linux/fs/devpts/inode.c
<<
>>
Prefs
   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 */
  45static int pty_limit = NR_UNIX98_PTY_DEFAULT;
  46static int pty_reserve = NR_UNIX98_PTY_RESERVE;
  47static int pty_limit_min;
  48static int pty_limit_max = INT_MAX;
  49static int pty_count;
  50
  51static 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
  78static struct ctl_table pty_kern_table[] = {
  79        {
  80                .procname       = "pty",
  81                .mode           = 0555,
  82                .child          = pty_table,
  83        },
  84        {}
  85};
  86
  87static struct ctl_table pty_root_table[] = {
  88        {
  89                .procname       = "kernel",
  90                .mode           = 0555,
  91                .child          = pty_kern_table,
  92        },
  93        {}
  94};
  95
  96static DEFINE_MUTEX(allocated_ptys_lock);
  97
  98struct 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
 109enum {
 110        Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,  Opt_max,
 111        Opt_err
 112};
 113
 114static 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
 124struct 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
 131static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
 132{
 133        return sb->s_fs_info;
 134}
 135
 136struct 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
 170out:
 171        path_put(&path);
 172        return result;
 173}
 174
 175void 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 */
 190static 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
 266static 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;
 319out:
 320        inode_unlock(d_inode(root));
 321        return rc;
 322}
 323
 324static 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
 333static 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
 353static 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
 372static const struct super_operations devpts_sops = {
 373        .statfs         = simple_statfs,
 374        .remount_fs     = devpts_remount,
 375        .show_options   = devpts_show_options,
 376};
 377
 378static 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
 394static int
 395devpts_fill_super(struct super_block *s, void *data, int silent)
 396{
 397        struct inode *inode;
 398
 399        s->s_iflags &= ~SB_I_NODEV;
 400        s->s_blocksize = 1024;
 401        s->s_blocksize_bits = 10;
 402        s->s_magic = DEVPTS_SUPER_MAGIC;
 403        s->s_op = &devpts_sops;
 404        s->s_time_gran = 1;
 405
 406        s->s_fs_info = new_pts_fs_info(s);
 407        if (!s->s_fs_info)
 408                goto fail;
 409
 410        inode = new_inode(s);
 411        if (!inode)
 412                goto fail;
 413        inode->i_ino = 1;
 414        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 415        inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 416        inode->i_op = &simple_dir_inode_operations;
 417        inode->i_fop = &simple_dir_operations;
 418        set_nlink(inode, 2);
 419
 420        s->s_root = d_make_root(inode);
 421        if (s->s_root)
 422                return 0;
 423
 424        pr_err("get root dentry failed\n");
 425
 426fail:
 427        return -ENOMEM;
 428}
 429
 430/*
 431 * devpts_mount()
 432 *
 433 *     Mount a new (private) instance of devpts.  PTYs created in this
 434 *     instance are independent of the PTYs in other devpts instances.
 435 */
 436static struct dentry *devpts_mount(struct file_system_type *fs_type,
 437        int flags, const char *dev_name, void *data)
 438{
 439        int error;
 440        struct pts_mount_opts opts;
 441        struct super_block *s;
 442
 443        error = parse_mount_options(data, PARSE_MOUNT, &opts);
 444        if (error)
 445                return ERR_PTR(error);
 446
 447        s = sget(fs_type, NULL, set_anon_super, flags, NULL);
 448        if (IS_ERR(s))
 449                return ERR_CAST(s);
 450
 451        if (!s->s_root) {
 452                error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 453                if (error)
 454                        goto out_undo_sget;
 455                s->s_flags |= MS_ACTIVE;
 456        }
 457
 458        memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
 459
 460        error = mknod_ptmx(s);
 461        if (error)
 462                goto out_undo_sget;
 463
 464        return dget(s->s_root);
 465
 466out_undo_sget:
 467        deactivate_locked_super(s);
 468        return ERR_PTR(error);
 469}
 470
 471static void devpts_kill_sb(struct super_block *sb)
 472{
 473        struct pts_fs_info *fsi = DEVPTS_SB(sb);
 474
 475        ida_destroy(&fsi->allocated_ptys);
 476        kfree(fsi);
 477        kill_litter_super(sb);
 478}
 479
 480static struct file_system_type devpts_fs_type = {
 481        .name           = "devpts",
 482        .mount          = devpts_mount,
 483        .kill_sb        = devpts_kill_sb,
 484        .fs_flags       = FS_USERNS_MOUNT,
 485};
 486
 487/*
 488 * The normal naming convention is simply /dev/pts/<number>; this conforms
 489 * to the System V naming convention
 490 */
 491
 492int devpts_new_index(struct pts_fs_info *fsi)
 493{
 494        int index;
 495        int ida_ret;
 496
 497retry:
 498        if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
 499                return -ENOMEM;
 500
 501        mutex_lock(&allocated_ptys_lock);
 502        if (pty_count >= (pty_limit -
 503                          (fsi->mount_opts.reserve ? 0 : pty_reserve))) {
 504                mutex_unlock(&allocated_ptys_lock);
 505                return -ENOSPC;
 506        }
 507
 508        ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 509        if (ida_ret < 0) {
 510                mutex_unlock(&allocated_ptys_lock);
 511                if (ida_ret == -EAGAIN)
 512                        goto retry;
 513                return -EIO;
 514        }
 515
 516        if (index >= fsi->mount_opts.max) {
 517                ida_remove(&fsi->allocated_ptys, index);
 518                mutex_unlock(&allocated_ptys_lock);
 519                return -ENOSPC;
 520        }
 521        pty_count++;
 522        mutex_unlock(&allocated_ptys_lock);
 523        return index;
 524}
 525
 526void devpts_kill_index(struct pts_fs_info *fsi, int idx)
 527{
 528        mutex_lock(&allocated_ptys_lock);
 529        ida_remove(&fsi->allocated_ptys, idx);
 530        pty_count--;
 531        mutex_unlock(&allocated_ptys_lock);
 532}
 533
 534/**
 535 * devpts_pty_new -- create a new inode in /dev/pts/
 536 * @ptmx_inode: inode of the master
 537 * @device: major+minor of the node to be created
 538 * @index: used as a name of the node
 539 * @priv: what's given back by devpts_get_priv
 540 *
 541 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill.
 542 */
 543struct dentry *devpts_pty_new(struct pts_fs_info *fsi, int index, void *priv)
 544{
 545        struct dentry *dentry;
 546        struct super_block *sb = fsi->sb;
 547        struct inode *inode;
 548        struct dentry *root;
 549        struct pts_mount_opts *opts;
 550        char s[12];
 551
 552        root = sb->s_root;
 553        opts = &fsi->mount_opts;
 554
 555        inode = new_inode(sb);
 556        if (!inode)
 557                return ERR_PTR(-ENOMEM);
 558
 559        inode->i_ino = index + 3;
 560        inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 561        inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 562        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 563        init_special_inode(inode, S_IFCHR|opts->mode, MKDEV(UNIX98_PTY_SLAVE_MAJOR, index));
 564
 565        sprintf(s, "%d", index);
 566
 567        dentry = d_alloc_name(root, s);
 568        if (dentry) {
 569                dentry->d_fsdata = priv;
 570                d_add(dentry, inode);
 571                fsnotify_create(d_inode(root), dentry);
 572        } else {
 573                iput(inode);
 574                dentry = ERR_PTR(-ENOMEM);
 575        }
 576
 577        return dentry;
 578}
 579
 580/**
 581 * devpts_get_priv -- get private data for a slave
 582 * @pts_inode: inode of the slave
 583 *
 584 * Returns whatever was passed as priv in devpts_pty_new for a given inode.
 585 */
 586void *devpts_get_priv(struct dentry *dentry)
 587{
 588        if (dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC)
 589                return NULL;
 590        return dentry->d_fsdata;
 591}
 592
 593/**
 594 * devpts_pty_kill -- remove inode form /dev/pts/
 595 * @inode: inode of the slave to be removed
 596 *
 597 * This is an inverse operation of devpts_pty_new.
 598 */
 599void devpts_pty_kill(struct dentry *dentry)
 600{
 601        WARN_ON_ONCE(dentry->d_sb->s_magic != DEVPTS_SUPER_MAGIC);
 602
 603        dentry->d_fsdata = NULL;
 604        drop_nlink(dentry->d_inode);
 605        d_delete(dentry);
 606        dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 607}
 608
 609static int __init init_devpts_fs(void)
 610{
 611        int err = register_filesystem(&devpts_fs_type);
 612        if (!err) {
 613                register_sysctl_table(pty_root_table);
 614        }
 615        return err;
 616}
 617module_init(init_devpts_fs)
 618