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/stat.h>
  30#include <linux/fcntl.h>
  31#include <linux/swap.h>
  32#include <linux/string.h>
  33#include <linux/init.h>
 
 
 
 
 
  34#include <linux/pagemap.h>
  35#include <linux/perf_event.h>
  36#include <linux/highmem.h>
  37#include <linux/spinlock.h>
  38#include <linux/key.h>
  39#include <linux/personality.h>
  40#include <linux/binfmts.h>
  41#include <linux/utsname.h>
  42#include <linux/pid_namespace.h>
  43#include <linux/module.h>
  44#include <linux/namei.h>
  45#include <linux/mount.h>
  46#include <linux/security.h>
  47#include <linux/syscalls.h>
  48#include <linux/tsacct_kern.h>
  49#include <linux/cn_proc.h>
  50#include <linux/audit.h>
  51#include <linux/tracehook.h>
  52#include <linux/kmod.h>
  53#include <linux/fsnotify.h>
  54#include <linux/fs_struct.h>
  55#include <linux/pipe_fs_i.h>
  56#include <linux/oom.h>
  57#include <linux/compat.h>
 
 
 
 
 
 
 
 
  58
  59#include <asm/uaccess.h>
  60#include <asm/mmu_context.h>
  61#include <asm/tlb.h>
  62#include <asm/exec.h>
  63
  64#include <trace/events/task.h>
  65#include "internal.h"
  66
  67#include <trace/events/sched.h>
  68
  69int core_uses_pid;
  70char core_pattern[CORENAME_MAX_SIZE] = "core";
  71unsigned int core_pipe_limit;
  72int suid_dumpable = 0;
  73
  74struct core_name {
  75	char *corename;
  76	int used, size;
  77};
  78static atomic_t call_count = ATOMIC_INIT(1);
  79
  80/* The maximal length of core_pattern is also specified in sysctl.c */
  81
  82static LIST_HEAD(formats);
  83static DEFINE_RWLOCK(binfmt_lock);
  84
  85void __register_binfmt(struct linux_binfmt * fmt, int insert)
  86{
  87	BUG_ON(!fmt);
  88	write_lock(&binfmt_lock);
  89	insert ? list_add(&fmt->lh, &formats) :
  90		 list_add_tail(&fmt->lh, &formats);
  91	write_unlock(&binfmt_lock);
  92}
  93
  94EXPORT_SYMBOL(__register_binfmt);
  95
  96void unregister_binfmt(struct linux_binfmt * fmt)
  97{
  98	write_lock(&binfmt_lock);
  99	list_del(&fmt->lh);
 100	write_unlock(&binfmt_lock);
 101}
 102
 103EXPORT_SYMBOL(unregister_binfmt);
 104
 105static inline void put_binfmt(struct linux_binfmt * fmt)
 106{
 107	module_put(fmt->module);
 108}
 109
 
 
 
 
 
 
 
 110/*
 111 * Note that a shared library must be both readable and executable due to
 112 * security reasons.
 113 *
 114 * Also note that we take the address to load from from the file itself.
 115 */
 116SYSCALL_DEFINE1(uselib, const char __user *, library)
 117{
 
 118	struct file *file;
 119	char *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 | MAY_OPEN,
 124		.intent = LOOKUP_OPEN
 
 125	};
 126
 127	if (IS_ERR(tmp))
 128		goto out;
 129
 130	file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 131	putname(tmp);
 132	error = PTR_ERR(file);
 133	if (IS_ERR(file))
 134		goto out;
 135
 136	error = -EINVAL;
 137	if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 138		goto exit;
 139
 140	error = -EACCES;
 141	if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 
 142		goto exit;
 143
 144	fsnotify_open(file);
 145
 146	error = -ENOEXEC;
 147	if(file->f_op) {
 148		struct linux_binfmt * fmt;
 149
 150		read_lock(&binfmt_lock);
 151		list_for_each_entry(fmt, &formats, lh) {
 152			if (!fmt->load_shlib)
 153				continue;
 154			if (!try_module_get(fmt->module))
 155				continue;
 156			read_unlock(&binfmt_lock);
 157			error = fmt->load_shlib(file);
 158			read_lock(&binfmt_lock);
 159			put_binfmt(fmt);
 160			if (error != -ENOEXEC)
 161				break;
 162		}
 163		read_unlock(&binfmt_lock);
 
 
 
 
 
 164	}
 
 165exit:
 166	fput(file);
 167out:
 168  	return error;
 169}
 
 170
 171#ifdef CONFIG_MMU
 172/*
 173 * The nascent bprm->mm is not visible until exec_mmap() but it can
 174 * use a lot of memory, account these pages in current->mm temporary
 175 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 176 * change the counter back via acct_arg_size(0).
 177 */
 178static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 179{
 180	struct mm_struct *mm = current->mm;
 181	long diff = (long)(pages - bprm->vma_pages);
 182
 183	if (!mm || !diff)
 184		return;
 185
 186	bprm->vma_pages = pages;
 187	add_mm_counter(mm, MM_ANONPAGES, diff);
 188}
 189
 190static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 191		int write)
 192{
 193	struct page *page;
 
 
 194	int ret;
 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	ret = get_user_pages(current, bprm->mm, pos,
 204			1, write, 1, &page, NULL);
 
 
 
 
 
 
 
 
 
 205	if (ret <= 0)
 206		return NULL;
 207
 208	if (write) {
 209		unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 210		struct rlimit *rlim;
 211
 212		acct_arg_size(bprm, size / PAGE_SIZE);
 213
 214		/*
 215		 * We've historically supported up to 32 pages (ARG_MAX)
 216		 * of argument strings even with small stacks
 217		 */
 218		if (size <= ARG_MAX)
 219			return page;
 220
 221		/*
 222		 * Limit to 1/4-th the stack size for the argv+env strings.
 223		 * This ensures that:
 224		 *  - the remaining binfmt code will not run out of stack space,
 225		 *  - the program will have a reasonable amount of stack left
 226		 *    to work from.
 227		 */
 228		rlim = current->signal->rlim;
 229		if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 230			put_page(page);
 231			return NULL;
 232		}
 233	}
 234
 235	return page;
 236}
 237
 238static void put_arg_page(struct page *page)
 239{
 240	put_page(page);
 241}
 242
 243static void free_arg_page(struct linux_binprm *bprm, int i)
 244{
 245}
 246
 247static void free_arg_pages(struct linux_binprm *bprm)
 248{
 249}
 250
 251static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 252		struct page *page)
 253{
 254	flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 255}
 256
 257static int __bprm_mm_init(struct linux_binprm *bprm)
 258{
 259	int err;
 260	struct vm_area_struct *vma = NULL;
 261	struct mm_struct *mm = bprm->mm;
 262
 263	bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 264	if (!vma)
 265		return -ENOMEM;
 
 
 
 
 
 
 266
 267	down_write(&mm->mmap_sem);
 268	vma->vm_mm = mm;
 
 
 
 
 
 269
 270	/*
 271	 * Place the stack at the largest stack address the architecture
 272	 * supports. Later, we'll move this to an appropriate place. We don't
 273	 * use STACK_TOP because that can depend on attributes which aren't
 274	 * configured yet.
 275	 */
 276	BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 277	vma->vm_end = STACK_TOP_MAX;
 278	vma->vm_start = vma->vm_end - PAGE_SIZE;
 279	vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 280	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 281	INIT_LIST_HEAD(&vma->anon_vma_chain);
 282
 283	err = insert_vm_struct(mm, vma);
 284	if (err)
 285		goto err;
 286
 287	mm->stack_vm = mm->total_vm = 1;
 288	up_write(&mm->mmap_sem);
 289	bprm->p = vma->vm_end - sizeof(void *);
 290	return 0;
 291err:
 292	up_write(&mm->mmap_sem);
 
 
 
 293	bprm->vma = NULL;
 294	kmem_cache_free(vm_area_cachep, vma);
 295	return err;
 296}
 297
 298static bool valid_arg_len(struct linux_binprm *bprm, long len)
 299{
 300	return len <= MAX_ARG_STRLEN;
 301}
 302
 303#else
 304
 305static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 306{
 307}
 308
 309static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 310		int write)
 311{
 312	struct page *page;
 313
 314	page = bprm->page[pos / PAGE_SIZE];
 315	if (!page && write) {
 316		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 317		if (!page)
 318			return NULL;
 319		bprm->page[pos / PAGE_SIZE] = page;
 320	}
 321
 322	return page;
 323}
 324
 325static void put_arg_page(struct page *page)
 326{
 327}
 328
 329static void free_arg_page(struct linux_binprm *bprm, int i)
 330{
 331	if (bprm->page[i]) {
 332		__free_page(bprm->page[i]);
 333		bprm->page[i] = NULL;
 334	}
 335}
 336
 337static void free_arg_pages(struct linux_binprm *bprm)
 338{
 339	int i;
 340
 341	for (i = 0; i < MAX_ARG_PAGES; i++)
 342		free_arg_page(bprm, i);
 343}
 344
 345static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 346		struct page *page)
 347{
 348}
 349
 350static int __bprm_mm_init(struct linux_binprm *bprm)
 351{
 352	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 353	return 0;
 354}
 355
 356static bool valid_arg_len(struct linux_binprm *bprm, long len)
 357{
 358	return len <= bprm->p;
 359}
 360
 361#endif /* CONFIG_MMU */
 362
 363/*
 364 * Create a new mm_struct and populate it with a temporary stack
 365 * vm_area_struct.  We don't have enough context at this point to set the stack
 366 * flags, permissions, and offset, so we use temporary values.  We'll update
 367 * them later in setup_arg_pages().
 368 */
 369int bprm_mm_init(struct linux_binprm *bprm)
 370{
 371	int err;
 372	struct mm_struct *mm = NULL;
 373
 374	bprm->mm = mm = mm_alloc();
 375	err = -ENOMEM;
 376	if (!mm)
 377		goto err;
 378
 379	err = init_new_context(current, mm);
 380	if (err)
 381		goto err;
 
 382
 383	err = __bprm_mm_init(bprm);
 384	if (err)
 385		goto err;
 386
 387	return 0;
 388
 389err:
 390	if (mm) {
 391		bprm->mm = NULL;
 392		mmdrop(mm);
 393	}
 394
 395	return err;
 396}
 397
 398struct user_arg_ptr {
 399#ifdef CONFIG_COMPAT
 400	bool is_compat;
 401#endif
 402	union {
 403		const char __user *const __user *native;
 404#ifdef CONFIG_COMPAT
 405		compat_uptr_t __user *compat;
 406#endif
 407	} ptr;
 408};
 409
 410static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 411{
 412	const char __user *native;
 413
 414#ifdef CONFIG_COMPAT
 415	if (unlikely(argv.is_compat)) {
 416		compat_uptr_t compat;
 417
 418		if (get_user(compat, argv.ptr.compat + nr))
 419			return ERR_PTR(-EFAULT);
 420
 421		return compat_ptr(compat);
 422	}
 423#endif
 424
 425	if (get_user(native, argv.ptr.native + nr))
 426		return ERR_PTR(-EFAULT);
 427
 428	return native;
 429}
 430
 431/*
 432 * count() counts the number of strings in array ARGV.
 433 */
 434static int count(struct user_arg_ptr argv, int max)
 435{
 436	int i = 0;
 437
 438	if (argv.ptr.native != NULL) {
 439		for (;;) {
 440			const char __user *p = get_user_arg_ptr(argv, i);
 441
 442			if (!p)
 443				break;
 444
 445			if (IS_ERR(p))
 446				return -EFAULT;
 447
 448			if (i++ >= max)
 449				return -E2BIG;
 
 450
 451			if (fatal_signal_pending(current))
 452				return -ERESTARTNOHAND;
 453			cond_resched();
 454		}
 455	}
 456	return i;
 457}
 458
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 459/*
 460 * 'copy_strings()' copies argument/environment strings from the old
 461 * processes's memory to the new process's stack.  The call to get_user_pages()
 462 * ensures the destination page is created and not swapped out.
 463 */
 464static int copy_strings(int argc, struct user_arg_ptr argv,
 465			struct linux_binprm *bprm)
 466{
 467	struct page *kmapped_page = NULL;
 468	char *kaddr = NULL;
 469	unsigned long kpos = 0;
 470	int ret;
 471
 472	while (argc-- > 0) {
 473		const char __user *str;
 474		int len;
 475		unsigned long pos;
 476
 477		ret = -EFAULT;
 478		str = get_user_arg_ptr(argv, argc);
 479		if (IS_ERR(str))
 480			goto out;
 481
 482		len = strnlen_user(str, MAX_ARG_STRLEN);
 483		if (!len)
 484			goto out;
 485
 486		ret = -E2BIG;
 487		if (!valid_arg_len(bprm, len))
 488			goto out;
 489
 490		/* We're going to work our way backwords. */
 491		pos = bprm->p;
 492		str += len;
 493		bprm->p -= len;
 
 
 494
 495		while (len > 0) {
 496			int offset, bytes_to_copy;
 497
 498			if (fatal_signal_pending(current)) {
 499				ret = -ERESTARTNOHAND;
 500				goto out;
 501			}
 502			cond_resched();
 503
 504			offset = pos % PAGE_SIZE;
 505			if (offset == 0)
 506				offset = PAGE_SIZE;
 507
 508			bytes_to_copy = offset;
 509			if (bytes_to_copy > len)
 510				bytes_to_copy = len;
 511
 512			offset -= bytes_to_copy;
 513			pos -= bytes_to_copy;
 514			str -= bytes_to_copy;
 515			len -= bytes_to_copy;
 516
 517			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 518				struct page *page;
 519
 520				page = get_arg_page(bprm, pos, 1);
 521				if (!page) {
 522					ret = -E2BIG;
 523					goto out;
 524				}
 525
 526				if (kmapped_page) {
 527					flush_kernel_dcache_page(kmapped_page);
 528					kunmap(kmapped_page);
 529					put_arg_page(kmapped_page);
 530				}
 531				kmapped_page = page;
 532				kaddr = kmap(kmapped_page);
 533				kpos = pos & PAGE_MASK;
 534				flush_arg_page(bprm, kpos, kmapped_page);
 535			}
 536			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 537				ret = -EFAULT;
 538				goto out;
 539			}
 540		}
 541	}
 542	ret = 0;
 543out:
 544	if (kmapped_page) {
 545		flush_kernel_dcache_page(kmapped_page);
 546		kunmap(kmapped_page);
 547		put_arg_page(kmapped_page);
 548	}
 549	return ret;
 550}
 551
 552/*
 553 * Like copy_strings, but get argv and its values from kernel memory.
 554 */
 555int copy_strings_kernel(int argc, const char *const *__argv,
 556			struct linux_binprm *bprm)
 557{
 558	int r;
 559	mm_segment_t oldfs = get_fs();
 560	struct user_arg_ptr argv = {
 561		.ptr.native = (const char __user *const  __user *)__argv,
 562	};
 563
 564	set_fs(KERNEL_DS);
 565	r = copy_strings(argc, argv, bprm);
 566	set_fs(oldfs);
 567
 568	return r;
 569}
 570EXPORT_SYMBOL(copy_strings_kernel);
 571
 572#ifdef CONFIG_MMU
 573
 574/*
 575 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 576 * the binfmt code determines where the new stack should reside, we shift it to
 577 * its final location.  The process proceeds as follows:
 578 *
 579 * 1) Use shift to calculate the new vma endpoints.
 580 * 2) Extend vma to cover both the old and new ranges.  This ensures the
 581 *    arguments passed to subsequent functions are consistent.
 582 * 3) Move vma's page tables to the new range.
 583 * 4) Free up any cleared pgd range.
 584 * 5) Shrink the vma to cover only the new range.
 585 */
 586static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 587{
 588	struct mm_struct *mm = vma->vm_mm;
 589	unsigned long old_start = vma->vm_start;
 590	unsigned long old_end = vma->vm_end;
 591	unsigned long length = old_end - old_start;
 592	unsigned long new_start = old_start - shift;
 593	unsigned long new_end = old_end - shift;
 594	struct mmu_gather tlb;
 595
 596	BUG_ON(new_start > new_end);
 597
 598	/*
 599	 * ensure there are no vmas between where we want to go
 600	 * and where we are
 601	 */
 602	if (vma != find_vma(mm, new_start))
 603		return -EFAULT;
 
 
 604
 605	/*
 606	 * cover the whole range: [new_start, old_end)
 607	 */
 608	if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 609		return -ENOMEM;
 
 
 
 
 
 
 
 
 
 610
 611	/*
 612	 * move the page tables downwards, on failure we rely on
 613	 * process cleanup to remove whatever mess we made.
 614	 */
 615	if (length != move_page_tables(vma, old_start,
 616				       vma, new_start, length))
 617		return -ENOMEM;
 618
 619	lru_add_drain();
 620	tlb_gather_mmu(&tlb, mm, 0);
 621	if (new_end > old_start) {
 622		/*
 623		 * when the old and new regions overlap clear from new_end.
 624		 */
 625		free_pgd_range(&tlb, new_end, old_end, new_end,
 626			vma->vm_next ? vma->vm_next->vm_start : 0);
 627	} else {
 628		/*
 629		 * otherwise, clean from old_start; this is done to not touch
 630		 * the address space in [new_end, old_start) some architectures
 631		 * have constraints on va-space that make this illegal (IA64) -
 632		 * for the others its just a little faster.
 633		 */
 634		free_pgd_range(&tlb, old_start, old_end, new_end,
 635			vma->vm_next ? vma->vm_next->vm_start : 0);
 636	}
 637	tlb_finish_mmu(&tlb, new_end, old_end);
 638
 639	/*
 640	 * Shrink the vma to just the new range.  Always succeeds.
 641	 */
 642	vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 643
 
 
 
 
 
 
 
 
 
 
 
 644	return 0;
 645}
 646
 
 
 647/*
 648 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 649 * the stack is optionally relocated, and some extra space is added.
 650 */
 651int setup_arg_pages(struct linux_binprm *bprm,
 652		    unsigned long stack_top,
 653		    int executable_stack)
 654{
 655	unsigned long ret;
 656	unsigned long stack_shift;
 657	struct mm_struct *mm = current->mm;
 658	struct vm_area_struct *vma = bprm->vma;
 659	struct vm_area_struct *prev = NULL;
 660	unsigned long vm_flags;
 661	unsigned long stack_base;
 662	unsigned long stack_size;
 663	unsigned long stack_expand;
 664	unsigned long rlim_stack;
 
 
 665
 666#ifdef CONFIG_STACK_GROWSUP
 667	/* Limit stack size to 1GB */
 668	stack_base = rlimit_max(RLIMIT_STACK);
 669	if (stack_base > (1 << 30))
 670		stack_base = 1 << 30;
 
 
 
 
 671
 672	/* Make sure we didn't let the argument array grow too large. */
 673	if (vma->vm_end - vma->vm_start > stack_base)
 674		return -ENOMEM;
 675
 676	stack_base = PAGE_ALIGN(stack_top - stack_base);
 677
 678	stack_shift = vma->vm_start - stack_base;
 679	mm->arg_start = bprm->p - stack_shift;
 680	bprm->p = vma->vm_end - stack_shift;
 681#else
 682	stack_top = arch_align_stack(stack_top);
 683	stack_top = PAGE_ALIGN(stack_top);
 684
 685	if (unlikely(stack_top < mmap_min_addr) ||
 686	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 687		return -ENOMEM;
 688
 689	stack_shift = vma->vm_end - stack_top;
 690
 691	bprm->p -= stack_shift;
 692	mm->arg_start = bprm->p;
 693#endif
 694
 695	if (bprm->loader)
 696		bprm->loader -= stack_shift;
 697	bprm->exec -= stack_shift;
 698
 699	down_write(&mm->mmap_sem);
 
 
 700	vm_flags = VM_STACK_FLAGS;
 701
 702	/*
 703	 * Adjust stack execute permissions; explicitly enable for
 704	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 705	 * (arch default) otherwise.
 706	 */
 707	if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 708		vm_flags |= VM_EXEC;
 709	else if (executable_stack == EXSTACK_DISABLE_X)
 710		vm_flags &= ~VM_EXEC;
 711	vm_flags |= mm->def_flags;
 712	vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 713
 714	ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 
 
 
 715			vm_flags);
 
 
 716	if (ret)
 717		goto out_unlock;
 718	BUG_ON(prev != vma);
 719
 
 
 
 
 
 720	/* Move stack pages down in memory. */
 721	if (stack_shift) {
 722		ret = shift_arg_pages(vma, stack_shift);
 
 
 
 
 
 723		if (ret)
 724			goto out_unlock;
 725	}
 726
 727	/* mprotect_fixup is overkill to remove the temporary stack flags */
 728	vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 729
 730	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 731	stack_size = vma->vm_end - vma->vm_start;
 732	/*
 733	 * Align this down to a page boundary as expand_stack
 734	 * will align it up.
 735	 */
 736	rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 
 
 
 737#ifdef CONFIG_STACK_GROWSUP
 738	if (stack_size + stack_expand > rlim_stack)
 739		stack_base = vma->vm_start + rlim_stack;
 740	else
 741		stack_base = vma->vm_end + stack_expand;
 742#else
 743	if (stack_size + stack_expand > rlim_stack)
 744		stack_base = vma->vm_end - rlim_stack;
 745	else
 746		stack_base = vma->vm_start - stack_expand;
 747#endif
 748	current->mm->start_stack = bprm->p;
 749	ret = expand_stack(vma, stack_base);
 750	if (ret)
 751		ret = -EFAULT;
 752
 753out_unlock:
 754	up_write(&mm->mmap_sem);
 755	return ret;
 756}
 757EXPORT_SYMBOL(setup_arg_pages);
 758
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 759#endif /* CONFIG_MMU */
 760
 761struct file *open_exec(const char *name)
 
 
 
 
 762{
 763	struct file *file;
 764	int err;
 765	static const struct open_flags open_exec_flags = {
 
 766		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 767		.acc_mode = MAY_EXEC | MAY_OPEN,
 768		.intent = LOOKUP_OPEN
 
 769	};
 770
 771	file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 772	if (IS_ERR(file))
 773		goto out;
 774
 775	err = -EACCES;
 776	if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 777		goto exit;
 778
 779	if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 780		goto exit;
 
 781
 782	fsnotify_open(file);
 
 
 
 
 
 
 
 783
 784	err = deny_write_access(file);
 785	if (err)
 786		goto exit;
 787
 788out:
 789	return file;
 790
 791exit:
 792	fput(file);
 793	return ERR_PTR(err);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 794}
 795EXPORT_SYMBOL(open_exec);
 796
 797int kernel_read(struct file *file, loff_t offset,
 798		char *addr, unsigned long count)
 799{
 800	mm_segment_t old_fs;
 801	loff_t pos = offset;
 802	int result;
 803
 804	old_fs = get_fs();
 805	set_fs(get_ds());
 806	/* The cast to a user pointer is valid due to the set_fs() */
 807	result = vfs_read(file, (void __user *)addr, count, &pos);
 808	set_fs(old_fs);
 809	return result;
 810}
 
 
 811
 812EXPORT_SYMBOL(kernel_read);
 813
 
 
 
 814static int exec_mmap(struct mm_struct *mm)
 815{
 816	struct task_struct *tsk;
 817	struct mm_struct * old_mm, *active_mm;
 
 818
 819	/* Notify parent that we're no longer interested in the old VM */
 820	tsk = current;
 821	old_mm = current->mm;
 822	mm_release(tsk, old_mm);
 
 
 
 
 823
 824	if (old_mm) {
 825		sync_mm_rss(old_mm);
 826		/*
 827		 * Make sure that if there is a core dump in progress
 828		 * for the old mm, we get out and die instead of going
 829		 * through with the exec.  We must hold mmap_sem around
 830		 * checking core_state and changing tsk->mm.
 831		 */
 832		down_read(&old_mm->mmap_sem);
 833		if (unlikely(old_mm->core_state)) {
 834			up_read(&old_mm->mmap_sem);
 835			return -EINTR;
 836		}
 837	}
 
 838	task_lock(tsk);
 
 
 
 839	active_mm = tsk->active_mm;
 840	tsk->mm = mm;
 841	tsk->active_mm = mm;
 
 
 
 
 
 
 
 
 
 
 
 842	activate_mm(active_mm, mm);
 
 
 
 843	task_unlock(tsk);
 844	arch_pick_mmap_layout(mm);
 845	if (old_mm) {
 846		up_read(&old_mm->mmap_sem);
 847		BUG_ON(active_mm != old_mm);
 848		setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 849		mm_update_next_owner(old_mm);
 850		mmput(old_mm);
 851		return 0;
 852	}
 853	mmdrop(active_mm);
 854	return 0;
 855}
 856
 857/*
 858 * This function makes sure the current process has its own signal table,
 859 * so that flush_signal_handlers can later reset the handlers without
 860 * disturbing other processes.  (Other processes might share the signal
 861 * table via the CLONE_SIGHAND option to clone().)
 862 */
 863static int de_thread(struct task_struct *tsk)
 864{
 865	struct signal_struct *sig = tsk->signal;
 866	struct sighand_struct *oldsighand = tsk->sighand;
 867	spinlock_t *lock = &oldsighand->siglock;
 868
 869	if (thread_group_empty(tsk))
 870		goto no_thread_group;
 871
 872	/*
 873	 * Kill all other threads in the thread group.
 874	 */
 875	spin_lock_irq(lock);
 876	if (signal_group_exit(sig)) {
 877		/*
 878		 * Another group action in progress, just
 879		 * return so that the signal is processed.
 880		 */
 881		spin_unlock_irq(lock);
 882		return -EAGAIN;
 883	}
 884
 885	sig->group_exit_task = tsk;
 886	sig->notify_count = zap_other_threads(tsk);
 887	if (!thread_group_leader(tsk))
 888		sig->notify_count--;
 889
 890	while (sig->notify_count) {
 891		__set_current_state(TASK_UNINTERRUPTIBLE);
 892		spin_unlock_irq(lock);
 893		schedule();
 
 
 894		spin_lock_irq(lock);
 895	}
 896	spin_unlock_irq(lock);
 897
 898	/*
 899	 * At this point all other threads have exited, all we have to
 900	 * do is to wait for the thread group leader to become inactive,
 901	 * and to assume its PID:
 902	 */
 903	if (!thread_group_leader(tsk)) {
 904		struct task_struct *leader = tsk->group_leader;
 905
 906		sig->notify_count = -1;	/* for exit_notify() */
 907		for (;;) {
 
 908			write_lock_irq(&tasklist_lock);
 
 
 
 
 
 909			if (likely(leader->exit_state))
 910				break;
 911			__set_current_state(TASK_UNINTERRUPTIBLE);
 912			write_unlock_irq(&tasklist_lock);
 
 913			schedule();
 
 
 914		}
 915
 916		/*
 917		 * The only record we have of the real-time age of a
 918		 * process, regardless of execs it's done, is start_time.
 919		 * All the past CPU time is accumulated in signal_struct
 920		 * from sister threads now dead.  But in this non-leader
 921		 * exec, nothing survives from the original leader thread,
 922		 * whose birth marks the true age of this process now.
 923		 * When we take on its identity by switching to its PID, we
 924		 * also take its birthdate (always earlier than our own).
 925		 */
 926		tsk->start_time = leader->start_time;
 
 927
 928		BUG_ON(!same_thread_group(leader, tsk));
 929		BUG_ON(has_group_leader_pid(tsk));
 930		/*
 931		 * An exec() starts a new thread group with the
 932		 * TGID of the previous thread group. Rehash the
 933		 * two threads with a switched PID, and release
 934		 * the former thread group leader:
 935		 */
 936
 937		/* Become a process group leader with the old leader's pid.
 938		 * The old leader becomes a thread of the this thread group.
 939		 * Note: The old leader also uses this pid until release_task
 940		 *       is called.  Odd but simple and correct.
 941		 */
 942		detach_pid(tsk, PIDTYPE_PID);
 943		tsk->pid = leader->pid;
 944		attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 945		transfer_pid(leader, tsk, PIDTYPE_PGID);
 946		transfer_pid(leader, tsk, PIDTYPE_SID);
 947
 948		list_replace_rcu(&leader->tasks, &tsk->tasks);
 949		list_replace_init(&leader->sibling, &tsk->sibling);
 950
 951		tsk->group_leader = tsk;
 952		leader->group_leader = tsk;
 953
 954		tsk->exit_signal = SIGCHLD;
 955		leader->exit_signal = -1;
 956
 957		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 958		leader->exit_state = EXIT_DEAD;
 959
 960		/*
 961		 * We are going to release_task()->ptrace_unlink() silently,
 962		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 963		 * the tracer wont't block again waiting for this thread.
 964		 */
 965		if (unlikely(leader->ptrace))
 966			__wake_up_parent(leader, leader->parent);
 967		write_unlock_irq(&tasklist_lock);
 
 968
 969		release_task(leader);
 970	}
 971
 972	sig->group_exit_task = NULL;
 973	sig->notify_count = 0;
 974
 975no_thread_group:
 976	/* we have changed execution domain */
 977	tsk->exit_signal = SIGCHLD;
 978
 979	exit_itimers(sig);
 980	flush_itimer_signals();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 981
 982	if (atomic_read(&oldsighand->count) != 1) {
 983		struct sighand_struct *newsighand;
 984		/*
 985		 * This ->sighand is shared with the CLONE_SIGHAND
 986		 * but not CLONE_THREAD task, switch to the new one.
 987		 */
 988		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 989		if (!newsighand)
 990			return -ENOMEM;
 991
 992		atomic_set(&newsighand->count, 1);
 993		memcpy(newsighand->action, oldsighand->action,
 994		       sizeof(newsighand->action));
 995
 996		write_lock_irq(&tasklist_lock);
 997		spin_lock(&oldsighand->siglock);
 998		rcu_assign_pointer(tsk->sighand, newsighand);
 
 
 999		spin_unlock(&oldsighand->siglock);
1000		write_unlock_irq(&tasklist_lock);
1001
1002		__cleanup_sighand(oldsighand);
1003	}
1004
1005	BUG_ON(!thread_group_leader(tsk));
1006	return 0;
1007}
1008
1009/*
1010 * These functions flushes out all traces of the currently running executable
1011 * so that a new one can be started
1012 */
1013static void flush_old_files(struct files_struct * files)
1014{
1015	long j = -1;
1016	struct fdtable *fdt;
1017
1018	spin_lock(&files->file_lock);
1019	for (;;) {
1020		unsigned long set, i;
1021
1022		j++;
1023		i = j * BITS_PER_LONG;
1024		fdt = files_fdtable(files);
1025		if (i >= fdt->max_fds)
1026			break;
1027		set = fdt->close_on_exec[j];
1028		if (!set)
1029			continue;
1030		fdt->close_on_exec[j] = 0;
1031		spin_unlock(&files->file_lock);
1032		for ( ; set ; i++,set >>= 1) {
1033			if (set & 1) {
1034				sys_close(i);
1035			}
1036		}
1037		spin_lock(&files->file_lock);
1038
1039	}
1040	spin_unlock(&files->file_lock);
1041}
1042
1043char *get_task_comm(char *buf, struct task_struct *tsk)
1044{
1045	/* buf must be at least sizeof(tsk->comm) in size */
1046	task_lock(tsk);
1047	strncpy(buf, tsk->comm, sizeof(tsk->comm));
 
1048	task_unlock(tsk);
1049	return buf;
1050}
1051EXPORT_SYMBOL_GPL(get_task_comm);
1052
1053void set_task_comm(struct task_struct *tsk, char *buf)
 
 
 
 
 
 
1054{
1055	task_lock(tsk);
 
1056
1057	trace_task_rename(tsk, buf);
 
 
 
1058
1059	/*
1060	 * Threads may access current->comm without holding
1061	 * the task lock, so write the string carefully.
1062	 * Readers without a lock may see incomplete new
1063	 * names but are safe from non-terminating string reads.
1064	 */
1065	memset(tsk->comm, 0, TASK_COMM_LEN);
1066	wmb();
1067	strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1068	task_unlock(tsk);
1069	perf_event_comm(tsk);
1070}
1071
1072static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1073{
1074	int i, ch;
 
1075
1076	/* Copies the binary name from after last slash */
1077	for (i = 0; (ch = *(fn++)) != '\0';) {
1078		if (ch == '/')
1079			i = 0; /* overwrite what we wrote */
1080		else
1081			if (i < len - 1)
1082				tcomm[i++] = ch;
1083	}
1084	tcomm[i] = '\0';
1085}
1086
1087int flush_old_exec(struct linux_binprm * bprm)
1088{
1089	int retval;
 
 
 
 
 
 
1090
1091	/*
1092	 * Make sure we have a private signal table and that
1093	 * we are unassociated from the previous thread group.
 
1094	 */
1095	retval = de_thread(current);
1096	if (retval)
1097		goto out;
1098
1099	set_mm_exe_file(bprm->mm, bprm->file);
 
 
 
1100
1101	filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1102	/*
1103	 * Release all of the old mmap stuff
1104	 */
1105	acct_arg_size(bprm, 0);
1106	retval = exec_mmap(bprm->mm);
1107	if (retval)
1108		goto out;
1109
1110	bprm->mm = NULL;		/* We're using it now */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1111
1112	set_fs(USER_DS);
1113	current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1114	flush_thread();
1115	current->personality &= ~bprm->per_clear;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1116
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1117	return 0;
1118
 
 
 
 
 
1119out:
1120	return retval;
1121}
1122EXPORT_SYMBOL(flush_old_exec);
1123
1124void would_dump(struct linux_binprm *bprm, struct file *file)
1125{
1126	if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
 
 
 
1127		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
 
 
 
 
 
 
 
 
 
 
 
 
1128}
1129EXPORT_SYMBOL(would_dump);
1130
1131void setup_new_exec(struct linux_binprm * bprm)
1132{
1133	arch_pick_mmap_layout(current->mm);
 
1134
1135	/* This is the point of no return */
1136	current->sas_ss_sp = current->sas_ss_size = 0;
1137
1138	if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1139		set_dumpable(current->mm, 1);
1140	else
1141		set_dumpable(current->mm, suid_dumpable);
1142
1143	set_task_comm(current, bprm->tcomm);
1144
1145	/* Set the new mm task size. We have to do that late because it may
1146	 * depend on TIF_32BIT which is only updated in flush_thread() on
1147	 * some architectures like powerpc
1148	 */
1149	current->mm->task_size = TASK_SIZE;
1150
1151	/* install the new credentials */
1152	if (!uid_eq(bprm->cred->uid, current_euid()) ||
1153	    !gid_eq(bprm->cred->gid, current_egid())) {
1154		current->pdeath_signal = 0;
1155	} else {
1156		would_dump(bprm, bprm->file);
1157		if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1158			set_dumpable(current->mm, suid_dumpable);
1159	}
1160
1161	/*
1162	 * Flush performance counters when crossing a
1163	 * security domain:
1164	 */
1165	if (!get_dumpable(current->mm))
1166		perf_event_exit_task(current);
1167
1168	/* An exec changes our domain. We are no longer part of the thread
1169	   group */
1170
1171	current->self_exec_id++;
1172			
1173	flush_signal_handlers(current, 0);
1174	flush_old_files(current->files);
1175}
1176EXPORT_SYMBOL(setup_new_exec);
1177
 
 
 
 
 
 
 
 
 
 
1178/*
1179 * Prepare credentials and lock ->cred_guard_mutex.
1180 * install_exec_creds() commits the new creds and drops the lock.
1181 * Or, if exec fails before, free_bprm() should release ->cred and
1182 * and unlock.
1183 */
1184int prepare_bprm_creds(struct linux_binprm *bprm)
1185{
1186	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1187		return -ERESTARTNOINTR;
1188
1189	bprm->cred = prepare_exec_creds();
1190	if (likely(bprm->cred))
1191		return 0;
1192
1193	mutex_unlock(&current->signal->cred_guard_mutex);
1194	return -ENOMEM;
1195}
1196
1197void free_bprm(struct linux_binprm *bprm)
 
 
 
 
 
 
 
 
 
1198{
 
 
 
 
1199	free_arg_pages(bprm);
1200	if (bprm->cred) {
1201		mutex_unlock(&current->signal->cred_guard_mutex);
1202		abort_creds(bprm->cred);
1203	}
 
 
 
 
 
 
 
1204	kfree(bprm);
1205}
1206
1207/*
1208 * install the new credentials for this executable
1209 */
1210void install_exec_creds(struct linux_binprm *bprm)
1211{
1212	security_bprm_committing_creds(bprm);
 
 
1213
1214	commit_creds(bprm->cred);
1215	bprm->cred = NULL;
1216	/*
1217	 * cred_guard_mutex must be held at least to this point to prevent
1218	 * ptrace_attach() from altering our determination of the task's
1219	 * credentials; any time after this it may be unlocked.
1220	 */
1221	security_bprm_committed_creds(bprm);
1222	mutex_unlock(&current->signal->cred_guard_mutex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1223}
1224EXPORT_SYMBOL(install_exec_creds);
 
 
 
 
 
 
 
 
 
 
 
1225
1226/*
1227 * determine how safe it is to execute the proposed program
1228 * - the caller must hold ->cred_guard_mutex to protect against
1229 *   PTRACE_ATTACH
1230 */
1231static int check_unsafe_exec(struct linux_binprm *bprm)
1232{
1233	struct task_struct *p = current, *t;
1234	unsigned n_fs;
1235	int res = 0;
1236
1237	if (p->ptrace) {
1238		if (p->ptrace & PT_PTRACE_CAP)
1239			bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1240		else
1241			bprm->unsafe |= LSM_UNSAFE_PTRACE;
1242	}
1243
1244	/*
1245	 * This isn't strictly necessary, but it makes it harder for LSMs to
1246	 * mess up.
1247	 */
1248	if (current->no_new_privs)
1249		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1250
 
 
 
 
 
 
1251	n_fs = 1;
1252	spin_lock(&p->fs->lock);
1253	rcu_read_lock();
1254	for (t = next_thread(p); t != p; t = next_thread(t)) {
1255		if (t->fs == p->fs)
1256			n_fs++;
1257	}
1258	rcu_read_unlock();
1259
1260	if (p->fs->users > n_fs) {
 
1261		bprm->unsafe |= LSM_UNSAFE_SHARE;
1262	} else {
1263		res = -EAGAIN;
1264		if (!p->fs->in_exec) {
1265			p->fs->in_exec = 1;
1266			res = 1;
1267		}
1268	}
1269	spin_unlock(&p->fs->lock);
1270
1271	return res;
1272}
1273
1274/* 
1275 * Fill the binprm structure from the inode. 
1276 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1277 *
1278 * This may be called multiple times for binary chains (scripts for example).
1279 */
1280int prepare_binprm(struct linux_binprm *bprm)
1281{
1282	umode_t mode;
1283	struct inode * inode = bprm->file->f_path.dentry->d_inode;
1284	int retval;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1285
 
 
 
 
1286	mode = inode->i_mode;
1287	if (bprm->file->f_op == NULL)
1288		return -EACCES;
 
 
1289
1290	/* clear any previous set[ug]id data from a previous binary */
1291	bprm->cred->euid = current_euid();
1292	bprm->cred->egid = current_egid();
1293
1294	if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1295	    !current->no_new_privs) {
1296		/* Set-uid? */
1297		if (mode & S_ISUID) {
1298			if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1299				return -EPERM;
1300			bprm->per_clear |= PER_CLEAR_ON_SETID;
1301			bprm->cred->euid = inode->i_uid;
1302
1303		}
 
 
 
1304
1305		/* Set-gid? */
1306		/*
1307		 * If setgid is set but no group execute bit then this
1308		 * is a candidate for mandatory locking, not a setgid
1309		 * executable.
1310		 */
1311		if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1312			if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1313				return -EPERM;
1314			bprm->per_clear |= PER_CLEAR_ON_SETID;
1315			bprm->cred->egid = inode->i_gid;
1316		}
1317	}
1318
1319	/* fill in binprm security blob */
1320	retval = security_bprm_set_creds(bprm);
1321	if (retval)
1322		return retval;
1323	bprm->cred_prepared = 1;
1324
1325	memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1326	return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
 
 
 
 
 
 
 
 
1327}
1328
1329EXPORT_SYMBOL(prepare_binprm);
 
 
 
 
 
 
 
 
 
 
 
 
1330
1331/*
1332 * Arguments are '\0' separated strings found at the location bprm->p
1333 * points to; chop off the first by relocating brpm->p to right after
1334 * the first '\0' encountered.
1335 */
1336int remove_arg_zero(struct linux_binprm *bprm)
1337{
1338	int ret = 0;
1339	unsigned long offset;
1340	char *kaddr;
1341	struct page *page;
1342
1343	if (!bprm->argc)
1344		return 0;
1345
1346	do {
1347		offset = bprm->p & ~PAGE_MASK;
1348		page = get_arg_page(bprm, bprm->p, 0);
1349		if (!page) {
1350			ret = -EFAULT;
1351			goto out;
1352		}
1353		kaddr = kmap_atomic(page);
1354
1355		for (; offset < PAGE_SIZE && kaddr[offset];
1356				offset++, bprm->p++)
1357			;
1358
1359		kunmap_atomic(kaddr);
1360		put_arg_page(page);
1361
1362		if (offset == PAGE_SIZE)
1363			free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1364	} while (offset == PAGE_SIZE);
1365
1366	bprm->p++;
1367	bprm->argc--;
1368	ret = 0;
1369
1370out:
1371	return ret;
1372}
1373EXPORT_SYMBOL(remove_arg_zero);
1374
 
1375/*
1376 * cycle the list of binary formats handler, until one recognizes the image
1377 */
1378int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1379{
1380	unsigned int depth = bprm->recursion_depth;
1381	int try,retval;
1382	struct linux_binfmt *fmt;
1383	pid_t old_pid, old_vpid;
1384
1385	retval = security_bprm_check(bprm);
1386	if (retval)
1387		return retval;
1388
1389	retval = audit_bprm(bprm);
1390	if (retval)
1391		return retval;
1392
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1393	/* Need to fetch pid before load_binary changes it */
1394	old_pid = current->pid;
1395	rcu_read_lock();
1396	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1397	rcu_read_unlock();
1398
1399	retval = -ENOENT;
1400	for (try=0; try<2; try++) {
1401		read_lock(&binfmt_lock);
1402		list_for_each_entry(fmt, &formats, lh) {
1403			int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1404			if (!fn)
1405				continue;
1406			if (!try_module_get(fmt->module))
1407				continue;
1408			read_unlock(&binfmt_lock);
1409			retval = fn(bprm, regs);
1410			/*
1411			 * Restore the depth counter to its starting value
1412			 * in this call, so we don't have to rely on every
1413			 * load_binary function to restore it on return.
1414			 */
1415			bprm->recursion_depth = depth;
1416			if (retval >= 0) {
1417				if (depth == 0) {
1418					trace_sched_process_exec(current, old_pid, bprm);
1419					ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1420				}
1421				put_binfmt(fmt);
1422				allow_write_access(bprm->file);
1423				if (bprm->file)
1424					fput(bprm->file);
1425				bprm->file = NULL;
1426				current->did_exec = 1;
1427				proc_exec_connector(current);
1428				return retval;
1429			}
1430			read_lock(&binfmt_lock);
1431			put_binfmt(fmt);
1432			if (retval != -ENOEXEC || bprm->mm == NULL)
1433				break;
1434			if (!bprm->file) {
1435				read_unlock(&binfmt_lock);
1436				return retval;
1437			}
1438		}
1439		read_unlock(&binfmt_lock);
1440#ifdef CONFIG_MODULES
1441		if (retval != -ENOEXEC || bprm->mm == NULL) {
1442			break;
1443		} else {
1444#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1445			if (printable(bprm->buf[0]) &&
1446			    printable(bprm->buf[1]) &&
1447			    printable(bprm->buf[2]) &&
1448			    printable(bprm->buf[3]))
1449				break; /* -ENOEXEC */
1450			if (try)
1451				break; /* -ENOEXEC */
1452			request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1453		}
1454#else
1455		break;
1456#endif
1457	}
1458	return retval;
 
 
 
 
 
1459}
1460
1461EXPORT_SYMBOL(search_binary_handler);
 
 
1462
1463/*
1464 * sys_execve() executes a new program.
1465 */
1466static int do_execve_common(const char *filename,
1467				struct user_arg_ptr argv,
1468				struct user_arg_ptr envp,
1469				struct pt_regs *regs)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1470{
1471	struct linux_binprm *bprm;
1472	struct file *file;
1473	struct files_struct *displaced;
1474	bool clear_in_exec;
1475	int retval;
1476	const struct cred *cred = current_cred();
 
 
1477
1478	/*
1479	 * We move the actual failure in case of RLIMIT_NPROC excess from
1480	 * set*uid() to execve() because too many poorly written programs
1481	 * don't check setuid() return code.  Here we additionally recheck
1482	 * whether NPROC limit is still exceeded.
1483	 */
1484	if ((current->flags & PF_NPROC_EXCEEDED) &&
1485	    atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1486		retval = -EAGAIN;
1487		goto out_ret;
1488	}
1489
1490	/* We're below the limit (still or again), so we don't want to make
1491	 * further execve() calls fail. */
1492	current->flags &= ~PF_NPROC_EXCEEDED;
1493
1494	retval = unshare_files(&displaced);
1495	if (retval)
 
1496		goto out_ret;
 
1497
1498	retval = -ENOMEM;
1499	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1500	if (!bprm)
1501		goto out_files;
1502
1503	retval = prepare_bprm_creds(bprm);
1504	if (retval)
1505		goto out_free;
 
1506
1507	retval = check_unsafe_exec(bprm);
1508	if (retval < 0)
1509		goto out_free;
1510	clear_in_exec = retval;
1511	current->in_execve = 1;
1512
1513	file = open_exec(filename);
1514	retval = PTR_ERR(file);
1515	if (IS_ERR(file))
1516		goto out_unmark;
1517
1518	sched_exec();
1519
1520	bprm->file = file;
1521	bprm->filename = filename;
1522	bprm->interp = filename;
1523
1524	retval = bprm_mm_init(bprm);
1525	if (retval)
1526		goto out_file;
1527
1528	bprm->argc = count(argv, MAX_ARG_STRINGS);
1529	if ((retval = bprm->argc) < 0)
1530		goto out;
1531
1532	bprm->envc = count(envp, MAX_ARG_STRINGS);
1533	if ((retval = bprm->envc) < 0)
1534		goto out;
1535
1536	retval = prepare_binprm(bprm);
1537	if (retval < 0)
1538		goto out;
1539
1540	retval = copy_strings_kernel(1, &bprm->filename, bprm);
1541	if (retval < 0)
1542		goto out;
1543
1544	bprm->exec = bprm->p;
 
1545	retval = copy_strings(bprm->envc, envp, bprm);
1546	if (retval < 0)
1547		goto out;
1548
1549	retval = copy_strings(bprm->argc, argv, bprm);
1550	if (retval < 0)
1551		goto out;
1552
1553	retval = search_binary_handler(bprm,regs);
1554	if (retval < 0)
1555		goto out;
 
 
 
 
 
 
 
 
 
1556
1557	/* execve succeeded */
1558	current->fs->in_exec = 0;
1559	current->in_execve = 0;
1560	acct_update_integrals(current);
1561	free_bprm(bprm);
1562	if (displaced)
1563		put_files_struct(displaced);
 
1564	return retval;
 
1565
1566out:
1567	if (bprm->mm) {
1568		acct_arg_size(bprm, 0);
1569		mmput(bprm->mm);
1570	}
 
 
1571
1572out_file:
1573	if (bprm->file) {
1574		allow_write_access(bprm->file);
1575		fput(bprm->file);
 
 
 
 
 
 
 
 
1576	}
1577
1578out_unmark:
1579	if (clear_in_exec)
1580		current->fs->in_exec = 0;
1581	current->in_execve = 0;
 
 
 
 
 
 
 
1582
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1583out_free:
1584	free_bprm(bprm);
1585
1586out_files:
1587	if (displaced)
1588		reset_files_struct(displaced);
1589out_ret:
 
1590	return retval;
1591}
1592
1593int do_execve(const char *filename,
1594	const char __user *const __user *__argv,
1595	const char __user *const __user *__envp,
1596	struct pt_regs *regs)
 
 
 
 
 
 
 
 
 
1597{
1598	struct user_arg_ptr argv = { .ptr.native = __argv };
1599	struct user_arg_ptr envp = { .ptr.native = __envp };
1600	return do_execve_common(filename, argv, envp, regs);
 
1601}
1602
1603#ifdef CONFIG_COMPAT
1604int compat_do_execve(char *filename,
1605	compat_uptr_t __user *__argv,
1606	compat_uptr_t __user *__envp,
1607	struct pt_regs *regs)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1608{
1609	struct user_arg_ptr argv = {
1610		.is_compat = true,
1611		.ptr.compat = __argv,
1612	};
1613	struct user_arg_ptr envp = {
1614		.is_compat = true,
1615		.ptr.compat = __envp,
1616	};
1617	return do_execve_common(filename, argv, envp, regs);
1618}
1619#endif
1620
1621void set_binfmt(struct linux_binfmt *new)
1622{
1623	struct mm_struct *mm = current->mm;
1624
1625	if (mm->binfmt)
1626		module_put(mm->binfmt->module);
1627
1628	mm->binfmt = new;
1629	if (new)
1630		__module_get(new->module);
1631}
1632
1633EXPORT_SYMBOL(set_binfmt);
1634
1635static int expand_corename(struct core_name *cn)
1636{
1637	char *old_corename = cn->corename;
1638
1639	cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1640	cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1641
1642	if (!cn->corename) {
1643		kfree(old_corename);
1644		return -ENOMEM;
1645	}
1646
1647	return 0;
1648}
1649
1650static int cn_printf(struct core_name *cn, const char *fmt, ...)
1651{
1652	char *cur;
1653	int need;
1654	int ret;
1655	va_list arg;
1656
1657	va_start(arg, fmt);
1658	need = vsnprintf(NULL, 0, fmt, arg);
1659	va_end(arg);
1660
1661	if (likely(need < cn->size - cn->used - 1))
1662		goto out_printf;
1663
1664	ret = expand_corename(cn);
1665	if (ret)
1666		goto expand_fail;
1667
1668out_printf:
1669	cur = cn->corename + cn->used;
1670	va_start(arg, fmt);
1671	vsnprintf(cur, need + 1, fmt, arg);
1672	va_end(arg);
1673	cn->used += need;
1674	return 0;
1675
1676expand_fail:
1677	return ret;
1678}
1679
1680static void cn_escape(char *str)
1681{
1682	for (; *str; str++)
1683		if (*str == '/')
1684			*str = '!';
1685}
1686
1687static int cn_print_exe_file(struct core_name *cn)
1688{
1689	struct file *exe_file;
1690	char *pathbuf, *path;
1691	int ret;
1692
1693	exe_file = get_mm_exe_file(current->mm);
1694	if (!exe_file) {
1695		char *commstart = cn->corename + cn->used;
1696		ret = cn_printf(cn, "%s (path unknown)", current->comm);
1697		cn_escape(commstart);
1698		return ret;
1699	}
1700
1701	pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1702	if (!pathbuf) {
1703		ret = -ENOMEM;
1704		goto put_exe_file;
1705	}
1706
1707	path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1708	if (IS_ERR(path)) {
1709		ret = PTR_ERR(path);
1710		goto free_buf;
1711	}
1712
1713	cn_escape(path);
1714
1715	ret = cn_printf(cn, "%s", path);
1716
1717free_buf:
1718	kfree(pathbuf);
1719put_exe_file:
1720	fput(exe_file);
1721	return ret;
1722}
1723
1724/* format_corename will inspect the pattern parameter, and output a
1725 * name into corename, which must have space for at least
1726 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1727 */
1728static int format_corename(struct core_name *cn, long signr)
1729{
1730	const struct cred *cred = current_cred();
1731	const char *pat_ptr = core_pattern;
1732	int ispipe = (*pat_ptr == '|');
1733	int pid_in_pattern = 0;
1734	int err = 0;
1735
1736	cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1737	cn->corename = kmalloc(cn->size, GFP_KERNEL);
1738	cn->used = 0;
1739
1740	if (!cn->corename)
1741		return -ENOMEM;
1742
1743	/* Repeat as long as we have more pattern to process and more output
1744	   space */
1745	while (*pat_ptr) {
1746		if (*pat_ptr != '%') {
1747			if (*pat_ptr == 0)
1748				goto out;
1749			err = cn_printf(cn, "%c", *pat_ptr++);
1750		} else {
1751			switch (*++pat_ptr) {
1752			/* single % at the end, drop that */
1753			case 0:
1754				goto out;
1755			/* Double percent, output one percent */
1756			case '%':
1757				err = cn_printf(cn, "%c", '%');
1758				break;
1759			/* pid */
1760			case 'p':
1761				pid_in_pattern = 1;
1762				err = cn_printf(cn, "%d",
1763					      task_tgid_vnr(current));
1764				break;
1765			/* uid */
1766			case 'u':
1767				err = cn_printf(cn, "%d", cred->uid);
1768				break;
1769			/* gid */
1770			case 'g':
1771				err = cn_printf(cn, "%d", cred->gid);
1772				break;
1773			/* signal that caused the coredump */
1774			case 's':
1775				err = cn_printf(cn, "%ld", signr);
1776				break;
1777			/* UNIX time of coredump */
1778			case 't': {
1779				struct timeval tv;
1780				do_gettimeofday(&tv);
1781				err = cn_printf(cn, "%lu", tv.tv_sec);
1782				break;
1783			}
1784			/* hostname */
1785			case 'h': {
1786				char *namestart = cn->corename + cn->used;
1787				down_read(&uts_sem);
1788				err = cn_printf(cn, "%s",
1789					      utsname()->nodename);
1790				up_read(&uts_sem);
1791				cn_escape(namestart);
1792				break;
1793			}
1794			/* executable */
1795			case 'e': {
1796				char *commstart = cn->corename + cn->used;
1797				err = cn_printf(cn, "%s", current->comm);
1798				cn_escape(commstart);
1799				break;
1800			}
1801			case 'E':
1802				err = cn_print_exe_file(cn);
1803				break;
1804			/* core limit size */
1805			case 'c':
1806				err = cn_printf(cn, "%lu",
1807					      rlimit(RLIMIT_CORE));
1808				break;
1809			default:
1810				break;
1811			}
1812			++pat_ptr;
1813		}
1814
1815		if (err)
1816			return err;
1817	}
1818
1819	/* Backward compatibility with core_uses_pid:
1820	 *
1821	 * If core_pattern does not include a %p (as is the default)
1822	 * and core_uses_pid is set, then .%pid will be appended to
1823	 * the filename. Do not do this for piped commands. */
1824	if (!ispipe && !pid_in_pattern && core_uses_pid) {
1825		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1826		if (err)
1827			return err;
1828	}
1829out:
1830	return ispipe;
1831}
1832
1833static int zap_process(struct task_struct *start, int exit_code)
1834{
1835	struct task_struct *t;
1836	int nr = 0;
1837
1838	start->signal->flags = SIGNAL_GROUP_EXIT;
1839	start->signal->group_exit_code = exit_code;
1840	start->signal->group_stop_count = 0;
1841
1842	t = start;
1843	do {
1844		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1845		if (t != current && t->mm) {
1846			sigaddset(&t->pending.signal, SIGKILL);
1847			signal_wake_up(t, 1);
1848			nr++;
1849		}
1850	} while_each_thread(start, t);
1851
1852	return nr;
1853}
1854
1855static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1856				struct core_state *core_state, int exit_code)
1857{
1858	struct task_struct *g, *p;
1859	unsigned long flags;
1860	int nr = -EAGAIN;
1861
1862	spin_lock_irq(&tsk->sighand->siglock);
1863	if (!signal_group_exit(tsk->signal)) {
1864		mm->core_state = core_state;
1865		nr = zap_process(tsk, exit_code);
1866	}
1867	spin_unlock_irq(&tsk->sighand->siglock);
1868	if (unlikely(nr < 0))
1869		return nr;
1870
1871	if (atomic_read(&mm->mm_users) == nr + 1)
1872		goto done;
1873	/*
1874	 * We should find and kill all tasks which use this mm, and we should
1875	 * count them correctly into ->nr_threads. We don't take tasklist
1876	 * lock, but this is safe wrt:
1877	 *
1878	 * fork:
1879	 *	None of sub-threads can fork after zap_process(leader). All
1880	 *	processes which were created before this point should be
1881	 *	visible to zap_threads() because copy_process() adds the new
1882	 *	process to the tail of init_task.tasks list, and lock/unlock
1883	 *	of ->siglock provides a memory barrier.
1884	 *
1885	 * do_exit:
1886	 *	The caller holds mm->mmap_sem. This means that the task which
1887	 *	uses this mm can't pass exit_mm(), so it can't exit or clear
1888	 *	its ->mm.
1889	 *
1890	 * de_thread:
1891	 *	It does list_replace_rcu(&leader->tasks, &current->tasks),
1892	 *	we must see either old or new leader, this does not matter.
1893	 *	However, it can change p->sighand, so lock_task_sighand(p)
1894	 *	must be used. Since p->mm != NULL and we hold ->mmap_sem
1895	 *	it can't fail.
1896	 *
1897	 *	Note also that "g" can be the old leader with ->mm == NULL
1898	 *	and already unhashed and thus removed from ->thread_group.
1899	 *	This is OK, __unhash_process()->list_del_rcu() does not
1900	 *	clear the ->next pointer, we will find the new leader via
1901	 *	next_thread().
1902	 */
1903	rcu_read_lock();
1904	for_each_process(g) {
1905		if (g == tsk->group_leader)
1906			continue;
1907		if (g->flags & PF_KTHREAD)
1908			continue;
1909		p = g;
1910		do {
1911			if (p->mm) {
1912				if (unlikely(p->mm == mm)) {
1913					lock_task_sighand(p, &flags);
1914					nr += zap_process(p, exit_code);
1915					unlock_task_sighand(p, &flags);
1916				}
1917				break;
1918			}
1919		} while_each_thread(g, p);
1920	}
1921	rcu_read_unlock();
1922done:
1923	atomic_set(&core_state->nr_threads, nr);
1924	return nr;
1925}
1926
1927static int coredump_wait(int exit_code, struct core_state *core_state)
1928{
1929	struct task_struct *tsk = current;
1930	struct mm_struct *mm = tsk->mm;
1931	int core_waiters = -EBUSY;
1932
1933	init_completion(&core_state->startup);
1934	core_state->dumper.task = tsk;
1935	core_state->dumper.next = NULL;
1936
1937	down_write(&mm->mmap_sem);
1938	if (!mm->core_state)
1939		core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1940	up_write(&mm->mmap_sem);
1941
1942	if (core_waiters > 0) {
1943		struct core_thread *ptr;
1944
1945		wait_for_completion(&core_state->startup);
1946		/*
1947		 * Wait for all the threads to become inactive, so that
1948		 * all the thread context (extended register state, like
1949		 * fpu etc) gets copied to the memory.
1950		 */
1951		ptr = core_state->dumper.next;
1952		while (ptr != NULL) {
1953			wait_task_inactive(ptr->task, 0);
1954			ptr = ptr->next;
1955		}
1956	}
1957
1958	return core_waiters;
1959}
1960
1961static void coredump_finish(struct mm_struct *mm)
1962{
1963	struct core_thread *curr, *next;
1964	struct task_struct *task;
1965
1966	next = mm->core_state->dumper.next;
1967	while ((curr = next) != NULL) {
1968		next = curr->next;
1969		task = curr->task;
1970		/*
1971		 * see exit_mm(), curr->task must not see
1972		 * ->task == NULL before we read ->next.
1973		 */
1974		smp_mb();
1975		curr->task = NULL;
1976		wake_up_process(task);
1977	}
1978
1979	mm->core_state = NULL;
1980}
1981
1982/*
1983 * set_dumpable converts traditional three-value dumpable to two flags and
1984 * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1985 * these bits are not changed atomically.  So get_dumpable can observe the
1986 * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1987 * return either old dumpable or new one by paying attention to the order of
1988 * modifying the bits.
1989 *
1990 * dumpable |   mm->flags (binary)
1991 * old  new | initial interim  final
1992 * ---------+-----------------------
1993 *  0    1  |   00      01      01
1994 *  0    2  |   00      10(*)   11
1995 *  1    0  |   01      00      00
1996 *  1    2  |   01      11      11
1997 *  2    0  |   11      10(*)   00
1998 *  2    1  |   11      11      01
1999 *
2000 * (*) get_dumpable regards interim value of 10 as 11.
2001 */
2002void set_dumpable(struct mm_struct *mm, int value)
2003{
2004	switch (value) {
2005	case 0:
2006		clear_bit(MMF_DUMPABLE, &mm->flags);
2007		smp_wmb();
2008		clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2009		break;
2010	case 1:
2011		set_bit(MMF_DUMPABLE, &mm->flags);
2012		smp_wmb();
2013		clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2014		break;
2015	case 2:
2016		set_bit(MMF_DUMP_SECURELY, &mm->flags);
2017		smp_wmb();
2018		set_bit(MMF_DUMPABLE, &mm->flags);
2019		break;
2020	}
2021}
2022
2023static int __get_dumpable(unsigned long mm_flags)
2024{
2025	int ret;
2026
2027	ret = mm_flags & MMF_DUMPABLE_MASK;
2028	return (ret >= 2) ? 2 : ret;
2029}
2030
2031int get_dumpable(struct mm_struct *mm)
 
 
 
2032{
2033	return __get_dumpable(mm->flags);
2034}
2035
2036static void wait_for_dump_helpers(struct file *file)
 
 
 
 
2037{
2038	struct pipe_inode_info *pipe;
2039
2040	pipe = file->f_path.dentry->d_inode->i_pipe;
2041
2042	pipe_lock(pipe);
2043	pipe->readers++;
2044	pipe->writers--;
2045
2046	while ((pipe->readers > 1) && (!signal_pending(current))) {
2047		wake_up_interruptible_sync(&pipe->wait);
2048		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2049		pipe_wait(pipe);
2050	}
2051
2052	pipe->readers--;
2053	pipe->writers++;
2054	pipe_unlock(pipe);
2055
2056}
2057
2058
2059/*
2060 * umh_pipe_setup
2061 * helper function to customize the process used
2062 * to collect the core in userspace.  Specifically
2063 * it sets up a pipe and installs it as fd 0 (stdin)
2064 * for the process.  Returns 0 on success, or
2065 * PTR_ERR on failure.
2066 * Note that it also sets the core limit to 1.  This
2067 * is a special value that we use to trap recursive
2068 * core dumps
2069 */
2070static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2071{
2072	struct file *rp, *wp;
2073	struct fdtable *fdt;
2074	struct coredump_params *cp = (struct coredump_params *)info->data;
2075	struct files_struct *cf = current->files;
2076
2077	wp = create_write_pipe(0);
2078	if (IS_ERR(wp))
2079		return PTR_ERR(wp);
2080
2081	rp = create_read_pipe(wp, 0);
2082	if (IS_ERR(rp)) {
2083		free_write_pipe(wp);
2084		return PTR_ERR(rp);
2085	}
2086
2087	cp->file = wp;
2088
2089	sys_close(0);
2090	fd_install(0, rp);
2091	spin_lock(&cf->file_lock);
2092	fdt = files_fdtable(cf);
2093	__set_open_fd(0, fdt);
2094	__clear_close_on_exec(0, fdt);
2095	spin_unlock(&cf->file_lock);
2096
2097	/* and disallow core files too */
2098	current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2099
2100	return 0;
2101}
 
2102
2103void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2104{
2105	struct core_state core_state;
2106	struct core_name cn;
2107	struct mm_struct *mm = current->mm;
2108	struct linux_binfmt * binfmt;
2109	const struct cred *old_cred;
2110	struct cred *cred;
2111	int retval = 0;
2112	int flag = 0;
2113	int ispipe;
2114	static atomic_t core_dump_count = ATOMIC_INIT(0);
2115	struct coredump_params cprm = {
2116		.signr = signr,
2117		.regs = regs,
2118		.limit = rlimit(RLIMIT_CORE),
2119		/*
2120		 * We must use the same mm->flags while dumping core to avoid
2121		 * inconsistency of bit flags, since this flag is not protected
2122		 * by any locks.
2123		 */
2124		.mm_flags = mm->flags,
2125	};
2126
2127	audit_core_dumps(signr);
2128
2129	binfmt = mm->binfmt;
2130	if (!binfmt || !binfmt->core_dump)
2131		goto fail;
2132	if (!__get_dumpable(cprm.mm_flags))
2133		goto fail;
2134
2135	cred = prepare_creds();
2136	if (!cred)
2137		goto fail;
2138	/*
2139	 *	We cannot trust fsuid as being the "true" uid of the
2140	 *	process nor do we know its entire history. We only know it
2141	 *	was tainted so we dump it as root in mode 2.
2142	 */
2143	if (__get_dumpable(cprm.mm_flags) == 2) {
2144		/* Setuid core dump mode */
2145		flag = O_EXCL;		/* Stop rewrite attacks */
2146		cred->fsuid = GLOBAL_ROOT_UID;	/* Dump root private */
2147	}
2148
2149	retval = coredump_wait(exit_code, &core_state);
2150	if (retval < 0)
2151		goto fail_creds;
2152
2153	old_cred = override_creds(cred);
2154
2155	/*
2156	 * Clear any false indication of pending signals that might
2157	 * be seen by the filesystem code called to write the core file.
2158	 */
2159	clear_thread_flag(TIF_SIGPENDING);
2160
2161	ispipe = format_corename(&cn, signr);
2162
2163 	if (ispipe) {
2164		int dump_count;
2165		char **helper_argv;
2166
2167		if (ispipe < 0) {
2168			printk(KERN_WARNING "format_corename failed\n");
2169			printk(KERN_WARNING "Aborting core\n");
2170			goto fail_corename;
2171		}
2172
2173		if (cprm.limit == 1) {
2174			/*
2175			 * Normally core limits are irrelevant to pipes, since
2176			 * we're not writing to the file system, but we use
2177			 * cprm.limit of 1 here as a speacial value. Any
2178			 * non-1 limit gets set to RLIM_INFINITY below, but
2179			 * a limit of 0 skips the dump.  This is a consistent
2180			 * way to catch recursive crashes.  We can still crash
2181			 * if the core_pattern binary sets RLIM_CORE =  !1
2182			 * but it runs as root, and can do lots of stupid things
2183			 * Note that we use task_tgid_vnr here to grab the pid
2184			 * of the process group leader.  That way we get the
2185			 * right pid if a thread in a multi-threaded
2186			 * core_pattern process dies.
2187			 */
2188			printk(KERN_WARNING
2189				"Process %d(%s) has RLIMIT_CORE set to 1\n",
2190				task_tgid_vnr(current), current->comm);
2191			printk(KERN_WARNING "Aborting core\n");
2192			goto fail_unlock;
2193		}
2194		cprm.limit = RLIM_INFINITY;
2195
2196		dump_count = atomic_inc_return(&core_dump_count);
2197		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2198			printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2199			       task_tgid_vnr(current), current->comm);
2200			printk(KERN_WARNING "Skipping core dump\n");
2201			goto fail_dropcount;
2202		}
2203
2204		helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2205		if (!helper_argv) {
2206			printk(KERN_WARNING "%s failed to allocate memory\n",
2207			       __func__);
2208			goto fail_dropcount;
2209		}
2210
2211		retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2212					NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2213					NULL, &cprm);
2214		argv_free(helper_argv);
2215		if (retval) {
2216 			printk(KERN_INFO "Core dump to %s pipe failed\n",
2217			       cn.corename);
2218			goto close_fail;
2219 		}
2220	} else {
2221		struct inode *inode;
2222
2223		if (cprm.limit < binfmt->min_coredump)
2224			goto fail_unlock;
2225
2226		cprm.file = filp_open(cn.corename,
2227				 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2228				 0600);
2229		if (IS_ERR(cprm.file))
2230			goto fail_unlock;
2231
2232		inode = cprm.file->f_path.dentry->d_inode;
2233		if (inode->i_nlink > 1)
2234			goto close_fail;
2235		if (d_unhashed(cprm.file->f_path.dentry))
2236			goto close_fail;
2237		/*
2238		 * AK: actually i see no reason to not allow this for named
2239		 * pipes etc, but keep the previous behaviour for now.
2240		 */
2241		if (!S_ISREG(inode->i_mode))
2242			goto close_fail;
2243		/*
2244		 * Dont allow local users get cute and trick others to coredump
2245		 * into their pre-created files.
2246		 */
2247		if (!uid_eq(inode->i_uid, current_fsuid()))
2248			goto close_fail;
2249		if (!cprm.file->f_op || !cprm.file->f_op->write)
2250			goto close_fail;
2251		if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2252			goto close_fail;
2253	}
2254
2255	retval = binfmt->core_dump(&cprm);
2256	if (retval)
2257		current->signal->group_exit_code |= 0x80;
2258
2259	if (ispipe && core_pipe_limit)
2260		wait_for_dump_helpers(cprm.file);
2261close_fail:
2262	if (cprm.file)
2263		filp_close(cprm.file, NULL);
2264fail_dropcount:
2265	if (ispipe)
2266		atomic_dec(&core_dump_count);
2267fail_unlock:
2268	kfree(cn.corename);
2269fail_corename:
2270	coredump_finish(mm);
2271	revert_creds(old_cred);
2272fail_creds:
2273	put_cred(cred);
2274fail:
2275	return;
2276}
2277
2278/*
2279 * Core dumping helper functions.  These are the only things you should
2280 * do on a core-file: use only these functions to write out all the
2281 * necessary info.
2282 */
2283int dump_write(struct file *file, const void *addr, int nr)
2284{
2285	return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
 
2286}
2287EXPORT_SYMBOL(dump_write);
2288
2289int dump_seek(struct file *file, loff_t off)
2290{
2291	int ret = 1;
2292
2293	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2294		if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2295			return 0;
2296	} else {
2297		char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2298
2299		if (!buf)
2300			return 0;
2301		while (off > 0) {
2302			unsigned long n = off;
2303
2304			if (n > PAGE_SIZE)
2305				n = PAGE_SIZE;
2306			if (!dump_write(file, buf, n)) {
2307				ret = 0;
2308				break;
2309			}
2310			off -= n;
2311		}
2312		free_page((unsigned long)buf);
2313	}
2314	return ret;
2315}
2316EXPORT_SYMBOL(dump_seek);