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/slab.h> 19 #include <linux/mount.h> 20 #include <linux/tty.h> 21 #include <linux/mutex.h> 22 #include <linux/magic.h> 23 #include <linux/idr.h> 24 #include <linux/devpts_fs.h> 25 #include <linux/parser.h> 26 #include <linux/fsnotify.h> 27 #include <linux/seq_file.h> 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 /* 40 * sysctl support for setting limits on the number of Unix98 ptys allocated. 41 * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly. 42 */ 43 static int pty_limit = NR_UNIX98_PTY_DEFAULT; 44 static int pty_reserve = NR_UNIX98_PTY_RESERVE; 45 static int pty_limit_min; 46 static int pty_limit_max = INT_MAX; 47 static int pty_count; 48 49 static struct ctl_table pty_table[] = { 50 { 51 .procname = "max", 52 .maxlen = sizeof(int), 53 .mode = 0644, 54 .data = &pty_limit, 55 .proc_handler = proc_dointvec_minmax, 56 .extra1 = &pty_limit_min, 57 .extra2 = &pty_limit_max, 58 }, { 59 .procname = "reserve", 60 .maxlen = sizeof(int), 61 .mode = 0644, 62 .data = &pty_reserve, 63 .proc_handler = proc_dointvec_minmax, 64 .extra1 = &pty_limit_min, 65 .extra2 = &pty_limit_max, 66 }, { 67 .procname = "nr", 68 .maxlen = sizeof(int), 69 .mode = 0444, 70 .data = &pty_count, 71 .proc_handler = proc_dointvec, 72 }, 73 {} 74 }; 75 76 static struct ctl_table pty_kern_table[] = { 77 { 78 .procname = "pty", 79 .mode = 0555, 80 .child = pty_table, 81 }, 82 {} 83 }; 84 85 static struct ctl_table pty_root_table[] = { 86 { 87 .procname = "kernel", 88 .mode = 0555, 89 .child = pty_kern_table, 90 }, 91 {} 92 }; 93 94 static DEFINE_MUTEX(allocated_ptys_lock); 95 96 static struct vfsmount *devpts_mnt; 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 newinstance; 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 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES 119 {Opt_ptmxmode, "ptmxmode=%o"}, 120 {Opt_newinstance, "newinstance"}, 121 {Opt_max, "max=%d"}, 122 #endif 123 {Opt_err, NULL} 124 }; 125 126 struct pts_fs_info { 127 struct ida allocated_ptys; 128 struct pts_mount_opts mount_opts; 129 struct dentry *ptmx_dentry; 130 }; 131 132 static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb) 133 { 134 return sb->s_fs_info; 135 } 136 137 static inline struct super_block *pts_sb_from_inode(struct inode *inode) 138 { 139 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES 140 if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC) 141 return inode->i_sb; 142 #endif 143 return devpts_mnt->mnt_sb; 144 } 145 146 #define PARSE_MOUNT 0 147 #define PARSE_REMOUNT 1 148 149 /* 150 * parse_mount_options(): 151 * Set @opts to mount options specified in @data. If an option is not 152 * specified in @data, set it to its default value. The exception is 153 * 'newinstance' option which can only be set/cleared on a mount (i.e. 154 * cannot be changed during remount). 155 * 156 * Note: @data may be NULL (in which case all options are set to default). 157 */ 158 static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts) 159 { 160 char *p; 161 kuid_t uid; 162 kgid_t gid; 163 164 opts->setuid = 0; 165 opts->setgid = 0; 166 opts->uid = GLOBAL_ROOT_UID; 167 opts->gid = GLOBAL_ROOT_GID; 168 opts->mode = DEVPTS_DEFAULT_MODE; 169 opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE; 170 opts->max = NR_UNIX98_PTY_MAX; 171 172 /* newinstance makes sense only on initial mount */ 173 if (op == PARSE_MOUNT) 174 opts->newinstance = 0; 175 176 while ((p = strsep(&data, ",")) != NULL) { 177 substring_t args[MAX_OPT_ARGS]; 178 int token; 179 int option; 180 181 if (!*p) 182 continue; 183 184 token = match_token(p, tokens, args); 185 switch (token) { 186 case Opt_uid: 187 if (match_int(&args[0], &option)) 188 return -EINVAL; 189 uid = make_kuid(current_user_ns(), option); 190 if (!uid_valid(uid)) 191 return -EINVAL; 192 opts->uid = uid; 193 opts->setuid = 1; 194 break; 195 case Opt_gid: 196 if (match_int(&args[0], &option)) 197 return -EINVAL; 198 gid = make_kgid(current_user_ns(), option); 199 if (!gid_valid(gid)) 200 return -EINVAL; 201 opts->gid = gid; 202 opts->setgid = 1; 203 break; 204 case Opt_mode: 205 if (match_octal(&args[0], &option)) 206 return -EINVAL; 207 opts->mode = option & S_IALLUGO; 208 break; 209 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES 210 case Opt_ptmxmode: 211 if (match_octal(&args[0], &option)) 212 return -EINVAL; 213 opts->ptmxmode = option & S_IALLUGO; 214 break; 215 case Opt_newinstance: 216 /* newinstance makes sense only on initial mount */ 217 if (op == PARSE_MOUNT) 218 opts->newinstance = 1; 219 break; 220 case Opt_max: 221 if (match_int(&args[0], &option) || 222 option < 0 || option > NR_UNIX98_PTY_MAX) 223 return -EINVAL; 224 opts->max = option; 225 break; 226 #endif 227 default: 228 printk(KERN_ERR "devpts: called with bogus options\n"); 229 return -EINVAL; 230 } 231 } 232 233 return 0; 234 } 235 236 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES 237 static int mknod_ptmx(struct super_block *sb) 238 { 239 int mode; 240 int rc = -ENOMEM; 241 struct dentry *dentry; 242 struct inode *inode; 243 struct dentry *root = sb->s_root; 244 struct pts_fs_info *fsi = DEVPTS_SB(sb); 245 struct pts_mount_opts *opts = &fsi->mount_opts; 246 kuid_t root_uid; 247 kgid_t root_gid; 248 249 root_uid = make_kuid(current_user_ns(), 0); 250 root_gid = make_kgid(current_user_ns(), 0); 251 if (!uid_valid(root_uid) || !gid_valid(root_gid)) 252 return -EINVAL; 253 254 mutex_lock(&root->d_inode->i_mutex); 255 256 /* If we have already created ptmx node, return */ 257 if (fsi->ptmx_dentry) { 258 rc = 0; 259 goto out; 260 } 261 262 dentry = d_alloc_name(root, "ptmx"); 263 if (!dentry) { 264 printk(KERN_NOTICE "Unable to alloc dentry for ptmx node\n"); 265 goto out; 266 } 267 268 /* 269 * Create a new 'ptmx' node in this mount of devpts. 270 */ 271 inode = new_inode(sb); 272 if (!inode) { 273 printk(KERN_ERR "Unable to alloc inode for ptmx node\n"); 274 dput(dentry); 275 goto out; 276 } 277 278 inode->i_ino = 2; 279 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 280 281 mode = S_IFCHR|opts->ptmxmode; 282 init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2)); 283 inode->i_uid = root_uid; 284 inode->i_gid = root_gid; 285 286 d_add(dentry, inode); 287 288 fsi->ptmx_dentry = dentry; 289 rc = 0; 290 out: 291 mutex_unlock(&root->d_inode->i_mutex); 292 return rc; 293 } 294 295 static void update_ptmx_mode(struct pts_fs_info *fsi) 296 { 297 struct inode *inode; 298 if (fsi->ptmx_dentry) { 299 inode = fsi->ptmx_dentry->d_inode; 300 inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode; 301 } 302 } 303 #else 304 static inline void update_ptmx_mode(struct pts_fs_info *fsi) 305 { 306 return; 307 } 308 #endif 309 310 static int devpts_remount(struct super_block *sb, int *flags, char *data) 311 { 312 int err; 313 struct pts_fs_info *fsi = DEVPTS_SB(sb); 314 struct pts_mount_opts *opts = &fsi->mount_opts; 315 316 err = parse_mount_options(data, PARSE_REMOUNT, opts); 317 318 /* 319 * parse_mount_options() restores options to default values 320 * before parsing and may have changed ptmxmode. So, update the 321 * mode in the inode too. Bogus options don't fail the remount, 322 * so do this even on error return. 323 */ 324 update_ptmx_mode(fsi); 325 326 return err; 327 } 328 329 static int devpts_show_options(struct seq_file *seq, struct dentry *root) 330 { 331 struct pts_fs_info *fsi = DEVPTS_SB(root->d_sb); 332 struct pts_mount_opts *opts = &fsi->mount_opts; 333 334 if (opts->setuid) 335 seq_printf(seq, ",uid=%u", from_kuid_munged(&init_user_ns, opts->uid)); 336 if (opts->setgid) 337 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, opts->gid)); 338 seq_printf(seq, ",mode=%03o", opts->mode); 339 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES 340 seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode); 341 if (opts->max < NR_UNIX98_PTY_MAX) 342 seq_printf(seq, ",max=%d", opts->max); 343 #endif 344 345 return 0; 346 } 347 348 static const struct super_operations devpts_sops = { 349 .statfs = simple_statfs, 350 .remount_fs = devpts_remount, 351 .show_options = devpts_show_options, 352 }; 353 354 static void *new_pts_fs_info(void) 355 { 356 struct pts_fs_info *fsi; 357 358 fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL); 359 if (!fsi) 360 return NULL; 361 362 ida_init(&fsi->allocated_ptys); 363 fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE; 364 fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE; 365 366 return fsi; 367 } 368 369 static int 370 devpts_fill_super(struct super_block *s, void *data, int silent) 371 { 372 struct inode *inode; 373 374 s->s_blocksize = 1024; 375 s->s_blocksize_bits = 10; 376 s->s_magic = DEVPTS_SUPER_MAGIC; 377 s->s_op = &devpts_sops; 378 s->s_time_gran = 1; 379 380 s->s_fs_info = new_pts_fs_info(); 381 if (!s->s_fs_info) 382 goto fail; 383 384 inode = new_inode(s); 385 if (!inode) 386 goto fail; 387 inode->i_ino = 1; 388 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 389 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR; 390 inode->i_op = &simple_dir_inode_operations; 391 inode->i_fop = &simple_dir_operations; 392 set_nlink(inode, 2); 393 394 s->s_root = d_make_root(inode); 395 if (s->s_root) 396 return 0; 397 398 printk(KERN_ERR "devpts: get root dentry failed\n"); 399 400 fail: 401 return -ENOMEM; 402 } 403 404 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES 405 static int compare_init_pts_sb(struct super_block *s, void *p) 406 { 407 if (devpts_mnt) 408 return devpts_mnt->mnt_sb == s; 409 return 0; 410 } 411 412 /* 413 * devpts_mount() 414 * 415 * If the '-o newinstance' mount option was specified, mount a new 416 * (private) instance of devpts. PTYs created in this instance are 417 * independent of the PTYs in other devpts instances. 418 * 419 * If the '-o newinstance' option was not specified, mount/remount the 420 * initial kernel mount of devpts. This type of mount gives the 421 * legacy, single-instance semantics. 422 * 423 * The 'newinstance' option is needed to support multiple namespace 424 * semantics in devpts while preserving backward compatibility of the 425 * current 'single-namespace' semantics. i.e all mounts of devpts 426 * without the 'newinstance' mount option should bind to the initial 427 * kernel mount, like mount_single(). 428 * 429 * Mounts with 'newinstance' option create a new, private namespace. 430 * 431 * NOTE: 432 * 433 * For single-mount semantics, devpts cannot use mount_single(), 434 * because mount_single()/sget() find and use the super-block from 435 * the most recent mount of devpts. But that recent mount may be a 436 * 'newinstance' mount and mount_single() would pick the newinstance 437 * super-block instead of the initial super-block. 438 */ 439 static struct dentry *devpts_mount(struct file_system_type *fs_type, 440 int flags, const char *dev_name, void *data) 441 { 442 int error; 443 struct pts_mount_opts opts; 444 struct super_block *s; 445 446 error = parse_mount_options(data, PARSE_MOUNT, &opts); 447 if (error) 448 return ERR_PTR(error); 449 450 /* Require newinstance for all user namespace mounts to ensure 451 * the mount options are not changed. 452 */ 453 if ((current_user_ns() != &init_user_ns) && !opts.newinstance) 454 return ERR_PTR(-EINVAL); 455 456 if (opts.newinstance) 457 s = sget(fs_type, NULL, set_anon_super, flags, NULL); 458 else 459 s = sget(fs_type, compare_init_pts_sb, set_anon_super, flags, 460 NULL); 461 462 if (IS_ERR(s)) 463 return ERR_CAST(s); 464 465 if (!s->s_root) { 466 error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0); 467 if (error) 468 goto out_undo_sget; 469 s->s_flags |= MS_ACTIVE; 470 } 471 472 memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts)); 473 474 error = mknod_ptmx(s); 475 if (error) 476 goto out_undo_sget; 477 478 return dget(s->s_root); 479 480 out_undo_sget: 481 deactivate_locked_super(s); 482 return ERR_PTR(error); 483 } 484 485 #else 486 /* 487 * This supports only the legacy single-instance semantics (no 488 * multiple-instance semantics) 489 */ 490 static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags, 491 const char *dev_name, void *data) 492 { 493 return mount_single(fs_type, flags, data, devpts_fill_super); 494 } 495 #endif 496 497 static void devpts_kill_sb(struct super_block *sb) 498 { 499 struct pts_fs_info *fsi = DEVPTS_SB(sb); 500 501 kfree(fsi); 502 kill_litter_super(sb); 503 } 504 505 static struct file_system_type devpts_fs_type = { 506 .name = "devpts", 507 .mount = devpts_mount, 508 .kill_sb = devpts_kill_sb, 509 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES 510 .fs_flags = FS_USERNS_MOUNT | FS_USERNS_DEV_MOUNT, 511 #endif 512 }; 513 514 /* 515 * The normal naming convention is simply /dev/pts/<number>; this conforms 516 * to the System V naming convention 517 */ 518 519 int devpts_new_index(struct inode *ptmx_inode) 520 { 521 struct super_block *sb = pts_sb_from_inode(ptmx_inode); 522 struct pts_fs_info *fsi = DEVPTS_SB(sb); 523 int index; 524 int ida_ret; 525 526 retry: 527 if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL)) 528 return -ENOMEM; 529 530 mutex_lock(&allocated_ptys_lock); 531 if (pty_count >= pty_limit - 532 (fsi->mount_opts.newinstance ? pty_reserve : 0)) { 533 mutex_unlock(&allocated_ptys_lock); 534 return -ENOSPC; 535 } 536 537 ida_ret = ida_get_new(&fsi->allocated_ptys, &index); 538 if (ida_ret < 0) { 539 mutex_unlock(&allocated_ptys_lock); 540 if (ida_ret == -EAGAIN) 541 goto retry; 542 return -EIO; 543 } 544 545 if (index >= fsi->mount_opts.max) { 546 ida_remove(&fsi->allocated_ptys, index); 547 mutex_unlock(&allocated_ptys_lock); 548 return -ENOSPC; 549 } 550 pty_count++; 551 mutex_unlock(&allocated_ptys_lock); 552 return index; 553 } 554 555 void devpts_kill_index(struct inode *ptmx_inode, int idx) 556 { 557 struct super_block *sb = pts_sb_from_inode(ptmx_inode); 558 struct pts_fs_info *fsi = DEVPTS_SB(sb); 559 560 mutex_lock(&allocated_ptys_lock); 561 ida_remove(&fsi->allocated_ptys, idx); 562 pty_count--; 563 mutex_unlock(&allocated_ptys_lock); 564 } 565 566 /** 567 * devpts_pty_new -- create a new inode in /dev/pts/ 568 * @ptmx_inode: inode of the master 569 * @device: major+minor of the node to be created 570 * @index: used as a name of the node 571 * @priv: what's given back by devpts_get_priv 572 * 573 * The created inode is returned. Remove it from /dev/pts/ by devpts_pty_kill. 574 */ 575 struct inode *devpts_pty_new(struct inode *ptmx_inode, dev_t device, int index, 576 void *priv) 577 { 578 struct dentry *dentry; 579 struct super_block *sb = pts_sb_from_inode(ptmx_inode); 580 struct inode *inode; 581 struct dentry *root = sb->s_root; 582 struct pts_fs_info *fsi = DEVPTS_SB(sb); 583 struct pts_mount_opts *opts = &fsi->mount_opts; 584 char s[12]; 585 586 inode = new_inode(sb); 587 if (!inode) 588 return ERR_PTR(-ENOMEM); 589 590 inode->i_ino = index + 3; 591 inode->i_uid = opts->setuid ? opts->uid : current_fsuid(); 592 inode->i_gid = opts->setgid ? opts->gid : current_fsgid(); 593 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 594 init_special_inode(inode, S_IFCHR|opts->mode, device); 595 inode->i_private = priv; 596 597 sprintf(s, "%d", index); 598 599 mutex_lock(&root->d_inode->i_mutex); 600 601 dentry = d_alloc_name(root, s); 602 if (dentry) { 603 d_add(dentry, inode); 604 fsnotify_create(root->d_inode, dentry); 605 } else { 606 iput(inode); 607 inode = ERR_PTR(-ENOMEM); 608 } 609 610 mutex_unlock(&root->d_inode->i_mutex); 611 612 return inode; 613 } 614 615 /** 616 * devpts_get_priv -- get private data for a slave 617 * @pts_inode: inode of the slave 618 * 619 * Returns whatever was passed as priv in devpts_pty_new for a given inode. 620 */ 621 void *devpts_get_priv(struct inode *pts_inode) 622 { 623 struct dentry *dentry; 624 void *priv = NULL; 625 626 BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR)); 627 628 /* Ensure dentry has not been deleted by devpts_pty_kill() */ 629 dentry = d_find_alias(pts_inode); 630 if (!dentry) 631 return NULL; 632 633 if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC) 634 priv = pts_inode->i_private; 635 636 dput(dentry); 637 638 return priv; 639 } 640 641 /** 642 * devpts_pty_kill -- remove inode form /dev/pts/ 643 * @inode: inode of the slave to be removed 644 * 645 * This is an inverse operation of devpts_pty_new. 646 */ 647 void devpts_pty_kill(struct inode *inode) 648 { 649 struct super_block *sb = pts_sb_from_inode(inode); 650 struct dentry *root = sb->s_root; 651 struct dentry *dentry; 652 653 BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR)); 654 655 mutex_lock(&root->d_inode->i_mutex); 656 657 dentry = d_find_alias(inode); 658 659 drop_nlink(inode); 660 d_delete(dentry); 661 dput(dentry); /* d_alloc_name() in devpts_pty_new() */ 662 dput(dentry); /* d_find_alias above */ 663 664 mutex_unlock(&root->d_inode->i_mutex); 665 } 666 667 static int __init init_devpts_fs(void) 668 { 669 int err = register_filesystem(&devpts_fs_type); 670 struct ctl_table_header *table; 671 672 if (!err) { 673 table = register_sysctl_table(pty_root_table); 674 devpts_mnt = kern_mount(&devpts_fs_type); 675 if (IS_ERR(devpts_mnt)) { 676 err = PTR_ERR(devpts_mnt); 677 unregister_filesystem(&devpts_fs_type); 678 unregister_sysctl_table(table); 679 } 680 } 681 return err; 682 } 683 module_init(init_devpts_fs) 684