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_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 193
 194         mode = S_IFCHR|opts->ptmxmode;
 195         init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
 196
 197         d_add(dentry, inode);
 198
 199         fsi->ptmx_dentry = dentry;
 200         rc = 0;
 201 out:
 202         mutex_unlock(&root->d_inode->i_mutex);
 203         return rc;
 204 }
 205
 206 static void update_ptmx_mode(struct pts_fs_info *fsi)
 207 {
 208         struct inode *inode;
 209         if (fsi->ptmx_dentry) {
 210                 inode = fsi->ptmx_dentry->d_inode;
 211                 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
 212         }
 213 }
 214 #else
 215 static inline void update_ptmx_mode(struct pts_fs_info *fsi)
 216 {
 217        return;
 218 }
 219 #endif
 220
 221 static int devpts_remount(struct super_block *sb, int *flags, char *data)
 222 {
 223         int err;
 224         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 225         struct pts_mount_opts *opts = &fsi->mount_opts;
 226
 227         err = parse_mount_options(data, PARSE_REMOUNT, opts);
 228
 229         /*
 230          * parse_mount_options() restores options to default values
 231          * before parsing and may have changed ptmxmode. So, update the
 232          * mode in the inode too. Bogus options don't fail the remount,
 233          * so do this even on error return.
 234          */
 235         update_ptmx_mode(fsi);
 236
 237         return err;
 238 }
 239
 240 static int devpts_show_options(struct seq_file *seq, struct vfsmount *vfs)
 241 {
 242         struct pts_fs_info *fsi = DEVPTS_SB(vfs->mnt_sb);
 243         struct pts_mount_opts *opts = &fsi->mount_opts;
 244
 245         if (opts->setuid)
 246                 seq_printf(seq, ",uid=%u", opts->uid);
 247         if (opts->setgid)
 248                 seq_printf(seq, ",gid=%u", opts->gid);
 249         seq_printf(seq, ",mode=%03o", opts->mode);
 250 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 251         seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
 252 #endif
 253
 254         return 0;
 255 }
 256
 257 static const struct super_operations devpts_sops = {
 258         .statfs         = simple_statfs,
 259         .remount_fs     = devpts_remount,
 260         .show_options   = devpts_show_options,
 261 };
 262
 263 static void *new_pts_fs_info(void)
 264 {
 265         struct pts_fs_info *fsi;
 266
 267         fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
 268         if (!fsi)
 269                 return NULL;
 270
 271         ida_init(&fsi->allocated_ptys);
 272         fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
 273         fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
 274
 275         return fsi;
 276 }
 277
 278 static int
 279 devpts_fill_super(struct super_block *s, void *data, int silent)
 280 {
 281         struct inode *inode;
 282
 283         s->s_blocksize = 1024;
 284         s->s_blocksize_bits = 10;
 285         s->s_magic = DEVPTS_SUPER_MAGIC;
 286         s->s_op = &devpts_sops;
 287         s->s_time_gran = 1;
 288
 289         s->s_fs_info = new_pts_fs_info();
 290         if (!s->s_fs_info)
 291                 goto fail;
 292
 293         inode = new_inode(s);
 294         if (!inode)
 295                 goto free_fsi;
 296         inode->i_ino = 1;
 297         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 298         inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 299         inode->i_op = &simple_dir_inode_operations;
 300         inode->i_fop = &simple_dir_operations;
 301         inode->i_nlink = 2;
 302
 303         s->s_root = d_alloc_root(inode);
 304         if (s->s_root)
 305                 return 0;
 306
 307         printk(KERN_ERR "devpts: get root dentry failed\n");
 308         iput(inode);
 309
 310 free_fsi:
 311         kfree(s->s_fs_info);
 312 fail:
 313         return -ENOMEM;
 314 }
 315
 316 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 317 static int compare_init_pts_sb(struct super_block *s, void *p)
 318 {
 319         if (devpts_mnt)
 320                 return devpts_mnt->mnt_sb == s;
 321         return 0;
 322 }
 323
 324 /*
 325  * Safely parse the mount options in @data and update @opts.
 326  *
 327  * devpts ends up parsing options two times during mount, due to the
 328  * two modes of operation it supports. The first parse occurs in
 329  * devpts_get_sb() when determining the mode (single-instance or
 330  * multi-instance mode). The second parse happens in devpts_remount()
 331  * or new_pts_mount() depending on the mode.
 332  *
 333  * Parsing of options modifies the @data making subsequent parsing
 334  * incorrect. So make a local copy of @data and parse it.
 335  *
 336  * Return: 0 On success, -errno on error
 337  */
 338 static int safe_parse_mount_options(void *data, struct pts_mount_opts *opts)
 339 {
 340         int rc;
 341         void *datacp;
 342
 343         if (!data)
 344                 return 0;
 345
 346         /* Use kstrdup() ?  */
 347         datacp = kmalloc(PAGE_SIZE, GFP_KERNEL);
 348         if (!datacp)
 349                 return -ENOMEM;
 350
 351         memcpy(datacp, data, PAGE_SIZE);
 352         rc = parse_mount_options((char *)datacp, PARSE_MOUNT, opts);
 353         kfree(datacp);
 354
 355         return rc;
 356 }
 357
 358 /*
 359  * Mount a new (private) instance of devpts.  PTYs created in this
 360  * instance are independent of the PTYs in other devpts instances.
 361  */
 362 static int new_pts_mount(struct file_system_type *fs_type, int flags,
 363                 void *data, struct vfsmount *mnt)
 364 {
 365         int err;
 366         struct pts_fs_info *fsi;
 367         struct pts_mount_opts *opts;
 368
 369         err = get_sb_nodev(fs_type, flags, data, devpts_fill_super, mnt);
 370         if (err)
 371                 return err;
 372
 373         fsi = DEVPTS_SB(mnt->mnt_sb);
 374         opts = &fsi->mount_opts;
 375
 376         err = parse_mount_options(data, PARSE_MOUNT, opts);
 377         if (err)
 378                 goto fail;
 379
 380         err = mknod_ptmx(mnt->mnt_sb);
 381         if (err)
 382                 goto fail;
 383
 384         return 0;
 385
 386 fail:
 387         dput(mnt->mnt_sb->s_root);
 388         deactivate_super(mnt->mnt_sb);
 389         return err;
 390 }
 391
 392 /*
 393  * Check if 'newinstance' mount option was specified in @data.
 394  *
 395  * Return: -errno       on error (eg: invalid mount options specified)
 396  *       : 1            if 'newinstance' mount option was specified
 397  *       : 0            if 'newinstance' mount option was NOT specified
 398  */
 399 static int is_new_instance_mount(void *data)
 400 {
 401         int rc;
 402         struct pts_mount_opts opts;
 403
 404         if (!data)
 405                 return 0;
 406
 407         rc = safe_parse_mount_options(data, &opts);
 408         if (!rc)
 409                 rc = opts.newinstance;
 410
 411         return rc;
 412 }
 413
 414 /*
 415  * get_init_pts_sb()
 416  *
 417  *     This interface is needed to support multiple namespace semantics in
 418  *     devpts while preserving backward compatibility of the current 'single-
 419  *     namespace' semantics. i.e all mounts of devpts without the 'newinstance'
 420  *     mount option should bind to the initial kernel mount, like
 421  *     get_sb_single().
 422  *
 423  *     Mounts with 'newinstance' option create a new private namespace.
 424  *
 425  *     But for single-mount semantics, devpts cannot use get_sb_single(),
 426  *     because get_sb_single()/sget() find and use the super-block from
 427  *     the most recent mount of devpts. But that recent mount may be a
 428  *     'newinstance' mount and get_sb_single() would pick the newinstance
 429  *     super-block instead of the initial super-block.
 430  *
 431  *     This interface is identical to get_sb_single() except that it
 432  *     consistently selects the 'single-namespace' superblock even in the
 433  *     presence of the private namespace (i.e 'newinstance') super-blocks.
 434  */
 435 static int get_init_pts_sb(struct file_system_type *fs_type, int flags,
 436                 void *data, struct vfsmount *mnt)
 437 {
 438         struct super_block *s;
 439         int error;
 440
 441         s = sget(fs_type, compare_init_pts_sb, set_anon_super, NULL);
 442         if (IS_ERR(s))
 443                 return PTR_ERR(s);
 444
 445         if (!s->s_root) {
 446                 s->s_flags = flags;
 447                 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 448                 if (error) {
 449                         up_write(&s->s_umount);
 450                         deactivate_super(s);
 451                         return error;
 452                 }
 453                 s->s_flags |= MS_ACTIVE;
 454         }
 455         do_remount_sb(s, flags, data, 0);
 456         return simple_set_mnt(mnt, s);
 457 }
 458
 459 /*
 460  * Mount or remount the initial kernel mount of devpts. This type of
 461  * mount maintains the legacy, single-instance semantics, while the
 462  * kernel still allows multiple-instances.
 463  */
 464 static int init_pts_mount(struct file_system_type *fs_type, int flags,
 465                 void *data, struct vfsmount *mnt)
 466 {
 467         int err;
 468
 469         err = get_init_pts_sb(fs_type, flags, data, mnt);
 470         if (err)
 471                 return err;
 472
 473         err = mknod_ptmx(mnt->mnt_sb);
 474         if (err) {
 475                 dput(mnt->mnt_sb->s_root);
 476                 deactivate_super(mnt->mnt_sb);
 477         }
 478
 479         return err;
 480 }
 481
 482 static int devpts_get_sb(struct file_system_type *fs_type,
 483         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 484 {
 485         int new;
 486
 487         new = is_new_instance_mount(data);
 488         if (new < 0)
 489                 return new;
 490
 491         if (new)
 492                 return new_pts_mount(fs_type, flags, data, mnt);
 493
 494         return init_pts_mount(fs_type, flags, data, mnt);
 495 }
 496 #else
 497 /*
 498  * This supports only the legacy single-instance semantics (no
 499  * multiple-instance semantics)
 500  */
 501 static int devpts_get_sb(struct file_system_type *fs_type, int flags,
 502                 const char *dev_name, void *data, struct vfsmount *mnt)
 503 {
 504         return get_sb_single(fs_type, flags, data, devpts_fill_super, mnt);
 505 }
 506 #endif
 507
 508 static void devpts_kill_sb(struct super_block *sb)
 509 {
 510         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 511
 512         kfree(fsi);
 513         kill_litter_super(sb);
 514 }
 515
 516 static struct file_system_type devpts_fs_type = {
 517         .owner          = THIS_MODULE,
 518         .name           = "devpts",
 519         .get_sb         = devpts_get_sb,
 520         .kill_sb        = devpts_kill_sb,
 521 };
 522
 523 /*
 524  * The normal naming convention is simply /dev/pts/<number>; this conforms
 525  * to the System V naming convention
 526  */
 527
 528 int devpts_new_index(struct inode *ptmx_inode)
 529 {
 530         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 531         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 532         int index;
 533         int ida_ret;
 534
 535 retry:
 536         if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
 537                 return -ENOMEM;
 538
 539         mutex_lock(&allocated_ptys_lock);
 540         ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 541         if (ida_ret < 0) {
 542                 mutex_unlock(&allocated_ptys_lock);
 543                 if (ida_ret == -EAGAIN)
 544                         goto retry;
 545                 return -EIO;
 546         }
 547
 548         if (index >= pty_limit) {
 549                 ida_remove(&fsi->allocated_ptys, index);
 550                 mutex_unlock(&allocated_ptys_lock);
 551                 return -EIO;
 552         }
 553         mutex_unlock(&allocated_ptys_lock);
 554         return index;
 555 }
 556
 557 void devpts_kill_index(struct inode *ptmx_inode, int idx)
 558 {
 559         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 560         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 561
 562         mutex_lock(&allocated_ptys_lock);
 563         ida_remove(&fsi->allocated_ptys, idx);
 564         mutex_unlock(&allocated_ptys_lock);
 565 }
 566
 567 int devpts_pty_new(struct inode *ptmx_inode, struct tty_struct *tty)
 568 {
 569         /* tty layer puts index from devpts_new_index() in here */
 570         int number = tty->index;
 571         struct tty_driver *driver = tty->driver;
 572         dev_t device = MKDEV(driver->major, driver->minor_start+number);
 573         struct dentry *dentry;
 574         struct super_block *sb = pts_sb_from_inode(ptmx_inode);
 575         struct inode *inode = new_inode(sb);
 576         struct dentry *root = sb->s_root;
 577         struct pts_fs_info *fsi = DEVPTS_SB(sb);
 578         struct pts_mount_opts *opts = &fsi->mount_opts;
 579         char s[12];
 580
 581         /* We're supposed to be given the slave end of a pty */
 582         BUG_ON(driver->type != TTY_DRIVER_TYPE_PTY);
 583         BUG_ON(driver->subtype != PTY_TYPE_SLAVE);
 584
 585         if (!inode)
 586                 return -ENOMEM;
 587
 588         inode->i_ino = number + 3;
 589         inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
 590         inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
 591         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 592         init_special_inode(inode, S_IFCHR|opts->mode, device);
 593         inode->i_private = tty;
 594         tty->driver_data = inode;
 595
 596         sprintf(s, "%d", number);
 597
 598         mutex_lock(&root->d_inode->i_mutex);
 599
 600         dentry = d_alloc_name(root, s);
 601         if (!IS_ERR(dentry)) {
 602                 d_add(dentry, inode);
 603                 fsnotify_create(root->d_inode, dentry);
 604         }
 605
 606         mutex_unlock(&root->d_inode->i_mutex);
 607
 608         return 0;
 609 }
 610
 611 struct tty_struct *devpts_get_tty(struct inode *pts_inode, int number)
 612 {
 613         BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 614
 615         if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
 616                 return (struct tty_struct *)pts_inode->i_private;
 617         return NULL;
 618 }
 619
 620 void devpts_pty_kill(struct tty_struct *tty)
 621 {
 622         struct inode *inode = tty->driver_data;
 623         struct super_block *sb = pts_sb_from_inode(inode);
 624         struct dentry *root = sb->s_root;
 625         struct dentry *dentry;
 626
 627         BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 628
 629         mutex_lock(&root->d_inode->i_mutex);
 630
 631         dentry = d_find_alias(inode);
 632         if (IS_ERR(dentry))
 633                 goto out;
 634
 635         if (dentry) {
 636                 inode->i_nlink--;
 637                 d_delete(dentry);
 638                 dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
 639         }
 640
 641         dput(dentry);           /* d_find_alias above */
 642 out:
 643         mutex_unlock(&root->d_inode->i_mutex);
 644 }
 645
 646 static int __init init_devpts_fs(void)
 647 {
 648         int err = register_filesystem(&devpts_fs_type);
 649         if (!err) {
 650                 devpts_mnt = kern_mount(&devpts_fs_type);
 651                 if (IS_ERR(devpts_mnt))
 652                         err = PTR_ERR(devpts_mnt);
 653         }
 654         return err;
 655 }
 656
 657 static void __exit exit_devpts_fs(void)
 658 {
 659         unregister_filesystem(&devpts_fs_type);
 660         mntput(devpts_mnt);
 661 }
 662
 663 module_init(init_devpts_fs)
 664 module_exit(exit_devpts_fs)
 665 MODULE_LICENSE("GPL");