vm_alloc.cpp

/*
 *  vm_alloc.cpp - Wrapper to various virtual memory allocation schemes
 *                 (supports mmap, vm_allocate or fallbacks to malloc)
 *
 *  Basilisk II (C) 1997-2008 Christian Bauer
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#ifdef HAVE_FCNTL_H
#include <fcntl.h>
#endif

#ifdef HAVE_WIN32_VM
#define WIN32_LEAN_AND_MEAN /* avoid including junk */
#include <windows.h>
#endif

#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#include "vm_alloc.h"

#if defined(__APPLE__) && defined(__MACH__)
#include <sys/utsname.h>
#endif

#ifdef HAVE_MACH_VM
#ifndef HAVE_MACH_TASK_SELF
#ifdef HAVE_TASK_SELF
#define mach_task_self task_self
#else
#error "No task_self(), you lose."
#endif
#endif
#endif

#ifdef HAVE_WIN32_VM
/* Windows is either ILP32 or LLP64 */
typedef UINT_PTR vm_uintptr_t;
#else
/* Other systems are sane as they are either ILP32 or LP64 */
typedef unsigned long vm_uintptr_t;
#endif

/* We want MAP_32BIT, if available, for SheepShaver and BasiliskII
   because the emulated target is 32-bit and this helps to allocate
   memory so that branches could be resolved more easily (32-bit
   displacement to code in .text), on AMD64 for example.  */
#if defined(__hpux)
#define MAP_32BIT MAP_ADDR32
#endif
#ifndef MAP_32BIT
#define MAP_32BIT 0
#endif
#ifdef __FreeBSD__
#define FORCE_MAP_32BIT MAP_FIXED
#else
#define FORCE_MAP_32BIT MAP_32BIT
#endif
#ifndef MAP_ANON
#define MAP_ANON 0
#endif
#ifndef MAP_ANONYMOUS
#define MAP_ANONYMOUS 0
#endif

/* NOTE: on linux MAP_32BIT is only implemented on AMD64
   it is a null op on all other architectures
   thus the MAP_BASE setting below is the only thing
   ensuring low addresses on aarch64 for example */
#define MAP_EXTRA_FLAGS (MAP_32BIT)

#ifdef HAVE_MMAP_VM
#if (defined(__linux__) && defined(__i386__)) || defined(__FreeBSD__) || HAVE_LINKER_SCRIPT
/* Force a reasonnable address below 0x80000000 on x86 so that we
   don't get addresses above when the program is run on AMD64.
   NOTE: this is empirically determined on Linux/x86.  */
#define MAP_BASE	0x10000000
#elif DIRECT_ADDRESSING
/* linux does not implement any useful fallback behavior
   such as allocating the next available address
   and the first 4k-64k of address space is marked unavailable
   for security reasons (see https://wiki.debian.org/mmap_min_addr)
   so we must start requesting after the first page
   or we get a high 64bit address that will crash direct addressing

   leaving NULL unmapped is a good idea anyway for debugging reasons */
#define MAP_BASE	0x00010000
#else
#define MAP_BASE	0x00000000
#endif
static char * next_address = (char *)MAP_BASE;
#ifdef HAVE_MMAP_ANON
#define map_flags	(MAP_ANON | MAP_EXTRA_FLAGS)
#define zero_fd		-1
#else
#ifdef HAVE_MMAP_ANONYMOUS
#define map_flags	(MAP_ANONYMOUS | MAP_EXTRA_FLAGS)
#define zero_fd		-1
#else
#define map_flags	(MAP_EXTRA_FLAGS)
static int zero_fd	= -1;
#endif
#endif
#endif

/* Translate generic VM map flags to host values.  */

#ifdef HAVE_MMAP_VM
static int translate_map_flags(int vm_flags)
{
	int flags = 0;
	if (vm_flags & VM_MAP_SHARED)
		flags |= MAP_SHARED;
	if (vm_flags & VM_MAP_PRIVATE)
		flags |= MAP_PRIVATE;
	if (vm_flags & VM_MAP_FIXED)
		flags |= MAP_FIXED;
	if (vm_flags & VM_MAP_32BIT)
		flags |= FORCE_MAP_32BIT;
	return flags;
}
#endif

/* Align ADDR and SIZE to 64K boundaries.  */

#ifdef HAVE_WIN32_VM
static inline LPVOID align_addr_segment(LPVOID addr)
{
	return LPVOID(vm_uintptr_t(addr) & ~vm_uintptr_t(0xFFFF));
}

static inline DWORD align_size_segment(LPVOID addr, DWORD size)
{
	return size + ((vm_uintptr_t)addr - (vm_uintptr_t)align_addr_segment(addr));
}
#endif

/* Translate generic VM prot flags to host values.  */

#ifdef HAVE_WIN32_VM
static int translate_prot_flags(int prot_flags)
{
	int prot = PAGE_READWRITE;
	if (prot_flags == (VM_PAGE_EXECUTE | VM_PAGE_READ | VM_PAGE_WRITE))
		prot = PAGE_EXECUTE_READWRITE;
	else if (prot_flags == (VM_PAGE_EXECUTE | VM_PAGE_READ))
		prot = PAGE_EXECUTE_READ;
	else if (prot_flags == (VM_PAGE_READ | VM_PAGE_WRITE))
		prot = PAGE_READWRITE;
	else if (prot_flags == VM_PAGE_READ)
		prot = PAGE_READONLY;
	else if (prot_flags == 0)
		prot = PAGE_NOACCESS;
	return prot;
}
#endif

/* Translate Mach return codes to POSIX errno values. */
#ifdef HAVE_MACH_VM
static int vm_error(kern_return_t ret_code)
{
	switch (ret_code) {
		case KERN_SUCCESS:
			return 0;
		case KERN_INVALID_ADDRESS:
		case KERN_NO_SPACE:
			return ENOMEM;
		case KERN_PROTECTION_FAILURE:
			return EACCES;
		default:
			return EINVAL;
	}
}
#endif

/* Initialize the VM system. Returns 0 if successful, -1 for errors.  */

int vm_init(void)
{
#ifdef HAVE_MMAP_VM
#ifndef zero_fd
	zero_fd = open("/dev/zero", O_RDWR);
	if (zero_fd < 0)
		return -1;
#endif
#endif

// On 10.4 and earlier, reset CrashReporter's task signal handler to
// avoid having it show up for signals that get handled.
#if defined(__APPLE__) && defined(__MACH__)
	struct utsname info;

	if (!uname(&info) && atoi(info.release) <= 8) {
		task_set_exception_ports(mach_task_self(),
			EXC_MASK_BAD_ACCESS | EXC_MASK_ARITHMETIC,
			MACH_PORT_NULL,
			EXCEPTION_STATE_IDENTITY,
			MACHINE_THREAD_STATE);
	}
#endif

	return 0;
}

/* Deallocate all internal data used to wrap virtual memory allocators.  */

void vm_exit(void)
{
#ifdef HAVE_MMAP_VM
#ifndef zero_fd
	if (zero_fd != -1) {
		close(zero_fd);
		zero_fd = -1;
	}
#endif
#endif
}

/* Allocate zero-filled memory of SIZE bytes. The mapping is private
   and default protection bits are read / write. The return value
   is the actual mapping address chosen or VM_MAP_FAILED for errors.  */

void * vm_acquire(size_t size, int options)
{
	void * addr;
	
	errno = 0;

	// VM_MAP_FIXED are to be used with vm_acquire_fixed() only
	if (options & VM_MAP_FIXED)
		return VM_MAP_FAILED;

#ifndef HAVE_VM_WRITE_WATCH
	if (options & VM_MAP_WRITE_WATCH)
		return VM_MAP_FAILED;
#endif

#if defined(HAVE_MACH_VM)
	// vm_allocate() returns a zero-filled memory region
	kern_return_t ret_code = vm_allocate(mach_task_self(), (vm_address_t *)&addr, size, TRUE);
	if (ret_code != KERN_SUCCESS) {
		errno = vm_error(ret_code);
		return VM_MAP_FAILED;
	}
#elif defined(HAVE_MMAP_VM)
	int fd = zero_fd;
	int the_map_flags = translate_map_flags(options) | map_flags;

	if ((addr = mmap((caddr_t)next_address, size, VM_PAGE_DEFAULT, the_map_flags, fd, 0)) == (void *)MAP_FAILED)
		return VM_MAP_FAILED;
	printf("mmap next=%p got=%p\n",next_address,addr);
	
#if DIRECT_ADDRESSING
	// Sanity check to prevent crash on 64-bit when direct addressing
	// FIXME: this results in a misleading "out of memory" error to the user when it fails
	if (sizeof(void *) == 8 && (options & VM_MAP_32BIT) && !((char *)addr <= (char *)0xffffffff))
		return VM_MAP_FAILED;
#endif

	next_address = (char *)addr + size;
#elif defined(HAVE_WIN32_VM)
	int alloc_type = MEM_RESERVE | MEM_COMMIT;
	if (options & VM_MAP_WRITE_WATCH)
	  alloc_type |= MEM_WRITE_WATCH;

	if ((addr = VirtualAlloc(NULL, size, alloc_type, PAGE_EXECUTE_READWRITE)) == NULL)
		return VM_MAP_FAILED;
#else
	if ((addr = calloc(size, 1)) == 0)
		return VM_MAP_FAILED;
	
	// Omit changes for protections because they are not supported in this mode
	return addr;
#endif

	// Explicitely protect the newly mapped region here because on some systems,
	// say MacOS X, mmap() doesn't honour the requested protection flags.
	if (vm_protect(addr, size, VM_PAGE_DEFAULT) != 0)
		return VM_MAP_FAILED;
	
	return addr;
}

/* Allocate zero-filled memory at exactly ADDR (which must be page-aligned).
   Retuns 0 if successful, -1 on errors.  */

int vm_acquire_fixed(void * addr, size_t size, int options)
{
	errno = 0;
	
	// Fixed mappings are required to be private
	if (options & VM_MAP_SHARED)
		return -1;

#ifndef HAVE_VM_WRITE_WATCH
	if (options & VM_MAP_WRITE_WATCH)
		return -1;
#endif

#if defined(HAVE_MACH_VM)
	// vm_allocate() returns a zero-filled memory region
	kern_return_t ret_code = vm_allocate(mach_task_self(), (vm_address_t *)&addr, size, 0);
	if (ret_code != KERN_SUCCESS) {
		errno = vm_error(ret_code);
		return -1;
	}
#elif defined(HAVE_MMAP_VM)
	int fd = zero_fd;
	int the_map_flags = translate_map_flags(options) | map_flags | MAP_FIXED;

	if (mmap((caddr_t)addr, size, VM_PAGE_DEFAULT, the_map_flags, fd, 0) == (void *)MAP_FAILED)
		return -1;
#elif defined(HAVE_WIN32_VM)
	// Windows cannot allocate Low Memory
	if (addr == NULL)
		return -1;

	int alloc_type = MEM_RESERVE | MEM_COMMIT;
	if (options & VM_MAP_WRITE_WATCH)
	  alloc_type |= MEM_WRITE_WATCH;

	// Allocate a possibly offset region to align on 64K boundaries
	LPVOID req_addr = align_addr_segment(addr);
	DWORD  req_size = align_size_segment(addr, size);
	LPVOID ret_addr = VirtualAlloc(req_addr, req_size, alloc_type, PAGE_EXECUTE_READWRITE);
	if (ret_addr != req_addr)
		return -1;
#else
	// Unsupported
	return -1;
#endif

	// Explicitely protect the newly mapped region here because on some systems,
	// say MacOS X, mmap() doesn't honour the requested protection flags.
	if (vm_protect(addr, size, VM_PAGE_DEFAULT) != 0)
		return -1;

	return 0;
}

/* Deallocate any mapping for the region starting at ADDR and extending
   LEN bytes. Returns 0 if successful, -1 on errors.  */

int vm_release(void * addr, size_t size)
{
	// Safety check: don't try to release memory that was not allocated
	if (addr == VM_MAP_FAILED)
		return 0;

#ifdef HAVE_MACH_VM
	if (vm_deallocate(mach_task_self(), (vm_address_t)addr, size) != KERN_SUCCESS)
		return -1;
#else
#ifdef HAVE_MMAP_VM
	if (munmap((caddr_t)addr, size) != 0)
		return -1;
#else
#ifdef HAVE_WIN32_VM
	if (VirtualFree(align_addr_segment(addr), 0, MEM_RELEASE) == 0)
		return -1;
#else
	free(addr);
#endif
#endif
#endif
	
	return 0;
}

/* Change the memory protection of the region starting at ADDR and
   extending LEN bytes to PROT. Returns 0 if successful, -1 for errors.  */

int vm_protect(void * addr, size_t size, int prot)
{
#ifdef HAVE_MACH_VM
	int ret_code = vm_protect(mach_task_self(), (vm_address_t)addr, size, 0, prot);
	return ret_code == KERN_SUCCESS ? 0 : -1;
#else
#ifdef HAVE_MMAP_VM
	int ret_code = mprotect((caddr_t)addr, size, prot);
	return ret_code == 0 ? 0 : -1;
#else
#ifdef HAVE_WIN32_VM
	DWORD old_prot;
	int ret_code = VirtualProtect(addr, size, translate_prot_flags(prot), &old_prot);
	return ret_code != 0 ? 0 : -1;
#else
	// Unsupported
	return -1;
#endif
#endif
#endif
}

/* Return the addresses of the pages that got modified in the
   specified range [ ADDR, ADDR + SIZE [ since the last reset of the watch
   bits. Returns 0 if successful, -1 for errors.  */

int vm_get_write_watch(void * addr, size_t size,
					   void ** pages, unsigned int * n_pages,
					   int options)
{
#ifdef HAVE_VM_WRITE_WATCH
#ifdef HAVE_WIN32_VM
	DWORD flags = 0;
	if (options & VM_WRITE_WATCH_RESET)
		flags |= WRITE_WATCH_FLAG_RESET;

	ULONG page_size;
	ULONG_PTR count = *n_pages;
	int ret_code = GetWriteWatch(flags, addr, size, pages, &count, &page_size);
	if (ret_code != 0)
		return -1;

	*n_pages = count;
	return 0;
#endif
#endif
	// Unsupported
	return -1;
}

/* Reset the write-tracking state for the specified range [ ADDR, ADDR
   + SIZE [. Returns 0 if successful, -1 for errors.  */

int vm_reset_write_watch(void * addr, size_t size)
{
#ifdef HAVE_VM_WRITE_WATCH
#ifdef HAVE_WIN32_VM
	int ret_code = ResetWriteWatch(addr, size);
	return ret_code == 0 ? 0 : -1;
#endif
#endif
	// Unsupported
	return -1;
}

/* Returns the size of a page.  */

int vm_get_page_size(void)
{
#ifdef HAVE_WIN32_VM
	static vm_uintptr_t page_size = 0;
	if (page_size == 0) {
		SYSTEM_INFO si;
		GetSystemInfo(&si);
		page_size = si.dwAllocationGranularity;
	}
	return page_size;
#else
	return getpagesize();
#endif
}

#ifdef CONFIGURE_TEST_VM_WRITE_WATCH
int main(void)
{
	int i, j;

	vm_init();

	vm_uintptr_t page_size = vm_get_page_size();
	
	char *area;
	const int n_pages = 7;
	const int area_size = n_pages * page_size;
	const int map_options = VM_MAP_DEFAULT | VM_MAP_WRITE_WATCH;
	if ((area = (char *)vm_acquire(area_size, map_options)) == VM_MAP_FAILED)
		return 1;

	unsigned int n_modified_pages_expected = 0;
	static const int touch_page[n_pages] = { 0, 1, 1, 0, 1, 0, 1 };
	for (i = 0; i < n_pages; i++) {
		if (touch_page[i]) {
			area[i * page_size] = 1;
			++n_modified_pages_expected;
		}
	}

	char *modified_pages[n_pages];
	unsigned int n_modified_pages = n_pages;
	if (vm_get_write_watch(area, area_size, (void **)modified_pages, &n_modified_pages) < 0)
		return 2;
	if (n_modified_pages != n_modified_pages_expected)
		return 3;
	for (i = 0, j = 0; i < n_pages; i++) {
		char v = area[i * page_size];
		if ((touch_page[i] && !v) || (!touch_page[i] && v))
			return 4;
		if (!touch_page[i])
			continue;
		if (modified_pages[j] != (area + i * page_size))
			return 5;
		++j;
	}

	vm_release(area, area_size);
	return 0;
}
#endif

#ifdef CONFIGURE_TEST_VM_MAP
#include <stdlib.h>
#include <signal.h>

static void fault_handler(int sig)
{
	exit(1);
}

/* Tests covered here:
   - TEST_VM_PROT_* program slices actually succeeds when a crash occurs
   - TEST_VM_MAP_ANON* program slices succeeds when it could be compiled
*/
int main(void)
{
	vm_init();

	signal(SIGSEGV, fault_handler);
#ifdef SIGBUS
	signal(SIGBUS,  fault_handler);
#endif
	
#define page_align(address) ((char *)((vm_uintptr_t)(address) & -page_size))
	vm_uintptr_t page_size = vm_get_page_size();
	
	const int area_size = 6 * page_size;
	volatile char * area = (volatile char *) vm_acquire(area_size);
	volatile char * fault_address = area + (page_size * 7) / 2;

#if defined(TEST_VM_MMAP_ANON) || defined(TEST_VM_MMAP_ANONYMOUS)
	if (area == VM_MAP_FAILED)
		return 1;

	if (vm_release((char *)area, area_size) < 0)
		return 1;
	
	return 0;
#endif

#if defined(TEST_VM_PROT_NONE_READ) || defined(TEST_VM_PROT_NONE_WRITE)
	if (area == VM_MAP_FAILED)
		return 0;
	
	if (vm_protect(page_align(fault_address), page_size, VM_PAGE_NOACCESS) < 0)
		return 0;
#endif

#if defined(TEST_VM_PROT_RDWR_WRITE)
	if (area == VM_MAP_FAILED)
		return 1;
	
	if (vm_protect(page_align(fault_address), page_size, VM_PAGE_READ) < 0)
		return 1;
	
	if (vm_protect(page_align(fault_address), page_size, VM_PAGE_READ | VM_PAGE_WRITE) < 0)
		return 1;
#endif

#if defined(TEST_VM_PROT_READ_WRITE)
	if (vm_protect(page_align(fault_address), page_size, VM_PAGE_READ) < 0)
		return 0;
#endif

#if defined(TEST_VM_PROT_NONE_READ)
	// this should cause a core dump
	char foo = *fault_address;
	return 0;
#endif

#if defined(TEST_VM_PROT_NONE_WRITE) || defined(TEST_VM_PROT_READ_WRITE)
	// this should cause a core dump
	*fault_address = 'z';
	return 0;
#endif

#if defined(TEST_VM_PROT_RDWR_WRITE)
	// this should not cause a core dump
	*fault_address = 'z';
	return 0;
#endif
}
#endif