575         inode->i_private = tty;
 576         tty->driver_data = inode;
 577
 578         sprintf(s, "%d", number);
 579
 580         mutex_lock(&root->d_inode->i_mutex);
 581
 582         dentry = d_alloc_name(root, s);
 583         if (dentry) {
 584                 d_add(dentry, inode);
 585                 fsnotify_create(root->d_inode, dentry);
 586         } else {
 587                 iput(inode);
 588                 ret = -ENOMEM;
 589         }
 590
 591         mutex_unlock(&root->d_inode->i_mutex);
 592
 593         return ret;
 594 }
 595
 596 struct tty_struct *devpts_get_tty(struct inode *pts_inode, int number)
 597 {
 598         struct dentry *dentry;
 599         struct tty_struct *tty;
 600
 601         BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 602
 603         /* Ensure dentry has not been deleted by devpts_pty_kill() */
 604         dentry = d_find_alias(pts_inode);
 605         if (!dentry)
 606                 return NULL;
 607
 608         tty = NULL;
 609         if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
 610                 tty = (struct tty_struct *)pts_inode->i_private;
 611
 612         dput(dentry);
 613
 614         return tty;
 615 }
 616
 617 void devpts_pty_kill(struct tty_struct *tty)
 618 {
 619         struct inode *inode = tty->driver_data;
 620         struct super_block *sb = pts_sb_from_inode(inode);
 621         struct dentry *root = sb->s_root;
 622         struct dentry *dentry;
 623
 624         BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 625
 626         mutex_lock(&root->d_inode->i_mutex);
 627
 628         dentry = d_find_alias(inode);
 629
 630         drop_nlink(inode);
 631         d_delete(dentry);
 632         dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 633         dput(dentry);           /* d_find_alias above */
 634
 635         mutex_unlock(&root->d_inode->i_mutex);
 636 }
 637
 638 static int __init init_devpts_fs(void)
 639 {
 640         int err = register_filesystem(&devpts_fs_type);
 641         struct ctl_table_header *table;
 642
 643         if (!err) {
 644                 table = register_sysctl_table(pty_root_table);
 645                 devpts_mnt = kern_mount(&devpts_fs_type);
 646                 if (IS_ERR(devpts_mnt)) {
 647                         err = PTR_ERR(devpts_mnt);
 648                         unregister_filesystem(&devpts_fs_type);
 649                         unregister_sysctl_table(table);
 650                 }
 651         }
 652         return err;
 653 }
 654 module_init(init_devpts_fs)