binfmt_misc.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * binfmt_misc.c
   4 *
   5 * Copyright (C) 1997 Richard Günther
   6 *
   7 * binfmt_misc detects binaries via a magic or filename extension and invokes
   8 * a specified wrapper. See Documentation/admin-guide/binfmt-misc.rst for more details.
   9 */
  10
  11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13#include <linux/kernel.h>
  14#include <linux/module.h>
  15#include <linux/init.h>
  16#include <linux/sched/mm.h>
  17#include <linux/magic.h>
  18#include <linux/binfmts.h>
  19#include <linux/slab.h>
  20#include <linux/ctype.h>
  21#include <linux/string_helpers.h>
  22#include <linux/file.h>
  23#include <linux/pagemap.h>
  24#include <linux/namei.h>
  25#include <linux/mount.h>
  26#include <linux/fs_context.h>
  27#include <linux/syscalls.h>
  28#include <linux/fs.h>
  29#include <linux/uaccess.h>
  30
  31#include "internal.h"
  32
  33#ifdef DEBUG
  34# define USE_DEBUG 1
  35#else
  36# define USE_DEBUG 0
  37#endif
  38
  39enum {
  40	VERBOSE_STATUS = 1 /* make it zero to save 400 bytes kernel memory */
  41};
  42
  43enum {Enabled, Magic};
  44#define MISC_FMT_PRESERVE_ARGV0 (1UL << 31)
  45#define MISC_FMT_OPEN_BINARY (1UL << 30)
  46#define MISC_FMT_CREDENTIALS (1UL << 29)
  47#define MISC_FMT_OPEN_FILE (1UL << 28)
  48
  49typedef struct {
  50	struct list_head list;
  51	unsigned long flags;		/* type, status, etc. */
  52	int offset;			/* offset of magic */
  53	int size;			/* size of magic/mask */
  54	char *magic;			/* magic or filename extension */
  55	char *mask;			/* mask, NULL for exact match */
  56	const char *interpreter;	/* filename of interpreter */
  57	char *name;
  58	struct dentry *dentry;
  59	struct file *interp_file;
  60	refcount_t users;		/* sync removal with load_misc_binary() */
  61} Node;
  62
  63static struct file_system_type bm_fs_type;
  64
  65/*
  66 * Max length of the register string.  Determined by:
  67 *  - 7 delimiters
  68 *  - name:   ~50 bytes
  69 *  - type:   1 byte
  70 *  - offset: 3 bytes (has to be smaller than BINPRM_BUF_SIZE)
  71 *  - magic:  128 bytes (512 in escaped form)
  72 *  - mask:   128 bytes (512 in escaped form)
  73 *  - interp: ~50 bytes
  74 *  - flags:  5 bytes
  75 * Round that up a bit, and then back off to hold the internal data
  76 * (like struct Node).
  77 */
  78#define MAX_REGISTER_LENGTH 1920
  79
  80/**
  81 * search_binfmt_handler - search for a binary handler for @bprm
  82 * @misc: handle to binfmt_misc instance
  83 * @bprm: binary for which we are looking for a handler
  84 *
  85 * Search for a binary type handler for @bprm in the list of registered binary
  86 * type handlers.
  87 *
  88 * Return: binary type list entry on success, NULL on failure
  89 */
  90static Node *search_binfmt_handler(struct binfmt_misc *misc,
  91				   struct linux_binprm *bprm)
  92{
  93	char *p = strrchr(bprm->interp, '.');
  94	Node *e;
  95
  96	/* Walk all the registered handlers. */
  97	list_for_each_entry(e, &misc->entries, list) {
  98		char *s;
  99		int j;
 100
 101		/* Make sure this one is currently enabled. */
 102		if (!test_bit(Enabled, &e->flags))
 103			continue;
 104
 105		/* Do matching based on extension if applicable. */
 106		if (!test_bit(Magic, &e->flags)) {
 107			if (p && !strcmp(e->magic, p + 1))
 108				return e;
 109			continue;
 110		}
 111
 112		/* Do matching based on magic & mask. */
 113		s = bprm->buf + e->offset;
 114		if (e->mask) {
 115			for (j = 0; j < e->size; j++)
 116				if ((*s++ ^ e->magic[j]) & e->mask[j])
 117					break;
 118		} else {
 119			for (j = 0; j < e->size; j++)
 120				if ((*s++ ^ e->magic[j]))
 121					break;
 122		}
 123		if (j == e->size)
 124			return e;
 125	}
 126
 127	return NULL;
 128}
 129
 130/**
 131 * get_binfmt_handler - try to find a binary type handler
 132 * @misc: handle to binfmt_misc instance
 133 * @bprm: binary for which we are looking for a handler
 134 *
 135 * Try to find a binfmt handler for the binary type. If one is found take a
 136 * reference to protect against removal via bm_{entry,status}_write().
 137 *
 138 * Return: binary type list entry on success, NULL on failure
 139 */
 140static Node *get_binfmt_handler(struct binfmt_misc *misc,
 141				struct linux_binprm *bprm)
 142{
 143	Node *e;
 144
 145	read_lock(&misc->entries_lock);
 146	e = search_binfmt_handler(misc, bprm);
 147	if (e)
 148		refcount_inc(&e->users);
 149	read_unlock(&misc->entries_lock);
 150	return e;
 151}
 152
 153/**
 154 * put_binfmt_handler - put binary handler node
 155 * @e: node to put
 156 *
 157 * Free node syncing with load_misc_binary() and defer final free to
 158 * load_misc_binary() in case it is using the binary type handler we were
 159 * requested to remove.
 160 */
 161static void put_binfmt_handler(Node *e)
 162{
 163	if (refcount_dec_and_test(&e->users)) {
 164		if (e->flags & MISC_FMT_OPEN_FILE)
 165			filp_close(e->interp_file, NULL);
 166		kfree(e);
 167	}
 168}
 169
 170/**
 171 * load_binfmt_misc - load the binfmt_misc of the caller's user namespace
 172 *
 173 * To be called in load_misc_binary() to load the relevant struct binfmt_misc.
 174 * If a user namespace doesn't have its own binfmt_misc mount it can make use
 175 * of its ancestor's binfmt_misc handlers. This mimicks the behavior of
 176 * pre-namespaced binfmt_misc where all registered binfmt_misc handlers where
 177 * available to all user and user namespaces on the system.
 178 *
 179 * Return: the binfmt_misc instance of the caller's user namespace
 180 */
 181static struct binfmt_misc *load_binfmt_misc(void)
 182{
 183	const struct user_namespace *user_ns;
 184	struct binfmt_misc *misc;
 185
 186	user_ns = current_user_ns();
 187	while (user_ns) {
 188		/* Pairs with smp_store_release() in bm_fill_super(). */
 189		misc = smp_load_acquire(&user_ns->binfmt_misc);
 190		if (misc)
 191			return misc;
 192
 193		user_ns = user_ns->parent;
 194	}
 195
 196	return &init_binfmt_misc;
 197}
 198
 199/*
 200 * the loader itself
 201 */
 202static int load_misc_binary(struct linux_binprm *bprm)
 203{
 204	Node *fmt;
 205	struct file *interp_file = NULL;
 206	int retval = -ENOEXEC;
 207	struct binfmt_misc *misc;
 208
 209	misc = load_binfmt_misc();
 210	if (!misc->enabled)
 211		return retval;
 212
 213	fmt = get_binfmt_handler(misc, bprm);
 214	if (!fmt)
 215		return retval;
 216
 217	/* Need to be able to load the file after exec */
 218	retval = -ENOENT;
 219	if (bprm->interp_flags & BINPRM_FLAGS_PATH_INACCESSIBLE)
 220		goto ret;
 221
 222	if (fmt->flags & MISC_FMT_PRESERVE_ARGV0) {
 223		bprm->interp_flags |= BINPRM_FLAGS_PRESERVE_ARGV0;
 224	} else {
 225		retval = remove_arg_zero(bprm);
 226		if (retval)
 227			goto ret;
 228	}
 229
 230	if (fmt->flags & MISC_FMT_OPEN_BINARY)
 231		bprm->have_execfd = 1;
 232
 233	/* make argv[1] be the path to the binary */
 234	retval = copy_string_kernel(bprm->interp, bprm);
 235	if (retval < 0)
 236		goto ret;
 237	bprm->argc++;
 238
 239	/* add the interp as argv[0] */
 240	retval = copy_string_kernel(fmt->interpreter, bprm);
 241	if (retval < 0)
 242		goto ret;
 243	bprm->argc++;
 244
 245	/* Update interp in case binfmt_script needs it. */
 246	retval = bprm_change_interp(fmt->interpreter, bprm);
 247	if (retval < 0)
 248		goto ret;
 249
 250	if (fmt->flags & MISC_FMT_OPEN_FILE) {
 251		interp_file = file_clone_open(fmt->interp_file);
 252		if (!IS_ERR(interp_file))
 253			deny_write_access(interp_file);
 254	} else {
 255		interp_file = open_exec(fmt->interpreter);
 256	}
 257	retval = PTR_ERR(interp_file);
 258	if (IS_ERR(interp_file))
 259		goto ret;
 260
 261	bprm->interpreter = interp_file;
 262	if (fmt->flags & MISC_FMT_CREDENTIALS)
 263		bprm->execfd_creds = 1;
 264
 265	retval = 0;
 266ret:
 267
 268	/*
 269	 * If we actually put the node here all concurrent calls to
 270	 * load_misc_binary() will have finished. We also know
 271	 * that for the refcount to be zero someone must have concurently
 272	 * removed the binary type handler from the list and it's our job to
 273	 * free it.
 274	 */
 275	put_binfmt_handler(fmt);
 276
 277	return retval;
 278}
 279
 280/* Command parsers */
 281
 282/*
 283 * parses and copies one argument enclosed in del from *sp to *dp,
 284 * recognising the \x special.
 285 * returns pointer to the copied argument or NULL in case of an
 286 * error (and sets err) or null argument length.
 287 */
 288static char *scanarg(char *s, char del)
 289{
 290	char c;
 291
 292	while ((c = *s++) != del) {
 293		if (c == '\\' && *s == 'x') {
 294			s++;
 295			if (!isxdigit(*s++))
 296				return NULL;
 297			if (!isxdigit(*s++))
 298				return NULL;
 299		}
 300	}
 301	s[-1] ='\0';
 302	return s;
 303}
 304
 305static char *check_special_flags(char *sfs, Node *e)
 306{
 307	char *p = sfs;
 308	int cont = 1;
 309
 310	/* special flags */
 311	while (cont) {
 312		switch (*p) {
 313		case 'P':
 314			pr_debug("register: flag: P (preserve argv0)\n");
 315			p++;
 316			e->flags |= MISC_FMT_PRESERVE_ARGV0;
 317			break;
 318		case 'O':
 319			pr_debug("register: flag: O (open binary)\n");
 320			p++;
 321			e->flags |= MISC_FMT_OPEN_BINARY;
 322			break;
 323		case 'C':
 324			pr_debug("register: flag: C (preserve creds)\n");
 325			p++;
 326			/* this flags also implies the
 327			   open-binary flag */
 328			e->flags |= (MISC_FMT_CREDENTIALS |
 329					MISC_FMT_OPEN_BINARY);
 330			break;
 331		case 'F':
 332			pr_debug("register: flag: F: open interpreter file now\n");
 333			p++;
 334			e->flags |= MISC_FMT_OPEN_FILE;
 335			break;
 336		default:
 337			cont = 0;
 338		}
 339	}
 340
 341	return p;
 342}
 343
 344/*
 345 * This registers a new binary format, it recognises the syntax
 346 * ':name:type:offset:magic:mask:interpreter:flags'
 347 * where the ':' is the IFS, that can be chosen with the first char
 348 */
 349static Node *create_entry(const char __user *buffer, size_t count)
 350{
 351	Node *e;
 352	int memsize, err;
 353	char *buf, *p;
 354	char del;
 355
 356	pr_debug("register: received %zu bytes\n", count);
 357
 358	/* some sanity checks */
 359	err = -EINVAL;
 360	if ((count < 11) || (count > MAX_REGISTER_LENGTH))
 361		goto out;
 362
 363	err = -ENOMEM;
 364	memsize = sizeof(Node) + count + 8;
 365	e = kmalloc(memsize, GFP_KERNEL_ACCOUNT);
 366	if (!e)
 367		goto out;
 368
 369	p = buf = (char *)e + sizeof(Node);
 370
 371	memset(e, 0, sizeof(Node));
 372	if (copy_from_user(buf, buffer, count))
 373		goto efault;
 374
 375	del = *p++;	/* delimeter */
 376
 377	pr_debug("register: delim: %#x {%c}\n", del, del);
 378
 379	/* Pad the buffer with the delim to simplify parsing below. */
 380	memset(buf + count, del, 8);
 381
 382	/* Parse the 'name' field. */
 383	e->name = p;
 384	p = strchr(p, del);
 385	if (!p)
 386		goto einval;
 387	*p++ = '\0';
 388	if (!e->name[0] ||
 389	    !strcmp(e->name, ".") ||
 390	    !strcmp(e->name, "..") ||
 391	    strchr(e->name, '/'))
 392		goto einval;
 393
 394	pr_debug("register: name: {%s}\n", e->name);
 395
 396	/* Parse the 'type' field. */
 397	switch (*p++) {
 398	case 'E':
 399		pr_debug("register: type: E (extension)\n");
 400		e->flags = 1 << Enabled;
 401		break;
 402	case 'M':
 403		pr_debug("register: type: M (magic)\n");
 404		e->flags = (1 << Enabled) | (1 << Magic);
 405		break;
 406	default:
 407		goto einval;
 408	}
 409	if (*p++ != del)
 410		goto einval;
 411
 412	if (test_bit(Magic, &e->flags)) {
 413		/* Handle the 'M' (magic) format. */
 414		char *s;
 415
 416		/* Parse the 'offset' field. */
 417		s = strchr(p, del);
 418		if (!s)
 419			goto einval;
 420		*s = '\0';
 421		if (p != s) {
 422			int r = kstrtoint(p, 10, &e->offset);
 423			if (r != 0 || e->offset < 0)
 424				goto einval;
 425		}
 426		p = s;
 427		if (*p++)
 428			goto einval;
 429		pr_debug("register: offset: %#x\n", e->offset);
 430
 431		/* Parse the 'magic' field. */
 432		e->magic = p;
 433		p = scanarg(p, del);
 434		if (!p)
 435			goto einval;
 436		if (!e->magic[0])
 437			goto einval;
 438		if (USE_DEBUG)
 439			print_hex_dump_bytes(
 440				KBUILD_MODNAME ": register: magic[raw]: ",
 441				DUMP_PREFIX_NONE, e->magic, p - e->magic);
 442
 443		/* Parse the 'mask' field. */
 444		e->mask = p;
 445		p = scanarg(p, del);
 446		if (!p)
 447			goto einval;
 448		if (!e->mask[0]) {
 449			e->mask = NULL;
 450			pr_debug("register:  mask[raw]: none\n");
 451		} else if (USE_DEBUG)
 452			print_hex_dump_bytes(
 453				KBUILD_MODNAME ": register:  mask[raw]: ",
 454				DUMP_PREFIX_NONE, e->mask, p - e->mask);
 455
 456		/*
 457		 * Decode the magic & mask fields.
 458		 * Note: while we might have accepted embedded NUL bytes from
 459		 * above, the unescape helpers here will stop at the first one
 460		 * it encounters.
 461		 */
 462		e->size = string_unescape_inplace(e->magic, UNESCAPE_HEX);
 463		if (e->mask &&
 464		    string_unescape_inplace(e->mask, UNESCAPE_HEX) != e->size)
 465			goto einval;
 466		if (e->size > BINPRM_BUF_SIZE ||
 467		    BINPRM_BUF_SIZE - e->size < e->offset)
 468			goto einval;
 469		pr_debug("register: magic/mask length: %i\n", e->size);
 470		if (USE_DEBUG) {
 471			print_hex_dump_bytes(
 472				KBUILD_MODNAME ": register: magic[decoded]: ",
 473				DUMP_PREFIX_NONE, e->magic, e->size);
 474
 475			if (e->mask) {
 476				int i;
 477				char *masked = kmalloc(e->size, GFP_KERNEL_ACCOUNT);
 478
 479				print_hex_dump_bytes(
 480					KBUILD_MODNAME ": register:  mask[decoded]: ",
 481					DUMP_PREFIX_NONE, e->mask, e->size);
 482
 483				if (masked) {
 484					for (i = 0; i < e->size; ++i)
 485						masked[i] = e->magic[i] & e->mask[i];
 486					print_hex_dump_bytes(
 487						KBUILD_MODNAME ": register:  magic[masked]: ",
 488						DUMP_PREFIX_NONE, masked, e->size);
 489
 490					kfree(masked);
 491				}
 492			}
 493		}
 494	} else {
 495		/* Handle the 'E' (extension) format. */
 496
 497		/* Skip the 'offset' field. */
 498		p = strchr(p, del);
 499		if (!p)
 500			goto einval;
 501		*p++ = '\0';
 502
 503		/* Parse the 'magic' field. */
 504		e->magic = p;
 505		p = strchr(p, del);
 506		if (!p)
 507			goto einval;
 508		*p++ = '\0';
 509		if (!e->magic[0] || strchr(e->magic, '/'))
 510			goto einval;
 511		pr_debug("register: extension: {%s}\n", e->magic);
 512
 513		/* Skip the 'mask' field. */
 514		p = strchr(p, del);
 515		if (!p)
 516			goto einval;
 517		*p++ = '\0';
 518	}
 519
 520	/* Parse the 'interpreter' field. */
 521	e->interpreter = p;
 522	p = strchr(p, del);
 523	if (!p)
 524		goto einval;
 525	*p++ = '\0';
 526	if (!e->interpreter[0])
 527		goto einval;
 528	pr_debug("register: interpreter: {%s}\n", e->interpreter);
 529
 530	/* Parse the 'flags' field. */
 531	p = check_special_flags(p, e);
 532	if (*p == '\n')
 533		p++;
 534	if (p != buf + count)
 535		goto einval;
 536
 537	return e;
 538
 539out:
 540	return ERR_PTR(err);
 541
 542efault:
 543	kfree(e);
 544	return ERR_PTR(-EFAULT);
 545einval:
 546	kfree(e);
 547	return ERR_PTR(-EINVAL);
 548}
 549
 550/*
 551 * Set status of entry/binfmt_misc:
 552 * '1' enables, '0' disables and '-1' clears entry/binfmt_misc
 553 */
 554static int parse_command(const char __user *buffer, size_t count)
 555{
 556	char s[4];
 557
 558	if (count > 3)
 559		return -EINVAL;
 560	if (copy_from_user(s, buffer, count))
 561		return -EFAULT;
 562	if (!count)
 563		return 0;
 564	if (s[count - 1] == '\n')
 565		count--;
 566	if (count == 1 && s[0] == '0')
 567		return 1;
 568	if (count == 1 && s[0] == '1')
 569		return 2;
 570	if (count == 2 && s[0] == '-' && s[1] == '1')
 571		return 3;
 572	return -EINVAL;
 573}
 574
 575/* generic stuff */
 576
 577static void entry_status(Node *e, char *page)
 578{
 579	char *dp = page;
 580	const char *status = "disabled";
 581
 582	if (test_bit(Enabled, &e->flags))
 583		status = "enabled";
 584
 585	if (!VERBOSE_STATUS) {
 586		sprintf(page, "%s\n", status);
 587		return;
 588	}
 589
 590	dp += sprintf(dp, "%s\ninterpreter %s\n", status, e->interpreter);
 591
 592	/* print the special flags */
 593	dp += sprintf(dp, "flags: ");
 594	if (e->flags & MISC_FMT_PRESERVE_ARGV0)
 595		*dp++ = 'P';
 596	if (e->flags & MISC_FMT_OPEN_BINARY)
 597		*dp++ = 'O';
 598	if (e->flags & MISC_FMT_CREDENTIALS)
 599		*dp++ = 'C';
 600	if (e->flags & MISC_FMT_OPEN_FILE)
 601		*dp++ = 'F';
 602	*dp++ = '\n';
 603
 604	if (!test_bit(Magic, &e->flags)) {
 605		sprintf(dp, "extension .%s\n", e->magic);
 606	} else {
 607		dp += sprintf(dp, "offset %i\nmagic ", e->offset);
 608		dp = bin2hex(dp, e->magic, e->size);
 609		if (e->mask) {
 610			dp += sprintf(dp, "\nmask ");
 611			dp = bin2hex(dp, e->mask, e->size);
 612		}
 613		*dp++ = '\n';
 614		*dp = '\0';
 615	}
 616}
 617
 618static struct inode *bm_get_inode(struct super_block *sb, int mode)
 619{
 620	struct inode *inode = new_inode(sb);
 621
 622	if (inode) {
 623		inode->i_ino = get_next_ino();
 624		inode->i_mode = mode;
 625		simple_inode_init_ts(inode);
 626	}
 627	return inode;
 628}
 629
 630/**
 631 * i_binfmt_misc - retrieve struct binfmt_misc from a binfmt_misc inode
 632 * @inode: inode of the relevant binfmt_misc instance
 633 *
 634 * This helper retrieves struct binfmt_misc from a binfmt_misc inode. This can
 635 * be done without any memory barriers because we are guaranteed that
 636 * user_ns->binfmt_misc is fully initialized. It was fully initialized when the
 637 * binfmt_misc mount was first created.
 638 *
 639 * Return: struct binfmt_misc of the relevant binfmt_misc instance
 640 */
 641static struct binfmt_misc *i_binfmt_misc(struct inode *inode)
 642{
 643	return inode->i_sb->s_user_ns->binfmt_misc;
 644}
 645
 646/**
 647 * bm_evict_inode - cleanup data associated with @inode
 648 * @inode: inode to which the data is attached
 649 *
 650 * Cleanup the binary type handler data associated with @inode if a binary type
 651 * entry is removed or the filesystem is unmounted and the super block is
 652 * shutdown.
 653 *
 654 * If the ->evict call was not caused by a super block shutdown but by a write
 655 * to remove the entry or all entries via bm_{entry,status}_write() the entry
 656 * will have already been removed from the list. We keep the list_empty() check
 657 * to make that explicit.
 658*/
 659static void bm_evict_inode(struct inode *inode)
 660{
 661	Node *e = inode->i_private;
 662
 663	clear_inode(inode);
 664
 665	if (e) {
 666		struct binfmt_misc *misc;
 667
 668		misc = i_binfmt_misc(inode);
 669		write_lock(&misc->entries_lock);
 670		if (!list_empty(&e->list))
 671			list_del_init(&e->list);
 672		write_unlock(&misc->entries_lock);
 673		put_binfmt_handler(e);
 674	}
 675}
 676
 677/**
 678 * remove_binfmt_handler - remove a binary type handler
 679 * @misc: handle to binfmt_misc instance
 680 * @e: binary type handler to remove
 681 *
 682 * Remove a binary type handler from the list of binary type handlers and
 683 * remove its associated dentry. This is called from
 684 * binfmt_{entry,status}_write(). In the future, we might want to think about
 685 * adding a proper ->unlink() method to binfmt_misc instead of forcing caller's
 686 * to use writes to files in order to delete binary type handlers. But it has
 687 * worked for so long that it's not a pressing issue.
 688 */
 689static void remove_binfmt_handler(struct binfmt_misc *misc, Node *e)
 690{
 691	write_lock(&misc->entries_lock);
 692	list_del_init(&e->list);
 693	write_unlock(&misc->entries_lock);
 694	locked_recursive_removal(e->dentry, NULL);
 695}
 696
 697/* /<entry> */
 698
 699static ssize_t
 700bm_entry_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
 701{
 702	Node *e = file_inode(file)->i_private;
 703	ssize_t res;
 704	char *page;
 705
 706	page = (char *) __get_free_page(GFP_KERNEL);
 707	if (!page)
 708		return -ENOMEM;
 709
 710	entry_status(e, page);
 711
 712	res = simple_read_from_buffer(buf, nbytes, ppos, page, strlen(page));
 713
 714	free_page((unsigned long) page);
 715	return res;
 716}
 717
 718static ssize_t bm_entry_write(struct file *file, const char __user *buffer,
 719				size_t count, loff_t *ppos)
 720{
 721	struct inode *inode = file_inode(file);
 722	Node *e = inode->i_private;
 723	int res = parse_command(buffer, count);
 724
 725	switch (res) {
 726	case 1:
 727		/* Disable this handler. */
 728		clear_bit(Enabled, &e->flags);
 729		break;
 730	case 2:
 731		/* Enable this handler. */
 732		set_bit(Enabled, &e->flags);
 733		break;
 734	case 3:
 735		/* Delete this handler. */
 736		inode = d_inode(inode->i_sb->s_root);
 737		inode_lock_nested(inode, I_MUTEX_PARENT);
 738
 739		/*
 740		 * In order to add new element or remove elements from the list
 741		 * via bm_{entry,register,status}_write() inode_lock() on the
 742		 * root inode must be held.
 743		 * The lock is exclusive ensuring that the list can't be
 744		 * modified. Only load_misc_binary() can access but does so
 745		 * read-only. So we only need to take the write lock when we
 746		 * actually remove the entry from the list.
 747		 */
 748		if (!list_empty(&e->list))
 749			remove_binfmt_handler(i_binfmt_misc(inode), e);
 750
 751		inode_unlock(inode);
 752		break;
 753	default:
 754		return res;
 755	}
 756
 757	return count;
 758}
 759
 760static const struct file_operations bm_entry_operations = {
 761	.read		= bm_entry_read,
 762	.write		= bm_entry_write,
 763	.llseek		= default_llseek,
 764};
 765
 766/* /register */
 767
 768/* add to filesystem */
 769static int add_entry(Node *e, struct super_block *sb)
 770{
 771	struct dentry *dentry = simple_start_creating(sb->s_root, e->name);
 772	struct inode *inode;
 773	struct binfmt_misc *misc;
 774
 775	if (IS_ERR(dentry))
 776		return PTR_ERR(dentry);
 777
 778	inode = bm_get_inode(sb, S_IFREG | 0644);
 779	if (unlikely(!inode)) {
 780		simple_done_creating(dentry);
 781		return -ENOMEM;
 782	}
 783
 784	refcount_set(&e->users, 1);
 785	e->dentry = dentry;
 786	inode->i_private = e;
 787	inode->i_fop = &bm_entry_operations;
 788
 789	d_make_persistent(dentry, inode);
 790	misc = i_binfmt_misc(inode);
 791	write_lock(&misc->entries_lock);
 792	list_add(&e->list, &misc->entries);
 793	write_unlock(&misc->entries_lock);
 794	simple_done_creating(dentry);
 795	return 0;
 796}
 797
 798static ssize_t bm_register_write(struct file *file, const char __user *buffer,
 799			       size_t count, loff_t *ppos)
 800{
 801	Node *e;
 802	struct super_block *sb = file_inode(file)->i_sb;
 803	int err = 0;
 804	struct file *f = NULL;
 805
 806	e = create_entry(buffer, count);
 807
 808	if (IS_ERR(e))
 809		return PTR_ERR(e);
 810
 811	if (e->flags & MISC_FMT_OPEN_FILE) {
 812		/*
 813		 * Now that we support unprivileged binfmt_misc mounts make
 814		 * sure we use the credentials that the register @file was
 815		 * opened with to also open the interpreter. Before that this
 816		 * didn't matter much as only a privileged process could open
 817		 * the register file.
 818		 */
 819		scoped_with_creds(file->f_cred)
 820			f = open_exec(e->interpreter);
 821		if (IS_ERR(f)) {
 822			pr_notice("register: failed to install interpreter file %s\n",
 823				 e->interpreter);
 824			kfree(e);
 825			return PTR_ERR(f);
 826		}
 827		e->interp_file = f;
 828	}
 829
 830	err = add_entry(e, sb);
 831	if (err) {
 832		if (f) {
 833			exe_file_allow_write_access(f);
 834			filp_close(f, NULL);
 835		}
 836		kfree(e);
 837		return err;
 838	}
 839	return count;
 840}
 841
 842static const struct file_operations bm_register_operations = {
 843	.write		= bm_register_write,
 844	.llseek		= noop_llseek,
 845};
 846
 847/* /status */
 848
 849static ssize_t
 850bm_status_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
 851{
 852	struct binfmt_misc *misc;
 853	char *s;
 854
 855	misc = i_binfmt_misc(file_inode(file));
 856	s = misc->enabled ? "enabled\n" : "disabled\n";
 857	return simple_read_from_buffer(buf, nbytes, ppos, s, strlen(s));
 858}
 859
 860static ssize_t bm_status_write(struct file *file, const char __user *buffer,
 861		size_t count, loff_t *ppos)
 862{
 863	struct binfmt_misc *misc;
 864	int res = parse_command(buffer, count);
 865	Node *e, *next;
 866	struct inode *inode;
 867
 868	misc = i_binfmt_misc(file_inode(file));
 869	switch (res) {
 870	case 1:
 871		/* Disable all handlers. */
 872		misc->enabled = false;
 873		break;
 874	case 2:
 875		/* Enable all handlers. */
 876		misc->enabled = true;
 877		break;
 878	case 3:
 879		/* Delete all handlers. */
 880		inode = d_inode(file_inode(file)->i_sb->s_root);
 881		inode_lock_nested(inode, I_MUTEX_PARENT);
 882
 883		/*
 884		 * In order to add new element or remove elements from the list
 885		 * via bm_{entry,register,status}_write() inode_lock() on the
 886		 * root inode must be held.
 887		 * The lock is exclusive ensuring that the list can't be
 888		 * modified. Only load_misc_binary() can access but does so
 889		 * read-only. So we only need to take the write lock when we
 890		 * actually remove the entry from the list.
 891		 */
 892		list_for_each_entry_safe(e, next, &misc->entries, list)
 893			remove_binfmt_handler(misc, e);
 894
 895		inode_unlock(inode);
 896		break;
 897	default:
 898		return res;
 899	}
 900
 901	return count;
 902}
 903
 904static const struct file_operations bm_status_operations = {
 905	.read		= bm_status_read,
 906	.write		= bm_status_write,
 907	.llseek		= default_llseek,
 908};
 909
 910/* Superblock handling */
 911
 912static void bm_put_super(struct super_block *sb)
 913{
 914	struct user_namespace *user_ns = sb->s_fs_info;
 915
 916	sb->s_fs_info = NULL;
 917	put_user_ns(user_ns);
 918}
 919
 920static const struct super_operations s_ops = {
 921	.statfs		= simple_statfs,
 922	.evict_inode	= bm_evict_inode,
 923	.put_super	= bm_put_super,
 924};
 925
 926static int bm_fill_super(struct super_block *sb, struct fs_context *fc)
 927{
 928	int err;
 929	struct user_namespace *user_ns = sb->s_user_ns;
 930	struct binfmt_misc *misc;
 931	static const struct tree_descr bm_files[] = {
 932		[2] = {"status", &bm_status_operations, S_IWUSR|S_IRUGO},
 933		[3] = {"register", &bm_register_operations, S_IWUSR},
 934		/* last one */ {""}
 935	};
 936
 937	if (WARN_ON(user_ns != current_user_ns()))
 938		return -EINVAL;
 939
 940	/*
 941	 * Lazily allocate a new binfmt_misc instance for this namespace, i.e.
 942	 * do it here during the first mount of binfmt_misc. We don't need to
 943	 * waste memory for every user namespace allocation. It's likely much
 944	 * more common to not mount a separate binfmt_misc instance than it is
 945	 * to mount one.
 946	 *
 947	 * While multiple superblocks can exist they are keyed by userns in
 948	 * s_fs_info for binfmt_misc. Hence, the vfs guarantees that
 949	 * bm_fill_super() is called exactly once whenever a binfmt_misc
 950	 * superblock for a userns is created. This in turn lets us conclude
 951	 * that when a binfmt_misc superblock is created for the first time for
 952	 * a userns there's no one racing us. Therefore we don't need any
 953	 * barriers when we dereference binfmt_misc.
 954	 */
 955	misc = user_ns->binfmt_misc;
 956	if (!misc) {
 957		/*
 958		 * If it turns out that most user namespaces actually want to
 959		 * register their own binary type handler and therefore all
 960		 * create their own separate binfmt_misc mounts we should
 961		 * consider turning this into a kmem cache.
 962		 */
 963		misc = kzalloc(sizeof(struct binfmt_misc), GFP_KERNEL);
 964		if (!misc)
 965			return -ENOMEM;
 966
 967		INIT_LIST_HEAD(&misc->entries);
 968		rwlock_init(&misc->entries_lock);
 969
 970		/* Pairs with smp_load_acquire() in load_binfmt_misc(). */
 971		smp_store_release(&user_ns->binfmt_misc, misc);
 972	}
 973
 974	/*
 975	 * When the binfmt_misc superblock for this userns is shutdown
 976	 * ->enabled might have been set to false and we don't reinitialize
 977	 * ->enabled again in put_super() as someone might already be mounting
 978	 * binfmt_misc again. It also would be pointless since by the time
 979	 * ->put_super() is called we know that the binary type list for this
 980	 * bintfmt_misc mount is empty making load_misc_binary() return
 981	 * -ENOEXEC independent of whether ->enabled is true. Instead, if
 982	 * someone mounts binfmt_misc for the first time or again we simply
 983	 * reset ->enabled to true.
 984	 */
 985	misc->enabled = true;
 986
 987	err = simple_fill_super(sb, BINFMTFS_MAGIC, bm_files);
 988	if (!err)
 989		sb->s_op = &s_ops;
 990	return err;
 991}
 992
 993static void bm_free(struct fs_context *fc)
 994{
 995	if (fc->s_fs_info)
 996		put_user_ns(fc->s_fs_info);
 997}
 998
 999static int bm_get_tree(struct fs_context *fc)
1000{
1001	return get_tree_keyed(fc, bm_fill_super, get_user_ns(fc->user_ns));
1002}
1003
1004static const struct fs_context_operations bm_context_ops = {
1005	.free		= bm_free,
1006	.get_tree	= bm_get_tree,
1007};
1008
1009static int bm_init_fs_context(struct fs_context *fc)
1010{
1011	fc->ops = &bm_context_ops;
1012	return 0;
1013}
1014
1015static struct linux_binfmt misc_format = {
1016	.module = THIS_MODULE,
1017	.load_binary = load_misc_binary,
1018};
1019
1020static struct file_system_type bm_fs_type = {
1021	.owner		= THIS_MODULE,
1022	.name		= "binfmt_misc",
1023	.init_fs_context = bm_init_fs_context,
1024	.fs_flags	= FS_USERNS_MOUNT,
1025	.kill_sb	= kill_anon_super,
1026};
1027MODULE_ALIAS_FS("binfmt_misc");
1028
1029static int __init init_misc_binfmt(void)
1030{
1031	int err = register_filesystem(&bm_fs_type);
1032	if (!err)
1033		insert_binfmt(&misc_format);
1034	return err;
1035}
1036
1037static void __exit exit_misc_binfmt(void)
1038{
1039	unregister_binfmt(&misc_format);
1040	unregister_filesystem(&bm_fs_type);
1041}
1042
1043core_initcall(init_misc_binfmt);
1044module_exit(exit_misc_binfmt);
1045MODULE_DESCRIPTION("Kernel support for miscellaneous binaries");
1046MODULE_LICENSE("GPL");