Linux Audio

Check our new training course

Loading...
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/exec.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 */
   7
   8/*
   9 * #!-checking implemented by tytso.
  10 */
  11/*
  12 * Demand-loading implemented 01.12.91 - no need to read anything but
  13 * the header into memory. The inode of the executable is put into
  14 * "current->executable", and page faults do the actual loading. Clean.
  15 *
  16 * Once more I can proudly say that linux stood up to being changed: it
  17 * was less than 2 hours work to get demand-loading completely implemented.
  18 *
  19 * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20 * current->executable is only used by the procfs.  This allows a dispatch
  21 * table to check for several different types  of binary formats.  We keep
  22 * trying until we recognize the file or we run out of supported binary
  23 * formats.
  24 */
  25
  26#include <linux/slab.h>
  27#include <linux/file.h>
  28#include <linux/fdtable.h>
  29#include <linux/mm.h>
  30#include <linux/vmacache.h>
  31#include <linux/stat.h>
  32#include <linux/fcntl.h>
  33#include <linux/swap.h>
  34#include <linux/string.h>
  35#include <linux/init.h>
  36#include <linux/sched/mm.h>
  37#include <linux/sched/coredump.h>
  38#include <linux/sched/signal.h>
  39#include <linux/sched/numa_balancing.h>
  40#include <linux/sched/task.h>
  41#include <linux/pagemap.h>
  42#include <linux/perf_event.h>
  43#include <linux/highmem.h>
  44#include <linux/spinlock.h>
  45#include <linux/key.h>
  46#include <linux/personality.h>
  47#include <linux/binfmts.h>
  48#include <linux/utsname.h>
  49#include <linux/pid_namespace.h>
  50#include <linux/module.h>
  51#include <linux/namei.h>
  52#include <linux/mount.h>
  53#include <linux/security.h>
  54#include <linux/syscalls.h>
  55#include <linux/tsacct_kern.h>
  56#include <linux/cn_proc.h>
  57#include <linux/audit.h>
  58#include <linux/tracehook.h>
  59#include <linux/kmod.h>
  60#include <linux/fsnotify.h>
  61#include <linux/fs_struct.h>
  62#include <linux/oom.h>
  63#include <linux/compat.h>
  64#include <linux/vmalloc.h>
  65
  66#include <linux/uaccess.h>
  67#include <asm/mmu_context.h>
  68#include <asm/tlb.h>
  69
  70#include <trace/events/task.h>
  71#include "internal.h"
  72
  73#include <trace/events/sched.h>
  74
  75static int bprm_creds_from_file(struct linux_binprm *bprm);
  76
  77int suid_dumpable = 0;
  78
  79static LIST_HEAD(formats);
  80static DEFINE_RWLOCK(binfmt_lock);
  81
  82void __register_binfmt(struct linux_binfmt * fmt, int insert)
  83{
  84	BUG_ON(!fmt);
  85	if (WARN_ON(!fmt->load_binary))
  86		return;
  87	write_lock(&binfmt_lock);
  88	insert ? list_add(&fmt->lh, &formats) :
  89		 list_add_tail(&fmt->lh, &formats);
  90	write_unlock(&binfmt_lock);
  91}
  92
  93EXPORT_SYMBOL(__register_binfmt);
  94
  95void unregister_binfmt(struct linux_binfmt * fmt)
  96{
  97	write_lock(&binfmt_lock);
  98	list_del(&fmt->lh);
  99	write_unlock(&binfmt_lock);
 100}
 101
 102EXPORT_SYMBOL(unregister_binfmt);
 103
 104static inline void put_binfmt(struct linux_binfmt * fmt)
 105{
 106	module_put(fmt->module);
 107}
 108
 109bool path_noexec(const struct path *path)
 110{
 111	return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 112	       (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 113}
 114
 115#ifdef CONFIG_USELIB
 116/*
 117 * Note that a shared library must be both readable and executable due to
 118 * security reasons.
 119 *
 120 * Also note that we take the address to load from from the file itself.
 121 */
 122SYSCALL_DEFINE1(uselib, const char __user *, library)
 123{
 124	struct linux_binfmt *fmt;
 125	struct file *file;
 126	struct filename *tmp = getname(library);
 127	int error = PTR_ERR(tmp);
 128	static const struct open_flags uselib_flags = {
 129		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 130		.acc_mode = MAY_READ | MAY_EXEC,
 131		.intent = LOOKUP_OPEN,
 132		.lookup_flags = LOOKUP_FOLLOW,
 133	};
 134
 135	if (IS_ERR(tmp))
 136		goto out;
 137
 138	file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 139	putname(tmp);
 140	error = PTR_ERR(file);
 141	if (IS_ERR(file))
 142		goto out;
 143
 144	/*
 145	 * may_open() has already checked for this, so it should be
 146	 * impossible to trip now. But we need to be extra cautious
 147	 * and check again at the very end too.
 148	 */
 149	error = -EACCES;
 150	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 151			 path_noexec(&file->f_path)))
 152		goto exit;
 153
 154	fsnotify_open(file);
 155
 156	error = -ENOEXEC;
 157
 158	read_lock(&binfmt_lock);
 159	list_for_each_entry(fmt, &formats, lh) {
 160		if (!fmt->load_shlib)
 161			continue;
 162		if (!try_module_get(fmt->module))
 163			continue;
 164		read_unlock(&binfmt_lock);
 165		error = fmt->load_shlib(file);
 166		read_lock(&binfmt_lock);
 167		put_binfmt(fmt);
 168		if (error != -ENOEXEC)
 169			break;
 170	}
 171	read_unlock(&binfmt_lock);
 172exit:
 173	fput(file);
 174out:
 175  	return error;
 176}
 177#endif /* #ifdef CONFIG_USELIB */
 178
 179#ifdef CONFIG_MMU
 180/*
 181 * The nascent bprm->mm is not visible until exec_mmap() but it can
 182 * use a lot of memory, account these pages in current->mm temporary
 183 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 184 * change the counter back via acct_arg_size(0).
 185 */
 186static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 187{
 188	struct mm_struct *mm = current->mm;
 189	long diff = (long)(pages - bprm->vma_pages);
 190
 191	if (!mm || !diff)
 192		return;
 193
 194	bprm->vma_pages = pages;
 195	add_mm_counter(mm, MM_ANONPAGES, diff);
 196}
 197
 198static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 199		int write)
 200{
 201	struct page *page;
 202	int ret;
 203	unsigned int gup_flags = FOLL_FORCE;
 204
 205#ifdef CONFIG_STACK_GROWSUP
 206	if (write) {
 207		ret = expand_downwards(bprm->vma, pos);
 208		if (ret < 0)
 209			return NULL;
 210	}
 211#endif
 212
 213	if (write)
 214		gup_flags |= FOLL_WRITE;
 215
 216	/*
 217	 * We are doing an exec().  'current' is the process
 218	 * doing the exec and bprm->mm is the new process's mm.
 219	 */
 220	ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 221			&page, NULL, NULL);
 222	if (ret <= 0)
 223		return NULL;
 224
 225	if (write)
 226		acct_arg_size(bprm, vma_pages(bprm->vma));
 227
 228	return page;
 229}
 230
 231static void put_arg_page(struct page *page)
 232{
 233	put_page(page);
 234}
 235
 236static void free_arg_pages(struct linux_binprm *bprm)
 237{
 238}
 239
 240static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 241		struct page *page)
 242{
 243	flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 244}
 245
 246static int __bprm_mm_init(struct linux_binprm *bprm)
 247{
 248	int err;
 249	struct vm_area_struct *vma = NULL;
 250	struct mm_struct *mm = bprm->mm;
 251
 252	bprm->vma = vma = vm_area_alloc(mm);
 253	if (!vma)
 254		return -ENOMEM;
 255	vma_set_anonymous(vma);
 256
 257	if (mmap_write_lock_killable(mm)) {
 258		err = -EINTR;
 259		goto err_free;
 260	}
 261
 262	/*
 263	 * Place the stack at the largest stack address the architecture
 264	 * supports. Later, we'll move this to an appropriate place. We don't
 265	 * use STACK_TOP because that can depend on attributes which aren't
 266	 * configured yet.
 267	 */
 268	BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 269	vma->vm_end = STACK_TOP_MAX;
 270	vma->vm_start = vma->vm_end - PAGE_SIZE;
 271	vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 272	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 273
 274	err = insert_vm_struct(mm, vma);
 275	if (err)
 276		goto err;
 277
 278	mm->stack_vm = mm->total_vm = 1;
 279	mmap_write_unlock(mm);
 280	bprm->p = vma->vm_end - sizeof(void *);
 281	return 0;
 282err:
 283	mmap_write_unlock(mm);
 284err_free:
 285	bprm->vma = NULL;
 286	vm_area_free(vma);
 287	return err;
 288}
 289
 290static bool valid_arg_len(struct linux_binprm *bprm, long len)
 291{
 292	return len <= MAX_ARG_STRLEN;
 293}
 294
 295#else
 296
 297static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 298{
 299}
 300
 301static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 302		int write)
 303{
 304	struct page *page;
 305
 306	page = bprm->page[pos / PAGE_SIZE];
 307	if (!page && write) {
 308		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 309		if (!page)
 310			return NULL;
 311		bprm->page[pos / PAGE_SIZE] = page;
 312	}
 313
 314	return page;
 315}
 316
 317static void put_arg_page(struct page *page)
 318{
 319}
 320
 321static void free_arg_page(struct linux_binprm *bprm, int i)
 322{
 323	if (bprm->page[i]) {
 324		__free_page(bprm->page[i]);
 325		bprm->page[i] = NULL;
 326	}
 327}
 328
 329static void free_arg_pages(struct linux_binprm *bprm)
 330{
 331	int i;
 332
 333	for (i = 0; i < MAX_ARG_PAGES; i++)
 334		free_arg_page(bprm, i);
 335}
 336
 337static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 338		struct page *page)
 339{
 340}
 341
 342static int __bprm_mm_init(struct linux_binprm *bprm)
 343{
 344	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 345	return 0;
 346}
 347
 348static bool valid_arg_len(struct linux_binprm *bprm, long len)
 349{
 350	return len <= bprm->p;
 351}
 352
 353#endif /* CONFIG_MMU */
 354
 355/*
 356 * Create a new mm_struct and populate it with a temporary stack
 357 * vm_area_struct.  We don't have enough context at this point to set the stack
 358 * flags, permissions, and offset, so we use temporary values.  We'll update
 359 * them later in setup_arg_pages().
 360 */
 361static int bprm_mm_init(struct linux_binprm *bprm)
 362{
 363	int err;
 364	struct mm_struct *mm = NULL;
 365
 366	bprm->mm = mm = mm_alloc();
 367	err = -ENOMEM;
 368	if (!mm)
 369		goto err;
 370
 371	/* Save current stack limit for all calculations made during exec. */
 372	task_lock(current->group_leader);
 373	bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 374	task_unlock(current->group_leader);
 375
 376	err = __bprm_mm_init(bprm);
 377	if (err)
 378		goto err;
 379
 380	return 0;
 381
 382err:
 383	if (mm) {
 384		bprm->mm = NULL;
 385		mmdrop(mm);
 386	}
 387
 388	return err;
 389}
 390
 391struct user_arg_ptr {
 392#ifdef CONFIG_COMPAT
 393	bool is_compat;
 394#endif
 395	union {
 396		const char __user *const __user *native;
 397#ifdef CONFIG_COMPAT
 398		const compat_uptr_t __user *compat;
 399#endif
 400	} ptr;
 401};
 402
 403static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 404{
 405	const char __user *native;
 406
 407#ifdef CONFIG_COMPAT
 408	if (unlikely(argv.is_compat)) {
 409		compat_uptr_t compat;
 410
 411		if (get_user(compat, argv.ptr.compat + nr))
 412			return ERR_PTR(-EFAULT);
 413
 414		return compat_ptr(compat);
 415	}
 416#endif
 417
 418	if (get_user(native, argv.ptr.native + nr))
 419		return ERR_PTR(-EFAULT);
 420
 421	return native;
 422}
 423
 424/*
 425 * count() counts the number of strings in array ARGV.
 426 */
 427static int count(struct user_arg_ptr argv, int max)
 428{
 429	int i = 0;
 430
 431	if (argv.ptr.native != NULL) {
 432		for (;;) {
 433			const char __user *p = get_user_arg_ptr(argv, i);
 434
 435			if (!p)
 436				break;
 437
 438			if (IS_ERR(p))
 439				return -EFAULT;
 440
 441			if (i >= max)
 442				return -E2BIG;
 443			++i;
 444
 445			if (fatal_signal_pending(current))
 446				return -ERESTARTNOHAND;
 447			cond_resched();
 448		}
 449	}
 450	return i;
 451}
 452
 453static int count_strings_kernel(const char *const *argv)
 454{
 455	int i;
 456
 457	if (!argv)
 458		return 0;
 459
 460	for (i = 0; argv[i]; ++i) {
 461		if (i >= MAX_ARG_STRINGS)
 462			return -E2BIG;
 463		if (fatal_signal_pending(current))
 464			return -ERESTARTNOHAND;
 465		cond_resched();
 466	}
 467	return i;
 468}
 469
 470static int bprm_stack_limits(struct linux_binprm *bprm)
 471{
 472	unsigned long limit, ptr_size;
 473
 474	/*
 475	 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 476	 * (whichever is smaller) for the argv+env strings.
 477	 * This ensures that:
 478	 *  - the remaining binfmt code will not run out of stack space,
 479	 *  - the program will have a reasonable amount of stack left
 480	 *    to work from.
 481	 */
 482	limit = _STK_LIM / 4 * 3;
 483	limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 484	/*
 485	 * We've historically supported up to 32 pages (ARG_MAX)
 486	 * of argument strings even with small stacks
 487	 */
 488	limit = max_t(unsigned long, limit, ARG_MAX);
 489	/*
 490	 * We must account for the size of all the argv and envp pointers to
 491	 * the argv and envp strings, since they will also take up space in
 492	 * the stack. They aren't stored until much later when we can't
 493	 * signal to the parent that the child has run out of stack space.
 494	 * Instead, calculate it here so it's possible to fail gracefully.
 495	 */
 496	ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 497	if (limit <= ptr_size)
 498		return -E2BIG;
 499	limit -= ptr_size;
 500
 501	bprm->argmin = bprm->p - limit;
 502	return 0;
 503}
 504
 505/*
 506 * 'copy_strings()' copies argument/environment strings from the old
 507 * processes's memory to the new process's stack.  The call to get_user_pages()
 508 * ensures the destination page is created and not swapped out.
 509 */
 510static int copy_strings(int argc, struct user_arg_ptr argv,
 511			struct linux_binprm *bprm)
 512{
 513	struct page *kmapped_page = NULL;
 514	char *kaddr = NULL;
 515	unsigned long kpos = 0;
 516	int ret;
 517
 518	while (argc-- > 0) {
 519		const char __user *str;
 520		int len;
 521		unsigned long pos;
 522
 523		ret = -EFAULT;
 524		str = get_user_arg_ptr(argv, argc);
 525		if (IS_ERR(str))
 526			goto out;
 527
 528		len = strnlen_user(str, MAX_ARG_STRLEN);
 529		if (!len)
 530			goto out;
 531
 532		ret = -E2BIG;
 533		if (!valid_arg_len(bprm, len))
 534			goto out;
 535
 536		/* We're going to work our way backwords. */
 537		pos = bprm->p;
 538		str += len;
 539		bprm->p -= len;
 540#ifdef CONFIG_MMU
 541		if (bprm->p < bprm->argmin)
 542			goto out;
 543#endif
 544
 545		while (len > 0) {
 546			int offset, bytes_to_copy;
 547
 548			if (fatal_signal_pending(current)) {
 549				ret = -ERESTARTNOHAND;
 550				goto out;
 551			}
 552			cond_resched();
 553
 554			offset = pos % PAGE_SIZE;
 555			if (offset == 0)
 556				offset = PAGE_SIZE;
 557
 558			bytes_to_copy = offset;
 559			if (bytes_to_copy > len)
 560				bytes_to_copy = len;
 561
 562			offset -= bytes_to_copy;
 563			pos -= bytes_to_copy;
 564			str -= bytes_to_copy;
 565			len -= bytes_to_copy;
 566
 567			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 568				struct page *page;
 569
 570				page = get_arg_page(bprm, pos, 1);
 571				if (!page) {
 572					ret = -E2BIG;
 573					goto out;
 574				}
 575
 576				if (kmapped_page) {
 577					flush_kernel_dcache_page(kmapped_page);
 578					kunmap(kmapped_page);
 579					put_arg_page(kmapped_page);
 580				}
 581				kmapped_page = page;
 582				kaddr = kmap(kmapped_page);
 583				kpos = pos & PAGE_MASK;
 584				flush_arg_page(bprm, kpos, kmapped_page);
 585			}
 586			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 587				ret = -EFAULT;
 588				goto out;
 589			}
 590		}
 591	}
 592	ret = 0;
 593out:
 594	if (kmapped_page) {
 595		flush_kernel_dcache_page(kmapped_page);
 596		kunmap(kmapped_page);
 597		put_arg_page(kmapped_page);
 598	}
 599	return ret;
 600}
 601
 602/*
 603 * Copy and argument/environment string from the kernel to the processes stack.
 604 */
 605int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 606{
 607	int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 608	unsigned long pos = bprm->p;
 609
 610	if (len == 0)
 611		return -EFAULT;
 612	if (!valid_arg_len(bprm, len))
 613		return -E2BIG;
 614
 615	/* We're going to work our way backwards. */
 616	arg += len;
 617	bprm->p -= len;
 618	if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 619		return -E2BIG;
 620
 621	while (len > 0) {
 622		unsigned int bytes_to_copy = min_t(unsigned int, len,
 623				min_not_zero(offset_in_page(pos), PAGE_SIZE));
 624		struct page *page;
 625		char *kaddr;
 626
 627		pos -= bytes_to_copy;
 628		arg -= bytes_to_copy;
 629		len -= bytes_to_copy;
 630
 631		page = get_arg_page(bprm, pos, 1);
 632		if (!page)
 633			return -E2BIG;
 634		kaddr = kmap_atomic(page);
 635		flush_arg_page(bprm, pos & PAGE_MASK, page);
 636		memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 637		flush_kernel_dcache_page(page);
 638		kunmap_atomic(kaddr);
 639		put_arg_page(page);
 640	}
 641
 642	return 0;
 643}
 644EXPORT_SYMBOL(copy_string_kernel);
 645
 646static int copy_strings_kernel(int argc, const char *const *argv,
 647			       struct linux_binprm *bprm)
 648{
 649	while (argc-- > 0) {
 650		int ret = copy_string_kernel(argv[argc], bprm);
 651		if (ret < 0)
 652			return ret;
 653		if (fatal_signal_pending(current))
 654			return -ERESTARTNOHAND;
 655		cond_resched();
 656	}
 657	return 0;
 658}
 659
 660#ifdef CONFIG_MMU
 661
 662/*
 663 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 664 * the binfmt code determines where the new stack should reside, we shift it to
 665 * its final location.  The process proceeds as follows:
 666 *
 667 * 1) Use shift to calculate the new vma endpoints.
 668 * 2) Extend vma to cover both the old and new ranges.  This ensures the
 669 *    arguments passed to subsequent functions are consistent.
 670 * 3) Move vma's page tables to the new range.
 671 * 4) Free up any cleared pgd range.
 672 * 5) Shrink the vma to cover only the new range.
 673 */
 674static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 675{
 676	struct mm_struct *mm = vma->vm_mm;
 677	unsigned long old_start = vma->vm_start;
 678	unsigned long old_end = vma->vm_end;
 679	unsigned long length = old_end - old_start;
 680	unsigned long new_start = old_start - shift;
 681	unsigned long new_end = old_end - shift;
 682	struct mmu_gather tlb;
 683
 684	BUG_ON(new_start > new_end);
 685
 686	/*
 687	 * ensure there are no vmas between where we want to go
 688	 * and where we are
 689	 */
 690	if (vma != find_vma(mm, new_start))
 691		return -EFAULT;
 692
 693	/*
 694	 * cover the whole range: [new_start, old_end)
 695	 */
 696	if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 697		return -ENOMEM;
 698
 699	/*
 700	 * move the page tables downwards, on failure we rely on
 701	 * process cleanup to remove whatever mess we made.
 702	 */
 703	if (length != move_page_tables(vma, old_start,
 704				       vma, new_start, length, false))
 705		return -ENOMEM;
 706
 707	lru_add_drain();
 708	tlb_gather_mmu(&tlb, mm, old_start, old_end);
 709	if (new_end > old_start) {
 710		/*
 711		 * when the old and new regions overlap clear from new_end.
 712		 */
 713		free_pgd_range(&tlb, new_end, old_end, new_end,
 714			vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 715	} else {
 716		/*
 717		 * otherwise, clean from old_start; this is done to not touch
 718		 * the address space in [new_end, old_start) some architectures
 719		 * have constraints on va-space that make this illegal (IA64) -
 720		 * for the others its just a little faster.
 721		 */
 722		free_pgd_range(&tlb, old_start, old_end, new_end,
 723			vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 724	}
 725	tlb_finish_mmu(&tlb, old_start, old_end);
 726
 727	/*
 728	 * Shrink the vma to just the new range.  Always succeeds.
 729	 */
 730	vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 731
 732	return 0;
 733}
 734
 735/*
 736 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 737 * the stack is optionally relocated, and some extra space is added.
 738 */
 739int setup_arg_pages(struct linux_binprm *bprm,
 740		    unsigned long stack_top,
 741		    int executable_stack)
 742{
 743	unsigned long ret;
 744	unsigned long stack_shift;
 745	struct mm_struct *mm = current->mm;
 746	struct vm_area_struct *vma = bprm->vma;
 747	struct vm_area_struct *prev = NULL;
 748	unsigned long vm_flags;
 749	unsigned long stack_base;
 750	unsigned long stack_size;
 751	unsigned long stack_expand;
 752	unsigned long rlim_stack;
 753
 754#ifdef CONFIG_STACK_GROWSUP
 755	/* Limit stack size */
 756	stack_base = bprm->rlim_stack.rlim_max;
 757	if (stack_base > STACK_SIZE_MAX)
 758		stack_base = STACK_SIZE_MAX;
 759
 760	/* Add space for stack randomization. */
 761	stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 762
 763	/* Make sure we didn't let the argument array grow too large. */
 764	if (vma->vm_end - vma->vm_start > stack_base)
 765		return -ENOMEM;
 766
 767	stack_base = PAGE_ALIGN(stack_top - stack_base);
 768
 769	stack_shift = vma->vm_start - stack_base;
 770	mm->arg_start = bprm->p - stack_shift;
 771	bprm->p = vma->vm_end - stack_shift;
 772#else
 773	stack_top = arch_align_stack(stack_top);
 774	stack_top = PAGE_ALIGN(stack_top);
 775
 776	if (unlikely(stack_top < mmap_min_addr) ||
 777	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 778		return -ENOMEM;
 779
 780	stack_shift = vma->vm_end - stack_top;
 781
 782	bprm->p -= stack_shift;
 783	mm->arg_start = bprm->p;
 784#endif
 785
 786	if (bprm->loader)
 787		bprm->loader -= stack_shift;
 788	bprm->exec -= stack_shift;
 789
 790	if (mmap_write_lock_killable(mm))
 791		return -EINTR;
 792
 793	vm_flags = VM_STACK_FLAGS;
 794
 795	/*
 796	 * Adjust stack execute permissions; explicitly enable for
 797	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 798	 * (arch default) otherwise.
 799	 */
 800	if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 801		vm_flags |= VM_EXEC;
 802	else if (executable_stack == EXSTACK_DISABLE_X)
 803		vm_flags &= ~VM_EXEC;
 804	vm_flags |= mm->def_flags;
 805	vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 806
 807	ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 808			vm_flags);
 809	if (ret)
 810		goto out_unlock;
 811	BUG_ON(prev != vma);
 812
 813	if (unlikely(vm_flags & VM_EXEC)) {
 814		pr_warn_once("process '%pD4' started with executable stack\n",
 815			     bprm->file);
 816	}
 817
 818	/* Move stack pages down in memory. */
 819	if (stack_shift) {
 820		ret = shift_arg_pages(vma, stack_shift);
 821		if (ret)
 822			goto out_unlock;
 823	}
 824
 825	/* mprotect_fixup is overkill to remove the temporary stack flags */
 826	vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 827
 828	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 829	stack_size = vma->vm_end - vma->vm_start;
 830	/*
 831	 * Align this down to a page boundary as expand_stack
 832	 * will align it up.
 833	 */
 834	rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 835#ifdef CONFIG_STACK_GROWSUP
 836	if (stack_size + stack_expand > rlim_stack)
 837		stack_base = vma->vm_start + rlim_stack;
 838	else
 839		stack_base = vma->vm_end + stack_expand;
 840#else
 841	if (stack_size + stack_expand > rlim_stack)
 842		stack_base = vma->vm_end - rlim_stack;
 843	else
 844		stack_base = vma->vm_start - stack_expand;
 845#endif
 846	current->mm->start_stack = bprm->p;
 847	ret = expand_stack(vma, stack_base);
 848	if (ret)
 849		ret = -EFAULT;
 850
 851out_unlock:
 852	mmap_write_unlock(mm);
 853	return ret;
 854}
 855EXPORT_SYMBOL(setup_arg_pages);
 856
 857#else
 858
 859/*
 860 * Transfer the program arguments and environment from the holding pages
 861 * onto the stack. The provided stack pointer is adjusted accordingly.
 862 */
 863int transfer_args_to_stack(struct linux_binprm *bprm,
 864			   unsigned long *sp_location)
 865{
 866	unsigned long index, stop, sp;
 867	int ret = 0;
 868
 869	stop = bprm->p >> PAGE_SHIFT;
 870	sp = *sp_location;
 871
 872	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 873		unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 874		char *src = kmap(bprm->page[index]) + offset;
 875		sp -= PAGE_SIZE - offset;
 876		if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 877			ret = -EFAULT;
 878		kunmap(bprm->page[index]);
 879		if (ret)
 880			goto out;
 881	}
 882
 883	*sp_location = sp;
 884
 885out:
 886	return ret;
 887}
 888EXPORT_SYMBOL(transfer_args_to_stack);
 889
 890#endif /* CONFIG_MMU */
 891
 892static struct file *do_open_execat(int fd, struct filename *name, int flags)
 893{
 894	struct file *file;
 895	int err;
 896	struct open_flags open_exec_flags = {
 897		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 898		.acc_mode = MAY_EXEC,
 899		.intent = LOOKUP_OPEN,
 900		.lookup_flags = LOOKUP_FOLLOW,
 901	};
 902
 903	if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 904		return ERR_PTR(-EINVAL);
 905	if (flags & AT_SYMLINK_NOFOLLOW)
 906		open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 907	if (flags & AT_EMPTY_PATH)
 908		open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 909
 910	file = do_filp_open(fd, name, &open_exec_flags);
 911	if (IS_ERR(file))
 912		goto out;
 913
 914	/*
 915	 * may_open() has already checked for this, so it should be
 916	 * impossible to trip now. But we need to be extra cautious
 917	 * and check again at the very end too.
 918	 */
 919	err = -EACCES;
 920	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 921			 path_noexec(&file->f_path)))
 922		goto exit;
 923
 924	err = deny_write_access(file);
 925	if (err)
 926		goto exit;
 927
 928	if (name->name[0] != '\0')
 929		fsnotify_open(file);
 930
 931out:
 932	return file;
 933
 934exit:
 935	fput(file);
 936	return ERR_PTR(err);
 937}
 938
 939struct file *open_exec(const char *name)
 940{
 941	struct filename *filename = getname_kernel(name);
 942	struct file *f = ERR_CAST(filename);
 943
 944	if (!IS_ERR(filename)) {
 945		f = do_open_execat(AT_FDCWD, filename, 0);
 946		putname(filename);
 947	}
 948	return f;
 949}
 950EXPORT_SYMBOL(open_exec);
 951
 952int kernel_read_file(struct file *file, void **buf, loff_t *size,
 953		     loff_t max_size, enum kernel_read_file_id id)
 954{
 955	loff_t i_size, pos;
 956	ssize_t bytes = 0;
 957	int ret;
 958
 959	if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 960		return -EINVAL;
 961
 962	ret = deny_write_access(file);
 963	if (ret)
 964		return ret;
 965
 966	ret = security_kernel_read_file(file, id);
 967	if (ret)
 968		goto out;
 969
 970	i_size = i_size_read(file_inode(file));
 971	if (i_size <= 0) {
 972		ret = -EINVAL;
 973		goto out;
 974	}
 975	if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 976		ret = -EFBIG;
 977		goto out;
 978	}
 979
 980	if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 981		*buf = vmalloc(i_size);
 982	if (!*buf) {
 983		ret = -ENOMEM;
 984		goto out;
 985	}
 986
 987	pos = 0;
 988	while (pos < i_size) {
 989		bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 990		if (bytes < 0) {
 991			ret = bytes;
 992			goto out_free;
 993		}
 994
 995		if (bytes == 0)
 996			break;
 997	}
 998
 999	if (pos != i_size) {
1000		ret = -EIO;
1001		goto out_free;
1002	}
1003
1004	ret = security_kernel_post_read_file(file, *buf, i_size, id);
1005	if (!ret)
1006		*size = pos;
1007
1008out_free:
1009	if (ret < 0) {
1010		if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
1011			vfree(*buf);
1012			*buf = NULL;
1013		}
1014	}
1015
1016out:
1017	allow_write_access(file);
1018	return ret;
1019}
1020EXPORT_SYMBOL_GPL(kernel_read_file);
1021
1022int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
1023			       loff_t max_size, enum kernel_read_file_id id)
1024{
1025	struct file *file;
1026	int ret;
1027
1028	if (!path || !*path)
1029		return -EINVAL;
1030
1031	file = filp_open(path, O_RDONLY, 0);
1032	if (IS_ERR(file))
1033		return PTR_ERR(file);
1034
1035	ret = kernel_read_file(file, buf, size, max_size, id);
1036	fput(file);
1037	return ret;
1038}
1039EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
1040
1041int kernel_read_file_from_path_initns(const char *path, void **buf,
1042				      loff_t *size, loff_t max_size,
1043				      enum kernel_read_file_id id)
1044{
1045	struct file *file;
1046	struct path root;
1047	int ret;
1048
1049	if (!path || !*path)
1050		return -EINVAL;
1051
1052	task_lock(&init_task);
1053	get_fs_root(init_task.fs, &root);
1054	task_unlock(&init_task);
1055
1056	file = file_open_root(root.dentry, root.mnt, path, O_RDONLY, 0);
1057	path_put(&root);
1058	if (IS_ERR(file))
1059		return PTR_ERR(file);
1060
1061	ret = kernel_read_file(file, buf, size, max_size, id);
1062	fput(file);
1063	return ret;
1064}
1065EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns);
1066
1067int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
1068			     enum kernel_read_file_id id)
1069{
1070	struct fd f = fdget(fd);
1071	int ret = -EBADF;
1072
1073	if (!f.file)
1074		goto out;
1075
1076	ret = kernel_read_file(f.file, buf, size, max_size, id);
1077out:
1078	fdput(f);
1079	return ret;
1080}
1081EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1082
1083#if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
1084    defined(CONFIG_BINFMT_ELF_FDPIC)
1085ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1086{
1087	ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1088	if (res > 0)
1089		flush_icache_user_range(addr, addr + len);
1090	return res;
1091}
1092EXPORT_SYMBOL(read_code);
1093#endif
1094
1095/*
1096 * Maps the mm_struct mm into the current task struct.
1097 * On success, this function returns with the mutex
1098 * exec_update_mutex locked.
1099 */
1100static int exec_mmap(struct mm_struct *mm)
1101{
1102	struct task_struct *tsk;
1103	struct mm_struct *old_mm, *active_mm;
1104	int ret;
1105
1106	/* Notify parent that we're no longer interested in the old VM */
1107	tsk = current;
1108	old_mm = current->mm;
1109	exec_mm_release(tsk, old_mm);
1110	if (old_mm)
1111		sync_mm_rss(old_mm);
1112
1113	ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
1114	if (ret)
1115		return ret;
1116
1117	if (old_mm) {
1118		/*
1119		 * Make sure that if there is a core dump in progress
1120		 * for the old mm, we get out and die instead of going
1121		 * through with the exec.  We must hold mmap_lock around
1122		 * checking core_state and changing tsk->mm.
1123		 */
1124		mmap_read_lock(old_mm);
1125		if (unlikely(old_mm->core_state)) {
1126			mmap_read_unlock(old_mm);
1127			mutex_unlock(&tsk->signal->exec_update_mutex);
1128			return -EINTR;
1129		}
1130	}
1131
1132	task_lock(tsk);
1133	active_mm = tsk->active_mm;
1134	membarrier_exec_mmap(mm);
1135	tsk->mm = mm;
1136	tsk->active_mm = mm;
1137	activate_mm(active_mm, mm);
1138	tsk->mm->vmacache_seqnum = 0;
1139	vmacache_flush(tsk);
1140	task_unlock(tsk);
1141	if (old_mm) {
1142		mmap_read_unlock(old_mm);
1143		BUG_ON(active_mm != old_mm);
1144		setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1145		mm_update_next_owner(old_mm);
1146		mmput(old_mm);
1147		return 0;
1148	}
1149	mmdrop(active_mm);
1150	return 0;
1151}
1152
1153static int de_thread(struct task_struct *tsk)
1154{
1155	struct signal_struct *sig = tsk->signal;
1156	struct sighand_struct *oldsighand = tsk->sighand;
1157	spinlock_t *lock = &oldsighand->siglock;
1158
1159	if (thread_group_empty(tsk))
1160		goto no_thread_group;
1161
1162	/*
1163	 * Kill all other threads in the thread group.
1164	 */
1165	spin_lock_irq(lock);
1166	if (signal_group_exit(sig)) {
1167		/*
1168		 * Another group action in progress, just
1169		 * return so that the signal is processed.
1170		 */
1171		spin_unlock_irq(lock);
1172		return -EAGAIN;
1173	}
1174
1175	sig->group_exit_task = tsk;
1176	sig->notify_count = zap_other_threads(tsk);
1177	if (!thread_group_leader(tsk))
1178		sig->notify_count--;
1179
1180	while (sig->notify_count) {
1181		__set_current_state(TASK_KILLABLE);
1182		spin_unlock_irq(lock);
1183		schedule();
1184		if (__fatal_signal_pending(tsk))
1185			goto killed;
1186		spin_lock_irq(lock);
1187	}
1188	spin_unlock_irq(lock);
1189
1190	/*
1191	 * At this point all other threads have exited, all we have to
1192	 * do is to wait for the thread group leader to become inactive,
1193	 * and to assume its PID:
1194	 */
1195	if (!thread_group_leader(tsk)) {
1196		struct task_struct *leader = tsk->group_leader;
1197
1198		for (;;) {
1199			cgroup_threadgroup_change_begin(tsk);
1200			write_lock_irq(&tasklist_lock);
1201			/*
1202			 * Do this under tasklist_lock to ensure that
1203			 * exit_notify() can't miss ->group_exit_task
1204			 */
1205			sig->notify_count = -1;
1206			if (likely(leader->exit_state))
1207				break;
1208			__set_current_state(TASK_KILLABLE);
1209			write_unlock_irq(&tasklist_lock);
1210			cgroup_threadgroup_change_end(tsk);
1211			schedule();
1212			if (__fatal_signal_pending(tsk))
1213				goto killed;
1214		}
1215
1216		/*
1217		 * The only record we have of the real-time age of a
1218		 * process, regardless of execs it's done, is start_time.
1219		 * All the past CPU time is accumulated in signal_struct
1220		 * from sister threads now dead.  But in this non-leader
1221		 * exec, nothing survives from the original leader thread,
1222		 * whose birth marks the true age of this process now.
1223		 * When we take on its identity by switching to its PID, we
1224		 * also take its birthdate (always earlier than our own).
1225		 */
1226		tsk->start_time = leader->start_time;
1227		tsk->start_boottime = leader->start_boottime;
1228
1229		BUG_ON(!same_thread_group(leader, tsk));
1230		/*
1231		 * An exec() starts a new thread group with the
1232		 * TGID of the previous thread group. Rehash the
1233		 * two threads with a switched PID, and release
1234		 * the former thread group leader:
1235		 */
1236
1237		/* Become a process group leader with the old leader's pid.
1238		 * The old leader becomes a thread of the this thread group.
1239		 */
1240		exchange_tids(tsk, leader);
1241		transfer_pid(leader, tsk, PIDTYPE_TGID);
1242		transfer_pid(leader, tsk, PIDTYPE_PGID);
1243		transfer_pid(leader, tsk, PIDTYPE_SID);
1244
1245		list_replace_rcu(&leader->tasks, &tsk->tasks);
1246		list_replace_init(&leader->sibling, &tsk->sibling);
1247
1248		tsk->group_leader = tsk;
1249		leader->group_leader = tsk;
1250
1251		tsk->exit_signal = SIGCHLD;
1252		leader->exit_signal = -1;
1253
1254		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1255		leader->exit_state = EXIT_DEAD;
1256
1257		/*
1258		 * We are going to release_task()->ptrace_unlink() silently,
1259		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1260		 * the tracer wont't block again waiting for this thread.
1261		 */
1262		if (unlikely(leader->ptrace))
1263			__wake_up_parent(leader, leader->parent);
1264		write_unlock_irq(&tasklist_lock);
1265		cgroup_threadgroup_change_end(tsk);
1266
1267		release_task(leader);
1268	}
1269
1270	sig->group_exit_task = NULL;
1271	sig->notify_count = 0;
1272
1273no_thread_group:
1274	/* we have changed execution domain */
1275	tsk->exit_signal = SIGCHLD;
1276
1277	BUG_ON(!thread_group_leader(tsk));
1278	return 0;
1279
1280killed:
1281	/* protects against exit_notify() and __exit_signal() */
1282	read_lock(&tasklist_lock);
1283	sig->group_exit_task = NULL;
1284	sig->notify_count = 0;
1285	read_unlock(&tasklist_lock);
1286	return -EAGAIN;
1287}
1288
1289
1290/*
1291 * This function makes sure the current process has its own signal table,
1292 * so that flush_signal_handlers can later reset the handlers without
1293 * disturbing other processes.  (Other processes might share the signal
1294 * table via the CLONE_SIGHAND option to clone().)
1295 */
1296static int unshare_sighand(struct task_struct *me)
1297{
1298	struct sighand_struct *oldsighand = me->sighand;
1299
1300	if (refcount_read(&oldsighand->count) != 1) {
1301		struct sighand_struct *newsighand;
1302		/*
1303		 * This ->sighand is shared with the CLONE_SIGHAND
1304		 * but not CLONE_THREAD task, switch to the new one.
1305		 */
1306		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1307		if (!newsighand)
1308			return -ENOMEM;
1309
1310		refcount_set(&newsighand->count, 1);
1311		memcpy(newsighand->action, oldsighand->action,
1312		       sizeof(newsighand->action));
1313
1314		write_lock_irq(&tasklist_lock);
1315		spin_lock(&oldsighand->siglock);
1316		rcu_assign_pointer(me->sighand, newsighand);
1317		spin_unlock(&oldsighand->siglock);
1318		write_unlock_irq(&tasklist_lock);
1319
1320		__cleanup_sighand(oldsighand);
1321	}
1322	return 0;
1323}
1324
1325char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1326{
1327	task_lock(tsk);
1328	strncpy(buf, tsk->comm, buf_size);
1329	task_unlock(tsk);
1330	return buf;
1331}
1332EXPORT_SYMBOL_GPL(__get_task_comm);
1333
1334/*
1335 * These functions flushes out all traces of the currently running executable
1336 * so that a new one can be started
1337 */
1338
1339void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1340{
1341	task_lock(tsk);
1342	trace_task_rename(tsk, buf);
1343	strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1344	task_unlock(tsk);
1345	perf_event_comm(tsk, exec);
1346}
1347
1348/*
1349 * Calling this is the point of no return. None of the failures will be
1350 * seen by userspace since either the process is already taking a fatal
1351 * signal (via de_thread() or coredump), or will have SEGV raised
1352 * (after exec_mmap()) by search_binary_handler (see below).
1353 */
1354int begin_new_exec(struct linux_binprm * bprm)
1355{
1356	struct task_struct *me = current;
1357	int retval;
1358
1359	/* Once we are committed compute the creds */
1360	retval = bprm_creds_from_file(bprm);
1361	if (retval)
1362		return retval;
1363
1364	/*
1365	 * Ensure all future errors are fatal.
1366	 */
1367	bprm->point_of_no_return = true;
1368
1369	/*
1370	 * Make this the only thread in the thread group.
1371	 */
1372	retval = de_thread(me);
1373	if (retval)
1374		goto out;
1375
1376	/*
1377	 * Must be called _before_ exec_mmap() as bprm->mm is
1378	 * not visibile until then. This also enables the update
1379	 * to be lockless.
1380	 */
1381	set_mm_exe_file(bprm->mm, bprm->file);
1382
1383	/* If the binary is not readable then enforce mm->dumpable=0 */
1384	would_dump(bprm, bprm->file);
1385	if (bprm->have_execfd)
1386		would_dump(bprm, bprm->executable);
1387
1388	/*
1389	 * Release all of the old mmap stuff
1390	 */
1391	acct_arg_size(bprm, 0);
1392	retval = exec_mmap(bprm->mm);
1393	if (retval)
1394		goto out;
1395
1396	bprm->mm = NULL;
1397
1398#ifdef CONFIG_POSIX_TIMERS
1399	exit_itimers(me->signal);
1400	flush_itimer_signals();
1401#endif
1402
1403	/*
1404	 * Make the signal table private.
1405	 */
1406	retval = unshare_sighand(me);
1407	if (retval)
1408		goto out_unlock;
1409
1410	/*
1411	 * Ensure that the uaccess routines can actually operate on userspace
1412	 * pointers:
1413	 */
1414	force_uaccess_begin();
1415
1416	me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1417					PF_NOFREEZE | PF_NO_SETAFFINITY);
1418	flush_thread();
1419	me->personality &= ~bprm->per_clear;
1420
1421	/*
1422	 * We have to apply CLOEXEC before we change whether the process is
1423	 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1424	 * trying to access the should-be-closed file descriptors of a process
1425	 * undergoing exec(2).
1426	 */
1427	do_close_on_exec(me->files);
1428
1429	if (bprm->secureexec) {
1430		/* Make sure parent cannot signal privileged process. */
1431		me->pdeath_signal = 0;
1432
1433		/*
1434		 * For secureexec, reset the stack limit to sane default to
1435		 * avoid bad behavior from the prior rlimits. This has to
1436		 * happen before arch_pick_mmap_layout(), which examines
1437		 * RLIMIT_STACK, but after the point of no return to avoid
1438		 * needing to clean up the change on failure.
1439		 */
1440		if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1441			bprm->rlim_stack.rlim_cur = _STK_LIM;
1442	}
1443
1444	me->sas_ss_sp = me->sas_ss_size = 0;
1445
1446	/*
1447	 * Figure out dumpability. Note that this checking only of current
1448	 * is wrong, but userspace depends on it. This should be testing
1449	 * bprm->secureexec instead.
1450	 */
1451	if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1452	    !(uid_eq(current_euid(), current_uid()) &&
1453	      gid_eq(current_egid(), current_gid())))
1454		set_dumpable(current->mm, suid_dumpable);
1455	else
1456		set_dumpable(current->mm, SUID_DUMP_USER);
1457
1458	perf_event_exec();
1459	__set_task_comm(me, kbasename(bprm->filename), true);
1460
1461	/* An exec changes our domain. We are no longer part of the thread
1462	   group */
1463	WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1464	flush_signal_handlers(me, 0);
1465
1466	/*
1467	 * install the new credentials for this executable
1468	 */
1469	security_bprm_committing_creds(bprm);
1470
1471	commit_creds(bprm->cred);
1472	bprm->cred = NULL;
1473
1474	/*
1475	 * Disable monitoring for regular users
1476	 * when executing setuid binaries. Must
1477	 * wait until new credentials are committed
1478	 * by commit_creds() above
1479	 */
1480	if (get_dumpable(me->mm) != SUID_DUMP_USER)
1481		perf_event_exit_task(me);
1482	/*
1483	 * cred_guard_mutex must be held at least to this point to prevent
1484	 * ptrace_attach() from altering our determination of the task's
1485	 * credentials; any time after this it may be unlocked.
1486	 */
1487	security_bprm_committed_creds(bprm);
1488
1489	/* Pass the opened binary to the interpreter. */
1490	if (bprm->have_execfd) {
1491		retval = get_unused_fd_flags(0);
1492		if (retval < 0)
1493			goto out_unlock;
1494		fd_install(retval, bprm->executable);
1495		bprm->executable = NULL;
1496		bprm->execfd = retval;
1497	}
1498	return 0;
1499
1500out_unlock:
1501	mutex_unlock(&me->signal->exec_update_mutex);
1502out:
1503	return retval;
1504}
1505EXPORT_SYMBOL(begin_new_exec);
1506
1507void would_dump(struct linux_binprm *bprm, struct file *file)
1508{
1509	struct inode *inode = file_inode(file);
1510	if (inode_permission(inode, MAY_READ) < 0) {
1511		struct user_namespace *old, *user_ns;
1512		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1513
1514		/* Ensure mm->user_ns contains the executable */
1515		user_ns = old = bprm->mm->user_ns;
1516		while ((user_ns != &init_user_ns) &&
1517		       !privileged_wrt_inode_uidgid(user_ns, inode))
1518			user_ns = user_ns->parent;
1519
1520		if (old != user_ns) {
1521			bprm->mm->user_ns = get_user_ns(user_ns);
1522			put_user_ns(old);
1523		}
1524	}
1525}
1526EXPORT_SYMBOL(would_dump);
1527
1528void setup_new_exec(struct linux_binprm * bprm)
1529{
1530	/* Setup things that can depend upon the personality */
1531	struct task_struct *me = current;
1532
1533	arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1534
1535	arch_setup_new_exec();
1536
1537	/* Set the new mm task size. We have to do that late because it may
1538	 * depend on TIF_32BIT which is only updated in flush_thread() on
1539	 * some architectures like powerpc
1540	 */
1541	me->mm->task_size = TASK_SIZE;
1542	mutex_unlock(&me->signal->exec_update_mutex);
1543	mutex_unlock(&me->signal->cred_guard_mutex);
1544}
1545EXPORT_SYMBOL(setup_new_exec);
1546
1547/* Runs immediately before start_thread() takes over. */
1548void finalize_exec(struct linux_binprm *bprm)
1549{
1550	/* Store any stack rlimit changes before starting thread. */
1551	task_lock(current->group_leader);
1552	current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1553	task_unlock(current->group_leader);
1554}
1555EXPORT_SYMBOL(finalize_exec);
1556
1557/*
1558 * Prepare credentials and lock ->cred_guard_mutex.
1559 * setup_new_exec() commits the new creds and drops the lock.
1560 * Or, if exec fails before, free_bprm() should release ->cred and
1561 * and unlock.
1562 */
1563static int prepare_bprm_creds(struct linux_binprm *bprm)
1564{
1565	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1566		return -ERESTARTNOINTR;
1567
1568	bprm->cred = prepare_exec_creds();
1569	if (likely(bprm->cred))
1570		return 0;
1571
1572	mutex_unlock(&current->signal->cred_guard_mutex);
1573	return -ENOMEM;
1574}
1575
1576static void free_bprm(struct linux_binprm *bprm)
1577{
1578	if (bprm->mm) {
1579		acct_arg_size(bprm, 0);
1580		mmput(bprm->mm);
1581	}
1582	free_arg_pages(bprm);
1583	if (bprm->cred) {
1584		mutex_unlock(&current->signal->cred_guard_mutex);
1585		abort_creds(bprm->cred);
1586	}
1587	if (bprm->file) {
1588		allow_write_access(bprm->file);
1589		fput(bprm->file);
1590	}
1591	if (bprm->executable)
1592		fput(bprm->executable);
1593	/* If a binfmt changed the interp, free it. */
1594	if (bprm->interp != bprm->filename)
1595		kfree(bprm->interp);
1596	kfree(bprm->fdpath);
1597	kfree(bprm);
1598}
1599
1600static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1601{
1602	struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1603	int retval = -ENOMEM;
1604	if (!bprm)
1605		goto out;
1606
1607	if (fd == AT_FDCWD || filename->name[0] == '/') {
1608		bprm->filename = filename->name;
1609	} else {
1610		if (filename->name[0] == '\0')
1611			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1612		else
1613			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1614						  fd, filename->name);
1615		if (!bprm->fdpath)
1616			goto out_free;
1617
1618		bprm->filename = bprm->fdpath;
1619	}
1620	bprm->interp = bprm->filename;
1621
1622	retval = bprm_mm_init(bprm);
1623	if (retval)
1624		goto out_free;
1625	return bprm;
1626
1627out_free:
1628	free_bprm(bprm);
1629out:
1630	return ERR_PTR(retval);
1631}
1632
1633int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1634{
1635	/* If a binfmt changed the interp, free it first. */
1636	if (bprm->interp != bprm->filename)
1637		kfree(bprm->interp);
1638	bprm->interp = kstrdup(interp, GFP_KERNEL);
1639	if (!bprm->interp)
1640		return -ENOMEM;
1641	return 0;
1642}
1643EXPORT_SYMBOL(bprm_change_interp);
1644
1645/*
1646 * determine how safe it is to execute the proposed program
1647 * - the caller must hold ->cred_guard_mutex to protect against
1648 *   PTRACE_ATTACH or seccomp thread-sync
1649 */
1650static void check_unsafe_exec(struct linux_binprm *bprm)
1651{
1652	struct task_struct *p = current, *t;
1653	unsigned n_fs;
1654
1655	if (p->ptrace)
1656		bprm->unsafe |= LSM_UNSAFE_PTRACE;
1657
1658	/*
1659	 * This isn't strictly necessary, but it makes it harder for LSMs to
1660	 * mess up.
1661	 */
1662	if (task_no_new_privs(current))
1663		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1664
1665	t = p;
1666	n_fs = 1;
1667	spin_lock(&p->fs->lock);
1668	rcu_read_lock();
1669	while_each_thread(p, t) {
1670		if (t->fs == p->fs)
1671			n_fs++;
1672	}
1673	rcu_read_unlock();
1674
1675	if (p->fs->users > n_fs)
1676		bprm->unsafe |= LSM_UNSAFE_SHARE;
1677	else
1678		p->fs->in_exec = 1;
1679	spin_unlock(&p->fs->lock);
1680}
1681
1682static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1683{
1684	/* Handle suid and sgid on files */
1685	struct inode *inode;
1686	unsigned int mode;
1687	kuid_t uid;
1688	kgid_t gid;
1689
1690	if (!mnt_may_suid(file->f_path.mnt))
1691		return;
1692
1693	if (task_no_new_privs(current))
1694		return;
1695
1696	inode = file->f_path.dentry->d_inode;
1697	mode = READ_ONCE(inode->i_mode);
1698	if (!(mode & (S_ISUID|S_ISGID)))
1699		return;
1700
1701	/* Be careful if suid/sgid is set */
1702	inode_lock(inode);
1703
1704	/* reload atomically mode/uid/gid now that lock held */
1705	mode = inode->i_mode;
1706	uid = inode->i_uid;
1707	gid = inode->i_gid;
1708	inode_unlock(inode);
1709
1710	/* We ignore suid/sgid if there are no mappings for them in the ns */
1711	if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1712		 !kgid_has_mapping(bprm->cred->user_ns, gid))
1713		return;
1714
1715	if (mode & S_ISUID) {
1716		bprm->per_clear |= PER_CLEAR_ON_SETID;
1717		bprm->cred->euid = uid;
1718	}
1719
1720	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1721		bprm->per_clear |= PER_CLEAR_ON_SETID;
1722		bprm->cred->egid = gid;
1723	}
1724}
1725
1726/*
1727 * Compute brpm->cred based upon the final binary.
1728 */
1729static int bprm_creds_from_file(struct linux_binprm *bprm)
1730{
1731	/* Compute creds based on which file? */
1732	struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1733
1734	bprm_fill_uid(bprm, file);
1735	return security_bprm_creds_from_file(bprm, file);
1736}
1737
1738/*
1739 * Fill the binprm structure from the inode.
1740 * Read the first BINPRM_BUF_SIZE bytes
1741 *
1742 * This may be called multiple times for binary chains (scripts for example).
1743 */
1744static int prepare_binprm(struct linux_binprm *bprm)
1745{
1746	loff_t pos = 0;
1747
1748	memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1749	return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1750}
1751
1752/*
1753 * Arguments are '\0' separated strings found at the location bprm->p
1754 * points to; chop off the first by relocating brpm->p to right after
1755 * the first '\0' encountered.
1756 */
1757int remove_arg_zero(struct linux_binprm *bprm)
1758{
1759	int ret = 0;
1760	unsigned long offset;
1761	char *kaddr;
1762	struct page *page;
1763
1764	if (!bprm->argc)
1765		return 0;
1766
1767	do {
1768		offset = bprm->p & ~PAGE_MASK;
1769		page = get_arg_page(bprm, bprm->p, 0);
1770		if (!page) {
1771			ret = -EFAULT;
1772			goto out;
1773		}
1774		kaddr = kmap_atomic(page);
1775
1776		for (; offset < PAGE_SIZE && kaddr[offset];
1777				offset++, bprm->p++)
1778			;
1779
1780		kunmap_atomic(kaddr);
1781		put_arg_page(page);
1782	} while (offset == PAGE_SIZE);
1783
1784	bprm->p++;
1785	bprm->argc--;
1786	ret = 0;
1787
1788out:
1789	return ret;
1790}
1791EXPORT_SYMBOL(remove_arg_zero);
1792
1793#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1794/*
1795 * cycle the list of binary formats handler, until one recognizes the image
1796 */
1797static int search_binary_handler(struct linux_binprm *bprm)
1798{
1799	bool need_retry = IS_ENABLED(CONFIG_MODULES);
1800	struct linux_binfmt *fmt;
1801	int retval;
1802
1803	retval = prepare_binprm(bprm);
1804	if (retval < 0)
1805		return retval;
1806
1807	retval = security_bprm_check(bprm);
1808	if (retval)
1809		return retval;
1810
1811	retval = -ENOENT;
1812 retry:
1813	read_lock(&binfmt_lock);
1814	list_for_each_entry(fmt, &formats, lh) {
1815		if (!try_module_get(fmt->module))
1816			continue;
1817		read_unlock(&binfmt_lock);
1818
1819		retval = fmt->load_binary(bprm);
1820
1821		read_lock(&binfmt_lock);
1822		put_binfmt(fmt);
1823		if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1824			read_unlock(&binfmt_lock);
1825			return retval;
1826		}
1827	}
1828	read_unlock(&binfmt_lock);
1829
1830	if (need_retry) {
1831		if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1832		    printable(bprm->buf[2]) && printable(bprm->buf[3]))
1833			return retval;
1834		if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1835			return retval;
1836		need_retry = false;
1837		goto retry;
1838	}
1839
1840	return retval;
1841}
1842
1843static int exec_binprm(struct linux_binprm *bprm)
1844{
1845	pid_t old_pid, old_vpid;
1846	int ret, depth;
1847
1848	/* Need to fetch pid before load_binary changes it */
1849	old_pid = current->pid;
1850	rcu_read_lock();
1851	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1852	rcu_read_unlock();
1853
1854	/* This allows 4 levels of binfmt rewrites before failing hard. */
1855	for (depth = 0;; depth++) {
1856		struct file *exec;
1857		if (depth > 5)
1858			return -ELOOP;
1859
1860		ret = search_binary_handler(bprm);
1861		if (ret < 0)
1862			return ret;
1863		if (!bprm->interpreter)
1864			break;
1865
1866		exec = bprm->file;
1867		bprm->file = bprm->interpreter;
1868		bprm->interpreter = NULL;
1869
1870		allow_write_access(exec);
1871		if (unlikely(bprm->have_execfd)) {
1872			if (bprm->executable) {
1873				fput(exec);
1874				return -ENOEXEC;
1875			}
1876			bprm->executable = exec;
1877		} else
1878			fput(exec);
1879	}
1880
1881	audit_bprm(bprm);
1882	trace_sched_process_exec(current, old_pid, bprm);
1883	ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1884	proc_exec_connector(current);
1885	return 0;
1886}
1887
1888/*
1889 * sys_execve() executes a new program.
1890 */
1891static int bprm_execve(struct linux_binprm *bprm,
1892		       int fd, struct filename *filename, int flags)
1893{
1894	struct file *file;
1895	struct files_struct *displaced;
1896	int retval;
1897
1898	retval = unshare_files(&displaced);
1899	if (retval)
1900		return retval;
1901
1902	retval = prepare_bprm_creds(bprm);
1903	if (retval)
1904		goto out_files;
1905
1906	check_unsafe_exec(bprm);
1907	current->in_execve = 1;
1908
1909	file = do_open_execat(fd, filename, flags);
1910	retval = PTR_ERR(file);
1911	if (IS_ERR(file))
1912		goto out_unmark;
1913
1914	sched_exec();
1915
1916	bprm->file = file;
1917	/*
1918	 * Record that a name derived from an O_CLOEXEC fd will be
1919	 * inaccessible after exec. Relies on having exclusive access to
1920	 * current->files (due to unshare_files above).
1921	 */
1922	if (bprm->fdpath &&
1923	    close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1924		bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1925
1926	/* Set the unchanging part of bprm->cred */
1927	retval = security_bprm_creds_for_exec(bprm);
1928	if (retval)
1929		goto out;
1930
1931	retval = exec_binprm(bprm);
1932	if (retval < 0)
1933		goto out;
1934
1935	/* execve succeeded */
1936	current->fs->in_exec = 0;
1937	current->in_execve = 0;
1938	rseq_execve(current);
1939	acct_update_integrals(current);
1940	task_numa_free(current, false);
1941	if (displaced)
1942		put_files_struct(displaced);
1943	return retval;
1944
1945out:
1946	/*
1947	 * If past the point of no return ensure the the code never
1948	 * returns to the userspace process.  Use an existing fatal
1949	 * signal if present otherwise terminate the process with
1950	 * SIGSEGV.
1951	 */
1952	if (bprm->point_of_no_return && !fatal_signal_pending(current))
1953		force_sigsegv(SIGSEGV);
1954
1955out_unmark:
1956	current->fs->in_exec = 0;
1957	current->in_execve = 0;
1958
1959out_files:
1960	if (displaced)
1961		reset_files_struct(displaced);
1962
1963	return retval;
1964}
1965
1966static int do_execveat_common(int fd, struct filename *filename,
1967			      struct user_arg_ptr argv,
1968			      struct user_arg_ptr envp,
1969			      int flags)
1970{
1971	struct linux_binprm *bprm;
1972	int retval;
1973
1974	if (IS_ERR(filename))
1975		return PTR_ERR(filename);
1976
1977	/*
1978	 * We move the actual failure in case of RLIMIT_NPROC excess from
1979	 * set*uid() to execve() because too many poorly written programs
1980	 * don't check setuid() return code.  Here we additionally recheck
1981	 * whether NPROC limit is still exceeded.
1982	 */
1983	if ((current->flags & PF_NPROC_EXCEEDED) &&
1984	    atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1985		retval = -EAGAIN;
1986		goto out_ret;
1987	}
1988
1989	/* We're below the limit (still or again), so we don't want to make
1990	 * further execve() calls fail. */
1991	current->flags &= ~PF_NPROC_EXCEEDED;
1992
1993	bprm = alloc_bprm(fd, filename);
1994	if (IS_ERR(bprm)) {
1995		retval = PTR_ERR(bprm);
1996		goto out_ret;
1997	}
1998
1999	retval = count(argv, MAX_ARG_STRINGS);
2000	if (retval < 0)
2001		goto out_free;
2002	bprm->argc = retval;
2003
2004	retval = count(envp, MAX_ARG_STRINGS);
2005	if (retval < 0)
2006		goto out_free;
2007	bprm->envc = retval;
2008
2009	retval = bprm_stack_limits(bprm);
2010	if (retval < 0)
2011		goto out_free;
2012
2013	retval = copy_string_kernel(bprm->filename, bprm);
2014	if (retval < 0)
2015		goto out_free;
2016	bprm->exec = bprm->p;
2017
2018	retval = copy_strings(bprm->envc, envp, bprm);
2019	if (retval < 0)
2020		goto out_free;
2021
2022	retval = copy_strings(bprm->argc, argv, bprm);
2023	if (retval < 0)
2024		goto out_free;
2025
2026	retval = bprm_execve(bprm, fd, filename, flags);
2027out_free:
2028	free_bprm(bprm);
2029
2030out_ret:
2031	putname(filename);
2032	return retval;
2033}
2034
2035int kernel_execve(const char *kernel_filename,
2036		  const char *const *argv, const char *const *envp)
2037{
2038	struct filename *filename;
2039	struct linux_binprm *bprm;
2040	int fd = AT_FDCWD;
2041	int retval;
2042
2043	filename = getname_kernel(kernel_filename);
2044	if (IS_ERR(filename))
2045		return PTR_ERR(filename);
2046
2047	bprm = alloc_bprm(fd, filename);
2048	if (IS_ERR(bprm)) {
2049		retval = PTR_ERR(bprm);
2050		goto out_ret;
2051	}
2052
2053	retval = count_strings_kernel(argv);
2054	if (retval < 0)
2055		goto out_free;
2056	bprm->argc = retval;
2057
2058	retval = count_strings_kernel(envp);
2059	if (retval < 0)
2060		goto out_free;
2061	bprm->envc = retval;
2062
2063	retval = bprm_stack_limits(bprm);
2064	if (retval < 0)
2065		goto out_free;
2066
2067	retval = copy_string_kernel(bprm->filename, bprm);
2068	if (retval < 0)
2069		goto out_free;
2070	bprm->exec = bprm->p;
2071
2072	retval = copy_strings_kernel(bprm->envc, envp, bprm);
2073	if (retval < 0)
2074		goto out_free;
2075
2076	retval = copy_strings_kernel(bprm->argc, argv, bprm);
2077	if (retval < 0)
2078		goto out_free;
2079
2080	retval = bprm_execve(bprm, fd, filename, 0);
2081out_free:
2082	free_bprm(bprm);
2083out_ret:
2084	putname(filename);
2085	return retval;
2086}
2087
2088static int do_execve(struct filename *filename,
2089	const char __user *const __user *__argv,
2090	const char __user *const __user *__envp)
2091{
2092	struct user_arg_ptr argv = { .ptr.native = __argv };
2093	struct user_arg_ptr envp = { .ptr.native = __envp };
2094	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2095}
2096
2097static int do_execveat(int fd, struct filename *filename,
2098		const char __user *const __user *__argv,
2099		const char __user *const __user *__envp,
2100		int flags)
2101{
2102	struct user_arg_ptr argv = { .ptr.native = __argv };
2103	struct user_arg_ptr envp = { .ptr.native = __envp };
2104
2105	return do_execveat_common(fd, filename, argv, envp, flags);
2106}
2107
2108#ifdef CONFIG_COMPAT
2109static int compat_do_execve(struct filename *filename,
2110	const compat_uptr_t __user *__argv,
2111	const compat_uptr_t __user *__envp)
2112{
2113	struct user_arg_ptr argv = {
2114		.is_compat = true,
2115		.ptr.compat = __argv,
2116	};
2117	struct user_arg_ptr envp = {
2118		.is_compat = true,
2119		.ptr.compat = __envp,
2120	};
2121	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2122}
2123
2124static int compat_do_execveat(int fd, struct filename *filename,
2125			      const compat_uptr_t __user *__argv,
2126			      const compat_uptr_t __user *__envp,
2127			      int flags)
2128{
2129	struct user_arg_ptr argv = {
2130		.is_compat = true,
2131		.ptr.compat = __argv,
2132	};
2133	struct user_arg_ptr envp = {
2134		.is_compat = true,
2135		.ptr.compat = __envp,
2136	};
2137	return do_execveat_common(fd, filename, argv, envp, flags);
2138}
2139#endif
2140
2141void set_binfmt(struct linux_binfmt *new)
2142{
2143	struct mm_struct *mm = current->mm;
2144
2145	if (mm->binfmt)
2146		module_put(mm->binfmt->module);
2147
2148	mm->binfmt = new;
2149	if (new)
2150		__module_get(new->module);
2151}
2152EXPORT_SYMBOL(set_binfmt);
2153
2154/*
2155 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2156 */
2157void set_dumpable(struct mm_struct *mm, int value)
2158{
2159	if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2160		return;
2161
2162	set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2163}
2164
2165SYSCALL_DEFINE3(execve,
2166		const char __user *, filename,
2167		const char __user *const __user *, argv,
2168		const char __user *const __user *, envp)
2169{
2170	return do_execve(getname(filename), argv, envp);
2171}
2172
2173SYSCALL_DEFINE5(execveat,
2174		int, fd, const char __user *, filename,
2175		const char __user *const __user *, argv,
2176		const char __user *const __user *, envp,
2177		int, flags)
2178{
2179	int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2180
2181	return do_execveat(fd,
2182			   getname_flags(filename, lookup_flags, NULL),
2183			   argv, envp, flags);
2184}
2185
2186#ifdef CONFIG_COMPAT
2187COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2188	const compat_uptr_t __user *, argv,
2189	const compat_uptr_t __user *, envp)
2190{
2191	return compat_do_execve(getname(filename), argv, envp);
2192}
2193
2194COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2195		       const char __user *, filename,
2196		       const compat_uptr_t __user *, argv,
2197		       const compat_uptr_t __user *, envp,
2198		       int,  flags)
2199{
2200	int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2201
2202	return compat_do_execveat(fd,
2203				  getname_flags(filename, lookup_flags, NULL),
2204				  argv, envp, flags);
2205}
2206#endif
   1/*
   2 *  linux/fs/exec.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7/*
   8 * #!-checking implemented by tytso.
   9 */
  10/*
  11 * Demand-loading implemented 01.12.91 - no need to read anything but
  12 * the header into memory. The inode of the executable is put into
  13 * "current->executable", and page faults do the actual loading. Clean.
  14 *
  15 * Once more I can proudly say that linux stood up to being changed: it
  16 * was less than 2 hours work to get demand-loading completely implemented.
  17 *
  18 * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19 * current->executable is only used by the procfs.  This allows a dispatch
  20 * table to check for several different types  of binary formats.  We keep
  21 * trying until we recognize the file or we run out of supported binary
  22 * formats.
  23 */
  24
  25#include <linux/slab.h>
  26#include <linux/file.h>
  27#include <linux/fdtable.h>
  28#include <linux/mm.h>
  29#include <linux/vmacache.h>
  30#include <linux/stat.h>
  31#include <linux/fcntl.h>
  32#include <linux/swap.h>
  33#include <linux/string.h>
  34#include <linux/init.h>
  35#include <linux/pagemap.h>
  36#include <linux/perf_event.h>
  37#include <linux/highmem.h>
  38#include <linux/spinlock.h>
  39#include <linux/key.h>
  40#include <linux/personality.h>
  41#include <linux/binfmts.h>
  42#include <linux/utsname.h>
  43#include <linux/pid_namespace.h>
  44#include <linux/module.h>
  45#include <linux/namei.h>
  46#include <linux/mount.h>
  47#include <linux/security.h>
  48#include <linux/syscalls.h>
  49#include <linux/tsacct_kern.h>
  50#include <linux/cn_proc.h>
  51#include <linux/audit.h>
  52#include <linux/tracehook.h>
  53#include <linux/kmod.h>
  54#include <linux/fsnotify.h>
  55#include <linux/fs_struct.h>
  56#include <linux/pipe_fs_i.h>
  57#include <linux/oom.h>
  58#include <linux/compat.h>
  59#include <linux/vmalloc.h>
  60
  61#include <linux/uaccess.h>
  62#include <asm/mmu_context.h>
  63#include <asm/tlb.h>
  64
  65#include <trace/events/task.h>
  66#include "internal.h"
  67
  68#include <trace/events/sched.h>
  69
  70int suid_dumpable = 0;
  71
  72static LIST_HEAD(formats);
  73static DEFINE_RWLOCK(binfmt_lock);
  74
  75void __register_binfmt(struct linux_binfmt * fmt, int insert)
  76{
  77	BUG_ON(!fmt);
  78	if (WARN_ON(!fmt->load_binary))
  79		return;
  80	write_lock(&binfmt_lock);
  81	insert ? list_add(&fmt->lh, &formats) :
  82		 list_add_tail(&fmt->lh, &formats);
  83	write_unlock(&binfmt_lock);
  84}
  85
  86EXPORT_SYMBOL(__register_binfmt);
  87
  88void unregister_binfmt(struct linux_binfmt * fmt)
  89{
  90	write_lock(&binfmt_lock);
  91	list_del(&fmt->lh);
  92	write_unlock(&binfmt_lock);
  93}
  94
  95EXPORT_SYMBOL(unregister_binfmt);
  96
  97static inline void put_binfmt(struct linux_binfmt * fmt)
  98{
  99	module_put(fmt->module);
 100}
 101
 102bool path_noexec(const struct path *path)
 103{
 104	return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 105	       (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 106}
 107
 108#ifdef CONFIG_USELIB
 109/*
 110 * Note that a shared library must be both readable and executable due to
 111 * security reasons.
 112 *
 113 * Also note that we take the address to load from from the file itself.
 114 */
 115SYSCALL_DEFINE1(uselib, const char __user *, library)
 116{
 117	struct linux_binfmt *fmt;
 118	struct file *file;
 119	struct filename *tmp = getname(library);
 120	int error = PTR_ERR(tmp);
 121	static const struct open_flags uselib_flags = {
 122		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 123		.acc_mode = MAY_READ | MAY_EXEC,
 124		.intent = LOOKUP_OPEN,
 125		.lookup_flags = LOOKUP_FOLLOW,
 126	};
 127
 128	if (IS_ERR(tmp))
 129		goto out;
 130
 131	file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 132	putname(tmp);
 133	error = PTR_ERR(file);
 134	if (IS_ERR(file))
 135		goto out;
 136
 137	error = -EINVAL;
 138	if (!S_ISREG(file_inode(file)->i_mode))
 139		goto exit;
 140
 141	error = -EACCES;
 142	if (path_noexec(&file->f_path))
 143		goto exit;
 144
 145	fsnotify_open(file);
 146
 147	error = -ENOEXEC;
 148
 149	read_lock(&binfmt_lock);
 150	list_for_each_entry(fmt, &formats, lh) {
 151		if (!fmt->load_shlib)
 152			continue;
 153		if (!try_module_get(fmt->module))
 154			continue;
 155		read_unlock(&binfmt_lock);
 156		error = fmt->load_shlib(file);
 157		read_lock(&binfmt_lock);
 158		put_binfmt(fmt);
 159		if (error != -ENOEXEC)
 160			break;
 161	}
 162	read_unlock(&binfmt_lock);
 163exit:
 164	fput(file);
 165out:
 166  	return error;
 167}
 168#endif /* #ifdef CONFIG_USELIB */
 169
 170#ifdef CONFIG_MMU
 171/*
 172 * The nascent bprm->mm is not visible until exec_mmap() but it can
 173 * use a lot of memory, account these pages in current->mm temporary
 174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 175 * change the counter back via acct_arg_size(0).
 176 */
 177static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 178{
 179	struct mm_struct *mm = current->mm;
 180	long diff = (long)(pages - bprm->vma_pages);
 181
 182	if (!mm || !diff)
 183		return;
 184
 185	bprm->vma_pages = pages;
 186	add_mm_counter(mm, MM_ANONPAGES, diff);
 187}
 188
 189static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 190		int write)
 191{
 192	struct page *page;
 193	int ret;
 194	unsigned int gup_flags = FOLL_FORCE;
 195
 196#ifdef CONFIG_STACK_GROWSUP
 197	if (write) {
 198		ret = expand_downwards(bprm->vma, pos);
 199		if (ret < 0)
 200			return NULL;
 201	}
 202#endif
 203
 204	if (write)
 205		gup_flags |= FOLL_WRITE;
 206
 207	/*
 208	 * We are doing an exec().  'current' is the process
 209	 * doing the exec and bprm->mm is the new process's mm.
 210	 */
 211	ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 212			&page, NULL, NULL);
 213	if (ret <= 0)
 214		return NULL;
 215
 216	if (write) {
 217		unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 218		struct rlimit *rlim;
 219
 220		acct_arg_size(bprm, size / PAGE_SIZE);
 221
 222		/*
 223		 * We've historically supported up to 32 pages (ARG_MAX)
 224		 * of argument strings even with small stacks
 225		 */
 226		if (size <= ARG_MAX)
 227			return page;
 228
 229		/*
 230		 * Limit to 1/4-th the stack size for the argv+env strings.
 231		 * This ensures that:
 232		 *  - the remaining binfmt code will not run out of stack space,
 233		 *  - the program will have a reasonable amount of stack left
 234		 *    to work from.
 235		 */
 236		rlim = current->signal->rlim;
 237		if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 238			put_page(page);
 239			return NULL;
 240		}
 241	}
 242
 243	return page;
 244}
 245
 246static void put_arg_page(struct page *page)
 247{
 248	put_page(page);
 249}
 250
 251static void free_arg_pages(struct linux_binprm *bprm)
 252{
 253}
 254
 255static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 256		struct page *page)
 257{
 258	flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 259}
 260
 261static int __bprm_mm_init(struct linux_binprm *bprm)
 262{
 263	int err;
 264	struct vm_area_struct *vma = NULL;
 265	struct mm_struct *mm = bprm->mm;
 266
 267	bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 268	if (!vma)
 269		return -ENOMEM;
 270
 271	if (down_write_killable(&mm->mmap_sem)) {
 272		err = -EINTR;
 273		goto err_free;
 274	}
 275	vma->vm_mm = mm;
 276
 277	/*
 278	 * Place the stack at the largest stack address the architecture
 279	 * supports. Later, we'll move this to an appropriate place. We don't
 280	 * use STACK_TOP because that can depend on attributes which aren't
 281	 * configured yet.
 282	 */
 283	BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 284	vma->vm_end = STACK_TOP_MAX;
 285	vma->vm_start = vma->vm_end - PAGE_SIZE;
 286	vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 287	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 288	INIT_LIST_HEAD(&vma->anon_vma_chain);
 289
 290	err = insert_vm_struct(mm, vma);
 291	if (err)
 292		goto err;
 293
 294	mm->stack_vm = mm->total_vm = 1;
 295	arch_bprm_mm_init(mm, vma);
 296	up_write(&mm->mmap_sem);
 297	bprm->p = vma->vm_end - sizeof(void *);
 298	return 0;
 299err:
 300	up_write(&mm->mmap_sem);
 301err_free:
 302	bprm->vma = NULL;
 303	kmem_cache_free(vm_area_cachep, vma);
 304	return err;
 305}
 306
 307static bool valid_arg_len(struct linux_binprm *bprm, long len)
 308{
 309	return len <= MAX_ARG_STRLEN;
 310}
 311
 312#else
 313
 314static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 315{
 316}
 317
 318static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 319		int write)
 320{
 321	struct page *page;
 322
 323	page = bprm->page[pos / PAGE_SIZE];
 324	if (!page && write) {
 325		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 326		if (!page)
 327			return NULL;
 328		bprm->page[pos / PAGE_SIZE] = page;
 329	}
 330
 331	return page;
 332}
 333
 334static void put_arg_page(struct page *page)
 335{
 336}
 337
 338static void free_arg_page(struct linux_binprm *bprm, int i)
 339{
 340	if (bprm->page[i]) {
 341		__free_page(bprm->page[i]);
 342		bprm->page[i] = NULL;
 343	}
 344}
 345
 346static void free_arg_pages(struct linux_binprm *bprm)
 347{
 348	int i;
 349
 350	for (i = 0; i < MAX_ARG_PAGES; i++)
 351		free_arg_page(bprm, i);
 352}
 353
 354static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 355		struct page *page)
 356{
 357}
 358
 359static int __bprm_mm_init(struct linux_binprm *bprm)
 360{
 361	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 362	return 0;
 363}
 364
 365static bool valid_arg_len(struct linux_binprm *bprm, long len)
 366{
 367	return len <= bprm->p;
 368}
 369
 370#endif /* CONFIG_MMU */
 371
 372/*
 373 * Create a new mm_struct and populate it with a temporary stack
 374 * vm_area_struct.  We don't have enough context at this point to set the stack
 375 * flags, permissions, and offset, so we use temporary values.  We'll update
 376 * them later in setup_arg_pages().
 377 */
 378static int bprm_mm_init(struct linux_binprm *bprm)
 379{
 380	int err;
 381	struct mm_struct *mm = NULL;
 382
 383	bprm->mm = mm = mm_alloc();
 384	err = -ENOMEM;
 385	if (!mm)
 386		goto err;
 387
 388	err = __bprm_mm_init(bprm);
 389	if (err)
 390		goto err;
 391
 392	return 0;
 393
 394err:
 395	if (mm) {
 396		bprm->mm = NULL;
 397		mmdrop(mm);
 398	}
 399
 400	return err;
 401}
 402
 403struct user_arg_ptr {
 404#ifdef CONFIG_COMPAT
 405	bool is_compat;
 406#endif
 407	union {
 408		const char __user *const __user *native;
 409#ifdef CONFIG_COMPAT
 410		const compat_uptr_t __user *compat;
 411#endif
 412	} ptr;
 413};
 414
 415static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 416{
 417	const char __user *native;
 418
 419#ifdef CONFIG_COMPAT
 420	if (unlikely(argv.is_compat)) {
 421		compat_uptr_t compat;
 422
 423		if (get_user(compat, argv.ptr.compat + nr))
 424			return ERR_PTR(-EFAULT);
 425
 426		return compat_ptr(compat);
 427	}
 428#endif
 429
 430	if (get_user(native, argv.ptr.native + nr))
 431		return ERR_PTR(-EFAULT);
 432
 433	return native;
 434}
 435
 436/*
 437 * count() counts the number of strings in array ARGV.
 438 */
 439static int count(struct user_arg_ptr argv, int max)
 440{
 441	int i = 0;
 442
 443	if (argv.ptr.native != NULL) {
 444		for (;;) {
 445			const char __user *p = get_user_arg_ptr(argv, i);
 446
 447			if (!p)
 448				break;
 449
 450			if (IS_ERR(p))
 451				return -EFAULT;
 452
 453			if (i >= max)
 454				return -E2BIG;
 455			++i;
 456
 457			if (fatal_signal_pending(current))
 458				return -ERESTARTNOHAND;
 459			cond_resched();
 460		}
 461	}
 462	return i;
 463}
 464
 465/*
 466 * 'copy_strings()' copies argument/environment strings from the old
 467 * processes's memory to the new process's stack.  The call to get_user_pages()
 468 * ensures the destination page is created and not swapped out.
 469 */
 470static int copy_strings(int argc, struct user_arg_ptr argv,
 471			struct linux_binprm *bprm)
 472{
 473	struct page *kmapped_page = NULL;
 474	char *kaddr = NULL;
 475	unsigned long kpos = 0;
 476	int ret;
 477
 478	while (argc-- > 0) {
 479		const char __user *str;
 480		int len;
 481		unsigned long pos;
 482
 483		ret = -EFAULT;
 484		str = get_user_arg_ptr(argv, argc);
 485		if (IS_ERR(str))
 486			goto out;
 487
 488		len = strnlen_user(str, MAX_ARG_STRLEN);
 489		if (!len)
 490			goto out;
 491
 492		ret = -E2BIG;
 493		if (!valid_arg_len(bprm, len))
 494			goto out;
 495
 496		/* We're going to work our way backwords. */
 497		pos = bprm->p;
 498		str += len;
 499		bprm->p -= len;
 500
 501		while (len > 0) {
 502			int offset, bytes_to_copy;
 503
 504			if (fatal_signal_pending(current)) {
 505				ret = -ERESTARTNOHAND;
 506				goto out;
 507			}
 508			cond_resched();
 509
 510			offset = pos % PAGE_SIZE;
 511			if (offset == 0)
 512				offset = PAGE_SIZE;
 513
 514			bytes_to_copy = offset;
 515			if (bytes_to_copy > len)
 516				bytes_to_copy = len;
 517
 518			offset -= bytes_to_copy;
 519			pos -= bytes_to_copy;
 520			str -= bytes_to_copy;
 521			len -= bytes_to_copy;
 522
 523			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 524				struct page *page;
 525
 526				page = get_arg_page(bprm, pos, 1);
 527				if (!page) {
 528					ret = -E2BIG;
 529					goto out;
 530				}
 531
 532				if (kmapped_page) {
 533					flush_kernel_dcache_page(kmapped_page);
 534					kunmap(kmapped_page);
 535					put_arg_page(kmapped_page);
 536				}
 537				kmapped_page = page;
 538				kaddr = kmap(kmapped_page);
 539				kpos = pos & PAGE_MASK;
 540				flush_arg_page(bprm, kpos, kmapped_page);
 541			}
 542			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 543				ret = -EFAULT;
 544				goto out;
 545			}
 546		}
 547	}
 548	ret = 0;
 549out:
 550	if (kmapped_page) {
 551		flush_kernel_dcache_page(kmapped_page);
 552		kunmap(kmapped_page);
 553		put_arg_page(kmapped_page);
 554	}
 555	return ret;
 556}
 557
 558/*
 559 * Like copy_strings, but get argv and its values from kernel memory.
 560 */
 561int copy_strings_kernel(int argc, const char *const *__argv,
 562			struct linux_binprm *bprm)
 563{
 564	int r;
 565	mm_segment_t oldfs = get_fs();
 566	struct user_arg_ptr argv = {
 567		.ptr.native = (const char __user *const  __user *)__argv,
 568	};
 569
 570	set_fs(KERNEL_DS);
 571	r = copy_strings(argc, argv, bprm);
 572	set_fs(oldfs);
 573
 574	return r;
 575}
 576EXPORT_SYMBOL(copy_strings_kernel);
 577
 578#ifdef CONFIG_MMU
 579
 580/*
 581 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 582 * the binfmt code determines where the new stack should reside, we shift it to
 583 * its final location.  The process proceeds as follows:
 584 *
 585 * 1) Use shift to calculate the new vma endpoints.
 586 * 2) Extend vma to cover both the old and new ranges.  This ensures the
 587 *    arguments passed to subsequent functions are consistent.
 588 * 3) Move vma's page tables to the new range.
 589 * 4) Free up any cleared pgd range.
 590 * 5) Shrink the vma to cover only the new range.
 591 */
 592static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 593{
 594	struct mm_struct *mm = vma->vm_mm;
 595	unsigned long old_start = vma->vm_start;
 596	unsigned long old_end = vma->vm_end;
 597	unsigned long length = old_end - old_start;
 598	unsigned long new_start = old_start - shift;
 599	unsigned long new_end = old_end - shift;
 600	struct mmu_gather tlb;
 601
 602	BUG_ON(new_start > new_end);
 603
 604	/*
 605	 * ensure there are no vmas between where we want to go
 606	 * and where we are
 607	 */
 608	if (vma != find_vma(mm, new_start))
 609		return -EFAULT;
 610
 611	/*
 612	 * cover the whole range: [new_start, old_end)
 613	 */
 614	if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 615		return -ENOMEM;
 616
 617	/*
 618	 * move the page tables downwards, on failure we rely on
 619	 * process cleanup to remove whatever mess we made.
 620	 */
 621	if (length != move_page_tables(vma, old_start,
 622				       vma, new_start, length, false))
 623		return -ENOMEM;
 624
 625	lru_add_drain();
 626	tlb_gather_mmu(&tlb, mm, old_start, old_end);
 627	if (new_end > old_start) {
 628		/*
 629		 * when the old and new regions overlap clear from new_end.
 630		 */
 631		free_pgd_range(&tlb, new_end, old_end, new_end,
 632			vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 633	} else {
 634		/*
 635		 * otherwise, clean from old_start; this is done to not touch
 636		 * the address space in [new_end, old_start) some architectures
 637		 * have constraints on va-space that make this illegal (IA64) -
 638		 * for the others its just a little faster.
 639		 */
 640		free_pgd_range(&tlb, old_start, old_end, new_end,
 641			vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 642	}
 643	tlb_finish_mmu(&tlb, old_start, old_end);
 644
 645	/*
 646	 * Shrink the vma to just the new range.  Always succeeds.
 647	 */
 648	vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 649
 650	return 0;
 651}
 652
 653/*
 654 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 655 * the stack is optionally relocated, and some extra space is added.
 656 */
 657int setup_arg_pages(struct linux_binprm *bprm,
 658		    unsigned long stack_top,
 659		    int executable_stack)
 660{
 661	unsigned long ret;
 662	unsigned long stack_shift;
 663	struct mm_struct *mm = current->mm;
 664	struct vm_area_struct *vma = bprm->vma;
 665	struct vm_area_struct *prev = NULL;
 666	unsigned long vm_flags;
 667	unsigned long stack_base;
 668	unsigned long stack_size;
 669	unsigned long stack_expand;
 670	unsigned long rlim_stack;
 671
 672#ifdef CONFIG_STACK_GROWSUP
 673	/* Limit stack size */
 674	stack_base = rlimit_max(RLIMIT_STACK);
 675	if (stack_base > STACK_SIZE_MAX)
 676		stack_base = STACK_SIZE_MAX;
 677
 678	/* Add space for stack randomization. */
 679	stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 680
 681	/* Make sure we didn't let the argument array grow too large. */
 682	if (vma->vm_end - vma->vm_start > stack_base)
 683		return -ENOMEM;
 684
 685	stack_base = PAGE_ALIGN(stack_top - stack_base);
 686
 687	stack_shift = vma->vm_start - stack_base;
 688	mm->arg_start = bprm->p - stack_shift;
 689	bprm->p = vma->vm_end - stack_shift;
 690#else
 691	stack_top = arch_align_stack(stack_top);
 692	stack_top = PAGE_ALIGN(stack_top);
 693
 694	if (unlikely(stack_top < mmap_min_addr) ||
 695	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 696		return -ENOMEM;
 697
 698	stack_shift = vma->vm_end - stack_top;
 699
 700	bprm->p -= stack_shift;
 701	mm->arg_start = bprm->p;
 702#endif
 703
 704	if (bprm->loader)
 705		bprm->loader -= stack_shift;
 706	bprm->exec -= stack_shift;
 707
 708	if (down_write_killable(&mm->mmap_sem))
 709		return -EINTR;
 710
 711	vm_flags = VM_STACK_FLAGS;
 712
 713	/*
 714	 * Adjust stack execute permissions; explicitly enable for
 715	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 716	 * (arch default) otherwise.
 717	 */
 718	if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 719		vm_flags |= VM_EXEC;
 720	else if (executable_stack == EXSTACK_DISABLE_X)
 721		vm_flags &= ~VM_EXEC;
 722	vm_flags |= mm->def_flags;
 723	vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 724
 725	ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 726			vm_flags);
 727	if (ret)
 728		goto out_unlock;
 729	BUG_ON(prev != vma);
 730
 731	/* Move stack pages down in memory. */
 732	if (stack_shift) {
 733		ret = shift_arg_pages(vma, stack_shift);
 734		if (ret)
 735			goto out_unlock;
 736	}
 737
 738	/* mprotect_fixup is overkill to remove the temporary stack flags */
 739	vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 740
 741	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 742	stack_size = vma->vm_end - vma->vm_start;
 743	/*
 744	 * Align this down to a page boundary as expand_stack
 745	 * will align it up.
 746	 */
 747	rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 748#ifdef CONFIG_STACK_GROWSUP
 749	if (stack_size + stack_expand > rlim_stack)
 750		stack_base = vma->vm_start + rlim_stack;
 751	else
 752		stack_base = vma->vm_end + stack_expand;
 753#else
 754	if (stack_size + stack_expand > rlim_stack)
 755		stack_base = vma->vm_end - rlim_stack;
 756	else
 757		stack_base = vma->vm_start - stack_expand;
 758#endif
 759	current->mm->start_stack = bprm->p;
 760	ret = expand_stack(vma, stack_base);
 761	if (ret)
 762		ret = -EFAULT;
 763
 764out_unlock:
 765	up_write(&mm->mmap_sem);
 766	return ret;
 767}
 768EXPORT_SYMBOL(setup_arg_pages);
 769
 770#else
 771
 772/*
 773 * Transfer the program arguments and environment from the holding pages
 774 * onto the stack. The provided stack pointer is adjusted accordingly.
 775 */
 776int transfer_args_to_stack(struct linux_binprm *bprm,
 777			   unsigned long *sp_location)
 778{
 779	unsigned long index, stop, sp;
 780	int ret = 0;
 781
 782	stop = bprm->p >> PAGE_SHIFT;
 783	sp = *sp_location;
 784
 785	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 786		unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 787		char *src = kmap(bprm->page[index]) + offset;
 788		sp -= PAGE_SIZE - offset;
 789		if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 790			ret = -EFAULT;
 791		kunmap(bprm->page[index]);
 792		if (ret)
 793			goto out;
 794	}
 795
 796	*sp_location = sp;
 797
 798out:
 799	return ret;
 800}
 801EXPORT_SYMBOL(transfer_args_to_stack);
 802
 803#endif /* CONFIG_MMU */
 804
 805static struct file *do_open_execat(int fd, struct filename *name, int flags)
 806{
 807	struct file *file;
 808	int err;
 809	struct open_flags open_exec_flags = {
 810		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 811		.acc_mode = MAY_EXEC,
 812		.intent = LOOKUP_OPEN,
 813		.lookup_flags = LOOKUP_FOLLOW,
 814	};
 815
 816	if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 817		return ERR_PTR(-EINVAL);
 818	if (flags & AT_SYMLINK_NOFOLLOW)
 819		open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 820	if (flags & AT_EMPTY_PATH)
 821		open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 822
 823	file = do_filp_open(fd, name, &open_exec_flags);
 824	if (IS_ERR(file))
 825		goto out;
 826
 827	err = -EACCES;
 828	if (!S_ISREG(file_inode(file)->i_mode))
 829		goto exit;
 830
 831	if (path_noexec(&file->f_path))
 832		goto exit;
 833
 834	err = deny_write_access(file);
 835	if (err)
 836		goto exit;
 837
 838	if (name->name[0] != '\0')
 839		fsnotify_open(file);
 840
 841out:
 842	return file;
 843
 844exit:
 845	fput(file);
 846	return ERR_PTR(err);
 847}
 848
 849struct file *open_exec(const char *name)
 850{
 851	struct filename *filename = getname_kernel(name);
 852	struct file *f = ERR_CAST(filename);
 853
 854	if (!IS_ERR(filename)) {
 855		f = do_open_execat(AT_FDCWD, filename, 0);
 856		putname(filename);
 857	}
 858	return f;
 859}
 860EXPORT_SYMBOL(open_exec);
 861
 862int kernel_read(struct file *file, loff_t offset,
 863		char *addr, unsigned long count)
 864{
 865	mm_segment_t old_fs;
 866	loff_t pos = offset;
 867	int result;
 868
 869	old_fs = get_fs();
 870	set_fs(get_ds());
 871	/* The cast to a user pointer is valid due to the set_fs() */
 872	result = vfs_read(file, (void __user *)addr, count, &pos);
 873	set_fs(old_fs);
 874	return result;
 875}
 876
 877EXPORT_SYMBOL(kernel_read);
 878
 879int kernel_read_file(struct file *file, void **buf, loff_t *size,
 880		     loff_t max_size, enum kernel_read_file_id id)
 881{
 882	loff_t i_size, pos;
 883	ssize_t bytes = 0;
 884	int ret;
 885
 886	if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 887		return -EINVAL;
 888
 889	ret = security_kernel_read_file(file, id);
 890	if (ret)
 891		return ret;
 892
 893	ret = deny_write_access(file);
 894	if (ret)
 895		return ret;
 896
 897	i_size = i_size_read(file_inode(file));
 898	if (max_size > 0 && i_size > max_size) {
 899		ret = -EFBIG;
 900		goto out;
 901	}
 902	if (i_size <= 0) {
 903		ret = -EINVAL;
 904		goto out;
 905	}
 906
 907	if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 908		*buf = vmalloc(i_size);
 909	if (!*buf) {
 910		ret = -ENOMEM;
 911		goto out;
 912	}
 913
 914	pos = 0;
 915	while (pos < i_size) {
 916		bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 917				    i_size - pos);
 918		if (bytes < 0) {
 919			ret = bytes;
 920			goto out;
 921		}
 922
 923		if (bytes == 0)
 924			break;
 925		pos += bytes;
 926	}
 927
 928	if (pos != i_size) {
 929		ret = -EIO;
 930		goto out_free;
 931	}
 932
 933	ret = security_kernel_post_read_file(file, *buf, i_size, id);
 934	if (!ret)
 935		*size = pos;
 936
 937out_free:
 938	if (ret < 0) {
 939		if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 940			vfree(*buf);
 941			*buf = NULL;
 942		}
 943	}
 944
 945out:
 946	allow_write_access(file);
 947	return ret;
 948}
 949EXPORT_SYMBOL_GPL(kernel_read_file);
 950
 951int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 952			       loff_t max_size, enum kernel_read_file_id id)
 953{
 954	struct file *file;
 955	int ret;
 956
 957	if (!path || !*path)
 958		return -EINVAL;
 959
 960	file = filp_open(path, O_RDONLY, 0);
 961	if (IS_ERR(file))
 962		return PTR_ERR(file);
 963
 964	ret = kernel_read_file(file, buf, size, max_size, id);
 965	fput(file);
 966	return ret;
 967}
 968EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 969
 970int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 971			     enum kernel_read_file_id id)
 972{
 973	struct fd f = fdget(fd);
 974	int ret = -EBADF;
 975
 976	if (!f.file)
 977		goto out;
 978
 979	ret = kernel_read_file(f.file, buf, size, max_size, id);
 980out:
 981	fdput(f);
 982	return ret;
 983}
 984EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 985
 986ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 987{
 988	ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 989	if (res > 0)
 990		flush_icache_range(addr, addr + len);
 991	return res;
 992}
 993EXPORT_SYMBOL(read_code);
 994
 995static int exec_mmap(struct mm_struct *mm)
 996{
 997	struct task_struct *tsk;
 998	struct mm_struct *old_mm, *active_mm;
 999
1000	/* Notify parent that we're no longer interested in the old VM */
1001	tsk = current;
1002	old_mm = current->mm;
1003	mm_release(tsk, old_mm);
1004
1005	if (old_mm) {
1006		sync_mm_rss(old_mm);
1007		/*
1008		 * Make sure that if there is a core dump in progress
1009		 * for the old mm, we get out and die instead of going
1010		 * through with the exec.  We must hold mmap_sem around
1011		 * checking core_state and changing tsk->mm.
1012		 */
1013		down_read(&old_mm->mmap_sem);
1014		if (unlikely(old_mm->core_state)) {
1015			up_read(&old_mm->mmap_sem);
1016			return -EINTR;
1017		}
1018	}
1019	task_lock(tsk);
1020	active_mm = tsk->active_mm;
1021	tsk->mm = mm;
1022	tsk->active_mm = mm;
1023	activate_mm(active_mm, mm);
1024	tsk->mm->vmacache_seqnum = 0;
1025	vmacache_flush(tsk);
1026	task_unlock(tsk);
1027	if (old_mm) {
1028		up_read(&old_mm->mmap_sem);
1029		BUG_ON(active_mm != old_mm);
1030		setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1031		mm_update_next_owner(old_mm);
1032		mmput(old_mm);
1033		return 0;
1034	}
1035	mmdrop(active_mm);
1036	return 0;
1037}
1038
1039/*
1040 * This function makes sure the current process has its own signal table,
1041 * so that flush_signal_handlers can later reset the handlers without
1042 * disturbing other processes.  (Other processes might share the signal
1043 * table via the CLONE_SIGHAND option to clone().)
1044 */
1045static int de_thread(struct task_struct *tsk)
1046{
1047	struct signal_struct *sig = tsk->signal;
1048	struct sighand_struct *oldsighand = tsk->sighand;
1049	spinlock_t *lock = &oldsighand->siglock;
1050
1051	if (thread_group_empty(tsk))
1052		goto no_thread_group;
1053
1054	/*
1055	 * Kill all other threads in the thread group.
1056	 */
1057	spin_lock_irq(lock);
1058	if (signal_group_exit(sig)) {
1059		/*
1060		 * Another group action in progress, just
1061		 * return so that the signal is processed.
1062		 */
1063		spin_unlock_irq(lock);
1064		return -EAGAIN;
1065	}
1066
1067	sig->group_exit_task = tsk;
1068	sig->notify_count = zap_other_threads(tsk);
1069	if (!thread_group_leader(tsk))
1070		sig->notify_count--;
1071
1072	while (sig->notify_count) {
1073		__set_current_state(TASK_KILLABLE);
1074		spin_unlock_irq(lock);
1075		schedule();
1076		if (unlikely(__fatal_signal_pending(tsk)))
1077			goto killed;
1078		spin_lock_irq(lock);
1079	}
1080	spin_unlock_irq(lock);
1081
1082	/*
1083	 * At this point all other threads have exited, all we have to
1084	 * do is to wait for the thread group leader to become inactive,
1085	 * and to assume its PID:
1086	 */
1087	if (!thread_group_leader(tsk)) {
1088		struct task_struct *leader = tsk->group_leader;
1089
1090		for (;;) {
1091			threadgroup_change_begin(tsk);
1092			write_lock_irq(&tasklist_lock);
1093			/*
1094			 * Do this under tasklist_lock to ensure that
1095			 * exit_notify() can't miss ->group_exit_task
1096			 */
1097			sig->notify_count = -1;
1098			if (likely(leader->exit_state))
1099				break;
1100			__set_current_state(TASK_KILLABLE);
1101			write_unlock_irq(&tasklist_lock);
1102			threadgroup_change_end(tsk);
1103			schedule();
1104			if (unlikely(__fatal_signal_pending(tsk)))
1105				goto killed;
1106		}
1107
1108		/*
1109		 * The only record we have of the real-time age of a
1110		 * process, regardless of execs it's done, is start_time.
1111		 * All the past CPU time is accumulated in signal_struct
1112		 * from sister threads now dead.  But in this non-leader
1113		 * exec, nothing survives from the original leader thread,
1114		 * whose birth marks the true age of this process now.
1115		 * When we take on its identity by switching to its PID, we
1116		 * also take its birthdate (always earlier than our own).
1117		 */
1118		tsk->start_time = leader->start_time;
1119		tsk->real_start_time = leader->real_start_time;
1120
1121		BUG_ON(!same_thread_group(leader, tsk));
1122		BUG_ON(has_group_leader_pid(tsk));
1123		/*
1124		 * An exec() starts a new thread group with the
1125		 * TGID of the previous thread group. Rehash the
1126		 * two threads with a switched PID, and release
1127		 * the former thread group leader:
1128		 */
1129
1130		/* Become a process group leader with the old leader's pid.
1131		 * The old leader becomes a thread of the this thread group.
1132		 * Note: The old leader also uses this pid until release_task
1133		 *       is called.  Odd but simple and correct.
1134		 */
1135		tsk->pid = leader->pid;
1136		change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1137		transfer_pid(leader, tsk, PIDTYPE_PGID);
1138		transfer_pid(leader, tsk, PIDTYPE_SID);
1139
1140		list_replace_rcu(&leader->tasks, &tsk->tasks);
1141		list_replace_init(&leader->sibling, &tsk->sibling);
1142
1143		tsk->group_leader = tsk;
1144		leader->group_leader = tsk;
1145
1146		tsk->exit_signal = SIGCHLD;
1147		leader->exit_signal = -1;
1148
1149		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1150		leader->exit_state = EXIT_DEAD;
1151
1152		/*
1153		 * We are going to release_task()->ptrace_unlink() silently,
1154		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1155		 * the tracer wont't block again waiting for this thread.
1156		 */
1157		if (unlikely(leader->ptrace))
1158			__wake_up_parent(leader, leader->parent);
1159		write_unlock_irq(&tasklist_lock);
1160		threadgroup_change_end(tsk);
1161
1162		release_task(leader);
1163	}
1164
1165	sig->group_exit_task = NULL;
1166	sig->notify_count = 0;
1167
1168no_thread_group:
1169	/* we have changed execution domain */
1170	tsk->exit_signal = SIGCHLD;
1171
1172#ifdef CONFIG_POSIX_TIMERS
1173	exit_itimers(sig);
1174	flush_itimer_signals();
1175#endif
1176
1177	if (atomic_read(&oldsighand->count) != 1) {
1178		struct sighand_struct *newsighand;
1179		/*
1180		 * This ->sighand is shared with the CLONE_SIGHAND
1181		 * but not CLONE_THREAD task, switch to the new one.
1182		 */
1183		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1184		if (!newsighand)
1185			return -ENOMEM;
1186
1187		atomic_set(&newsighand->count, 1);
1188		memcpy(newsighand->action, oldsighand->action,
1189		       sizeof(newsighand->action));
1190
1191		write_lock_irq(&tasklist_lock);
1192		spin_lock(&oldsighand->siglock);
1193		rcu_assign_pointer(tsk->sighand, newsighand);
1194		spin_unlock(&oldsighand->siglock);
1195		write_unlock_irq(&tasklist_lock);
1196
1197		__cleanup_sighand(oldsighand);
1198	}
1199
1200	BUG_ON(!thread_group_leader(tsk));
1201	return 0;
1202
1203killed:
1204	/* protects against exit_notify() and __exit_signal() */
1205	read_lock(&tasklist_lock);
1206	sig->group_exit_task = NULL;
1207	sig->notify_count = 0;
1208	read_unlock(&tasklist_lock);
1209	return -EAGAIN;
1210}
1211
1212char *get_task_comm(char *buf, struct task_struct *tsk)
1213{
1214	/* buf must be at least sizeof(tsk->comm) in size */
1215	task_lock(tsk);
1216	strncpy(buf, tsk->comm, sizeof(tsk->comm));
1217	task_unlock(tsk);
1218	return buf;
1219}
1220EXPORT_SYMBOL_GPL(get_task_comm);
1221
1222/*
1223 * These functions flushes out all traces of the currently running executable
1224 * so that a new one can be started
1225 */
1226
1227void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1228{
1229	task_lock(tsk);
1230	trace_task_rename(tsk, buf);
1231	strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1232	task_unlock(tsk);
1233	perf_event_comm(tsk, exec);
1234}
1235
1236int flush_old_exec(struct linux_binprm * bprm)
1237{
1238	int retval;
1239
1240	/*
1241	 * Make sure we have a private signal table and that
1242	 * we are unassociated from the previous thread group.
1243	 */
1244	retval = de_thread(current);
1245	if (retval)
1246		goto out;
1247
1248	/*
1249	 * Must be called _before_ exec_mmap() as bprm->mm is
1250	 * not visibile until then. This also enables the update
1251	 * to be lockless.
1252	 */
1253	set_mm_exe_file(bprm->mm, bprm->file);
1254
1255	/*
1256	 * Release all of the old mmap stuff
1257	 */
1258	acct_arg_size(bprm, 0);
1259	retval = exec_mmap(bprm->mm);
1260	if (retval)
1261		goto out;
1262
1263	bprm->mm = NULL;		/* We're using it now */
1264
1265	set_fs(USER_DS);
1266	current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1267					PF_NOFREEZE | PF_NO_SETAFFINITY);
1268	flush_thread();
1269	current->personality &= ~bprm->per_clear;
1270
1271	/*
1272	 * We have to apply CLOEXEC before we change whether the process is
1273	 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1274	 * trying to access the should-be-closed file descriptors of a process
1275	 * undergoing exec(2).
1276	 */
1277	do_close_on_exec(current->files);
1278	return 0;
1279
1280out:
1281	return retval;
1282}
1283EXPORT_SYMBOL(flush_old_exec);
1284
1285void would_dump(struct linux_binprm *bprm, struct file *file)
1286{
1287	struct inode *inode = file_inode(file);
1288	if (inode_permission(inode, MAY_READ) < 0) {
1289		struct user_namespace *old, *user_ns;
1290		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1291
1292		/* Ensure mm->user_ns contains the executable */
1293		user_ns = old = bprm->mm->user_ns;
1294		while ((user_ns != &init_user_ns) &&
1295		       !privileged_wrt_inode_uidgid(user_ns, inode))
1296			user_ns = user_ns->parent;
1297
1298		if (old != user_ns) {
1299			bprm->mm->user_ns = get_user_ns(user_ns);
1300			put_user_ns(old);
1301		}
1302	}
1303}
1304EXPORT_SYMBOL(would_dump);
1305
1306void setup_new_exec(struct linux_binprm * bprm)
1307{
1308	arch_pick_mmap_layout(current->mm);
1309
1310	/* This is the point of no return */
1311	current->sas_ss_sp = current->sas_ss_size = 0;
1312
1313	if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1314		set_dumpable(current->mm, SUID_DUMP_USER);
1315	else
1316		set_dumpable(current->mm, suid_dumpable);
1317
1318	perf_event_exec();
1319	__set_task_comm(current, kbasename(bprm->filename), true);
1320
1321	/* Set the new mm task size. We have to do that late because it may
1322	 * depend on TIF_32BIT which is only updated in flush_thread() on
1323	 * some architectures like powerpc
1324	 */
1325	current->mm->task_size = TASK_SIZE;
1326
1327	/* install the new credentials */
1328	if (!uid_eq(bprm->cred->uid, current_euid()) ||
1329	    !gid_eq(bprm->cred->gid, current_egid())) {
1330		current->pdeath_signal = 0;
1331	} else {
1332		if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1333			set_dumpable(current->mm, suid_dumpable);
1334	}
1335
1336	/* An exec changes our domain. We are no longer part of the thread
1337	   group */
1338	current->self_exec_id++;
1339	flush_signal_handlers(current, 0);
1340}
1341EXPORT_SYMBOL(setup_new_exec);
1342
1343/*
1344 * Prepare credentials and lock ->cred_guard_mutex.
1345 * install_exec_creds() commits the new creds and drops the lock.
1346 * Or, if exec fails before, free_bprm() should release ->cred and
1347 * and unlock.
1348 */
1349int prepare_bprm_creds(struct linux_binprm *bprm)
1350{
1351	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1352		return -ERESTARTNOINTR;
1353
1354	bprm->cred = prepare_exec_creds();
1355	if (likely(bprm->cred))
1356		return 0;
1357
1358	mutex_unlock(&current->signal->cred_guard_mutex);
1359	return -ENOMEM;
1360}
1361
1362static void free_bprm(struct linux_binprm *bprm)
1363{
1364	free_arg_pages(bprm);
1365	if (bprm->cred) {
1366		mutex_unlock(&current->signal->cred_guard_mutex);
1367		abort_creds(bprm->cred);
1368	}
1369	if (bprm->file) {
1370		allow_write_access(bprm->file);
1371		fput(bprm->file);
1372	}
1373	/* If a binfmt changed the interp, free it. */
1374	if (bprm->interp != bprm->filename)
1375		kfree(bprm->interp);
1376	kfree(bprm);
1377}
1378
1379int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1380{
1381	/* If a binfmt changed the interp, free it first. */
1382	if (bprm->interp != bprm->filename)
1383		kfree(bprm->interp);
1384	bprm->interp = kstrdup(interp, GFP_KERNEL);
1385	if (!bprm->interp)
1386		return -ENOMEM;
1387	return 0;
1388}
1389EXPORT_SYMBOL(bprm_change_interp);
1390
1391/*
1392 * install the new credentials for this executable
1393 */
1394void install_exec_creds(struct linux_binprm *bprm)
1395{
1396	security_bprm_committing_creds(bprm);
1397
1398	commit_creds(bprm->cred);
1399	bprm->cred = NULL;
1400
1401	/*
1402	 * Disable monitoring for regular users
1403	 * when executing setuid binaries. Must
1404	 * wait until new credentials are committed
1405	 * by commit_creds() above
1406	 */
1407	if (get_dumpable(current->mm) != SUID_DUMP_USER)
1408		perf_event_exit_task(current);
1409	/*
1410	 * cred_guard_mutex must be held at least to this point to prevent
1411	 * ptrace_attach() from altering our determination of the task's
1412	 * credentials; any time after this it may be unlocked.
1413	 */
1414	security_bprm_committed_creds(bprm);
1415	mutex_unlock(&current->signal->cred_guard_mutex);
1416}
1417EXPORT_SYMBOL(install_exec_creds);
1418
1419/*
1420 * determine how safe it is to execute the proposed program
1421 * - the caller must hold ->cred_guard_mutex to protect against
1422 *   PTRACE_ATTACH or seccomp thread-sync
1423 */
1424static void check_unsafe_exec(struct linux_binprm *bprm)
1425{
1426	struct task_struct *p = current, *t;
1427	unsigned n_fs;
1428
1429	if (p->ptrace) {
1430		if (ptracer_capable(p, current_user_ns()))
1431			bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1432		else
1433			bprm->unsafe |= LSM_UNSAFE_PTRACE;
1434	}
1435
1436	/*
1437	 * This isn't strictly necessary, but it makes it harder for LSMs to
1438	 * mess up.
1439	 */
1440	if (task_no_new_privs(current))
1441		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1442
1443	t = p;
1444	n_fs = 1;
1445	spin_lock(&p->fs->lock);
1446	rcu_read_lock();
1447	while_each_thread(p, t) {
1448		if (t->fs == p->fs)
1449			n_fs++;
1450	}
1451	rcu_read_unlock();
1452
1453	if (p->fs->users > n_fs)
1454		bprm->unsafe |= LSM_UNSAFE_SHARE;
1455	else
1456		p->fs->in_exec = 1;
1457	spin_unlock(&p->fs->lock);
1458}
1459
1460static void bprm_fill_uid(struct linux_binprm *bprm)
1461{
1462	struct inode *inode;
1463	unsigned int mode;
1464	kuid_t uid;
1465	kgid_t gid;
1466
1467	/*
1468	 * Since this can be called multiple times (via prepare_binprm),
1469	 * we must clear any previous work done when setting set[ug]id
1470	 * bits from any earlier bprm->file uses (for example when run
1471	 * first for a setuid script then again for its interpreter).
1472	 */
1473	bprm->cred->euid = current_euid();
1474	bprm->cred->egid = current_egid();
1475
1476	if (!mnt_may_suid(bprm->file->f_path.mnt))
1477		return;
1478
1479	if (task_no_new_privs(current))
1480		return;
1481
1482	inode = file_inode(bprm->file);
1483	mode = READ_ONCE(inode->i_mode);
1484	if (!(mode & (S_ISUID|S_ISGID)))
1485		return;
1486
1487	/* Be careful if suid/sgid is set */
1488	inode_lock(inode);
1489
1490	/* reload atomically mode/uid/gid now that lock held */
1491	mode = inode->i_mode;
1492	uid = inode->i_uid;
1493	gid = inode->i_gid;
1494	inode_unlock(inode);
1495
1496	/* We ignore suid/sgid if there are no mappings for them in the ns */
1497	if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1498		 !kgid_has_mapping(bprm->cred->user_ns, gid))
1499		return;
1500
1501	if (mode & S_ISUID) {
1502		bprm->per_clear |= PER_CLEAR_ON_SETID;
1503		bprm->cred->euid = uid;
1504	}
1505
1506	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1507		bprm->per_clear |= PER_CLEAR_ON_SETID;
1508		bprm->cred->egid = gid;
1509	}
1510}
1511
1512/*
1513 * Fill the binprm structure from the inode.
1514 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1515 *
1516 * This may be called multiple times for binary chains (scripts for example).
1517 */
1518int prepare_binprm(struct linux_binprm *bprm)
1519{
1520	int retval;
1521
1522	bprm_fill_uid(bprm);
1523
1524	/* fill in binprm security blob */
1525	retval = security_bprm_set_creds(bprm);
1526	if (retval)
1527		return retval;
1528	bprm->cred_prepared = 1;
1529
1530	memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1531	return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1532}
1533
1534EXPORT_SYMBOL(prepare_binprm);
1535
1536/*
1537 * Arguments are '\0' separated strings found at the location bprm->p
1538 * points to; chop off the first by relocating brpm->p to right after
1539 * the first '\0' encountered.
1540 */
1541int remove_arg_zero(struct linux_binprm *bprm)
1542{
1543	int ret = 0;
1544	unsigned long offset;
1545	char *kaddr;
1546	struct page *page;
1547
1548	if (!bprm->argc)
1549		return 0;
1550
1551	do {
1552		offset = bprm->p & ~PAGE_MASK;
1553		page = get_arg_page(bprm, bprm->p, 0);
1554		if (!page) {
1555			ret = -EFAULT;
1556			goto out;
1557		}
1558		kaddr = kmap_atomic(page);
1559
1560		for (; offset < PAGE_SIZE && kaddr[offset];
1561				offset++, bprm->p++)
1562			;
1563
1564		kunmap_atomic(kaddr);
1565		put_arg_page(page);
1566	} while (offset == PAGE_SIZE);
1567
1568	bprm->p++;
1569	bprm->argc--;
1570	ret = 0;
1571
1572out:
1573	return ret;
1574}
1575EXPORT_SYMBOL(remove_arg_zero);
1576
1577#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1578/*
1579 * cycle the list of binary formats handler, until one recognizes the image
1580 */
1581int search_binary_handler(struct linux_binprm *bprm)
1582{
1583	bool need_retry = IS_ENABLED(CONFIG_MODULES);
1584	struct linux_binfmt *fmt;
1585	int retval;
1586
1587	/* This allows 4 levels of binfmt rewrites before failing hard. */
1588	if (bprm->recursion_depth > 5)
1589		return -ELOOP;
1590
1591	retval = security_bprm_check(bprm);
1592	if (retval)
1593		return retval;
1594
1595	retval = -ENOENT;
1596 retry:
1597	read_lock(&binfmt_lock);
1598	list_for_each_entry(fmt, &formats, lh) {
1599		if (!try_module_get(fmt->module))
1600			continue;
1601		read_unlock(&binfmt_lock);
1602		bprm->recursion_depth++;
1603		retval = fmt->load_binary(bprm);
1604		read_lock(&binfmt_lock);
1605		put_binfmt(fmt);
1606		bprm->recursion_depth--;
1607		if (retval < 0 && !bprm->mm) {
1608			/* we got to flush_old_exec() and failed after it */
1609			read_unlock(&binfmt_lock);
1610			force_sigsegv(SIGSEGV, current);
1611			return retval;
1612		}
1613		if (retval != -ENOEXEC || !bprm->file) {
1614			read_unlock(&binfmt_lock);
1615			return retval;
1616		}
1617	}
1618	read_unlock(&binfmt_lock);
1619
1620	if (need_retry) {
1621		if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1622		    printable(bprm->buf[2]) && printable(bprm->buf[3]))
1623			return retval;
1624		if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1625			return retval;
1626		need_retry = false;
1627		goto retry;
1628	}
1629
1630	return retval;
1631}
1632EXPORT_SYMBOL(search_binary_handler);
1633
1634static int exec_binprm(struct linux_binprm *bprm)
1635{
1636	pid_t old_pid, old_vpid;
1637	int ret;
1638
1639	/* Need to fetch pid before load_binary changes it */
1640	old_pid = current->pid;
1641	rcu_read_lock();
1642	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1643	rcu_read_unlock();
1644
1645	ret = search_binary_handler(bprm);
1646	if (ret >= 0) {
1647		audit_bprm(bprm);
1648		trace_sched_process_exec(current, old_pid, bprm);
1649		ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1650		proc_exec_connector(current);
1651	}
1652
1653	return ret;
1654}
1655
1656/*
1657 * sys_execve() executes a new program.
1658 */
1659static int do_execveat_common(int fd, struct filename *filename,
1660			      struct user_arg_ptr argv,
1661			      struct user_arg_ptr envp,
1662			      int flags)
1663{
1664	char *pathbuf = NULL;
1665	struct linux_binprm *bprm;
1666	struct file *file;
1667	struct files_struct *displaced;
1668	int retval;
1669
1670	if (IS_ERR(filename))
1671		return PTR_ERR(filename);
1672
1673	/*
1674	 * We move the actual failure in case of RLIMIT_NPROC excess from
1675	 * set*uid() to execve() because too many poorly written programs
1676	 * don't check setuid() return code.  Here we additionally recheck
1677	 * whether NPROC limit is still exceeded.
1678	 */
1679	if ((current->flags & PF_NPROC_EXCEEDED) &&
1680	    atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1681		retval = -EAGAIN;
1682		goto out_ret;
1683	}
1684
1685	/* We're below the limit (still or again), so we don't want to make
1686	 * further execve() calls fail. */
1687	current->flags &= ~PF_NPROC_EXCEEDED;
1688
1689	retval = unshare_files(&displaced);
1690	if (retval)
1691		goto out_ret;
1692
1693	retval = -ENOMEM;
1694	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1695	if (!bprm)
1696		goto out_files;
1697
1698	retval = prepare_bprm_creds(bprm);
1699	if (retval)
1700		goto out_free;
1701
1702	check_unsafe_exec(bprm);
1703	current->in_execve = 1;
1704
1705	file = do_open_execat(fd, filename, flags);
1706	retval = PTR_ERR(file);
1707	if (IS_ERR(file))
1708		goto out_unmark;
1709
1710	sched_exec();
1711
1712	bprm->file = file;
1713	if (fd == AT_FDCWD || filename->name[0] == '/') {
1714		bprm->filename = filename->name;
1715	} else {
1716		if (filename->name[0] == '\0')
1717			pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1718		else
1719			pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1720					    fd, filename->name);
1721		if (!pathbuf) {
1722			retval = -ENOMEM;
1723			goto out_unmark;
1724		}
1725		/*
1726		 * Record that a name derived from an O_CLOEXEC fd will be
1727		 * inaccessible after exec. Relies on having exclusive access to
1728		 * current->files (due to unshare_files above).
1729		 */
1730		if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1731			bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1732		bprm->filename = pathbuf;
1733	}
1734	bprm->interp = bprm->filename;
1735
1736	retval = bprm_mm_init(bprm);
1737	if (retval)
1738		goto out_unmark;
1739
1740	bprm->argc = count(argv, MAX_ARG_STRINGS);
1741	if ((retval = bprm->argc) < 0)
1742		goto out;
1743
1744	bprm->envc = count(envp, MAX_ARG_STRINGS);
1745	if ((retval = bprm->envc) < 0)
1746		goto out;
1747
1748	retval = prepare_binprm(bprm);
1749	if (retval < 0)
1750		goto out;
1751
1752	retval = copy_strings_kernel(1, &bprm->filename, bprm);
1753	if (retval < 0)
1754		goto out;
1755
1756	bprm->exec = bprm->p;
1757	retval = copy_strings(bprm->envc, envp, bprm);
1758	if (retval < 0)
1759		goto out;
1760
1761	retval = copy_strings(bprm->argc, argv, bprm);
1762	if (retval < 0)
1763		goto out;
1764
1765	would_dump(bprm, bprm->file);
1766
1767	retval = exec_binprm(bprm);
1768	if (retval < 0)
1769		goto out;
1770
1771	/* execve succeeded */
1772	current->fs->in_exec = 0;
1773	current->in_execve = 0;
1774	acct_update_integrals(current);
1775	task_numa_free(current);
1776	free_bprm(bprm);
1777	kfree(pathbuf);
1778	putname(filename);
1779	if (displaced)
1780		put_files_struct(displaced);
1781	return retval;
1782
1783out:
1784	if (bprm->mm) {
1785		acct_arg_size(bprm, 0);
1786		mmput(bprm->mm);
1787	}
1788
1789out_unmark:
1790	current->fs->in_exec = 0;
1791	current->in_execve = 0;
1792
1793out_free:
1794	free_bprm(bprm);
1795	kfree(pathbuf);
1796
1797out_files:
1798	if (displaced)
1799		reset_files_struct(displaced);
1800out_ret:
1801	putname(filename);
1802	return retval;
1803}
1804
1805int do_execve(struct filename *filename,
1806	const char __user *const __user *__argv,
1807	const char __user *const __user *__envp)
1808{
1809	struct user_arg_ptr argv = { .ptr.native = __argv };
1810	struct user_arg_ptr envp = { .ptr.native = __envp };
1811	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1812}
1813
1814int do_execveat(int fd, struct filename *filename,
1815		const char __user *const __user *__argv,
1816		const char __user *const __user *__envp,
1817		int flags)
1818{
1819	struct user_arg_ptr argv = { .ptr.native = __argv };
1820	struct user_arg_ptr envp = { .ptr.native = __envp };
1821
1822	return do_execveat_common(fd, filename, argv, envp, flags);
1823}
1824
1825#ifdef CONFIG_COMPAT
1826static int compat_do_execve(struct filename *filename,
1827	const compat_uptr_t __user *__argv,
1828	const compat_uptr_t __user *__envp)
1829{
1830	struct user_arg_ptr argv = {
1831		.is_compat = true,
1832		.ptr.compat = __argv,
1833	};
1834	struct user_arg_ptr envp = {
1835		.is_compat = true,
1836		.ptr.compat = __envp,
1837	};
1838	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1839}
1840
1841static int compat_do_execveat(int fd, struct filename *filename,
1842			      const compat_uptr_t __user *__argv,
1843			      const compat_uptr_t __user *__envp,
1844			      int flags)
1845{
1846	struct user_arg_ptr argv = {
1847		.is_compat = true,
1848		.ptr.compat = __argv,
1849	};
1850	struct user_arg_ptr envp = {
1851		.is_compat = true,
1852		.ptr.compat = __envp,
1853	};
1854	return do_execveat_common(fd, filename, argv, envp, flags);
1855}
1856#endif
1857
1858void set_binfmt(struct linux_binfmt *new)
1859{
1860	struct mm_struct *mm = current->mm;
1861
1862	if (mm->binfmt)
1863		module_put(mm->binfmt->module);
1864
1865	mm->binfmt = new;
1866	if (new)
1867		__module_get(new->module);
1868}
1869EXPORT_SYMBOL(set_binfmt);
1870
1871/*
1872 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1873 */
1874void set_dumpable(struct mm_struct *mm, int value)
1875{
1876	unsigned long old, new;
1877
1878	if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1879		return;
1880
1881	do {
1882		old = ACCESS_ONCE(mm->flags);
1883		new = (old & ~MMF_DUMPABLE_MASK) | value;
1884	} while (cmpxchg(&mm->flags, old, new) != old);
1885}
1886
1887SYSCALL_DEFINE3(execve,
1888		const char __user *, filename,
1889		const char __user *const __user *, argv,
1890		const char __user *const __user *, envp)
1891{
1892	return do_execve(getname(filename), argv, envp);
1893}
1894
1895SYSCALL_DEFINE5(execveat,
1896		int, fd, const char __user *, filename,
1897		const char __user *const __user *, argv,
1898		const char __user *const __user *, envp,
1899		int, flags)
1900{
1901	int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1902
1903	return do_execveat(fd,
1904			   getname_flags(filename, lookup_flags, NULL),
1905			   argv, envp, flags);
1906}
1907
1908#ifdef CONFIG_COMPAT
1909COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1910	const compat_uptr_t __user *, argv,
1911	const compat_uptr_t __user *, envp)
1912{
1913	return compat_do_execve(getname(filename), argv, envp);
1914}
1915
1916COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1917		       const char __user *, filename,
1918		       const compat_uptr_t __user *, argv,
1919		       const compat_uptr_t __user *, envp,
1920		       int,  flags)
1921{
1922	int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1923
1924	return compat_do_execveat(fd,
1925				  getname_flags(filename, lookup_flags, NULL),
1926				  argv, envp, flags);
1927}
1928#endif