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 #include <linux/module.h>
  14 #include <linux/init.h>
  15 #include <linux/fs.h>
  16 #include <linux/sched.h>
  17 #include <linux/namei.h>
  18 #include <linux/mount.h>
  19 #include <linux/tty.h>
  20 #include <linux/mutex.h>
  21 #include <linux/idr.h>
  22 #include <linux/devpts_fs.h>
  23 #include <linux/parser.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/seq_file.h>
  26
  27 #define DEVPTS_SUPER_MAGIC 0x1cd1
  28
  29 #define DEVPTS_DEFAULT_MODE 0600
  30 /*
  31  * ptmx is a new node in /dev/pts and will be unused in legacy (single-
  32  * instance) mode. To prevent surprises in user space, set permissions of
  33  * ptmx to 0. Use 'chmod' or remount with '-o ptmxmode' to set meaningful
  34  * permissions.
  35  */
  36 #define DEVPTS_DEFAULT_PTMX_MODE 0000
  37 #define PTMX_MINOR      2
  38
  39 extern int pty_limit;                   /* Config limit on Unix98 ptys */
  40 static DEFINE_MUTEX(allocated_ptys_lock);
  41
  42 static struct vfsmount *devpts_mnt;
  43
  44 struct pts_mount_opts {
  45         int setuid;
  46         int setgid;
  47         uid_t   uid;
  48         gid_t   gid;
  49         umode_t mode;
  50         umode_t ptmxmode;
  51         int newinstance;
  52 };
  53
  54 enum {
  55         Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,
  56         Opt_err
  57 };
  58
  59 static const match_table_t tokens = {
  60         {Opt_uid, "uid=%u"},
  61         {Opt_gid, "gid=%u"},
  62         {Opt_mode, "mode=%o"},
  63 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
  64         {Opt_ptmxmode, "ptmxmode=%o"},
  65         {Opt_newinstance, "newinstance"},
  66 #endif
  67         {Opt_err, NULL}
  68 };
  69
  70 struct pts_fs_info {
  71         struct ida allocated_ptys;
  72         struct pts_mount_opts mount_opts;
  73         struct dentry *ptmx_dentry;
  74 };
  75
  76 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
  77 {
  78         return sb->s_fs_info;
  79 }
  80
  81 static inline struct super_block *pts_sb_from_inode(struct inode *inode)
  82 {
  83 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
  84         if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
  85                 return inode->i_sb;
  86 #endif
  87         return devpts_mnt->mnt_sb;
  88 }
  89
  90 #define PARSE_MOUNT     0
  91 #define PARSE_REMOUNT   1
  92
  93 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
  94 {
  95         char *p;
  96
  97         opts->setuid  = 0;
  98         opts->setgid  = 0;
  99         opts->uid     = 0;
 100         opts->gid     = 0;
 101         opts->mode    = DEVPTS_DEFAULT_MODE;
 102         opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 103
 104         /* newinstance makes sense only on initial mount */
 105         if (op == PARSE_MOUNT)
 106                 opts->newinstance = 0;
 107
 108         while ((p = strsep(&data, ",")) != NULL) {
 109                 substring_t args[MAX_OPT_ARGS];
 110                 int token;
 111                 int option;
 112
 113                 if (!*p)
 114                         continue;
 115
 116                 token = match_token(p, tokens, args);
 117                 switch (token) {
 118                 case Opt_uid:
 119                         if (match_int(&args[0], &option))
 120                                 return -EINVAL;
 121                         opts->uid = option;
 122                         opts->setuid = 1;
 123                         break;
 124                 case Opt_gid:
 125                         if (match_int(&args[0], &option))
 126                                 return -EINVAL;
 127                         opts->gid = option;
 128                         opts->setgid = 1;
 129                         break;
 130                 case Opt_mode:
 131                         if (match_octal(&args[0], &option))
 132                                 return -EINVAL;
 133                         opts->mode = option & S_IALLUGO;
 134                         break;
 135 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 136                 case Opt_ptmxmode:
 137                         if (match_octal(&args[0], &option))
 138                                 return -EINVAL;
 139                         opts->ptmxmode = option & S_IALLUGO;
 140                         break;
 141                 case Opt_newinstance:
 142                         /* newinstance makes sense only on initial mount */
 143                         if (op == PARSE_MOUNT)
 144                                 opts->newinstance = 1;
 145                         break;
 146 #endif
 147                 default:
 148                         printk(KERN_ERR "devpts: called with bogus options\n");
 149                         return -EINVAL;
 150                 }
 151         }
 152
 153         return 0;
 154 }
 155
 156 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 157 static int mknod_ptmx(struct super_block *sb)
 158 {
 159         int mode;
 160         int rc = -ENOMEM;
 161         struct dentry *dentry;
 162         struct inode *inode;
 163         struct dentry *root = sb->s_root;
 164         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 165         struct pts_mount_opts *opts = &fsi->mount_opts;
 166
 167         mutex_lock(&root->d_inode->i_mutex);
 168
 169         /* If we have already created ptmx node, return */
 170         if (fsi->ptmx_dentry) {
 171                 rc = 0;
 172                 goto out;
 173         }
 174
 175         dentry = d_alloc_name(root, "ptmx");
 176         if (!dentry) {
 177                 printk(KERN_NOTICE "Unable to alloc dentry for ptmx node\n");
 178                 goto out;
 179         }
 180
 181         /*
 182          * Create a new 'ptmx' node in this mount of devpts.
 183          */
 184         inode = new_inode(sb);
 185         if (!inode) {
 186                 printk(KERN_ERR "Unable to alloc inode for ptmx node\n");
 187                 dput(dentry);
 188                 goto out;
 189         }
 190
 191         inode->i_ino = 2;
 192         inode->i_uid = inode->i_gid = 0;
 193         inode->i_blocks = 0;
 194         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 195
 196         mode = S_IFCHR|opts->ptmxmode;
 197         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 198
 199         d_add(dentry, inode);
 200
 201         fsi->ptmx_dentry = dentry;
 202         rc = 0;
 203
 204         printk(KERN_DEBUG "Created ptmx node in devpts ino %lu\n",
 205                         inode->i_ino);
 206 out:
 207         mutex_unlock(&root->d_inode->i_mutex);
 208         return rc;
 209 }
 210
 211 static void update_ptmx_mode(struct pts_fs_info *fsi)
 212 {
 213         struct inode *inode;
 214         if (fsi->ptmx_dentry) {
 215                 inode = fsi->ptmx_dentry->d_inode;
 216                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 217         }
 218 }
 219 #else
 220 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
 221 {
 222        return;
 223 }
 224 #endif
 225
 226 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 227 {
 228         int err;
 229         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 230         struct pts_mount_opts *opts = &fsi->mount_opts;
 231
 232         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 233
 234         /*
 235          * parse_mount_options() restores options to default values
 236          * before parsing and may have changed ptmxmode. So, update the
 237          * mode in the inode too. Bogus options don't fail the remount,
 238          * so do this even on error return.
 239          */
 240         update_ptmx_mode(fsi);
 241
 242         return err;
 243 }
 244
 245 static int devpts_show_options(struct seq_file *seq, struct vfsmount *vfs)
 246 {
 247         struct pts_fs_info *fsi = DEVPTS_SB(vfs->mnt_sb);
 248         struct pts_mount_opts *opts = &fsi->mount_opts;
 249
 250         if (opts->setuid)
 251                 seq_printf(seq, ",uid=%u", opts->uid);
 252         if (opts->setgid)
 253                 seq_printf(seq, ",gid=%u", opts->gid);
 254         seq_printf(seq, ",mode=%03o", opts->mode);
 255 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 256         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 257 #endif
 258
 259         return 0;
 260 }
 261
 262 static const struct super_operations devpts_sops = {
 263         .statfs         = simple_statfs,
 264         .remount_fs     = devpts_remount,
 265         .show_options   = devpts_show_options,
 266 };
 267
 268 static void *new_pts_fs_info(void)
 269 {
 270         struct pts_fs_info *fsi;
 271
 272         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 273         if (!fsi)
 274                 return NULL;
 275
 276         ida_init(&fsi->allocated_ptys);
 277         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 278         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 279
 280         return fsi;
 281 }
 282
 283 static int
 284 devpts_fill_super(struct super_block *s, void *data, int silent)
 285 {
 286         struct inode *inode;
 287
 288         s->s_blocksize = 1024;
 289         s->s_blocksize_bits = 10;
 290         s->s_magic = DEVPTS_SUPER_MAGIC;
 291         s->s_op = &devpts_sops;
 292         s->s_time_gran = 1;
 293
 294         s->s_fs_info = new_pts_fs_info();
 295         if (!s->s_fs_info)
 296                 goto fail;
 297
 298         inode = new_inode(s);
 299         if (!inode)
 300                 goto free_fsi;
 301         inode->i_ino = 1;
 302         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 303         inode->i_blocks = 0;
 304         inode->i_uid = inode->i_gid = 0;
 305         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 306         inode->i_op = &simple_dir_inode_operations;
 307         inode->i_fop = &simple_dir_operations;
 308         inode->i_nlink = 2;
 309
 310         s->s_root = d_alloc_root(inode);
 311         if (s->s_root)
 312                 return 0;
 313
 314         printk("devpts: get root dentry failed\n");
 315         iput(inode);
 316
 317 free_fsi:
 318         kfree(s->s_fs_info);
 319 fail:
 320         return -ENOMEM;
 321 }
 322
 323 static int compare_init_pts_sb(struct super_block *s, void *p)
 324 {
 325         if (devpts_mnt)
 326                 return devpts_mnt->mnt_sb == s;
 327         return 0;
 328 }
 329
 330 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 331 /*
 332  * Safely parse the mount options in @data and update @opts.
 333  *
 334  * devpts ends up parsing options two times during mount, due to the
 335  * two modes of operation it supports. The first parse occurs in
 336  * devpts_get_sb() when determining the mode (single-instance or
 337  * multi-instance mode). The second parse happens in devpts_remount()
 338  * or new_pts_mount() depending on the mode.
 339  *
 340  * Parsing of options modifies the @data making subsequent parsing
 341  * incorrect. So make a local copy of @data and parse it.
 342  *
 343  * Return: 0 On success, -errno on error
 344  */
 345 static int safe_parse_mount_options(void *data, struct pts_mount_opts *opts)
 346 {
 347         int rc;
 348         void *datacp;
 349
 350         if (!data)
 351                 return 0;
 352
 353         /* Use kstrdup() ?  */
 354         datacp = kmalloc(PAGE_SIZE, GFP_KERNEL);
 355         if (!datacp)
 356                 return -ENOMEM;
 357
 358         memcpy(datacp, data, PAGE_SIZE);
 359         rc = parse_mount_options((char *)datacp, PARSE_MOUNT, opts);
 360         kfree(datacp);
 361
 362         return rc;
 363 }
 364
 365 /*
 366  * Mount a new (private) instance of devpts.  PTYs created in this
 367  * instance are independent of the PTYs in other devpts instances.
 368  */
 369 static int new_pts_mount(struct file_system_type *fs_type, int flags,
 370                 void *data, struct vfsmount *mnt)
 371 {
 372         int err;
 373         struct pts_fs_info *fsi;
 374         struct pts_mount_opts *opts;
 375
 376         printk(KERN_NOTICE "devpts: newinstance mount\n");
 377
 378         err = get_sb_nodev(fs_type, flags, data, devpts_fill_super, mnt);
 379         if (err)
 380                 return err;
 381
 382         fsi = DEVPTS_SB(mnt->mnt_sb);
 383         opts = &fsi->mount_opts;
 384
 385         err = parse_mount_options(data, PARSE_MOUNT, opts);
 386         if (err)
 387                 goto fail;
 388
 389         err = mknod_ptmx(mnt->mnt_sb);
 390         if (err)
 391                 goto fail;
 392
 393         return 0;
 394
 395 fail:
 396         dput(mnt->mnt_sb->s_root);
 397         deactivate_super(mnt->mnt_sb);
 398         return err;
 399 }
 400
 401 /*
 402  * Check if 'newinstance' mount option was specified in @data.
 403  *
 404  * Return: -errno       on error (eg: invalid mount options specified)
 405  *       : 1            if 'newinstance' mount option was specified
 406  *       : 0            if 'newinstance' mount option was NOT specified
 407  */
 408 static int is_new_instance_mount(void *data)
 409 {
 410         int rc;
 411         struct pts_mount_opts opts;
 412
 413         if (!data)
 414                 return 0;
 415
 416         rc = safe_parse_mount_options(data, &opts);
 417         if (!rc)
 418                 rc = opts.newinstance;
 419
 420         return rc;
 421 }
 422
 423 /*
 424  * get_init_pts_sb()
 425  *
 426  *     This interface is needed to support multiple namespace semantics in
 427  *     devpts while preserving backward compatibility of the current 'single-
 428  *     namespace' semantics. i.e all mounts of devpts without the 'newinstance'
 429  *     mount option should bind to the initial kernel mount, like
 430  *     get_sb_single().
 431  *
 432  *     Mounts with 'newinstance' option create a new private namespace.
 433  *
 434  *     But for single-mount semantics, devpts cannot use get_sb_single(),
 435  *     because get_sb_single()/sget() find and use the super-block from
 436  *     the most recent mount of devpts. But that recent mount may be a
 437  *     'newinstance' mount and get_sb_single() would pick the newinstance
 438  *     super-block instead of the initial super-block.
 439  *
 440  *     This interface is identical to get_sb_single() except that it
 441  *     consistently selects the 'single-namespace' superblock even in the
 442  *     presence of the private namespace (i.e 'newinstance') super-blocks.
 443  */
 444 static int get_init_pts_sb(struct file_system_type *fs_type, int flags,
 445                 void *data, struct vfsmount *mnt)
 446 {
 447         struct super_block *s;
 448         int error;
 449
 450         s = sget(fs_type, compare_init_pts_sb, set_anon_super, NULL);
 451         if (IS_ERR(s))
 452                 return PTR_ERR(s);
 453
 454         if (!s->s_root) {
 455                 s->s_flags = flags;
 456                 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 457                 if (error) {
 458                         up_write(&s->s_umount);
 459                         deactivate_super(s);
 460                         return error;
 461                 }
 462                 s->s_flags |= MS_ACTIVE;
 463         }
 464         do_remount_sb(s, flags, data, 0);
 465         return simple_set_mnt(mnt, s);
 466 }
 467
 468 /*
 469  * Mount or remount the initial kernel mount of devpts. This type of
 470  * mount maintains the legacy, single-instance semantics, while the
 471  * kernel still allows multiple-instances.
 472  */
 473 static int init_pts_mount(struct file_system_type *fs_type, int flags,
 474                 void *data, struct vfsmount *mnt)
 475 {
 476         int err;
 477
 478         err = get_init_pts_sb(fs_type, flags, data, mnt);
 479         if (err)
 480                  return err;
 481
 482         err = mknod_ptmx(mnt->mnt_sb);
 483         if (err) {
 484                 dput(mnt->mnt_sb->s_root);
 485                 deactivate_super(mnt->mnt_sb);
 486         }
 487
 488         return err;
 489 }
 490
 491 static int devpts_get_sb(struct file_system_type *fs_type,
 492         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 493 {
 494         int new;
 495
 496         new = is_new_instance_mount(data);
 497         if (new < 0)
 498                 return new;
 499
 500         if (new)
 501                 return new_pts_mount(fs_type, flags, data, mnt);
 502
 503         return init_pts_mount(fs_type, flags, data, mnt);
 504 }
 505 #else
 506 /*
 507  * This supports only the legacy single-instance semantics (no
 508  * multiple-instance semantics)
 509  */
 510 static int devpts_get_sb(struct file_system_type *fs_type, int flags,
 511                 const char *dev_name, void *data, struct vfsmount *mnt)
 512 {
 513         return get_sb_single(fs_type, flags, data, devpts_fill_super, mnt);
 514 }
 515 #endif
 516
 517 static void devpts_kill_sb(struct super_block *sb)
 518 {
 519         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 520
 521         kfree(fsi);
 522         kill_litter_super(sb);
 523 }
 524
 525 static struct file_system_type devpts_fs_type = {
 526         .owner          = THIS_MODULE,
 527         .name           = "devpts",
 528         .get_sb         = devpts_get_sb,
 529         .kill_sb        = devpts_kill_sb,
 530 };
 531
 532 /*
 533  * The normal naming convention is simply /dev/pts/<number>; this conforms
 534  * to the System V naming convention
 535  */
 536
 537 int devpts_new_index(struct inode *ptmx_inode)
 538 {
 539         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 540         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 541         int index;
 542         int ida_ret;
 543
 544 retry:
 545         if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL)) {
 546                 return -ENOMEM;
 547         }
 548
 549         mutex_lock(&allocated_ptys_lock);
 550         ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 551         if (ida_ret < 0) {
 552                 mutex_unlock(&allocated_ptys_lock);
 553                 if (ida_ret == -EAGAIN)
 554                         goto retry;
 555                 return -EIO;
 556         }
 557
 558         if (index >= pty_limit) {
 559                 ida_remove(&fsi->allocated_ptys, index);
 560                 mutex_unlock(&allocated_ptys_lock);
 561                 return -EIO;
 562         }
 563         mutex_unlock(&allocated_ptys_lock);
 564         return index;
 565 }
 566
 567 void devpts_kill_index(struct inode *ptmx_inode, int idx)
 568 {
 569         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 570         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 571
 572         mutex_lock(&allocated_ptys_lock);
 573         ida_remove(&fsi->allocated_ptys, idx);
 574         mutex_unlock(&allocated_ptys_lock);
 575 }
 576
 577 int devpts_pty_new(struct inode *ptmx_inode, struct tty_struct *tty)
 578 {
 579         int number = tty->index; /* tty layer puts index from devpts_new_index() in here */
 580         struct tty_driver *driver = tty->driver;
 581         dev_t device = MKDEV(driver->major, driver->minor_start+number);
 582         struct dentry *dentry;
 583         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 584         struct inode *inode = new_inode(sb);
 585         struct dentry *root = sb->s_root;
 586         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 587         struct pts_mount_opts *opts = &fsi->mount_opts;
 588         char s[12];
 589
 590         /* We're supposed to be given the slave end of a pty */
 591         BUG_ON(driver->type != TTY_DRIVER_TYPE_PTY);
 592         BUG_ON(driver->subtype != PTY_TYPE_SLAVE);
 593
 594         if (!inode)
 595                 return -ENOMEM;
 596
 597         inode->i_ino = number+2;
 598         inode->i_uid = config.setuid ? config.uid : current_fsuid();
 599         inode->i_gid = config.setgid ? config.gid : current_fsgid();
 600         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 601         init_special_inode(inode, S_IFCHR|opts->mode, device);
 602         inode->i_private = tty;
 603         tty->driver_data = inode;
 604
 605         sprintf(s, "%d", number);
 606
 607         mutex_lock(&root->d_inode->i_mutex);
 608
 609         dentry = d_alloc_name(root, s);
 610         if (!IS_ERR(dentry)) {
 611                 d_add(dentry, inode);
 612                 fsnotify_create(root->d_inode, dentry);
 613         }
 614
 615         mutex_unlock(&root->d_inode->i_mutex);
 616
 617         return 0;
 618 }
 619
 620 struct tty_struct *devpts_get_tty(struct inode *pts_inode, int number)
 621 {
 622         BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 623
 624         if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
 625                 return (struct tty_struct *)pts_inode->i_private;
 626         return NULL;
 627 }
 628
 629 void devpts_pty_kill(struct tty_struct *tty)
 630 {
 631         struct inode *inode = tty->driver_data;
 632         struct super_block *sb = pts_sb_from_inode(inode);
 633         struct dentry *root = sb->s_root;
 634         struct dentry *dentry;
 635
 636         BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 637
 638         mutex_lock(&root->d_inode->i_mutex);
 639
 640         dentry = d_find_alias(inode);
 641         if (IS_ERR(dentry))
 642                 goto out;
 643
 644         if (dentry) {
 645                 inode->i_nlink--;
 646                 d_delete(dentry);
 647                 dput(dentry);   // d_alloc_name() in devpts_pty_new()
 648         }
 649
 650         dput(dentry);           // d_find_alias above
 651
 652 out:
 653         mutex_unlock(&root->d_inode->i_mutex);
 654 }
 655
 656 static int __init init_devpts_fs(void)
 657 {
 658         int err = register_filesystem(&devpts_fs_type);
 659         if (!err) {
 660                 devpts_mnt = kern_mount(&devpts_fs_type);
 661                 if (IS_ERR(devpts_mnt))
 662                         err = PTR_ERR(devpts_mnt);
 663         }
 664         return err;
 665 }
 666
 667 static void __exit exit_devpts_fs(void)
 668 {
 669         unregister_filesystem(&devpts_fs_type);
 670         mntput(devpts_mnt);
 671 }
 672
 673 module_init(init_devpts_fs)
 674 module_exit(exit_devpts_fs)
 675 MODULE_LICENSE("GPL");