fs/devpts/inode.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- linux-c -*- --------------------------------------------------------- *
   3  *
   4  * linux/fs/devpts/inode.c
   5  *
   6  *  Copyright 1998-2004 H. Peter Anvin -- All Rights Reserved
   7  *
   8  * ------------------------------------------------------------------------- */
   9
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/module.h>
  13 #include <linux/init.h>
  14 #include <linux/fs.h>
  15 #include <linux/sched.h>
  16 #include <linux/namei.h>
  17 #include <linux/slab.h>
  18 #include <linux/mount.h>
  19 #include <linux/tty.h>
  20 #include <linux/mutex.h>
  21 #include <linux/magic.h>
  22 #include <linux/idr.h>
  23 #include <linux/devpts_fs.h>
  24 #include <linux/parser.h>
  25 #include <linux/fsnotify.h>
  26 #include <linux/seq_file.h>
  27
  28 #define DEVPTS_DEFAULT_MODE 0600
  29 /*
  30  * ptmx is a new node in /dev/pts and will be unused in legacy (single-
  31  * instance) mode. To prevent surprises in user space, set permissions of
  32  * ptmx to 0. Use 'chmod' or remount with '-o ptmxmode' to set meaningful
  33  * permissions.
  34  */
  35 #define DEVPTS_DEFAULT_PTMX_MODE 0000
  36 #define PTMX_MINOR      2
  37
  38 /*
  39  * sysctl support for setting limits on the number of Unix98 ptys allocated.
  40  * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
  41  */
  42 static int pty_limit = NR_UNIX98_PTY_DEFAULT;
  43 static int pty_reserve = NR_UNIX98_PTY_RESERVE;
  44 static int pty_limit_min;
  45 static int pty_limit_max = INT_MAX;
  46 static atomic_t pty_count = ATOMIC_INIT(0);
  47
  48 static struct ctl_table pty_table[] = {
  49         {
  50                 .procname       = "max",
  51                 .maxlen         = sizeof(int),
  52                 .mode           = 0644,
  53                 .data           = &pty_limit,
  54                 .proc_handler   = proc_dointvec_minmax,
  55                 .extra1         = &pty_limit_min,
  56                 .extra2         = &pty_limit_max,
  57         }, {
  58                 .procname       = "reserve",
  59                 .maxlen         = sizeof(int),
  60                 .mode           = 0644,
  61                 .data           = &pty_reserve,
  62                 .proc_handler   = proc_dointvec_minmax,
  63                 .extra1         = &pty_limit_min,
  64                 .extra2         = &pty_limit_max,
  65         }, {
  66                 .procname       = "nr",
  67                 .maxlen         = sizeof(int),
  68                 .mode           = 0444,
  69                 .data           = &pty_count,
  70                 .proc_handler   = proc_dointvec,
  71         },
  72         {}
  73 };
  74
  75 static struct ctl_table pty_kern_table[] = {
  76         {
  77                 .procname       = "pty",
  78                 .mode           = 0555,
  79                 .child          = pty_table,
  80         },
  81         {}
  82 };
  83
  84 static struct ctl_table pty_root_table[] = {
  85         {
  86                 .procname       = "kernel",
  87                 .mode           = 0555,
  88                 .child          = pty_kern_table,
  89         },
  90         {}
  91 };
  92
  93 struct pts_mount_opts {
  94         int setuid;
  95         int setgid;
  96         kuid_t   uid;
  97         kgid_t   gid;
  98         umode_t mode;
  99         umode_t ptmxmode;
 100         int reserve;
 101         int max;
 102 };
 103
 104 enum {
 105         Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,  Opt_max,
 106         Opt_err
 107 };
 108
 109 static const match_table_t tokens = {
 110         {Opt_uid, "uid=%u"},
 111         {Opt_gid, "gid=%u"},
 112         {Opt_mode, "mode=%o"},
 113         {Opt_ptmxmode, "ptmxmode=%o"},
 114         {Opt_newinstance, "newinstance"},
 115         {Opt_max, "max=%d"},
 116         {Opt_err, NULL}
 117 };
 118
 119 struct pts_fs_info {
 120         struct ida allocated_ptys;
 121         struct pts_mount_opts mount_opts;
 122         struct super_block *sb;
 123         struct dentry *ptmx_dentry;
 124 };
 125
 126 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
 127 {
 128         return sb->s_fs_info;
 129 }
 130
 131 static int devpts_ptmx_path(struct path *path)
 132 {
 133         struct super_block *sb;
 134         int err;
 135
 136         /* Is a devpts filesystem at "pts" in the same directory? */
 137         err = path_pts(path);
 138         if (err)
 139                 return err;
 140
 141         /* Is the path the root of a devpts filesystem? */
 142         sb = path->mnt->mnt_sb;
 143         if ((sb->s_magic != DEVPTS_SUPER_MAGIC) ||
 144             (path->mnt->mnt_root != sb->s_root))
 145                 return -ENODEV;
 146
 147         return 0;
 148 }
 149
 150 /*
 151  * Try to find a suitable devpts filesystem. We support the following
 152  * scenarios:
 153  * - The ptmx device node is located in the same directory as the devpts
 154  *   mount where the pts device nodes are located.
 155  *   This is e.g. the case when calling open on the /dev/pts/ptmx device
 156  *   node when the devpts filesystem is mounted at /dev/pts.
 157  * - The ptmx device node is located outside the devpts filesystem mount
 158  *   where the pts device nodes are located. For example, the ptmx device
 159  *   is a symlink, separate device node, or bind-mount.
 160  *   A supported scenario is bind-mounting /dev/pts/ptmx to /dev/ptmx and
 161  *   then calling open on /dev/ptmx. In this case a suitable pts
 162  *   subdirectory can be found in the common parent directory /dev of the
 163  *   devpts mount and the ptmx bind-mount, after resolving the /dev/ptmx
 164  *   bind-mount.
 165  *   If no suitable pts subdirectory can be found this function will fail.
 166  *   This is e.g. the case when bind-mounting /dev/pts/ptmx to /ptmx.
 167  */
 168 struct vfsmount *devpts_mntget(struct file *filp, struct pts_fs_info *fsi)
 169 {
 170         struct path path;
 171         int err = 0;
 172
 173         path = filp->f_path;
 174         path_get(&path);
 175
 176         /* Walk upward while the start point is a bind mount of
 177          * a single file.
 178          */
 179         while (path.mnt->mnt_root == path.dentry)
 180                 if (follow_up(&path) == 0)
 181                         break;
 182
 183         /* devpts_ptmx_path() finds a devpts fs or returns an error. */
 184         if ((path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) ||
 185             (DEVPTS_SB(path.mnt->mnt_sb) != fsi))
 186                 err = devpts_ptmx_path(&path);
 187         dput(path.dentry);
 188         if (!err) {
 189                 if (DEVPTS_SB(path.mnt->mnt_sb) == fsi)
 190                         return path.mnt;
 191
 192                 err = -ENODEV;
 193         }
 194
 195         mntput(path.mnt);
 196         return ERR_PTR(err);
 197 }
 198
 199 struct pts_fs_info *devpts_acquire(struct file *filp)
 200 {
 201         struct pts_fs_info *result;
 202         struct path path;
 203         struct super_block *sb;
 204
 205         path = filp->f_path;
 206         path_get(&path);
 207
 208         /* Has the devpts filesystem already been found? */
 209         if (path.mnt->mnt_sb->s_magic != DEVPTS_SUPER_MAGIC) {
 210                 int err;
 211
 212                 err = devpts_ptmx_path(&path);
 213                 if (err) {
 214                         result = ERR_PTR(err);
 215                         goto out;
 216                 }
 217         }
 218
 219         /*
 220          * pty code needs to hold extra references in case of last /dev/tty close
 221          */
 222         sb = path.mnt->mnt_sb;
 223         atomic_inc(&sb->s_active);
 224         result = DEVPTS_SB(sb);
 225
 226 out:
 227         path_put(&path);
 228         return result;
 229 }
 230
 231 void devpts_release(struct pts_fs_info *fsi)
 232 {
 233         deactivate_super(fsi->sb);
 234 }
 235
 236 #define PARSE_MOUNT     0
 237 #define PARSE_REMOUNT   1
 238
 239 /*
 240  * parse_mount_options():
 241  *      Set @opts to mount options specified in @data. If an option is not
 242  *      specified in @data, set it to its default value.
 243  *
 244  * Note: @data may be NULL (in which case all options are set to default).
 245  */
 246 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 247 {
 248         char *p;
 249         kuid_t uid;
 250         kgid_t gid;
 251
 252         opts->setuid  = 0;
 253         opts->setgid  = 0;
 254         opts->uid     = GLOBAL_ROOT_UID;
 255         opts->gid     = GLOBAL_ROOT_GID;
 256         opts->mode    = DEVPTS_DEFAULT_MODE;
 257         opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 258         opts->max     = NR_UNIX98_PTY_MAX;
 259
 260         /* Only allow instances mounted from the initial mount
 261          * namespace to tap the reserve pool of ptys.
 262          */
 263         if (op == PARSE_MOUNT)
 264                 opts->reserve =
 265                         (current->nsproxy->mnt_ns == init_task.nsproxy->mnt_ns);
 266
 267         while ((p = strsep(&data, ",")) != NULL) {
 268                 substring_t args[MAX_OPT_ARGS];
 269                 int token;
 270                 int option;
 271
 272                 if (!*p)
 273                         continue;
 274
 275                 token = match_token(p, tokens, args);
 276                 switch (token) {
 277                 case Opt_uid:
 278                         if (match_int(&args[0], &option))
 279                                 return -EINVAL;
 280                         uid = make_kuid(current_user_ns(), option);
 281                         if (!uid_valid(uid))
 282                                 return -EINVAL;
 283                         opts->uid = uid;
 284                         opts->setuid = 1;
 285                         break;
 286                 case Opt_gid:
 287                         if (match_int(&args[0], &option))
 288                                 return -EINVAL;
 289                         gid = make_kgid(current_user_ns(), option);
 290                         if (!gid_valid(gid))
 291                                 return -EINVAL;
 292                         opts->gid = gid;
 293                         opts->setgid = 1;
 294                         break;
 295                 case Opt_mode:
 296                         if (match_octal(&args[0], &option))
 297                                 return -EINVAL;
 298                         opts->mode = option & S_IALLUGO;
 299                         break;
 300                 case Opt_ptmxmode:
 301                         if (match_octal(&args[0], &option))
 302                                 return -EINVAL;
 303                         opts->ptmxmode = option & S_IALLUGO;
 304                         break;
 305                 case Opt_newinstance:
 306                         break;
 307                 case Opt_max:
 308                         if (match_int(&args[0], &option) ||
 309                             option < 0 || option > NR_UNIX98_PTY_MAX)
 310                                 return -EINVAL;
 311                         opts->max = option;
 312                         break;
 313                 default:
 314                         pr_err("called with bogus options\n");
 315                         return -EINVAL;
 316                 }
 317         }
 318
 319         return 0;
 320 }
 321
 322 static int mknod_ptmx(struct super_block *sb)
 323 {
 324         int mode;
 325         int rc = -ENOMEM;
 326         struct dentry *dentry;
 327         struct inode *inode;
 328         struct dentry *root = sb->s_root;
 329         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 330         struct pts_mount_opts *opts = &fsi->mount_opts;
 331         kuid_t ptmx_uid = current_fsuid();
 332         kgid_t ptmx_gid = current_fsgid();
 333
 334         inode_lock(d_inode(root));
 335
 336         /* If we have already created ptmx node, return */
 337         if (fsi->ptmx_dentry) {
 338                 rc = 0;
 339                 goto out;
 340         }
 341
 342         dentry = d_alloc_name(root, "ptmx");
 343         if (!dentry) {
 344                 pr_err("Unable to alloc dentry for ptmx node\n");
 345                 goto out;
 346         }
 347
 348         /*
 349          * Create a new 'ptmx' node in this mount of devpts.
 350          */
 351         inode = new_inode(sb);
 352         if (!inode) {
 353                 pr_err("Unable to alloc inode for ptmx node\n");
 354                 dput(dentry);
 355                 goto out;
 356         }
 357
 358         inode->i_ino = 2;
 359         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 360
 361         mode = S_IFCHR|opts->ptmxmode;
 362         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 363         inode->i_uid = ptmx_uid;
 364         inode->i_gid = ptmx_gid;
 365
 366         d_add(dentry, inode);
 367
 368         fsi->ptmx_dentry = dentry;
 369         rc = 0;
 370 out:
 371         inode_unlock(d_inode(root));
 372         return rc;
 373 }
 374
 375 static void update_ptmx_mode(struct pts_fs_info *fsi)
 376 {
 377         struct inode *inode;
 378         if (fsi->ptmx_dentry) {
 379                 inode = d_inode(fsi->ptmx_dentry);
 380                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 381         }
 382 }
 383
 384 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 385 {
 386         int err;
 387         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 388         struct pts_mount_opts *opts = &fsi->mount_opts;
 389
 390         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 391
 392         /*
 393          * parse_mount_options() restores options to default values
 394          * before parsing and may have changed ptmxmode. So, update the
 395          * mode in the inode too. Bogus options don't fail the remount,
 396          * so do this even on error return.
 397          */
 398         update_ptmx_mode(fsi);
 399
 400         return err;
 401 }
 402
 403 static int devpts_show_options(struct seq_file *seq, struct dentry *root)
 404 {
 405         struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb);
 406         struct pts_mount_opts *opts = &fsi->mount_opts;
 407
 408         if (opts->setuid)
 409                 seq_printf(seq, ",uid=%u",
 410                            from_kuid_munged(&init_user_ns, opts->uid));
 411         if (opts->setgid)
 412                 seq_printf(seq, ",gid=%u",
 413                            from_kgid_munged(&init_user_ns, opts->gid));
 414         seq_printf(seq, ",mode=%03o", opts->mode);
 415         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 416         if (opts->max < NR_UNIX98_PTY_MAX)
 417                 seq_printf(seq, ",max=%d", opts->max);
 418
 419         return 0;
 420 }
 421
 422 static const struct super_operations devpts_sops = {
 423         .statfs         = simple_statfs,
 424         .remount_fs     = devpts_remount,
 425         .show_options   = devpts_show_options,
 426 };
 427
 428 static void *new_pts_fs_info(struct super_block *sb)
 429 {
 430         struct pts_fs_info *fsi;
 431
 432         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 433         if (!fsi)
 434                 return NULL;
 435
 436         ida_init(&fsi->allocated_ptys);
 437         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 438         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 439         fsi->sb = sb;
 440
 441         return fsi;
 442 }
 443
 444 static int
 445 devpts_fill_super(struct super_block *s, void *data, int silent)
 446 {
 447         struct inode *inode;
 448         int error;
 449
 450         s->s_iflags &= ~SB_I_NODEV;
 451         s->s_blocksize = 1024;
 452         s->s_blocksize_bits = 10;
 453         s->s_magic = DEVPTS_SUPER_MAGIC;
 454         s->s_op = &devpts_sops;
 455         s->s_d_op = &simple_dentry_operations;
 456         s->s_time_gran = 1;
 457
 458         error = -ENOMEM;
 459         s->s_fs_info = new_pts_fs_info(s);
 460         if (!s->s_fs_info)
 461                 goto fail;
 462
 463         error = parse_mount_options(data, PARSE_MOUNT, &DEVPTS_SB(s)->mount_opts);
 464         if (error)
 465                 goto fail;
 466
 467         error = -ENOMEM;
 468         inode = new_inode(s);
 469         if (!inode)
 470                 goto fail;
 471         inode->i_ino = 1;
 472         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 473         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 474         inode->i_op = &simple_dir_inode_operations;
 475         inode->i_fop = &simple_dir_operations;
 476         set_nlink(inode, 2);
 477
 478         s->s_root = d_make_root(inode);
 479         if (!s->s_root) {
 480                 pr_err("get root dentry failed\n");
 481                 goto fail;
 482         }
 483
 484         error = mknod_ptmx(s);
 485         if (error)
 486                 goto fail_dput;
 487
 488         return 0;
 489 fail_dput:
 490         dput(s->s_root);
 491         s->s_root = NULL;
 492 fail:
 493         return error;
 494 }
 495
 496 /*
 497  * devpts_mount()
 498  *
 499  *     Mount a new (private) instance of devpts.  PTYs created in this
 500  *     instance are independent of the PTYs in other devpts instances.
 501  */
 502 static struct dentry *devpts_mount(struct file_system_type *fs_type,
 503         int flags, const char *dev_name, void *data)
 504 {
 505         return mount_nodev(fs_type, flags, data, devpts_fill_super);
 506 }
 507
 508 static void devpts_kill_sb(struct super_block *sb)
 509 {
 510         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 511
 512         if (fsi)
 513                 ida_destroy(&fsi->allocated_ptys);
 514         kfree(fsi);
 515         kill_litter_super(sb);
 516 }
 517
 518 static struct file_system_type devpts_fs_type = {
 519         .name           = "devpts",
 520         .mount          = devpts_mount,
 521         .kill_sb        = devpts_kill_sb,
 522         .fs_flags       = FS_USERNS_MOUNT,
 523 };
 524
 525 /*
 526  * The normal naming convention is simply /dev/pts/<number>; this conforms
 527  * to the System V naming convention
 528  */
 529
 530 int devpts_new_index(struct pts_fs_info *fsi)
 531 {
 532         int index = -ENOSPC;
 533
 534         if (atomic_inc_return(&pty_count) >= (pty_limit -
 535                           (fsi->mount_opts.reserve ? 0 : pty_reserve)))
 536                 goto out;
 537
 538         index = ida_alloc_max(&fsi->allocated_ptys, fsi->mount_opts.max - 1,
 539                         GFP_KERNEL);
 540
 541 out:
 542         if (index < 0)
 543                 atomic_dec(&pty_count);
 544         return index;
 545 }
 546
 547 void devpts_kill_index(struct pts_fs_info *fsi, int idx)
 548 {
 549         ida_free(&fsi->allocated_ptys, idx);
 550         atomic_dec(&pty_count);
 551 }
 552
 553 /**
 554  * devpts_pty_new -- create a new inode in /dev/pts/
 555  * @ptmx_inode: inode of the master
 556  * @device: major+minor of the node to be created
 557  * @index: used as a name of the node
 558  * @priv: what's given back by devpts_get_priv
 559  *
 560  * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
 561  */
 562 struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
 563 {
 564         struct dentry *dentry;
 565         struct super_block *sb = fsi->sb;
 566         struct inode *inode;
 567         struct dentry *root;
 568         struct pts_mount_opts *opts;
 569         char s[12];
 570
 571         root = sb->s_root;
 572         opts = &fsi->mount_opts;
 573
 574         inode = new_inode(sb);
 575         if (!inode)
 576                 return ERR_PTR(-ENOMEM);
 577
 578         inode->i_ino = index + 3;
 579         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 580         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 581         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 582         init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
 583
 584         sprintf(s, "%d", index);
 585
 586         dentry = d_alloc_name(root, s);
 587         if (dentry) {
 588                 dentry->d_fsdata = priv;
 589                 d_add(dentry, inode);
 590                 fsnotify_create(d_inode(root), dentry);
 591         } else {
 592                 iput(inode);
 593                 dentry = ERR_PTR(-ENOMEM);
 594         }
 595
 596         return dentry;
 597 }
 598
 599 /**
 600  * devpts_get_priv -- get private data for a slave
 601  * @pts_inode: inode of the slave
 602  *
 603  * Returns whatever was passed as priv in devpts_pty_new for a given inode.
 604  */
 605 void *devpts_get_priv(struct dentry *dentry)
 606 {
 607         if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
 608                 return NULL;
 609         return dentry->d_fsdata;
 610 }
 611
 612 /**
 613  * devpts_pty_kill -- remove inode form /dev/pts/
 614  * @inode: inode of the slave to be removed
 615  *
 616  * This is an inverse operation of devpts_pty_new.
 617  */
 618 void devpts_pty_kill(struct dentry *dentry)
 619 {
 620         WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
 621
 622         dentry->d_fsdata = NULL;
 623         drop_nlink(dentry->d_inode);
 624         fsnotify_unlink(d_inode(dentry->d_parent), dentry);
 625         d_drop(dentry);
 626         dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 627 }
 628
 629 static int __init init_devpts_fs(void)
 630 {
 631         int err = register_filesystem(&devpts_fs_type);
 632         if (!err) {
 633                 register_sysctl_table(pty_root_table);
 634         }
 635         return err;
 636 }
 637 module_init(init_devpts_fs)