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         /* Has the devpts filesystem already been found? */
 142         if (path->mnt->mnt_sb->s_magic == DEVPTS_SUPER_MAGIC)
 143                 return 0;
 144
 145         /* Is a devpts filesystem at "pts" in the same directory? */
 146         err = path_pts(path);
 147         if (err)
 148                 return err;
 149
 150         /* Is the path the root of a devpts filesystem? */
 151         sb = path->mnt->mnt_sb;
 152         if ((sb->s_magic != DEVPTS_SUPER_MAGIC) ||
 153             (path->mnt->mnt_root != sb->s_root))
 154                 return -ENODEV;
 155
 156         return 0;
 157 }
 158
 159 struct vfsmount *devpts_mntget(struct file *filp, struct pts_fs_info *fsi)
 160 {
 161         struct path path;
 162         int err;
 163
 164         path = filp->f_path;
 165         path_get(&path);
 166
 167         err = devpts_ptmx_path(&path);
 168         dput(path.dentry);
 169         if (err) {
 170                 mntput(path.mnt);
 171                 return ERR_PTR(err);
 172         }
 173         if (DEVPTS_SB(path.mnt->mnt_sb) != fsi) {
 174                 mntput(path.mnt);
 175                 return ERR_PTR(-ENODEV);
 176         }
 177         return path.mnt;
 178 }
 179
 180 struct pts_fs_info *devpts_acquire(struct file *filp)
 181 {
 182         struct pts_fs_info *result;
 183         struct path path;
 184         struct super_block *sb;
 185         int err;
 186
 187         path = filp->f_path;
 188         path_get(&path);
 189
 190         err = devpts_ptmx_path(&path);
 191         if (err) {
 192                 result = ERR_PTR(err);
 193                 goto out;
 194         }
 195
 196         /*
 197          * pty code needs to hold extra references in case of last /dev/tty close
 198          */
 199         sb = path.mnt->mnt_sb;
 200         atomic_inc(&sb->s_active);
 201         result = DEVPTS_SB(sb);
 202
 203 out:
 204         path_put(&path);
 205         return result;
 206 }
 207
 208 void devpts_release(struct pts_fs_info *fsi)
 209 {
 210         deactivate_super(fsi->sb);
 211 }
 212
 213 #define PARSE_MOUNT     0
 214 #define PARSE_REMOUNT   1
 215
 216 /*
 217  * parse_mount_options():
 218  *      Set @opts to mount options specified in @data. If an option is not
 219  *      specified in @data, set it to its default value.
 220  *
 221  * Note: @data may be NULL (in which case all options are set to default).
 222  */
 223 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 224 {
 225         char *p;
 226         kuid_t uid;
 227         kgid_t gid;
 228
 229         opts->setuid  = 0;
 230         opts->setgid  = 0;
 231         opts->uid     = GLOBAL_ROOT_UID;
 232         opts->gid     = GLOBAL_ROOT_GID;
 233         opts->mode    = DEVPTS_DEFAULT_MODE;
 234         opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 235         opts->max     = NR_UNIX98_PTY_MAX;
 236
 237         /* Only allow instances mounted from the initial mount
 238          * namespace to tap the reserve pool of ptys.
 239          */
 240         if (op == PARSE_MOUNT)
 241                 opts->reserve =
 242                         (current->nsproxy->mnt_ns == init_task.nsproxy->mnt_ns);
 243
 244         while ((p = strsep(&data, ",")) != NULL) {
 245                 substring_t args[MAX_OPT_ARGS];
 246                 int token;
 247                 int option;
 248
 249                 if (!*p)
 250                         continue;
 251
 252                 token = match_token(p, tokens, args);
 253                 switch (token) {
 254                 case Opt_uid:
 255                         if (match_int(&args[0], &option))
 256                                 return -EINVAL;
 257                         uid = make_kuid(current_user_ns(), option);
 258                         if (!uid_valid(uid))
 259                                 return -EINVAL;
 260                         opts->uid = uid;
 261                         opts->setuid = 1;
 262                         break;
 263                 case Opt_gid:
 264                         if (match_int(&args[0], &option))
 265                                 return -EINVAL;
 266                         gid = make_kgid(current_user_ns(), option);
 267                         if (!gid_valid(gid))
 268                                 return -EINVAL;
 269                         opts->gid = gid;
 270                         opts->setgid = 1;
 271                         break;
 272                 case Opt_mode:
 273                         if (match_octal(&args[0], &option))
 274                                 return -EINVAL;
 275                         opts->mode = option & S_IALLUGO;
 276                         break;
 277                 case Opt_ptmxmode:
 278                         if (match_octal(&args[0], &option))
 279                                 return -EINVAL;
 280                         opts->ptmxmode = option & S_IALLUGO;
 281                         break;
 282                 case Opt_newinstance:
 283                         break;
 284                 case Opt_max:
 285                         if (match_int(&args[0], &option) ||
 286                             option < 0 || option > NR_UNIX98_PTY_MAX)
 287                                 return -EINVAL;
 288                         opts->max = option;
 289                         break;
 290                 default:
 291                         pr_err("called with bogus options\n");
 292                         return -EINVAL;
 293                 }
 294         }
 295
 296         return 0;
 297 }
 298
 299 static int mknod_ptmx(struct super_block *sb)
 300 {
 301         int mode;
 302         int rc = -ENOMEM;
 303         struct dentry *dentry;
 304         struct inode *inode;
 305         struct dentry *root = sb->s_root;
 306         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 307         struct pts_mount_opts *opts = &fsi->mount_opts;
 308         kuid_t ptmx_uid = current_fsuid();
 309         kgid_t ptmx_gid = current_fsgid();
 310
 311         inode_lock(d_inode(root));
 312
 313         /* If we have already created ptmx node, return */
 314         if (fsi->ptmx_dentry) {
 315                 rc = 0;
 316                 goto out;
 317         }
 318
 319         dentry = d_alloc_name(root, "ptmx");
 320         if (!dentry) {
 321                 pr_err("Unable to alloc dentry for ptmx node\n");
 322                 goto out;
 323         }
 324
 325         /*
 326          * Create a new 'ptmx' node in this mount of devpts.
 327          */
 328         inode = new_inode(sb);
 329         if (!inode) {
 330                 pr_err("Unable to alloc inode for ptmx node\n");
 331                 dput(dentry);
 332                 goto out;
 333         }
 334
 335         inode->i_ino = 2;
 336         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 337
 338         mode = S_IFCHR|opts->ptmxmode;
 339         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 340         inode->i_uid = ptmx_uid;
 341         inode->i_gid = ptmx_gid;
 342
 343         d_add(dentry, inode);
 344
 345         fsi->ptmx_dentry = dentry;
 346         rc = 0;
 347 out:
 348         inode_unlock(d_inode(root));
 349         return rc;
 350 }
 351
 352 static void update_ptmx_mode(struct pts_fs_info *fsi)
 353 {
 354         struct inode *inode;
 355         if (fsi->ptmx_dentry) {
 356                 inode = d_inode(fsi->ptmx_dentry);
 357                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 358         }
 359 }
 360
 361 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 362 {
 363         int err;
 364         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 365         struct pts_mount_opts *opts = &fsi->mount_opts;
 366
 367         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 368
 369         /*
 370          * parse_mount_options() restores options to default values
 371          * before parsing and may have changed ptmxmode. So, update the
 372          * mode in the inode too. Bogus options don't fail the remount,
 373          * so do this even on error return.
 374          */
 375         update_ptmx_mode(fsi);
 376
 377         return err;
 378 }
 379
 380 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
 381 {
 382         struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
 383         struct pts_mount_opts *opts = &fsi->mount_opts;
 384
 385         if (opts->setuid)
 386                 seq_printf(seq, ",uid=%u",
 387                            from_kuid_munged(&init_user_ns, opts->uid));
 388         if (opts->setgid)
 389                 seq_printf(seq, ",gid=%u",
 390                            from_kgid_munged(&init_user_ns, opts->gid));
 391         seq_printf(seq, ",mode=%03o", opts->mode);
 392         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 393         if (opts->max < NR_UNIX98_PTY_MAX)
 394                 seq_printf(seq, ",max=%d", opts->max);
 395
 396         return 0;
 397 }
 398
 399 static const struct super_operations devpts_sops = {
 400         .statfs         = simple_statfs,
 401         .remount_fs     = devpts_remount,
 402         .show_options   = devpts_show_options,
 403 };
 404
 405 static void *new_pts_fs_info(struct super_block *sb)
 406 {
 407         struct pts_fs_info *fsi;
 408
 409         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 410         if (!fsi)
 411                 return NULL;
 412
 413         ida_init(&fsi->allocated_ptys);
 414         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 415         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 416         fsi->sb = sb;
 417
 418         return fsi;
 419 }
 420
 421 static int
 422 devpts_fill_super(struct super_block *s, void *data, int silent)
 423 {
 424         struct inode *inode;
 425         int error;
 426
 427         s->s_iflags &= ~SB_I_NODEV;
 428         s->s_blocksize = 1024;
 429         s->s_blocksize_bits = 10;
 430         s->s_magic = DEVPTS_SUPER_MAGIC;
 431         s->s_op = &devpts_sops;
 432         s->s_time_gran = 1;
 433
 434         error = -ENOMEM;
 435         s->s_fs_info = new_pts_fs_info(s);
 436         if (!s->s_fs_info)
 437                 goto fail;
 438
 439         error = parse_mount_options(data, PARSE_MOUNT, &DEVPTS_SB(s)->mount_opts);
 440         if (error)
 441                 goto fail;
 442
 443         error = -ENOMEM;
 444         inode = new_inode(s);
 445         if (!inode)
 446                 goto fail;
 447         inode->i_ino = 1;
 448         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 449         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 450         inode->i_op = &simple_dir_inode_operations;
 451         inode->i_fop = &simple_dir_operations;
 452         set_nlink(inode, 2);
 453
 454         s->s_root = d_make_root(inode);
 455         if (!s->s_root) {
 456                 pr_err("get root dentry failed\n");
 457                 goto fail;
 458         }
 459
 460         error = mknod_ptmx(s);
 461         if (error)
 462                 goto fail_dput;
 463
 464         return 0;
 465 fail_dput:
 466         dput(s->s_root);
 467         s->s_root = NULL;
 468 fail:
 469         return error;
 470 }
 471
 472 /*
 473  * devpts_mount()
 474  *
 475  *     Mount a new (private) instance of devpts.  PTYs created in this
 476  *     instance are independent of the PTYs in other devpts instances.
 477  */
 478 static struct dentry *devpts_mount(struct file_system_type *fs_type,
 479         int flags, const char *dev_name, void *data)
 480 {
 481         return mount_nodev(fs_type, flags, data, devpts_fill_super);
 482 }
 483
 484 static void devpts_kill_sb(struct super_block *sb)
 485 {
 486         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 487
 488         if (fsi)
 489                 ida_destroy(&fsi->allocated_ptys);
 490         kfree(fsi);
 491         kill_litter_super(sb);
 492 }
 493
 494 static struct file_system_type devpts_fs_type = {
 495         .name           = "devpts",
 496         .mount          = devpts_mount,
 497         .kill_sb        = devpts_kill_sb,
 498         .fs_flags       = FS_USERNS_MOUNT,
 499 };
 500
 501 /*
 502  * The normal naming convention is simply /dev/pts/<number>; this conforms
 503  * to the System V naming convention
 504  */
 505
 506 int devpts_new_index(struct pts_fs_info *fsi)
 507 {
 508         int index;
 509         int ida_ret;
 510
 511 retry:
 512         if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
 513                 return -ENOMEM;
 514
 515         mutex_lock(&allocated_ptys_lock);
 516         if (pty_count >= (pty_limit -
 517                           (fsi->mount_opts.reserve ? 0 : pty_reserve))) {
 518                 mutex_unlock(&allocated_ptys_lock);
 519                 return -ENOSPC;
 520         }
 521
 522         ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 523         if (ida_ret < 0) {
 524                 mutex_unlock(&allocated_ptys_lock);
 525                 if (ida_ret == -EAGAIN)
 526                         goto retry;
 527                 return -EIO;
 528         }
 529
 530         if (index >= fsi->mount_opts.max) {
 531                 ida_remove(&fsi->allocated_ptys, index);
 532                 mutex_unlock(&allocated_ptys_lock);
 533                 return -ENOSPC;
 534         }
 535         pty_count++;
 536         mutex_unlock(&allocated_ptys_lock);
 537         return index;
 538 }
 539
 540 void devpts_kill_index(struct pts_fs_info *fsi, int idx)
 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 /**
 549  * devpts_pty_new -- create a new inode in /dev/pts/
 550  * @ptmx_inode: inode of the master
 551  * @device: major+minor of the node to be created
 552  * @index: used as a name of the node
 553  * @priv: what's given back by devpts_get_priv
 554  *
 555  * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
 556  */
 557 struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
 558 {
 559         struct dentry *dentry;
 560         struct super_block *sb = fsi->sb;
 561         struct inode *inode;
 562         struct dentry *root;
 563         struct pts_mount_opts *opts;
 564         char s[12];
 565
 566         root = sb->s_root;
 567         opts = &fsi->mount_opts;
 568
 569         inode = new_inode(sb);
 570         if (!inode)
 571                 return ERR_PTR(-ENOMEM);
 572
 573         inode->i_ino = index + 3;
 574         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 575         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 576         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 577         init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
 578
 579         sprintf(s, "%d", index);
 580
 581         dentry = d_alloc_name(root, s);
 582         if (dentry) {
 583                 dentry->d_fsdata = priv;
 584                 d_add(dentry, inode);
 585                 fsnotify_create(d_inode(root), dentry);
 586         } else {
 587                 iput(inode);
 588                 dentry = ERR_PTR(-ENOMEM);
 589         }
 590
 591         return dentry;
 592 }
 593
 594 /**
 595  * devpts_get_priv -- get private data for a slave
 596  * @pts_inode: inode of the slave
 597  *
 598  * Returns whatever was passed as priv in devpts_pty_new for a given inode.
 599  */
 600 void *devpts_get_priv(struct dentry *dentry)
 601 {
 602         if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
 603                 return NULL;
 604         return dentry->d_fsdata;
 605 }
 606
 607 /**
 608  * devpts_pty_kill -- remove inode form /dev/pts/
 609  * @inode: inode of the slave to be removed
 610  *
 611  * This is an inverse operation of devpts_pty_new.
 612  */
 613 void devpts_pty_kill(struct dentry *dentry)
 614 {
 615         WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
 616
 617         dentry->d_fsdata = NULL;
 618         drop_nlink(dentry->d_inode);
 619         d_delete(dentry);
 620         dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 621 }
 622
 623 static int __init init_devpts_fs(void)
 624 {
 625         int err = register_filesystem(&devpts_fs_type);
 626         if (!err) {
 627                 register_sysctl_table(pty_root_table);
 628         }
 629         return err;
 630 }
 631 module_init(init_devpts_fs)