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 struct pts_fs_info *devpts_acquire(struct file *filp)
 137 {
 138         struct pts_fs_info *result;
 139         struct path path;
 140         struct super_block *sb;
 141         int err;
 142
 143         path = filp->f_path;
 144         path_get(&path);
 145
 146         /* Has the devpts filesystem already been found? */
 147         sb = path.mnt->mnt_sb;
 148         if (sb->s_magic != DEVPTS_SUPER_MAGIC) {
 149                 /* Is a devpts filesystem at "pts" in the same directory? */
 150                 err = path_pts(&path);
 151                 if (err) {
 152                         result = ERR_PTR(err);
 153                         goto out;
 154                 }
 155
 156                 /* Is the path the root of a devpts filesystem? */
 157                 result = ERR_PTR(-ENODEV);
 158                 sb = path.mnt->mnt_sb;
 159                 if ((sb->s_magic != DEVPTS_SUPER_MAGIC) ||
 160                     (path.mnt->mnt_root != sb->s_root))
 161                         goto out;
 162         }
 163
 164         /*
 165          * pty code needs to hold extra references in case of last /dev/tty close
 166          */
 167         atomic_inc(&sb->s_active);
 168         result = DEVPTS_SB(sb);
 169
 170 out:
 171         path_put(&path);
 172         return result;
 173 }
 174
 175 void devpts_release(struct pts_fs_info *fsi)
 176 {
 177         deactivate_super(fsi->sb);
 178 }
 179
 180 #define PARSE_MOUNT     0
 181 #define PARSE_REMOUNT   1
 182
 183 /*
 184  * parse_mount_options():
 185  *      Set @opts to mount options specified in @data. If an option is not
 186  *      specified in @data, set it to its default value.
 187  *
 188  * Note: @data may be NULL (in which case all options are set to default).
 189  */
 190 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 191 {
 192         char *p;
 193         kuid_t uid;
 194         kgid_t gid;
 195
 196         opts->setuid  = 0;
 197         opts->setgid  = 0;
 198         opts->uid     = GLOBAL_ROOT_UID;
 199         opts->gid     = GLOBAL_ROOT_GID;
 200         opts->mode    = DEVPTS_DEFAULT_MODE;
 201         opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 202         opts->max     = NR_UNIX98_PTY_MAX;
 203
 204         /* Only allow instances mounted from the initial mount
 205          * namespace to tap the reserve pool of ptys.
 206          */
 207         if (op == PARSE_MOUNT)
 208                 opts->reserve =
 209                         (current->nsproxy->mnt_ns == init_task.nsproxy->mnt_ns);
 210
 211         while ((p = strsep(&data, ",")) != NULL) {
 212                 substring_t args[MAX_OPT_ARGS];
 213                 int token;
 214                 int option;
 215
 216                 if (!*p)
 217                         continue;
 218
 219                 token = match_token(p, tokens, args);
 220                 switch (token) {
 221                 case Opt_uid:
 222                         if (match_int(&args[0], &option))
 223                                 return -EINVAL;
 224                         uid = make_kuid(current_user_ns(), option);
 225                         if (!uid_valid(uid))
 226                                 return -EINVAL;
 227                         opts->uid = uid;
 228                         opts->setuid = 1;
 229                         break;
 230                 case Opt_gid:
 231                         if (match_int(&args[0], &option))
 232                                 return -EINVAL;
 233                         gid = make_kgid(current_user_ns(), option);
 234                         if (!gid_valid(gid))
 235                                 return -EINVAL;
 236                         opts->gid = gid;
 237                         opts->setgid = 1;
 238                         break;
 239                 case Opt_mode:
 240                         if (match_octal(&args[0], &option))
 241                                 return -EINVAL;
 242                         opts->mode = option & S_IALLUGO;
 243                         break;
 244                 case Opt_ptmxmode:
 245                         if (match_octal(&args[0], &option))
 246                                 return -EINVAL;
 247                         opts->ptmxmode = option & S_IALLUGO;
 248                         break;
 249                 case Opt_newinstance:
 250                         break;
 251                 case Opt_max:
 252                         if (match_int(&args[0], &option) ||
 253                             option < 0 || option > NR_UNIX98_PTY_MAX)
 254                                 return -EINVAL;
 255                         opts->max = option;
 256                         break;
 257                 default:
 258                         pr_err("called with bogus options\n");
 259                         return -EINVAL;
 260                 }
 261         }
 262
 263         return 0;
 264 }
 265
 266 static int mknod_ptmx(struct super_block *sb)
 267 {
 268         int mode;
 269         int rc = -ENOMEM;
 270         struct dentry *dentry;
 271         struct inode *inode;
 272         struct dentry *root = sb->s_root;
 273         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 274         struct pts_mount_opts *opts = &fsi->mount_opts;
 275         kuid_t root_uid;
 276         kgid_t root_gid;
 277
 278         root_uid = make_kuid(current_user_ns(), 0);
 279         root_gid = make_kgid(current_user_ns(), 0);
 280         if (!uid_valid(root_uid) || !gid_valid(root_gid))
 281                 return -EINVAL;
 282
 283         inode_lock(d_inode(root));
 284
 285         /* If we have already created ptmx node, return */
 286         if (fsi->ptmx_dentry) {
 287                 rc = 0;
 288                 goto out;
 289         }
 290
 291         dentry = d_alloc_name(root, "ptmx");
 292         if (!dentry) {
 293                 pr_err("Unable to alloc dentry for ptmx node\n");
 294                 goto out;
 295         }
 296
 297         /*
 298          * Create a new 'ptmx' node in this mount of devpts.
 299          */
 300         inode = new_inode(sb);
 301         if (!inode) {
 302                 pr_err("Unable to alloc inode for ptmx node\n");
 303                 dput(dentry);
 304                 goto out;
 305         }
 306
 307         inode->i_ino = 2;
 308         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 309
 310         mode = S_IFCHR|opts->ptmxmode;
 311         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 312         inode->i_uid = root_uid;
 313         inode->i_gid = root_gid;
 314
 315         d_add(dentry, inode);
 316
 317         fsi->ptmx_dentry = dentry;
 318         rc = 0;
 319 out:
 320         inode_unlock(d_inode(root));
 321         return rc;
 322 }
 323
 324 static void update_ptmx_mode(struct pts_fs_info *fsi)
 325 {
 326         struct inode *inode;
 327         if (fsi->ptmx_dentry) {
 328                 inode = d_inode(fsi->ptmx_dentry);
 329                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 330         }
 331 }
 332
 333 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 334 {
 335         int err;
 336         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 337         struct pts_mount_opts *opts = &fsi->mount_opts;
 338
 339         sync_filesystem(sb);
 340         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 341
 342         /*
 343          * parse_mount_options() restores options to default values
 344          * before parsing and may have changed ptmxmode. So, update the
 345          * mode in the inode too. Bogus options don't fail the remount,
 346          * so do this even on error return.
 347          */
 348         update_ptmx_mode(fsi);
 349
 350         return err;
 351 }
 352
 353 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
 354 {
 355         struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
 356         struct pts_mount_opts *opts = &fsi->mount_opts;
 357
 358         if (opts->setuid)
 359                 seq_printf(seq, ",uid=%u",
 360                            from_kuid_munged(&init_user_ns, opts->uid));
 361         if (opts->setgid)
 362                 seq_printf(seq, ",gid=%u",
 363                            from_kgid_munged(&init_user_ns, opts->gid));
 364         seq_printf(seq, ",mode=%03o", opts->mode);
 365         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 366         if (opts->max < NR_UNIX98_PTY_MAX)
 367                 seq_printf(seq, ",max=%d", opts->max);
 368
 369         return 0;
 370 }
 371
 372 static const struct super_operations devpts_sops = {
 373         .statfs         = simple_statfs,
 374         .remount_fs     = devpts_remount,
 375         .show_options   = devpts_show_options,
 376 };
 377
 378 static void *new_pts_fs_info(struct super_block *sb)
 379 {
 380         struct pts_fs_info *fsi;
 381
 382         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 383         if (!fsi)
 384                 return NULL;
 385
 386         ida_init(&fsi->allocated_ptys);
 387         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 388         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 389         fsi->sb = sb;
 390
 391         return fsi;
 392 }
 393
 394 static int
 395 devpts_fill_super(struct super_block *s, void *data, int silent)
 396 {
 397         struct inode *inode;
 398         int error;
 399
 400         s->s_iflags &= ~SB_I_NODEV;
 401         s->s_blocksize = 1024;
 402         s->s_blocksize_bits = 10;
 403         s->s_magic = DEVPTS_SUPER_MAGIC;
 404         s->s_op = &devpts_sops;
 405         s->s_time_gran = 1;
 406
 407         error = -ENOMEM;
 408         s->s_fs_info = new_pts_fs_info(s);
 409         if (!s->s_fs_info)
 410                 goto fail;
 411
 412         error = parse_mount_options(data, PARSE_MOUNT, &DEVPTS_SB(s)->mount_opts);
 413         if (error)
 414                 goto fail;
 415
 416         error = -ENOMEM;
 417         inode = new_inode(s);
 418         if (!inode)
 419                 goto fail;
 420         inode->i_ino = 1;
 421         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 422         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 423         inode->i_op = &simple_dir_inode_operations;
 424         inode->i_fop = &simple_dir_operations;
 425         set_nlink(inode, 2);
 426
 427         s->s_root = d_make_root(inode);
 428         if (!s->s_root) {
 429                 pr_err("get root dentry failed\n");
 430                 goto fail;
 431         }
 432
 433         error = mknod_ptmx(s);
 434         if (error)
 435                 goto fail_dput;
 436
 437         return 0;
 438 fail_dput:
 439         dput(s->s_root);
 440         s->s_root = NULL;
 441 fail:
 442         return error;
 443 }
 444
 445 /*
 446  * devpts_mount()
 447  *
 448  *     Mount a new (private) instance of devpts.  PTYs created in this
 449  *     instance are independent of the PTYs in other devpts instances.
 450  */
 451 static struct dentry *devpts_mount(struct file_system_type *fs_type,
 452         int flags, const char *dev_name, void *data)
 453 {
 454         return mount_nodev(fs_type, flags, data, devpts_fill_super);
 455 }
 456
 457 static void devpts_kill_sb(struct super_block *sb)
 458 {
 459         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 460
 461         ida_destroy(&fsi->allocated_ptys);
 462         kfree(fsi);
 463         kill_litter_super(sb);
 464 }
 465
 466 static struct file_system_type devpts_fs_type = {
 467         .name           = "devpts",
 468         .mount          = devpts_mount,
 469         .kill_sb        = devpts_kill_sb,
 470         .fs_flags       = FS_USERNS_MOUNT,
 471 };
 472
 473 /*
 474  * The normal naming convention is simply /dev/pts/<number>; this conforms
 475  * to the System V naming convention
 476  */
 477
 478 int devpts_new_index(struct pts_fs_info *fsi)
 479 {
 480         int index;
 481         int ida_ret;
 482
 483 retry:
 484         if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
 485                 return -ENOMEM;
 486
 487         mutex_lock(&allocated_ptys_lock);
 488         if (pty_count >= (pty_limit -
 489                           (fsi->mount_opts.reserve ? 0 : pty_reserve))) {
 490                 mutex_unlock(&allocated_ptys_lock);
 491                 return -ENOSPC;
 492         }
 493
 494         ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 495         if (ida_ret < 0) {
 496                 mutex_unlock(&allocated_ptys_lock);
 497                 if (ida_ret == -EAGAIN)
 498                         goto retry;
 499                 return -EIO;
 500         }
 501
 502         if (index >= fsi->mount_opts.max) {
 503                 ida_remove(&fsi->allocated_ptys, index);
 504                 mutex_unlock(&allocated_ptys_lock);
 505                 return -ENOSPC;
 506         }
 507         pty_count++;
 508         mutex_unlock(&allocated_ptys_lock);
 509         return index;
 510 }
 511
 512 void devpts_kill_index(struct pts_fs_info *fsi, int idx)
 513 {
 514         mutex_lock(&allocated_ptys_lock);
 515         ida_remove(&fsi->allocated_ptys, idx);
 516         pty_count--;
 517         mutex_unlock(&allocated_ptys_lock);
 518 }
 519
 520 /**
 521  * devpts_pty_new -- create a new inode in /dev/pts/
 522  * @ptmx_inode: inode of the master
 523  * @device: major+minor of the node to be created
 524  * @index: used as a name of the node
 525  * @priv: what's given back by devpts_get_priv
 526  *
 527  * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
 528  */
 529 struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
 530 {
 531         struct dentry *dentry;
 532         struct super_block *sb = fsi->sb;
 533         struct inode *inode;
 534         struct dentry *root;
 535         struct pts_mount_opts *opts;
 536         char s[12];
 537
 538         root = sb->s_root;
 539         opts = &fsi->mount_opts;
 540
 541         inode = new_inode(sb);
 542         if (!inode)
 543                 return ERR_PTR(-ENOMEM);
 544
 545         inode->i_ino = index + 3;
 546         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 547         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 548         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 549         init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
 550
 551         sprintf(s, "%d", index);
 552
 553         dentry = d_alloc_name(root, s);
 554         if (dentry) {
 555                 dentry->d_fsdata = priv;
 556                 d_add(dentry, inode);
 557                 fsnotify_create(d_inode(root), dentry);
 558         } else {
 559                 iput(inode);
 560                 dentry = ERR_PTR(-ENOMEM);
 561         }
 562
 563         return dentry;
 564 }
 565
 566 /**
 567  * devpts_get_priv -- get private data for a slave
 568  * @pts_inode: inode of the slave
 569  *
 570  * Returns whatever was passed as priv in devpts_pty_new for a given inode.
 571  */
 572 void *devpts_get_priv(struct dentry *dentry)
 573 {
 574         if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
 575                 return NULL;
 576         return dentry->d_fsdata;
 577 }
 578
 579 /**
 580  * devpts_pty_kill -- remove inode form /dev/pts/
 581  * @inode: inode of the slave to be removed
 582  *
 583  * This is an inverse operation of devpts_pty_new.
 584  */
 585 void devpts_pty_kill(struct dentry *dentry)
 586 {
 587         WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
 588
 589         dentry->d_fsdata = NULL;
 590         drop_nlink(dentry->d_inode);
 591         d_delete(dentry);
 592         dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 593 }
 594
 595 static int __init init_devpts_fs(void)
 596 {
 597         int err = register_filesystem(&devpts_fs_type);
 598         if (!err) {
 599                 register_sysctl_table(pty_root_table);
 600         }
 601         return err;
 602 }
 603 module_init(init_devpts_fs)