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