coreboot-kgpe-d16/util/cbfstool/cbfs_image.c

1936 lines
55 KiB
C

/*
* CBFS Image Manipulation
*
* Copyright (C) 2013 The Chromium OS Authors. All rights reserved.
* Copyright (C) 2016 Siemens AG. All rights reserved.
*
* 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; version 2 of the License.
*
* 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.
*/
#include <inttypes.h>
#include <libgen.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <strings.h>
#include <commonlib/endian.h>
#include "common.h"
#include "cbfs_image.h"
#include "elfparsing.h"
#include "rmodule.h"
/* Even though the file-adding functions---cbfs_add_entry() and
* cbfs_add_entry_at()---perform their sizing checks against the beginning of
* the subsequent section rather than a stable recorded value such as an empty
* file header's len field, it's possible to prove two interesting properties
* about their behavior:
* - Placing a new file within an empty entry located below an existing file
* entry will never leave an aligned flash address containing neither the
* beginning of a file header nor part of a file.
* - Placing a new file in an empty entry at the very end of the image such
* that it fits, but leaves no room for a final header, is guaranteed not to
* change the total amount of space for entries, even if that new file is
* later removed from the CBFS.
* These properties are somewhat nonobvious from the implementation, so the
* reader is encouraged to blame this comment and examine the full proofs
* in the commit message before making significant changes that would risk
* removing said guarantees.
*/
/* The file name align is not defined in CBFS spec -- only a preference by
* (old) cbfstool. */
#define CBFS_FILENAME_ALIGN (16)
/* Type and format */
static const struct typedesc_t types_cbfs_compression[] = {
{CBFS_COMPRESS_NONE, "none"},
{CBFS_COMPRESS_LZMA, "LZMA"},
{CBFS_COMPRESS_LZ4, "LZ4"},
{0, NULL},
};
static const char *lookup_name_by_type(const struct typedesc_t *desc, uint32_t type,
const char *default_value)
{
int i;
for (i = 0; desc[i].name; i++)
if (desc[i].type == type)
return desc[i].name;
return default_value;
}
static int lookup_type_by_name(const struct typedesc_t *desc, const char *name)
{
int i;
for (i = 0; desc[i].name && strcasecmp(name, desc[i].name); ++i);
return desc[i].name ? (int)desc[i].type : -1;
}
static const char *get_cbfs_entry_type_name(uint32_t type)
{
return lookup_name_by_type(filetypes, type, "(unknown)");
}
int cbfs_parse_comp_algo(const char *name)
{
return lookup_type_by_name(types_cbfs_compression, name);
}
static const char *get_hash_attr_name(uint16_t hash_type)
{
return lookup_name_by_type(types_cbfs_hash, hash_type, "(invalid)");
}
int cbfs_parse_hash_algo(const char *name)
{
return lookup_type_by_name(types_cbfs_hash, name);
}
/* CBFS image */
size_t cbfs_calculate_file_header_size(const char *name)
{
return (sizeof(struct cbfs_file) +
align_up(strlen(name) + 1, CBFS_FILENAME_ALIGN));
}
/* Only call on legacy CBFSes possessing a master header. */
static int cbfs_fix_legacy_size(struct cbfs_image *image, char *hdr_loc)
{
assert(image);
assert(cbfs_is_legacy_cbfs(image));
// A bug in old cbfstool may produce extra few bytes (by alignment) and
// cause cbfstool to overwrite things after free space -- which is
// usually CBFS header on x86. We need to workaround that.
// Except when we run across a file that contains the actual header,
// in which case this image is a safe, new-style
// `cbfstool add-master-header` based image.
struct cbfs_file *entry, *first = NULL, *last = NULL;
for (first = entry = cbfs_find_first_entry(image);
entry && cbfs_is_valid_entry(image, entry);
entry = cbfs_find_next_entry(image, entry)) {
/* Is the header guarded by a CBFS file entry? Then exit */
if (((char *)entry) + ntohl(entry->offset) == hdr_loc) {
return 0;
}
last = entry;
}
if ((char *)first < (char *)hdr_loc &&
(char *)entry > (char *)hdr_loc) {
WARN("CBFS image was created with old cbfstool with size bug. "
"Fixing size in last entry...\n");
last->len = htonl(ntohl(last->len) - image->header.align);
DEBUG("Last entry has been changed from 0x%x to 0x%x.\n",
cbfs_get_entry_addr(image, entry),
cbfs_get_entry_addr(image,
cbfs_find_next_entry(image, last)));
}
return 0;
}
void cbfs_put_header(void *dest, const struct cbfs_header *header)
{
struct buffer outheader;
outheader.data = dest;
outheader.size = 0;
xdr_be.put32(&outheader, header->magic);
xdr_be.put32(&outheader, header->version);
xdr_be.put32(&outheader, header->romsize);
xdr_be.put32(&outheader, header->bootblocksize);
xdr_be.put32(&outheader, header->align);
xdr_be.put32(&outheader, header->offset);
xdr_be.put32(&outheader, header->architecture);
}
static void cbfs_decode_payload_segment(struct cbfs_payload_segment *output,
struct cbfs_payload_segment *input)
{
struct buffer seg = {
.data = (void *)input,
.size = sizeof(*input),
};
output->type = xdr_be.get32(&seg);
output->compression = xdr_be.get32(&seg);
output->offset = xdr_be.get32(&seg);
output->load_addr = xdr_be.get64(&seg);
output->len = xdr_be.get32(&seg);
output->mem_len = xdr_be.get32(&seg);
assert(seg.size == 0);
}
static int cbfs_file_get_compression_info(struct cbfs_file *entry,
uint32_t *decompressed_size)
{
unsigned int compression = CBFS_COMPRESS_NONE;
*decompressed_size = ntohl(entry->len);
for (struct cbfs_file_attribute *attr = cbfs_file_first_attr(entry);
attr != NULL;
attr = cbfs_file_next_attr(entry, attr)) {
if (ntohl(attr->tag) == CBFS_FILE_ATTR_TAG_COMPRESSION) {
struct cbfs_file_attr_compression *ac =
(struct cbfs_file_attr_compression *)attr;
compression = ntohl(ac->compression);
if (decompressed_size)
*decompressed_size =
ntohl(ac->decompressed_size);
}
}
return compression;
}
static struct cbfs_file_attr_hash *cbfs_file_get_next_hash(
struct cbfs_file *entry, struct cbfs_file_attr_hash *cur)
{
struct cbfs_file_attribute *attr = (struct cbfs_file_attribute *)cur;
if (attr == NULL) {
attr = cbfs_file_first_attr(entry);
if (attr == NULL)
return NULL;
if (ntohl(attr->tag) == CBFS_FILE_ATTR_TAG_HASH)
return (struct cbfs_file_attr_hash *)attr;
}
while ((attr = cbfs_file_next_attr(entry, attr)) != NULL) {
if (ntohl(attr->tag) == CBFS_FILE_ATTR_TAG_HASH)
return (struct cbfs_file_attr_hash *)attr;
};
return NULL;
}
void cbfs_get_header(struct cbfs_header *header, void *src)
{
struct buffer outheader;
outheader.data = src; /* We're not modifying the data */
outheader.size = 0;
header->magic = xdr_be.get32(&outheader);
header->version = xdr_be.get32(&outheader);
header->romsize = xdr_be.get32(&outheader);
header->bootblocksize = xdr_be.get32(&outheader);
header->align = xdr_be.get32(&outheader);
header->offset = xdr_be.get32(&outheader);
header->architecture = xdr_be.get32(&outheader);
}
int cbfs_image_create(struct cbfs_image *image, size_t entries_size)
{
assert(image);
assert(image->buffer.data);
size_t empty_header_len = cbfs_calculate_file_header_size("");
uint32_t entries_offset = 0;
uint32_t align = CBFS_ENTRY_ALIGNMENT;
if (image->has_header) {
entries_offset = image->header.offset;
if (entries_offset > image->buffer.size) {
ERROR("CBFS file entries are located outside CBFS itself\n");
return -1;
}
align = image->header.align;
}
// This attribute must be given in order to prove that this module
// correctly preserves certain CBFS properties. See the block comment
// near the top of this file (and the associated commit message).
if (align < empty_header_len) {
ERROR("CBFS must be aligned to at least %zu bytes\n",
empty_header_len);
return -1;
}
if (entries_size > image->buffer.size - entries_offset) {
ERROR("CBFS doesn't have enough space to fit its file entries\n");
return -1;
}
if (empty_header_len > entries_size) {
ERROR("CBFS is too small to fit any header\n");
return -1;
}
struct cbfs_file *entry_header =
(struct cbfs_file *)(image->buffer.data + entries_offset);
// This alignment is necessary in order to prove that this module
// correctly preserves certain CBFS properties. See the block comment
// near the top of this file (and the associated commit message).
entries_size -= entries_size % align;
size_t capacity = entries_size - empty_header_len;
LOG("Created CBFS (capacity = %zu bytes)\n", capacity);
return cbfs_create_empty_entry(entry_header, CBFS_COMPONENT_NULL,
capacity, "");
}
int cbfs_legacy_image_create(struct cbfs_image *image,
uint32_t architecture,
uint32_t align,
struct buffer *bootblock,
uint32_t bootblock_offset,
uint32_t header_offset,
uint32_t entries_offset)
{
assert(image);
assert(image->buffer.data);
assert(bootblock);
int32_t *rel_offset;
uint32_t cbfs_len;
void *header_loc;
size_t size = image->buffer.size;
DEBUG("cbfs_image_create: bootblock=0x%x+0x%zx, "
"header=0x%x+0x%zx, entries_offset=0x%x\n",
bootblock_offset, bootblock->size, header_offset,
sizeof(image->header), entries_offset);
// Adjust legacy top-aligned address to ROM offset.
if (IS_TOP_ALIGNED_ADDRESS(entries_offset))
entries_offset = size + (int32_t)entries_offset;
if (IS_TOP_ALIGNED_ADDRESS(bootblock_offset))
bootblock_offset = size + (int32_t)bootblock_offset;
if (IS_TOP_ALIGNED_ADDRESS(header_offset))
header_offset = size + (int32_t)header_offset;
DEBUG("cbfs_create_image: (real offset) bootblock=0x%x, "
"header=0x%x, entries_offset=0x%x\n",
bootblock_offset, header_offset, entries_offset);
// Prepare bootblock
if (bootblock_offset + bootblock->size > size) {
ERROR("Bootblock (0x%x+0x%zx) exceed ROM size (0x%zx)\n",
bootblock_offset, bootblock->size, size);
return -1;
}
if (entries_offset > bootblock_offset &&
entries_offset < bootblock->size) {
ERROR("Bootblock (0x%x+0x%zx) overlap CBFS data (0x%x)\n",
bootblock_offset, bootblock->size, entries_offset);
return -1;
}
memcpy(image->buffer.data + bootblock_offset, bootblock->data,
bootblock->size);
// Prepare header
if (header_offset + sizeof(image->header) > size - sizeof(int32_t)) {
ERROR("Header (0x%x+0x%zx) exceed ROM size (0x%zx)\n",
header_offset, sizeof(image->header), size);
return -1;
}
image->header.magic = CBFS_HEADER_MAGIC;
image->header.version = CBFS_HEADER_VERSION;
image->header.romsize = size;
image->header.bootblocksize = bootblock->size;
image->header.align = align;
image->header.offset = entries_offset;
image->header.architecture = architecture;
header_loc = (image->buffer.data + header_offset);
cbfs_put_header(header_loc, &image->header);
image->has_header = true;
// The last 4 byte of the image contain the relative offset from the end
// of the image to the master header as a 32-bit signed integer. x86
// relies on this also being its (memory-mapped, top-aligned) absolute
// 32-bit address by virtue of how two's complement numbers work.
assert(size % sizeof(int32_t) == 0);
rel_offset = (int32_t *)(image->buffer.data + size - sizeof(int32_t));
*rel_offset = header_offset - size;
// Prepare entries
if (align_up(entries_offset, align) != entries_offset) {
ERROR("Offset (0x%x) must be aligned to 0x%x.\n",
entries_offset, align);
return -1;
}
// To calculate available length, find
// e = min(bootblock, header, rel_offset) where e > entries_offset.
cbfs_len = size - sizeof(int32_t);
if (bootblock_offset > entries_offset && bootblock_offset < cbfs_len)
cbfs_len = bootblock_offset;
if (header_offset > entries_offset && header_offset < cbfs_len)
cbfs_len = header_offset;
if (cbfs_image_create(image, cbfs_len - entries_offset))
return -1;
return 0;
}
int cbfs_image_from_buffer(struct cbfs_image *out, struct buffer *in,
uint32_t offset)
{
assert(out);
assert(in);
assert(in->data);
buffer_clone(&out->buffer, in);
out->has_header = false;
if (cbfs_is_valid_cbfs(out)) {
return 0;
}
void *header_loc = cbfs_find_header(in->data, in->size, offset);
if (header_loc) {
cbfs_get_header(&out->header, header_loc);
out->has_header = true;
cbfs_fix_legacy_size(out, header_loc);
return 0;
} else if (offset != ~0u) {
ERROR("The -H switch is only valid on legacy images having CBFS master headers.\n");
return 1;
}
ERROR("Selected image region is not a valid CBFS.\n");
return 1;
}
int cbfs_copy_instance(struct cbfs_image *image, struct buffer *dst)
{
assert(image);
struct cbfs_file *src_entry, *dst_entry;
size_t align;
ssize_t last_entry_size;
size_t copy_end = buffer_size(dst);
align = CBFS_ENTRY_ALIGNMENT;
dst_entry = (struct cbfs_file *)buffer_get(dst);
/* Copy non-empty files */
for (src_entry = cbfs_find_first_entry(image);
src_entry && cbfs_is_valid_entry(image, src_entry);
src_entry = cbfs_find_next_entry(image, src_entry)) {
size_t entry_size;
if ((src_entry->type == htonl(CBFS_COMPONENT_NULL)) ||
(src_entry->type == htonl(CBFS_COMPONENT_CBFSHEADER)) ||
(src_entry->type == htonl(CBFS_COMPONENT_DELETED)))
continue;
entry_size = htonl(src_entry->len) + htonl(src_entry->offset);
memcpy(dst_entry, src_entry, entry_size);
dst_entry = (struct cbfs_file *)(
(uintptr_t)dst_entry + align_up(entry_size, align));
if ((size_t)((void *)dst_entry - buffer_get(dst)) >=
copy_end) {
ERROR("Ran out of room in copy region.\n");
return 1;
}
}
/* Last entry size is all the room above it, except for top 4 bytes
* which may be used by the master header pointer. This messes with
* the ability to stash something "top-aligned" into the region, but
* keeps things simpler. */
last_entry_size = copy_end - ((void *)dst_entry - buffer_get(dst))
- cbfs_calculate_file_header_size("") - sizeof(int32_t);
if (last_entry_size < 0)
WARN("No room to create the last entry!\n")
else
cbfs_create_empty_entry(dst_entry, CBFS_COMPONENT_NULL,
last_entry_size, "");
return 0;
}
static size_t cbfs_file_entry_metadata_size(const struct cbfs_file *f)
{
return ntohl(f->offset);
}
static size_t cbfs_file_entry_data_size(const struct cbfs_file *f)
{
return ntohl(f->len);
}
static size_t cbfs_file_entry_size(const struct cbfs_file *f)
{
return cbfs_file_entry_metadata_size(f) + cbfs_file_entry_data_size(f);
}
int cbfs_compact_instance(struct cbfs_image *image)
{
assert(image);
struct cbfs_file *prev;
struct cbfs_file *cur;
/* The prev entry will always be an empty entry. */
prev = NULL;
/*
* Note: this function does not honor alignment or fixed location files.
* It's behavior is akin to cbfs_copy_instance() in that it expects
* the caller to understand the ramifications of compacting a
* fragmented CBFS image.
*/
for (cur = cbfs_find_first_entry(image);
cur && cbfs_is_valid_entry(image, cur);
cur = cbfs_find_next_entry(image, cur)) {
size_t prev_size;
size_t cur_size;
size_t empty_metadata_size;
size_t spill_size;
uint32_t type = htonl(cur->type);
/* Current entry is empty. Kepp track of it. */
if ((type == htonl(CBFS_COMPONENT_NULL)) ||
(type == htonl(CBFS_COMPONENT_DELETED))) {
prev = cur;
continue;
}
/* Need to ensure the previous entry is an empty one. */
if (prev == NULL)
continue;
/* At this point prev is an empty entry. Put the non-empty
* file in prev's location. Then add a new emptry entry. This
* essentialy bubbles empty entries towards the end. */
prev_size = cbfs_file_entry_size(prev);
cur_size = cbfs_file_entry_size(cur);
/*
* Adjust the empty file size by the actual space occupied
* bewtween the beginning of the empty file and the non-empty
* file.
*/
prev_size += (cbfs_get_entry_addr(image, cur) -
cbfs_get_entry_addr(image, prev)) - prev_size;
/* Move the non-empty file over the empty file. */
memmove(prev, cur, cur_size);
/*
* Get location of the empty file. Note that since prev was
* overwritten with the non-empty file the previously moved
* file needs to be used to calculate the empty file's location.
*/
cur = cbfs_find_next_entry(image, prev);
/*
* The total space to work with for swapping the 2 entries
* consists of the 2 files' sizes combined. However, the
* cbfs_file entries start on CBFS_ALIGNMENT boundaries.
* Because of this the empty file size may end up smaller
* because of the non-empty file's metadata and data length.
*
* Calculate the spill size which is the amount of data lost
* due to the alignment constraints after moving the non-empty
* file.
*/
spill_size = (cbfs_get_entry_addr(image, cur) -
cbfs_get_entry_addr(image, prev)) - cur_size;
empty_metadata_size = cbfs_calculate_file_header_size("");
/* Check if new empty size can contain the metadata. */
if (empty_metadata_size + spill_size > prev_size) {
ERROR("Unable to swap '%s' with prev empty entry.\n",
prev->filename);
return 1;
}
/* Update the empty file's size. */
prev_size -= spill_size + empty_metadata_size;
/* Create new empty file. */
cbfs_create_empty_entry(cur, CBFS_COMPONENT_NULL,
prev_size, "");
/* Merge any potential empty entries together. */
cbfs_walk(image, cbfs_merge_empty_entry, NULL);
/*
* Since current switched to an empty file keep track of it.
* Even if any empty files were merged the empty entry still
* starts at previously calculated location.
*/
prev = cur;
}
return 0;
}
int cbfs_image_delete(struct cbfs_image *image)
{
if (image == NULL)
return 0;
buffer_delete(&image->buffer);
return 0;
}
/* Tries to add an entry with its data (CBFS_SUBHEADER) at given offset. */
static int cbfs_add_entry_at(struct cbfs_image *image,
struct cbfs_file *entry,
const void *data,
uint32_t content_offset,
const struct cbfs_file *header)
{
struct cbfs_file *next = cbfs_find_next_entry(image, entry);
uint32_t addr = cbfs_get_entry_addr(image, entry),
addr_next = cbfs_get_entry_addr(image, next);
uint32_t min_entry_size = cbfs_calculate_file_header_size("");
uint32_t len, header_offset;
uint32_t align = image->has_header ? image->header.align :
CBFS_ENTRY_ALIGNMENT;
uint32_t header_size = ntohl(header->offset);
header_offset = content_offset - header_size;
if (header_offset % align)
header_offset -= header_offset % align;
if (header_offset < addr) {
ERROR("No space to hold cbfs_file header.");
return -1;
}
// Process buffer BEFORE content_offset.
if (header_offset - addr > min_entry_size) {
DEBUG("|min|...|header|content|... <create new entry>\n");
len = header_offset - addr - min_entry_size;
cbfs_create_empty_entry(entry, CBFS_COMPONENT_NULL, len, "");
if (verbose > 1) cbfs_print_entry_info(image, entry, stderr);
entry = cbfs_find_next_entry(image, entry);
addr = cbfs_get_entry_addr(image, entry);
}
len = content_offset - addr - header_size;
memcpy(entry, header, header_size);
if (len != 0) {
/* the header moved backwards a bit to accomodate cbfs_file
* alignment requirements, so patch up ->offset to still point
* to file data.
*/
DEBUG("|..|header|content|... <use offset to create entry>\n");
DEBUG("before: offset=0x%x\n", ntohl(entry->offset));
// TODO reset expanded name buffer to 0xFF.
entry->offset = htonl(ntohl(entry->offset) + len);
DEBUG("after: offset=0x%x\n", ntohl(entry->len));
}
// Ready to fill data into entry.
DEBUG("content_offset: 0x%x, entry location: %x\n",
content_offset, (int)((char*)CBFS_SUBHEADER(entry) -
image->buffer.data));
assert((char*)CBFS_SUBHEADER(entry) - image->buffer.data ==
(ptrdiff_t)content_offset);
memcpy(CBFS_SUBHEADER(entry), data, ntohl(entry->len));
if (verbose > 1) cbfs_print_entry_info(image, entry, stderr);
// Process buffer AFTER entry.
entry = cbfs_find_next_entry(image, entry);
addr = cbfs_get_entry_addr(image, entry);
if (addr == addr_next)
return 0;
assert(addr < addr_next);
if (addr_next - addr < min_entry_size) {
DEBUG("No need for new \"empty\" entry\n");
/* No need to increase the size of the just
* stored file to extend to next file. Alignment
* of next file takes care of this.
*/
return 0;
}
len = addr_next - addr - min_entry_size;
/* keep space for master header pointer */
if ((void *)entry + min_entry_size + len > buffer_get(&image->buffer) +
buffer_size(&image->buffer) - sizeof(int32_t)) {
len -= sizeof(int32_t);
}
cbfs_create_empty_entry(entry, CBFS_COMPONENT_NULL, len, "");
if (verbose > 1) cbfs_print_entry_info(image, entry, stderr);
return 0;
}
int cbfs_add_entry(struct cbfs_image *image, struct buffer *buffer,
uint32_t content_offset,
struct cbfs_file *header)
{
assert(image);
assert(buffer);
assert(buffer->data);
assert(!IS_TOP_ALIGNED_ADDRESS(content_offset));
const char *name = header->filename;
uint32_t entry_type;
uint32_t addr, addr_next;
struct cbfs_file *entry, *next;
uint32_t need_size;
uint32_t header_size = ntohl(header->offset);
need_size = header_size + buffer->size;
DEBUG("cbfs_add_entry('%s'@0x%x) => need_size = %u+%zu=%u\n",
name, content_offset, header_size, buffer->size, need_size);
// Merge empty entries.
DEBUG("(trying to merge empty entries...)\n");
cbfs_walk(image, cbfs_merge_empty_entry, NULL);
for (entry = cbfs_find_first_entry(image);
entry && cbfs_is_valid_entry(image, entry);
entry = cbfs_find_next_entry(image, entry)) {
entry_type = ntohl(entry->type);
if (entry_type != CBFS_COMPONENT_NULL)
continue;
addr = cbfs_get_entry_addr(image, entry);
next = cbfs_find_next_entry(image, entry);
addr_next = cbfs_get_entry_addr(image, next);
DEBUG("cbfs_add_entry: space at 0x%x+0x%x(%d) bytes\n",
addr, addr_next - addr, addr_next - addr);
/* Will the file fit? Don't yet worry if we have space for a new
* "empty" entry. We take care of that later.
*/
if (addr + need_size > addr_next)
continue;
// Test for complicated cases
if (content_offset > 0) {
if (addr_next < content_offset) {
DEBUG("Not for specified offset yet");
continue;
} else if (addr > content_offset) {
DEBUG("Exceed specified content_offset.");
break;
} else if (addr + header_size > content_offset) {
ERROR("Not enough space for header.\n");
break;
} else if (content_offset + buffer->size > addr_next) {
ERROR("Not enough space for content.\n");
break;
}
}
// TODO there are more few tricky cases that we may
// want to fit by altering offset.
if (content_offset == 0) {
// we tested every condition earlier under which
// placing the file there might fail
content_offset = addr + header_size;
}
DEBUG("section 0x%x+0x%x for content_offset 0x%x.\n",
addr, addr_next - addr, content_offset);
if (cbfs_add_entry_at(image, entry, buffer->data,
content_offset, header) == 0) {
return 0;
}
break;
}
ERROR("Could not add [%s, %zd bytes (%zd KB)@0x%x]; too big?\n",
buffer->name, buffer->size, buffer->size / 1024, content_offset);
return -1;
}
struct cbfs_file *cbfs_get_entry(struct cbfs_image *image, const char *name)
{
struct cbfs_file *entry;
for (entry = cbfs_find_first_entry(image);
entry && cbfs_is_valid_entry(image, entry);
entry = cbfs_find_next_entry(image, entry)) {
if (strcasecmp(entry->filename, name) == 0) {
DEBUG("cbfs_get_entry: found %s\n", name);
return entry;
}
}
return NULL;
}
static int cbfs_stage_decompress(struct cbfs_stage *stage, struct buffer *buff)
{
struct buffer reader;
char *orig_buffer;
char *new_buffer;
size_t new_buff_sz;
decomp_func_ptr decompress;
buffer_clone(&reader, buff);
/* The stage metadata is in little endian. */
stage->compression = xdr_le.get32(&reader);
stage->entry = xdr_le.get64(&reader);
stage->load = xdr_le.get64(&reader);
stage->len = xdr_le.get32(&reader);
stage->memlen = xdr_le.get32(&reader);
/* Create a buffer just with the uncompressed program now that the
* struct cbfs_stage has been peeled off. */
if (stage->compression == CBFS_COMPRESS_NONE) {
new_buff_sz = buffer_size(buff) - sizeof(struct cbfs_stage);
orig_buffer = buffer_get(buff);
new_buffer = calloc(1, new_buff_sz);
memcpy(new_buffer, orig_buffer + sizeof(struct cbfs_stage),
new_buff_sz);
buffer_init(buff, buff->name, new_buffer, new_buff_sz);
free(orig_buffer);
return 0;
}
decompress = decompression_function(stage->compression);
if (decompress == NULL)
return -1;
orig_buffer = buffer_get(buff);
/* This can be too big of a buffer needed, but there's no current
* field indicating decompressed size of data. */
new_buff_sz = stage->memlen;
new_buffer = calloc(1, new_buff_sz);
if (decompress(orig_buffer + sizeof(struct cbfs_stage),
(int)(buffer_size(buff) - sizeof(struct cbfs_stage)),
new_buffer, (int)new_buff_sz, &new_buff_sz)) {
ERROR("Couldn't decompress stage.\n");
free(new_buffer);
return -1;
}
/* Include correct size for full stage info. */
buffer_init(buff, buff->name, new_buffer, new_buff_sz);
/* True decompressed size is just the data size -- no metadata. */
stage->len = new_buff_sz;
/* Stage is not compressed. */
stage->compression = CBFS_COMPRESS_NONE;
free(orig_buffer);
return 0;
}
static int cbfs_payload_decompress(struct cbfs_payload_segment *segments,
struct buffer *buff, int num_seg)
{
struct buffer new_buffer;
struct buffer seg_buffer;
size_t new_buff_sz;
char *in_ptr;
char *out_ptr;
size_t new_offset;
decomp_func_ptr decompress;
new_offset = num_seg * sizeof(*segments);
new_buff_sz = num_seg * sizeof(*segments);
/* Find out and allocate the amount of memory occupied
* by the binary data */
for (int i = 0; i < num_seg; i++)
new_buff_sz += segments[i].mem_len;
buffer_create(&new_buffer, new_buff_sz, "decompressed_buff");
in_ptr = buffer_get(buff) + new_offset;
out_ptr = buffer_get(&new_buffer) + new_offset;
for (int i = 0; i < num_seg; i++) {
struct buffer tbuff;
size_t decomp_size;
/* The payload uses an unknown compression algorithm. */
decompress = decompression_function(segments[i].compression);
if (decompress == NULL) {
ERROR("Unknown decompression algorithm: %u",
segments[i].compression);
return -1;
}
/* Segments BSS and ENTRY do not have binary data. */
if (segments[i].type == PAYLOAD_SEGMENT_BSS ||
segments[i].type == PAYLOAD_SEGMENT_ENTRY) {
continue;
} else if (segments[i].type == PAYLOAD_SEGMENT_PARAMS) {
memcpy(out_ptr, in_ptr, segments[i].len);
segments[i].offset = new_offset;
new_offset += segments[i].len;
in_ptr += segments[i].len;
out_ptr += segments[i].len;
segments[i].compression = CBFS_COMPRESS_NONE;
continue;
}
buffer_create(&tbuff, segments[i].mem_len, "segment");
if (decompress(in_ptr, segments[i].len, buffer_get(&tbuff),
(int) buffer_size(&tbuff),
&decomp_size)) {
ERROR("Couldn't decompress payload segment %u\n", i);
buffer_delete(&new_buffer);
return -1;
}
memcpy(out_ptr, buffer_get(&tbuff), decomp_size);
in_ptr += segments[i].len;
/* Update the offset of the segment. */
segments[i].offset = new_offset;
/* True decompressed size is just the data size. No metadata */
segments[i].len = decomp_size;
/* Segment is not compressed. */
segments[i].compression = CBFS_COMPRESS_NONE;
/* Update the offset and output buffer pointer. */
new_offset += decomp_size;
out_ptr += decomp_size;
buffer_delete(&tbuff);
}
buffer_splice(&seg_buffer, &new_buffer, 0, 0);
xdr_segs(&seg_buffer, segments, num_seg);
buffer_delete(buff);
*buff = new_buffer;
return 0;
}
static int init_elf_from_arch(Elf64_Ehdr *ehdr, uint32_t cbfs_arch)
{
int endian;
int nbits;
int machine;
switch (cbfs_arch) {
case CBFS_ARCHITECTURE_X86:
endian = ELFDATA2LSB;
nbits = ELFCLASS32;
machine = EM_386;
break;
case CBFS_ARCHITECTURE_ARM:
endian = ELFDATA2LSB;
nbits = ELFCLASS32;
machine = EM_ARM;
break;
case CBFS_ARCHITECTURE_AARCH64:
endian = ELFDATA2LSB;
nbits = ELFCLASS64;
machine = EM_AARCH64;
break;
case CBFS_ARCHITECTURE_MIPS:
endian = ELFDATA2LSB;
nbits = ELFCLASS32;
machine = EM_MIPS;
break;
case CBFS_ARCHITECTURE_RISCV:
endian = ELFDATA2LSB;
nbits = ELFCLASS32;
machine = EM_RISCV;
break;
default:
ERROR("Unsupported arch: %x\n", cbfs_arch);
return -1;
}
elf_init_eheader(ehdr, machine, nbits, endian);
return 0;
}
static int cbfs_stage_make_elf(struct buffer *buff, uint32_t arch)
{
Elf64_Ehdr ehdr;
Elf64_Shdr shdr;
struct cbfs_stage stage;
struct elf_writer *ew;
struct buffer elf_out;
size_t empty_sz;
int rmod_ret;
if (cbfs_stage_decompress(&stage, buff)) {
ERROR("Failed to decompress stage.\n");
return -1;
}
if (init_elf_from_arch(&ehdr, arch))
return -1;
ehdr.e_entry = stage.entry;
/* Attempt rmodule translation first. */
rmod_ret = rmodule_stage_to_elf(&ehdr, buff);
if (rmod_ret < 0) {
ERROR("rmodule parsing failed\n");
return -1;
} else if (rmod_ret == 0)
return 0;
/* Rmodule couldn't do anything with the data. Continue on with SELF. */
ew = elf_writer_init(&ehdr);
if (ew == NULL) {
ERROR("Unable to init ELF writer.\n");
return -1;
}
memset(&shdr, 0, sizeof(shdr));
shdr.sh_type = SHT_PROGBITS;
shdr.sh_flags = SHF_WRITE | SHF_ALLOC | SHF_EXECINSTR;
shdr.sh_addr = stage.load;
shdr.sh_size = stage.len;
empty_sz = stage.memlen - stage.len;
if (elf_writer_add_section(ew, &shdr, buff, ".program")) {
ERROR("Unable to add ELF section: .program\n");
elf_writer_destroy(ew);
return -1;
}
if (empty_sz != 0) {
struct buffer b;
buffer_init(&b, NULL, NULL, 0);
memset(&shdr, 0, sizeof(shdr));
shdr.sh_type = SHT_NOBITS;
shdr.sh_flags = SHF_WRITE | SHF_ALLOC;
shdr.sh_addr = stage.load + stage.len;
shdr.sh_size = empty_sz;
if (elf_writer_add_section(ew, &shdr, &b, ".empty")) {
ERROR("Unable to add ELF section: .empty\n");
elf_writer_destroy(ew);
return -1;
}
}
if (elf_writer_serialize(ew, &elf_out)) {
ERROR("Unable to create ELF file from stage.\n");
elf_writer_destroy(ew);
return -1;
}
/* Flip buffer with the created ELF one. */
buffer_delete(buff);
*buff = elf_out;
elf_writer_destroy(ew);
return 0;
}
static int cbfs_payload_make_elf(struct buffer *buff, uint32_t arch)
{
Elf64_Ehdr ehdr;
Elf64_Shdr shdr;
struct cbfs_payload_segment *segs;
struct elf_writer *ew;
struct buffer elf_out;
size_t empty_sz;
int segments = 0;
/* Count the number of segments inside buffer */
while (true) {
uint32_t payload_type = 0;
struct cbfs_payload_segment *seg;
seg = buffer_get(buff);
payload_type = read_be32(&seg[segments].type);
if (payload_type == PAYLOAD_SEGMENT_CODE) {
segments++;
} else if (payload_type == PAYLOAD_SEGMENT_DATA) {
segments++;
} else if (payload_type == PAYLOAD_SEGMENT_BSS) {
segments++;
} else if (payload_type == PAYLOAD_SEGMENT_PARAMS) {
segments++;
} else if (payload_type == PAYLOAD_SEGMENT_ENTRY) {
/* The last segment in a payload is always ENTRY as
* specified by the parse_elf_to_payload() function.
* Therefore there is no need to continue looking for
* segments.*/
segments++;
break;
} else {
ERROR("Unknown payload segment type: %x\n",
payload_type);
return -1;
}
}
segs = malloc(segments * sizeof(*segs));
/* Decode xdr segments */
for (int i = 0; i < segments; i++) {
struct cbfs_payload_segment *serialized_seg = buffer_get(buff);
xdr_get_seg(&segs[i], &serialized_seg[i]);
}
if (cbfs_payload_decompress(segs, buff, segments)) {
ERROR("Failed to decompress payload.\n");
return -1;
}
if (init_elf_from_arch(&ehdr, arch))
return -1;
ehdr.e_entry = segs[segments-1].load_addr;
ew = elf_writer_init(&ehdr);
if (ew == NULL) {
ERROR("Unable to init ELF writer.\n");
return -1;
}
for (int i = 0; i < segments; i++) {
struct buffer tbuff;
memset(&shdr, 0, sizeof(shdr));
char *name = NULL;
if (segs[i].type == PAYLOAD_SEGMENT_CODE) {
shdr.sh_type = SHT_PROGBITS;
shdr.sh_flags = SHF_WRITE | SHF_ALLOC | SHF_EXECINSTR;
shdr.sh_addr = segs[i].load_addr;
shdr.sh_size = segs[i].len;
empty_sz = segs[i].mem_len - segs[i].len;
name = strdup(".text");
buffer_splice(&tbuff, buff, segs[i].offset,
segs[i].len);
} else if (segs[i].type == PAYLOAD_SEGMENT_DATA) {
shdr.sh_type = SHT_PROGBITS;
shdr.sh_flags = SHF_ALLOC | SHF_WRITE;
shdr.sh_addr = segs[i].load_addr;
shdr.sh_size = segs[i].len;
empty_sz = segs[i].mem_len - segs[i].len;
name = strdup(".data");
buffer_splice(&tbuff, buff, segs[i].offset,
segs[i].len);
} else if (segs[i].type == PAYLOAD_SEGMENT_BSS) {
shdr.sh_type = SHT_NOBITS;
shdr.sh_flags = SHF_ALLOC | SHF_WRITE;
shdr.sh_addr = segs[i].load_addr;
shdr.sh_size = segs[i].len;
name = strdup(".bss");
buffer_splice(&tbuff, buff, 0, 0);
} else if (segs[i].type == PAYLOAD_SEGMENT_PARAMS) {
shdr.sh_type = SHT_NOTE;
shdr.sh_flags = 0;
shdr.sh_size = segs[i].len;
name = strdup(".note.pinfo");
buffer_splice(&tbuff, buff, segs[i].offset,
segs[i].len);
} else if (segs[i].type == PAYLOAD_SEGMENT_ENTRY) {
break;
}
if (elf_writer_add_section(ew, &shdr, &tbuff, name)) {
ERROR("Unable to add ELF section: %s\n", name);
elf_writer_destroy(ew);
return -1;
}
if (empty_sz != 0) {
struct buffer b;
buffer_init(&b, NULL, NULL, 0);
memset(&shdr, 0, sizeof(shdr));
shdr.sh_type = SHT_NOBITS;
shdr.sh_flags = SHF_WRITE | SHF_ALLOC;
shdr.sh_addr = segs[i].load_addr + segs[i].len;
shdr.sh_size = empty_sz;
name = strdup(".empty");
if (elf_writer_add_section(ew, &shdr, &b, name)) {
ERROR("Unable to add ELF section: %s\n", name);
elf_writer_destroy(ew);
return -1;
}
}
}
if (elf_writer_serialize(ew, &elf_out)) {
ERROR("Unable to create ELF file from stage.\n");
elf_writer_destroy(ew);
return -1;
}
/* Flip buffer with the created ELF one. */
buffer_delete(buff);
*buff = elf_out;
elf_writer_destroy(ew);
return 0;
}
int cbfs_export_entry(struct cbfs_image *image, const char *entry_name,
const char *filename, uint32_t arch)
{
struct cbfs_file *entry = cbfs_get_entry(image, entry_name);
struct buffer buffer;
if (!entry) {
ERROR("File not found: %s\n", entry_name);
return -1;
}
unsigned int decompressed_size = 0;
unsigned int compression = cbfs_file_get_compression_info(entry,
&decompressed_size);
decomp_func_ptr decompress = decompression_function(compression);
if (!decompress) {
ERROR("looking up decompression routine failed\n");
return -1;
}
LOG("Found file %.30s at 0x%x, type %.12s, size %d\n",
entry_name, cbfs_get_entry_addr(image, entry),
get_cbfs_entry_type_name(ntohl(entry->type)), decompressed_size);
buffer_init(&buffer, strdup("(cbfs_export_entry)"), NULL, 0);
buffer.data = malloc(decompressed_size);
buffer.size = decompressed_size;
if (decompress(CBFS_SUBHEADER(entry), ntohl(entry->len),
buffer.data, buffer.size, NULL)) {
ERROR("decompression failed for %s\n", entry_name);
buffer_delete(&buffer);
return -1;
}
/*
* The stage metadata is never compressed proper for cbfs_stage
* files. The contents of the stage data can be though. Therefore
* one has to do a second pass for stages to potentially decompress
* the stage data to make it more meaningful.
*/
if (ntohl(entry->type) == CBFS_COMPONENT_STAGE) {
if (cbfs_stage_make_elf(&buffer, arch)) {
buffer_delete(&buffer);
return -1;
}
} else if (ntohl(entry->type) == CBFS_COMPONENT_PAYLOAD) {
if (cbfs_payload_make_elf(&buffer, arch)) {
buffer_delete(&buffer);
return -1;
}
}
if (buffer_write_file(&buffer, filename) != 0) {
ERROR("Failed to write %s into %s.\n",
entry_name, filename);
buffer_delete(&buffer);
return -1;
}
buffer_delete(&buffer);
INFO("Successfully dumped the file to: %s\n", filename);
return 0;
}
int cbfs_remove_entry(struct cbfs_image *image, const char *name)
{
struct cbfs_file *entry;
entry = cbfs_get_entry(image, name);
if (!entry) {
ERROR("CBFS file %s not found.\n", name);
return -1;
}
DEBUG("cbfs_remove_entry: Removed %s @ 0x%x\n",
entry->filename, cbfs_get_entry_addr(image, entry));
entry->type = htonl(CBFS_COMPONENT_DELETED);
cbfs_walk(image, cbfs_merge_empty_entry, NULL);
return 0;
}
int cbfs_print_header_info(struct cbfs_image *image)
{
char *name = strdup(image->buffer.name);
assert(image);
printf("%s: %zd kB, bootblocksize %d, romsize %d, offset 0x%x\n"
"alignment: %d bytes, architecture: %s\n\n",
basename(name),
image->buffer.size / 1024,
image->header.bootblocksize,
image->header.romsize,
image->header.offset,
image->header.align,
arch_to_string(image->header.architecture));
free(name);
return 0;
}
static int cbfs_print_stage_info(struct cbfs_stage *stage, FILE* fp)
{
fprintf(fp,
" %s compression, entry: 0x%" PRIx64 ", load: 0x%" PRIx64 ", "
"length: %d/%d\n",
lookup_name_by_type(types_cbfs_compression,
stage->compression, "(unknown)"),
stage->entry,
stage->load,
stage->len,
stage->memlen);
return 0;
}
static int cbfs_print_decoded_payload_segment_info(
struct cbfs_payload_segment *seg, FILE *fp)
{
/* The input (seg) must be already decoded by
* cbfs_decode_payload_segment.
*/
switch (seg->type) {
case PAYLOAD_SEGMENT_CODE:
case PAYLOAD_SEGMENT_DATA:
fprintf(fp, " %s (%s compression, offset: 0x%x, "
"load: 0x%" PRIx64 ", length: %d/%d)\n",
(seg->type == PAYLOAD_SEGMENT_CODE ?
"code " : "data"),
lookup_name_by_type(types_cbfs_compression,
seg->compression,
"(unknown)"),
seg->offset, seg->load_addr, seg->len,
seg->mem_len);
break;
case PAYLOAD_SEGMENT_ENTRY:
fprintf(fp, " entry (0x%" PRIx64 ")\n",
seg->load_addr);
break;
case PAYLOAD_SEGMENT_BSS:
fprintf(fp, " BSS (address 0x%016" PRIx64 ", "
"length 0x%x)\n",
seg->load_addr, seg->len);
break;
case PAYLOAD_SEGMENT_PARAMS:
fprintf(fp, " parameters\n");
break;
default:
fprintf(fp, " 0x%x (%s compression, offset: 0x%x, "
"load: 0x%" PRIx64 ", length: %d/%d\n",
seg->type,
lookup_name_by_type(types_cbfs_compression,
seg->compression,
"(unknown)"),
seg->offset, seg->load_addr, seg->len,
seg->mem_len);
break;
}
return 0;
}
int cbfs_print_entry_info(struct cbfs_image *image, struct cbfs_file *entry,
void *arg)
{
const char *name = entry->filename;
struct cbfs_payload_segment *payload;
FILE *fp = (FILE *)arg;
if (!cbfs_is_valid_entry(image, entry)) {
ERROR("cbfs_print_entry_info: Invalid entry at 0x%x\n",
cbfs_get_entry_addr(image, entry));
return -1;
}
if (!fp)
fp = stdout;
unsigned int decompressed_size = 0;
unsigned int compression = cbfs_file_get_compression_info(entry,
&decompressed_size);
if (compression == CBFS_COMPRESS_NONE) {
fprintf(fp, "%-30s 0x%-8x %-12s %d\n",
*name ? name : "(empty)",
cbfs_get_entry_addr(image, entry),
get_cbfs_entry_type_name(ntohl(entry->type)),
ntohl(entry->len));
} else {
fprintf(fp, "%-30s 0x%-8x %-12s %d (%d after %s decompression)\n",
*name ? name : "(empty)",
cbfs_get_entry_addr(image, entry),
get_cbfs_entry_type_name(ntohl(entry->type)),
ntohl(entry->len),
decompressed_size,
lookup_name_by_type(types_cbfs_compression,
compression, "(unknown)")
);
}
struct cbfs_file_attr_hash *hash = NULL;
while ((hash = cbfs_file_get_next_hash(entry, hash)) != NULL) {
unsigned int hash_type = ntohl(hash->hash_type);
if (hash_type > CBFS_NUM_SUPPORTED_HASHES) {
fprintf(fp, "invalid hash type %d\n", hash_type);
break;
}
size_t hash_len = widths_cbfs_hash[hash_type];
char *hash_str = bintohex(hash->hash_data, hash_len);
uint8_t local_hash[hash_len];
if (vb2_digest_buffer(CBFS_SUBHEADER(entry),
ntohl(entry->len), hash_type, local_hash,
hash_len) != VB2_SUCCESS) {
fprintf(fp, "failed to hash '%s'\n", name);
break;
}
int valid = memcmp(local_hash, hash->hash_data, hash_len) == 0;
const char *valid_str = valid ? "valid" : "invalid";
fprintf(fp, " hash %s:%s %s\n",
get_hash_attr_name(hash_type),
hash_str, valid_str);
free(hash_str);
}
if (!verbose)
return 0;
DEBUG(" cbfs_file=0x%x, offset=0x%x, content_address=0x%x+0x%x\n",
cbfs_get_entry_addr(image, entry), ntohl(entry->offset),
cbfs_get_entry_addr(image, entry) + ntohl(entry->offset),
ntohl(entry->len));
/* note the components of the subheader may be in host order ... */
switch (ntohl(entry->type)) {
case CBFS_COMPONENT_STAGE:
cbfs_print_stage_info((struct cbfs_stage *)
CBFS_SUBHEADER(entry), fp);
break;
case CBFS_COMPONENT_PAYLOAD:
payload = (struct cbfs_payload_segment *)
CBFS_SUBHEADER(entry);
while (payload) {
struct cbfs_payload_segment seg;
cbfs_decode_payload_segment(&seg, payload);
cbfs_print_decoded_payload_segment_info(
&seg, fp);
if (seg.type == PAYLOAD_SEGMENT_ENTRY)
break;
else
payload ++;
}
break;
default:
break;
}
return 0;
}
static int cbfs_print_parseable_entry_info(struct cbfs_image *image,
struct cbfs_file *entry, void *arg)
{
FILE *fp = (FILE *)arg;
const char *name;
const char *type;
size_t offset;
size_t metadata_size;
size_t data_size;
const char *sep = "\t";
if (!cbfs_is_valid_entry(image, entry)) {
ERROR("cbfs_print_entry_info: Invalid entry at 0x%x\n",
cbfs_get_entry_addr(image, entry));
return -1;
}
name = entry->filename;
if (*name == '\0')
name = "(empty)";
type = get_cbfs_entry_type_name(ntohl(entry->type)),
metadata_size = ntohl(entry->offset);
data_size = ntohl(entry->len);
offset = cbfs_get_entry_addr(image, entry);
fprintf(fp, "%s%s", name, sep);
fprintf(fp, "0x%zx%s", offset, sep);
fprintf(fp, "%s%s", type, sep);
fprintf(fp, "0x%zx%s", metadata_size, sep);
fprintf(fp, "0x%zx%s", data_size, sep);
fprintf(fp, "0x%zx\n", metadata_size + data_size);
return 0;
}
int cbfs_print_directory(struct cbfs_image *image)
{
if (cbfs_is_legacy_cbfs(image))
cbfs_print_header_info(image);
printf("%-30s %-10s %-12s Size\n", "Name", "Offset", "Type");
cbfs_walk(image, cbfs_print_entry_info, NULL);
return 0;
}
int cbfs_print_parseable_directory(struct cbfs_image *image)
{
size_t i;
const char *header[] = {
"Name",
"Offset",
"Type",
"Metadata Size",
"Data Size",
"Total Size",
};
const char *sep = "\t";
for (i = 0; i < ARRAY_SIZE(header) - 1; i++)
fprintf(stdout, "%s%s", header[i], sep);
fprintf(stdout, "%s\n", header[i]);
cbfs_walk(image, cbfs_print_parseable_entry_info, stdout);
return 0;
}
int cbfs_merge_empty_entry(struct cbfs_image *image, struct cbfs_file *entry,
unused void *arg)
{
struct cbfs_file *next;
uint8_t *name;
uint32_t type, addr, last_addr;
type = ntohl(entry->type);
if (type == CBFS_COMPONENT_DELETED) {
// Ready to be recycled.
type = CBFS_COMPONENT_NULL;
entry->type = htonl(type);
// Place NUL byte as first byte of name to be viewed as "empty".
name = (void *)&entry[1];
*name = '\0';
}
if (type != CBFS_COMPONENT_NULL)
return 0;
next = cbfs_find_next_entry(image, entry);
while (next && cbfs_is_valid_entry(image, next)) {
type = ntohl(next->type);
if (type == CBFS_COMPONENT_DELETED) {
type = CBFS_COMPONENT_NULL;
next->type = htonl(type);
}
if (type != CBFS_COMPONENT_NULL)
return 0;
addr = cbfs_get_entry_addr(image, entry);
last_addr = cbfs_get_entry_addr(
image, cbfs_find_next_entry(image, next));
// Now, we find two deleted/empty entries; try to merge now.
DEBUG("join_empty_entry: combine 0x%x+0x%x and 0x%x+0x%x.\n",
cbfs_get_entry_addr(image, entry), ntohl(entry->len),
cbfs_get_entry_addr(image, next), ntohl(next->len));
cbfs_create_empty_entry(entry, CBFS_COMPONENT_NULL,
(last_addr - addr -
cbfs_calculate_file_header_size("")),
"");
DEBUG("new empty entry: length=0x%x\n", ntohl(entry->len));
next = cbfs_find_next_entry(image, entry);
}
return 0;
}
int cbfs_walk(struct cbfs_image *image, cbfs_entry_callback callback,
void *arg)
{
int count = 0;
struct cbfs_file *entry;
for (entry = cbfs_find_first_entry(image);
entry && cbfs_is_valid_entry(image, entry);
entry = cbfs_find_next_entry(image, entry)) {
count ++;
if (callback(image, entry, arg) != 0)
break;
}
return count;
}
static int cbfs_header_valid(struct cbfs_header *header)
{
if ((ntohl(header->magic) == CBFS_HEADER_MAGIC) &&
((ntohl(header->version) == CBFS_HEADER_VERSION1) ||
(ntohl(header->version) == CBFS_HEADER_VERSION2)) &&
(ntohl(header->offset) < ntohl(header->romsize)))
return 1;
return 0;
}
struct cbfs_header *cbfs_find_header(char *data, size_t size,
uint32_t forced_offset)
{
size_t offset;
int found = 0;
int32_t rel_offset;
struct cbfs_header *header, *result = NULL;
if (forced_offset < (size - sizeof(struct cbfs_header))) {
/* Check if the forced header is valid. */
header = (struct cbfs_header *)(data + forced_offset);
if (cbfs_header_valid(header))
return header;
return NULL;
}
// Try finding relative offset of master header at end of file first.
rel_offset = *(int32_t *)(data + size - sizeof(int32_t));
offset = size + rel_offset;
DEBUG("relative offset: %#zx(-%#zx), offset: %#zx\n",
(size_t)rel_offset, (size_t)-rel_offset, offset);
if (offset >= size - sizeof(*header) ||
!cbfs_header_valid((struct cbfs_header *)(data + offset))) {
// Some use cases append non-CBFS data to the end of the ROM.
DEBUG("relative offset seems wrong, scanning whole image...\n");
offset = 0;
}
for (; offset + sizeof(*header) < size; offset++) {
header = (struct cbfs_header *)(data + offset);
if (!cbfs_header_valid(header))
continue;
if (!found++)
result = header;
}
if (found > 1)
// Top-aligned images usually have a working relative offset
// field, so this is more likely to happen on bottom-aligned
// ones (where the first header is the "outermost" one)
WARN("Multiple (%d) CBFS headers found, using the first one.\n",
found);
return result;
}
struct cbfs_file *cbfs_find_first_entry(struct cbfs_image *image)
{
assert(image);
if (image->has_header)
/* header.offset is relative to start of flash, not
* start of region, so use it with the full image.
*/
return (struct cbfs_file *)
(buffer_get_original_backing(&image->buffer) +
image->header.offset);
else
return (struct cbfs_file *)buffer_get(&image->buffer);
}
struct cbfs_file *cbfs_find_next_entry(struct cbfs_image *image,
struct cbfs_file *entry)
{
uint32_t addr = cbfs_get_entry_addr(image, entry);
int align = image->has_header ? image->header.align :
CBFS_ENTRY_ALIGNMENT;
assert(entry && cbfs_is_valid_entry(image, entry));
addr += ntohl(entry->offset) + ntohl(entry->len);
addr = align_up(addr, align);
return (struct cbfs_file *)(image->buffer.data + addr);
}
uint32_t cbfs_get_entry_addr(struct cbfs_image *image, struct cbfs_file *entry)
{
assert(image && image->buffer.data && entry);
return (int32_t)((char *)entry - image->buffer.data);
}
int cbfs_is_valid_cbfs(struct cbfs_image *image)
{
return buffer_check_magic(&image->buffer, CBFS_FILE_MAGIC,
strlen(CBFS_FILE_MAGIC));
}
int cbfs_is_legacy_cbfs(struct cbfs_image *image)
{
return image->has_header;
}
int cbfs_is_valid_entry(struct cbfs_image *image, struct cbfs_file *entry)
{
uint32_t offset = cbfs_get_entry_addr(image, entry);
if (offset >= image->buffer.size)
return 0;
struct buffer entry_data;
buffer_clone(&entry_data, &image->buffer);
buffer_seek(&entry_data, offset);
return buffer_check_magic(&entry_data, CBFS_FILE_MAGIC,
strlen(CBFS_FILE_MAGIC));
}
struct cbfs_file *cbfs_create_file_header(int type,
size_t len, const char *name)
{
struct cbfs_file *entry = malloc(MAX_CBFS_FILE_HEADER_BUFFER);
memset(entry, CBFS_CONTENT_DEFAULT_VALUE, MAX_CBFS_FILE_HEADER_BUFFER);
memcpy(entry->magic, CBFS_FILE_MAGIC, sizeof(entry->magic));
entry->type = htonl(type);
entry->len = htonl(len);
entry->attributes_offset = 0;
entry->offset = htonl(cbfs_calculate_file_header_size(name));
memset(entry->filename, 0, ntohl(entry->offset) - sizeof(*entry));
strcpy(entry->filename, name);
return entry;
}
int cbfs_create_empty_entry(struct cbfs_file *entry, int type,
size_t len, const char *name)
{
struct cbfs_file *tmp = cbfs_create_file_header(type, len, name);
memcpy(entry, tmp, ntohl(tmp->offset));
free(tmp);
memset(CBFS_SUBHEADER(entry), CBFS_CONTENT_DEFAULT_VALUE, len);
return 0;
}
struct cbfs_file_attribute *cbfs_file_first_attr(struct cbfs_file *file)
{
/* attributes_offset should be 0 when there is no attribute, but all
* values that point into the cbfs_file header are invalid, too. */
if (ntohl(file->attributes_offset) <= sizeof(*file))
return NULL;
/* There needs to be enough space for the file header and one
* attribute header for this to make sense. */
if (ntohl(file->offset) <=
sizeof(*file) + sizeof(struct cbfs_file_attribute))
return NULL;
return (struct cbfs_file_attribute *)
(((uint8_t *)file) + ntohl(file->attributes_offset));
}
struct cbfs_file_attribute *cbfs_file_next_attr(struct cbfs_file *file,
struct cbfs_file_attribute *attr)
{
/* ex falso sequitur quodlibet */
if (attr == NULL)
return NULL;
/* Is there enough space for another attribute? */
if ((uint8_t *)attr + ntohl(attr->len) +
sizeof(struct cbfs_file_attribute) >=
(uint8_t *)file + ntohl(file->offset))
return NULL;
struct cbfs_file_attribute *next = (struct cbfs_file_attribute *)
(((uint8_t *)attr) + ntohl(attr->len));
/* If any, "unused" attributes must come last. */
if (ntohl(next->tag) == CBFS_FILE_ATTR_TAG_UNUSED)
return NULL;
if (ntohl(next->tag) == CBFS_FILE_ATTR_TAG_UNUSED2)
return NULL;
return next;
}
struct cbfs_file_attribute *cbfs_add_file_attr(struct cbfs_file *header,
uint32_t tag,
uint32_t size)
{
struct cbfs_file_attribute *attr, *next;
next = cbfs_file_first_attr(header);
do {
attr = next;
next = cbfs_file_next_attr(header, attr);
} while (next != NULL);
uint32_t header_size = ntohl(header->offset) + size;
if (header_size > MAX_CBFS_FILE_HEADER_BUFFER) {
DEBUG("exceeding allocated space for cbfs_file headers");
return NULL;
}
/* attr points to the last valid attribute now.
* If NULL, we have to create the first one. */
if (attr == NULL) {
/* New attributes start where the header ends.
* header->offset is later set to accomodate the
* additional structure.
* No endianess translation necessary here, because both
* fields are encoded the same way. */
header->attributes_offset = header->offset;
attr = (struct cbfs_file_attribute *)
(((uint8_t *)header) +
ntohl(header->attributes_offset));
} else {
attr = (struct cbfs_file_attribute *)
(((uint8_t *)attr) +
ntohl(attr->len));
}
header->offset = htonl(header_size);
memset(attr, CBFS_CONTENT_DEFAULT_VALUE, size);
attr->tag = htonl(tag);
attr->len = htonl(size);
return attr;
}
int cbfs_add_file_hash(struct cbfs_file *header, struct buffer *buffer,
enum vb2_hash_algorithm hash_type)
{
uint32_t hash_index = hash_type;
if (hash_index >= CBFS_NUM_SUPPORTED_HASHES)
return -1;
unsigned hash_size = widths_cbfs_hash[hash_type];
if (hash_size == 0)
return -1;
struct cbfs_file_attr_hash *attrs =
(struct cbfs_file_attr_hash *)cbfs_add_file_attr(header,
CBFS_FILE_ATTR_TAG_HASH,
sizeof(struct cbfs_file_attr_hash) + hash_size);
if (attrs == NULL)
return -1;
attrs->hash_type = htonl(hash_type);
if (vb2_digest_buffer(buffer_get(buffer), buffer_size(buffer),
hash_type, attrs->hash_data, hash_size) != VB2_SUCCESS)
return -1;
return 0;
}
/* Finds a place to hold whole data in same memory page. */
static int is_in_same_page(uint32_t start, uint32_t size, uint32_t page)
{
if (!page)
return 1;
return (start / page) == (start + size - 1) / page;
}
/* Tests if data can fit in a range by given offset:
* start ->| metadata_size | offset (+ size) |<- end
*/
static int is_in_range(size_t start, size_t end, size_t metadata_size,
size_t offset, size_t size)
{
return (offset >= start + metadata_size && offset + size <= end);
}
static size_t absolute_align(const struct cbfs_image *image, size_t val,
size_t align)
{
const size_t region_offset = buffer_offset(&image->buffer);
/* To perform alignment on absolute address, take the region offset */
/* of the image into account. */
return align_up(val + region_offset, align) - region_offset;
}
int32_t cbfs_locate_entry(struct cbfs_image *image, size_t size,
size_t page_size, size_t align, size_t metadata_size)
{
struct cbfs_file *entry;
size_t need_len;
size_t addr, addr_next, addr2, addr3, offset;
/* Default values: allow fitting anywhere in ROM. */
if (!page_size)
page_size = image->has_header ? image->header.romsize :
image->buffer.size;
if (!align)
align = 1;
if (size > page_size)
ERROR("Input file size (%zd) greater than page size (%zd).\n",
size, page_size);
size_t image_align = image->has_header ? image->header.align :
CBFS_ENTRY_ALIGNMENT;
if (page_size % image_align)
WARN("%s: Page size (%#zx) not aligned with CBFS image (%#zx).\n",
__func__, page_size, image_align);
need_len = metadata_size + size;
// Merge empty entries to build get max available space.
cbfs_walk(image, cbfs_merge_empty_entry, NULL);
/* Three cases of content location on memory page:
* case 1.
* | PAGE 1 | PAGE 2 |
* | <header><content>| Fit. Return start of content.
*
* case 2.
* | PAGE 1 | PAGE 2 |
* | <header><content> | Fits when we shift content to align
* shift-> | <header>|<content> | at starting of PAGE 2.
*
* case 3. (large content filling whole page)
* | PAGE 1 | PAGE 2 | PAGE 3 |
* | <header>< content > | Can't fit. If we shift content to
* |trial-> <header>< content > | PAGE 2, header can't fit in free
* | shift-> <header><content> space, so we must use PAGE 3.
*
* The returned address can be then used as "base-address" (-b) in add-*
* commands (will be re-calculated and positioned by cbfs_add_entry_at).
* For stage targets, the address is also used to re-link stage before
* being added into CBFS.
*/
for (entry = cbfs_find_first_entry(image);
entry && cbfs_is_valid_entry(image, entry);
entry = cbfs_find_next_entry(image, entry)) {
uint32_t type = ntohl(entry->type);
if (type != CBFS_COMPONENT_NULL)
continue;
addr = cbfs_get_entry_addr(image, entry);
addr_next = cbfs_get_entry_addr(image, cbfs_find_next_entry(
image, entry));
if (addr_next - addr < need_len)
continue;
offset = absolute_align(image, addr + metadata_size, align);
if (is_in_same_page(offset, size, page_size) &&
is_in_range(addr, addr_next, metadata_size, offset, size)) {
DEBUG("cbfs_locate_entry: FIT (PAGE1).");
return offset;
}
addr2 = align_up(addr, page_size);
offset = absolute_align(image, addr2, align);
if (is_in_range(addr, addr_next, metadata_size, offset, size)) {
DEBUG("cbfs_locate_entry: OVERLAP (PAGE2).");
return offset;
}
/* Assume page_size >= metadata_size so adding one page will
* definitely provide the space for header. */
assert(page_size >= metadata_size);
addr3 = addr2 + page_size;
offset = absolute_align(image, addr3, align);
if (is_in_range(addr, addr_next, metadata_size, offset, size)) {
DEBUG("cbfs_locate_entry: OVERLAP+ (PAGE3).");
return offset;
}
}
return -1;
}