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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  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>
  30
  31 #define DEVPTS_DEFAULT_MODE 0600
  32 /*
  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.
  37  */
  38 #define DEVPTS_DEFAULT_PTMX_MODE 0000
  39 #define PTMX_MINOR      2
  40
  41 /*
  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.
  44  */
  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;
  50
  51 static struct ctl_table pty_table[] = {
  52         {
  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,
  74         },
  75         {}
  76 };
  77
  78 static struct ctl_table pty_kern_table[] = {
  79         {
  80                 .procname       = "pty",
  81                 .mode           = 0555,
  82                 .child          = pty_table,
  83         },
  84         {}
  85 };
  86
  87 static struct ctl_table pty_root_table[] = {
  88         {
  89                 .procname       = "kernel",
  90                 .mode           = 0555,
  91                 .child          = pty_kern_table,
  92         },
  93         {}
  94 };
  95
  96 static DEFINE_MUTEX(allocated_ptys_lock);
  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 reserve;
 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         {Opt_ptmxmode, "ptmxmode=%o"},
 119         {Opt_newinstance, "newinstance"},
 120         {Opt_max, "max=%d"},
 121         {Opt_err, NULL}
 122 };
 123
 124 struct pts_fs_info {
 125         struct ida allocated_ptys;
 126         struct pts_mount_opts mount_opts;
 127         struct super_block *sb;
 128         struct dentry *ptmx_dentry;
 129 };
 130
 131 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
 132 {
 133         return sb->s_fs_info;
 134 }
 135
 136 static int devpts_ptmx_path(struct path *path)
 137 {
 138         struct super_block *sb;
 139         int err;
 140
 141         /* Is a devpts filesystem at "pts" in the same directory? */
 142         err = path_pts(path);
 143         if (err)
 144                 return err;
 145
 146         /* Is the path the root of a devpts filesystem? */
 147         sb = path->mnt->mnt_sb;
 148         if ((sb->s_magic != DEVPTS_SUPER_MAGIC) ||
 149             (path->mnt->mnt_root != sb->s_root))
 150                 return -ENODEV;
 151
 152         return 0;
 153 }
 154
 155 /*
 156  * Try to find a suitable devpts filesystem. We support the following
 157  * scenarios:
 158  * - The ptmx device node is located in the same directory as the devpts
 159  *   mount where the pts device nodes are located.
 160  *   This is e.g. the case when calling open on the /dev/pts/ptmx device
 161  *   node when the devpts filesystem is mounted at /dev/pts.
 162  * - The ptmx device node is located outside the devpts filesystem mount
 163  *   where the pts device nodes are located. For example, the ptmx device
 164  *   is a symlink, separate device node, or bind-mount.
 165  *   A supported scenario is bind-mounting /dev/pts/ptmx to /dev/ptmx and
 166  *   then calling open on /dev/ptmx. In this case a suitable pts
 167  *   subdirectory can be found in the common parent directory /dev of the
 168  *   devpts mount and the ptmx bind-mount, after resolving the /dev/ptmx
 169  *   bind-mount.
 170  *   If no suitable pts subdirectory can be found this function will fail.
 171  *   This is e.g. the case when bind-mounting /dev/pts/ptmx to /ptmx.
 172  */
 173 struct vfsmount *devpts_mntget(struct file *filp, struct pts_fs_info *fsi)
 174 {
 175         struct path path;
 176         int err = 0;
 177
 178         path = filp->f_path;
 179         path_get(&path);
 180
 181         /* Walk upward while the start point is a bind mount of
 182          * a single file.
 183          */
 184         while (path.mnt->mnt_root == path.dentry)
 185                 if (follow_up(&path) == 0)
 186                         break;
 187
 188         /* devpts_ptmx_path() finds a devpts fs or returns an error. */
 189         if ((path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) ||
 190             (DEVPTS_SB(path.mnt->mnt_sb) != fsi))
 191                 err = devpts_ptmx_path(&path);
 192         dput(path.dentry);
 193         if (!err) {
 194                 if (DEVPTS_SB(path.mnt->mnt_sb) == fsi)
 195                         return path.mnt;
 196
 197                 err = -ENODEV;
 198         }
 199
 200         mntput(path.mnt);
 201         return ERR_PTR(err);
 202 }
 203
 204 struct pts_fs_info *devpts_acquire(struct file *filp)
 205 {
 206         struct pts_fs_info *result;
 207         struct path path;
 208         struct super_block *sb;
 209
 210         path = filp->f_path;
 211         path_get(&path);
 212
 213         /* Has the devpts filesystem already been found? */
 214         if (path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) {
 215                 int err;
 216
 217                 err = devpts_ptmx_path(&path);
 218                 if (err) {
 219                         result = ERR_PTR(err);
 220                         goto out;
 221                 }
 222         }
 223
 224         /*
 225          * pty code needs to hold extra references in case of last /dev/tty close
 226          */
 227         sb = path.mnt->mnt_sb;
 228         atomic_inc(&sb->s_active);
 229         result = DEVPTS_SB(sb);
 230
 231 out:
 232         path_put(&path);
 233         return result;
 234 }
 235
 236 void devpts_release(struct pts_fs_info *fsi)
 237 {
 238         deactivate_super(fsi->sb);
 239 }
 240
 241 #define PARSE_MOUNT     0
 242 #define PARSE_REMOUNT   1
 243
 244 /*
 245  * parse_mount_options():
 246  *      Set @opts to mount options specified in @data. If an option is not
 247  *      specified in @data, set it to its default value.
 248  *
 249  * Note: @data may be NULL (in which case all options are set to default).
 250  */
 251 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 252 {
 253         char *p;
 254         kuid_t uid;
 255         kgid_t gid;
 256
 257         opts->setuid  = 0;
 258         opts->setgid  = 0;
 259         opts->uid     = GLOBAL_ROOT_UID;
 260         opts->gid     = GLOBAL_ROOT_GID;
 261         opts->mode    = DEVPTS_DEFAULT_MODE;
 262         opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 263         opts->max     = NR_UNIX98_PTY_MAX;
 264
 265         /* Only allow instances mounted from the initial mount
 266          * namespace to tap the reserve pool of ptys.
 267          */
 268         if (op == PARSE_MOUNT)
 269                 opts->reserve =
 270                         (current->nsproxy->mnt_ns == init_task.nsproxy->mnt_ns);
 271
 272         while ((p = strsep(&data, ",")) != NULL) {
 273                 substring_t args[MAX_OPT_ARGS];
 274                 int token;
 275                 int option;
 276
 277                 if (!*p)
 278                         continue;
 279
 280                 token = match_token(p, tokens, args);
 281                 switch (token) {
 282                 case Opt_uid:
 283                         if (match_int(&args[0], &option))
 284                                 return -EINVAL;
 285                         uid = make_kuid(current_user_ns(), option);
 286                         if (!uid_valid(uid))
 287                                 return -EINVAL;
 288                         opts->uid = uid;
 289                         opts->setuid = 1;
 290                         break;
 291                 case Opt_gid:
 292                         if (match_int(&args[0], &option))
 293                                 return -EINVAL;
 294                         gid = make_kgid(current_user_ns(), option);
 295                         if (!gid_valid(gid))
 296                                 return -EINVAL;
 297                         opts->gid = gid;
 298                         opts->setgid = 1;
 299                         break;
 300                 case Opt_mode:
 301                         if (match_octal(&args[0], &option))
 302                                 return -EINVAL;
 303                         opts->mode = option & S_IALLUGO;
 304                         break;
 305                 case Opt_ptmxmode:
 306                         if (match_octal(&args[0], &option))
 307                                 return -EINVAL;
 308                         opts->ptmxmode = option & S_IALLUGO;
 309                         break;
 310                 case Opt_newinstance:
 311                         break;
 312                 case Opt_max:
 313                         if (match_int(&args[0], &option) ||
 314                             option < 0 || option > NR_UNIX98_PTY_MAX)
 315                                 return -EINVAL;
 316                         opts->max = option;
 317                         break;
 318                 default:
 319                         pr_err("called with bogus options\n");
 320                         return -EINVAL;
 321                 }
 322         }
 323
 324         return 0;
 325 }
 326
 327 static int mknod_ptmx(struct super_block *sb)
 328 {
 329         int mode;
 330         int rc = -ENOMEM;
 331         struct dentry *dentry;
 332         struct inode *inode;
 333         struct dentry *root = sb->s_root;
 334         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 335         struct pts_mount_opts *opts = &fsi->mount_opts;
 336         kuid_t ptmx_uid = current_fsuid();
 337         kgid_t ptmx_gid = current_fsgid();
 338
 339         inode_lock(d_inode(root));
 340
 341         /* If we have already created ptmx node, return */
 342         if (fsi->ptmx_dentry) {
 343                 rc = 0;
 344                 goto out;
 345         }
 346
 347         dentry = d_alloc_name(root, "ptmx");
 348         if (!dentry) {
 349                 pr_err("Unable to alloc dentry for ptmx node\n");
 350                 goto out;
 351         }
 352
 353         /*
 354          * Create a new 'ptmx' node in this mount of devpts.
 355          */
 356         inode = new_inode(sb);
 357         if (!inode) {
 358                 pr_err("Unable to alloc inode for ptmx node\n");
 359                 dput(dentry);
 360                 goto out;
 361         }
 362
 363         inode->i_ino = 2;
 364         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 365
 366         mode = S_IFCHR|opts->ptmxmode;
 367         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 368         inode->i_uid = ptmx_uid;
 369         inode->i_gid = ptmx_gid;
 370
 371         d_add(dentry, inode);
 372
 373         fsi->ptmx_dentry = dentry;
 374         rc = 0;
 375 out:
 376         inode_unlock(d_inode(root));
 377         return rc;
 378 }
 379
 380 static void update_ptmx_mode(struct pts_fs_info *fsi)
 381 {
 382         struct inode *inode;
 383         if (fsi->ptmx_dentry) {
 384                 inode = d_inode(fsi->ptmx_dentry);
 385                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 386         }
 387 }
 388
 389 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 390 {
 391         int err;
 392         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 393         struct pts_mount_opts *opts = &fsi->mount_opts;
 394
 395         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 396
 397         /*
 398          * parse_mount_options() restores options to default values
 399          * before parsing and may have changed ptmxmode. So, update the
 400          * mode in the inode too. Bogus options don't fail the remount,
 401          * so do this even on error return.
 402          */
 403         update_ptmx_mode(fsi);
 404
 405         return err;
 406 }
 407
 408 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
 409 {
 410         struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
 411         struct pts_mount_opts *opts = &fsi->mount_opts;
 412
 413         if (opts->setuid)
 414                 seq_printf(seq, ",uid=%u",
 415                            from_kuid_munged(&init_user_ns, opts->uid));
 416         if (opts->setgid)
 417                 seq_printf(seq, ",gid=%u",
 418                            from_kgid_munged(&init_user_ns, opts->gid));
 419         seq_printf(seq, ",mode=%03o", opts->mode);
 420         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 421         if (opts->max < NR_UNIX98_PTY_MAX)
 422                 seq_printf(seq, ",max=%d", opts->max);
 423
 424         return 0;
 425 }
 426
 427 static const struct super_operations devpts_sops = {
 428         .statfs         = simple_statfs,
 429         .remount_fs     = devpts_remount,
 430         .show_options   = devpts_show_options,
 431 };
 432
 433 static void *new_pts_fs_info(struct super_block *sb)
 434 {
 435         struct pts_fs_info *fsi;
 436
 437         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 438         if (!fsi)
 439                 return NULL;
 440
 441         ida_init(&fsi->allocated_ptys);
 442         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 443         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 444         fsi->sb = sb;
 445
 446         return fsi;
 447 }
 448
 449 static int
 450 devpts_fill_super(struct super_block *s, void *data, int silent)
 451 {
 452         struct inode *inode;
 453         int error;
 454
 455         s->s_iflags &= ~SB_I_NODEV;
 456         s->s_blocksize = 1024;
 457         s->s_blocksize_bits = 10;
 458         s->s_magic = DEVPTS_SUPER_MAGIC;
 459         s->s_op = &devpts_sops;
 460         s->s_d_op = &simple_dentry_operations;
 461         s->s_time_gran = 1;
 462
 463         error = -ENOMEM;
 464         s->s_fs_info = new_pts_fs_info(s);
 465         if (!s->s_fs_info)
 466                 goto fail;
 467
 468         error = parse_mount_options(data, PARSE_MOUNT, &DEVPTS_SB(s)->mount_opts);
 469         if (error)
 470                 goto fail;
 471
 472         error = -ENOMEM;
 473         inode = new_inode(s);
 474         if (!inode)
 475                 goto fail;
 476         inode->i_ino = 1;
 477         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 478         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 479         inode->i_op = &simple_dir_inode_operations;
 480         inode->i_fop = &simple_dir_operations;
 481         set_nlink(inode, 2);
 482
 483         s->s_root = d_make_root(inode);
 484         if (!s->s_root) {
 485                 pr_err("get root dentry failed\n");
 486                 goto fail;
 487         }
 488
 489         error = mknod_ptmx(s);
 490         if (error)
 491                 goto fail_dput;
 492
 493         return 0;
 494 fail_dput:
 495         dput(s->s_root);
 496         s->s_root = NULL;
 497 fail:
 498         return error;
 499 }
 500
 501 /*
 502  * devpts_mount()
 503  *
 504  *     Mount a new (private) instance of devpts.  PTYs created in this
 505  *     instance are independent of the PTYs in other devpts instances.
 506  */
 507 static struct dentry *devpts_mount(struct file_system_type *fs_type,
 508         int flags, const char *dev_name, void *data)
 509 {
 510         return mount_nodev(fs_type, flags, data, devpts_fill_super);
 511 }
 512
 513 static void devpts_kill_sb(struct super_block *sb)
 514 {
 515         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 516
 517         if (fsi)
 518                 ida_destroy(&fsi->allocated_ptys);
 519         kfree(fsi);
 520         kill_litter_super(sb);
 521 }
 522
 523 static struct file_system_type devpts_fs_type = {
 524         .name           = "devpts",
 525         .mount          = devpts_mount,
 526         .kill_sb        = devpts_kill_sb,
 527         .fs_flags       = FS_USERNS_MOUNT,
 528 };
 529
 530 /*
 531  * The normal naming convention is simply /dev/pts/<number>; this conforms
 532  * to the System V naming convention
 533  */
 534
 535 int devpts_new_index(struct pts_fs_info *fsi)
 536 {
 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         if (pty_count >= (pty_limit -
 546                           (fsi->mount_opts.reserve ? 0 : pty_reserve))) {
 547                 mutex_unlock(&allocated_ptys_lock);
 548                 return -ENOSPC;
 549         }
 550
 551         ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 552         if (ida_ret < 0) {
 553                 mutex_unlock(&allocated_ptys_lock);
 554                 if (ida_ret == -EAGAIN)
 555                         goto retry;
 556                 return -EIO;
 557         }
 558
 559         if (index >= fsi->mount_opts.max) {
 560                 ida_remove(&fsi->allocated_ptys, index);
 561                 mutex_unlock(&allocated_ptys_lock);
 562                 return -ENOSPC;
 563         }
 564         pty_count++;
 565         mutex_unlock(&allocated_ptys_lock);
 566         return index;
 567 }
 568
 569 void devpts_kill_index(struct pts_fs_info *fsi, int idx)
 570 {
 571         mutex_lock(&allocated_ptys_lock);
 572         ida_remove(&fsi->allocated_ptys, idx);
 573         pty_count--;
 574         mutex_unlock(&allocated_ptys_lock);
 575 }
 576
 577 /**
 578  * devpts_pty_new -- create a new inode in /dev/pts/
 579  * @ptmx_inode: inode of the master
 580  * @device: major+minor of the node to be created
 581  * @index: used as a name of the node
 582  * @priv: what's given back by devpts_get_priv
 583  *
 584  * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
 585  */
 586 struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
 587 {
 588         struct dentry *dentry;
 589         struct super_block *sb = fsi->sb;
 590         struct inode *inode;
 591         struct dentry *root;
 592         struct pts_mount_opts *opts;
 593         char s[12];
 594
 595         root = sb->s_root;
 596         opts = &fsi->mount_opts;
 597
 598         inode = new_inode(sb);
 599         if (!inode)
 600                 return ERR_PTR(-ENOMEM);
 601
 602         inode->i_ino = index + 3;
 603         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 604         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 605         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 606         init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
 607
 608         sprintf(s, "%d", index);
 609
 610         dentry = d_alloc_name(root, s);
 611         if (dentry) {
 612                 dentry->d_fsdata = priv;
 613                 d_add(dentry, inode);
 614                 fsnotify_create(d_inode(root), dentry);
 615         } else {
 616                 iput(inode);
 617                 dentry = ERR_PTR(-ENOMEM);
 618         }
 619
 620         return dentry;
 621 }
 622
 623 /**
 624  * devpts_get_priv -- get private data for a slave
 625  * @pts_inode: inode of the slave
 626  *
 627  * Returns whatever was passed as priv in devpts_pty_new for a given inode.
 628  */
 629 void *devpts_get_priv(struct dentry *dentry)
 630 {
 631         if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
 632                 return NULL;
 633         return dentry->d_fsdata;
 634 }
 635
 636 /**
 637  * devpts_pty_kill -- remove inode form /dev/pts/
 638  * @inode: inode of the slave to be removed
 639  *
 640  * This is an inverse operation of devpts_pty_new.
 641  */
 642 void devpts_pty_kill(struct dentry *dentry)
 643 {
 644         WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
 645
 646         dentry->d_fsdata = NULL;
 647         drop_nlink(dentry->d_inode);
 648         d_delete(dentry);
 649         dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 650 }
 651
 652 static int __init init_devpts_fs(void)
 653 {
 654         int err = register_filesystem(&devpts_fs_type);
 655         if (!err) {
 656                 register_sysctl_table(pty_root_table);
 657         }
 658         return err;
 659 }
 660 module_init(init_devpts_fs)