mirror of
https://github.com/Telecominfraproject/OpenCellular.git
synced 2025-11-24 02:05:01 +00:00
LoadKernel already returns the partition number for the selected kernel. This change makes it also return the GPT Entry's UniqueGuid, which will eventually be passed to the kernel itself, so the kernel can determine which of several possible devices it has booted from. It doesn't know for certain because the BIOS and the kernel may enumerate the devices in a different order. BUG=chromium-os:4984 Review URL: http://codereview.chromium.org/3056014
355 lines
11 KiB
C
355 lines
11 KiB
C
/* Copyright (c) 2010 The Chromium OS Authors. All rights reserved.
|
|
* Use of this source code is governed by a BSD-style license that can be
|
|
* found in the LICENSE file.
|
|
*/
|
|
|
|
#include "cgptlib.h"
|
|
#include "cgptlib_internal.h"
|
|
#include "crc32.h"
|
|
#include "gpt.h"
|
|
#include "utility.h"
|
|
|
|
|
|
int CheckParameters(GptData *gpt) {
|
|
/* Currently, we only support 512-byte sector. In the future, we may support
|
|
* larger sector. */
|
|
if (gpt->sector_bytes != 512)
|
|
return GPT_ERROR_INVALID_SECTOR_SIZE;
|
|
|
|
/* The sector number of a drive should be reasonable. If the given value is
|
|
* too small to contain basic GPT structure (PMBR + Headers + Entries),
|
|
* the value is wrong. */
|
|
if (gpt->drive_sectors < (1 + 2 * (1 + GPT_ENTRIES_SECTORS)))
|
|
return GPT_ERROR_INVALID_SECTOR_NUMBER;
|
|
|
|
return GPT_SUCCESS;
|
|
}
|
|
|
|
|
|
uint32_t HeaderCrc(GptHeader* h) {
|
|
uint32_t crc32, original_crc32;
|
|
|
|
/* Original CRC is calculated with the CRC field 0. */
|
|
original_crc32 = h->header_crc32;
|
|
h->header_crc32 = 0;
|
|
crc32 = Crc32((const uint8_t *)h, h->size);
|
|
h->header_crc32 = original_crc32;
|
|
|
|
return crc32;
|
|
}
|
|
|
|
|
|
int CheckHeader(GptHeader *h, int is_secondary, uint64_t drive_sectors) {
|
|
if (!h)
|
|
return 1;
|
|
|
|
/* Make sure we're looking at a header of reasonable size before
|
|
* attempting to calculate CRC. */
|
|
if (Memcmp(h->signature, GPT_HEADER_SIGNATURE, GPT_HEADER_SIGNATURE_SIZE))
|
|
return 1;
|
|
if (h->revision != GPT_HEADER_REVISION)
|
|
return 1;
|
|
if (h->size < MIN_SIZE_OF_HEADER || h->size > MAX_SIZE_OF_HEADER)
|
|
return 1;
|
|
|
|
/* Check CRC before looking at remaining fields */
|
|
if (HeaderCrc(h) != h->header_crc32)
|
|
return 1;
|
|
|
|
/* Reserved fields must be zero. */
|
|
if (h->reserved_zero)
|
|
return 1;
|
|
|
|
/* Could check that padding is zero, but that doesn't matter to us. */
|
|
|
|
/* If entry size is different than our struct, we won't be able to
|
|
* parse it. Technically, any size 2^N where N>=7 is valid. */
|
|
if (h->size_of_entry != sizeof(GptEntry))
|
|
return 1;
|
|
if ((h->number_of_entries < MIN_NUMBER_OF_ENTRIES) ||
|
|
(h->number_of_entries > MAX_NUMBER_OF_ENTRIES) ||
|
|
(h->number_of_entries * h->size_of_entry != TOTAL_ENTRIES_SIZE))
|
|
return 1;
|
|
|
|
/* Check locations for the header and its entries. The primary
|
|
* immediately follows the PMBR, and is followed by its entries.
|
|
* The secondary is at the end of the drive, preceded by its
|
|
* entries. */
|
|
if (is_secondary) {
|
|
if (h->my_lba != drive_sectors - 1)
|
|
return 1;
|
|
if (h->entries_lba != h->my_lba - GPT_ENTRIES_SECTORS)
|
|
return 1;
|
|
} else {
|
|
if (h->my_lba != 1)
|
|
return 1;
|
|
if (h->entries_lba != h->my_lba + 1)
|
|
return 1;
|
|
}
|
|
|
|
/* FirstUsableLBA must be after the end of the primary GPT table
|
|
* array. LastUsableLBA must be before the start of the secondary
|
|
* GPT table array. FirstUsableLBA <= LastUsableLBA. */
|
|
if (h->first_usable_lba < 2 + GPT_ENTRIES_SECTORS)
|
|
return 1;
|
|
if (h->last_usable_lba >= drive_sectors - 1 - GPT_ENTRIES_SECTORS)
|
|
return 1;
|
|
if (h->first_usable_lba > h->last_usable_lba)
|
|
return 1;
|
|
|
|
/* Success */
|
|
return 0;
|
|
}
|
|
|
|
|
|
/* Return non-zero if the entry is unused, 0 if it is used. */
|
|
int IsUnusedEntry(const GptEntry* e) {
|
|
static Guid zero = {{{0, 0, 0, 0, 0, {0, 0, 0, 0, 0, 0}}}};
|
|
return !Memcmp(&zero, (const uint8_t*)(&e->type), sizeof(zero));
|
|
}
|
|
|
|
/* Returns non-zero if the entry is a Chrome OS kernel partition, else 0. */
|
|
int IsKernelEntry(const GptEntry* e) {
|
|
static Guid chromeos_kernel = GPT_ENT_TYPE_CHROMEOS_KERNEL;
|
|
return !Memcmp(&e->type, &chromeos_kernel, sizeof(Guid));
|
|
}
|
|
|
|
|
|
int CheckEntries(GptEntry* entries, GptHeader* h) {
|
|
|
|
GptEntry* entry;
|
|
uint32_t crc32;
|
|
uint32_t i;
|
|
|
|
/* Check CRC before examining entries. */
|
|
crc32 = Crc32((const uint8_t *)entries,
|
|
h->size_of_entry * h->number_of_entries);
|
|
if (crc32 != h->entries_crc32)
|
|
return 1;
|
|
|
|
/* Check all entries. */
|
|
for (i = 0, entry = entries; i < h->number_of_entries; i++, entry++) {
|
|
GptEntry* e2;
|
|
uint32_t i2;
|
|
|
|
if (IsUnusedEntry(entry))
|
|
continue;
|
|
|
|
/* Entry must be in valid region. */
|
|
if ((entry->starting_lba < h->first_usable_lba) ||
|
|
(entry->ending_lba > h->last_usable_lba) ||
|
|
(entry->ending_lba < entry->starting_lba))
|
|
return 1;
|
|
|
|
/* Entry must not overlap other entries. */
|
|
for (i2 = 0, e2 = entries; i2 < h->number_of_entries; i2++, e2++) {
|
|
if (i2 == i || IsUnusedEntry(e2))
|
|
continue;
|
|
|
|
if ((entry->starting_lba >= e2->starting_lba) &&
|
|
(entry->starting_lba <= e2->ending_lba))
|
|
return 1;
|
|
if ((entry->ending_lba >= e2->starting_lba) &&
|
|
(entry->ending_lba <= e2->ending_lba))
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
/* Success */
|
|
return 0;
|
|
}
|
|
|
|
|
|
/* Returns 0 if the GptHeaders are the same for all fields which don't
|
|
* differ between the primary and secondary headers - that is, all
|
|
* fields other than:
|
|
*
|
|
* my_lba
|
|
* alternate_lba
|
|
* entries_lba */
|
|
int HeaderFieldsSame(GptHeader *h1, GptHeader *h2) {
|
|
if (Memcmp(h1->signature, h2->signature, sizeof(h1->signature)))
|
|
return 1;
|
|
if (h1->revision != h2->revision)
|
|
return 1;
|
|
if (h1->size != h2->size)
|
|
return 1;
|
|
if (h1->reserved_zero != h2->reserved_zero)
|
|
return 1;
|
|
if (h1->first_usable_lba != h2->first_usable_lba)
|
|
return 1;
|
|
if (h1->last_usable_lba != h2->last_usable_lba)
|
|
return 1;
|
|
if (Memcmp(&h1->disk_uuid, &h2->disk_uuid, sizeof(Guid)))
|
|
return 1;
|
|
if (h1->number_of_entries != h2->number_of_entries)
|
|
return 1;
|
|
if (h1->size_of_entry != h2->size_of_entry)
|
|
return 1;
|
|
if (h1->entries_crc32 != h2->entries_crc32)
|
|
return 1;
|
|
|
|
return 0;
|
|
}
|
|
|
|
|
|
int GptSanityCheck(GptData *gpt) {
|
|
int retval;
|
|
GptHeader* header1 = (GptHeader*)(gpt->primary_header);
|
|
GptHeader* header2 = (GptHeader*)(gpt->secondary_header);
|
|
GptEntry* entries1 = (GptEntry*)(gpt->primary_entries);
|
|
GptEntry* entries2 = (GptEntry*)(gpt->secondary_entries);
|
|
GptHeader* goodhdr = NULL;
|
|
|
|
gpt->valid_headers = 0;
|
|
gpt->valid_entries = 0;
|
|
|
|
retval = CheckParameters(gpt);
|
|
if (retval != GPT_SUCCESS)
|
|
return retval;
|
|
|
|
/* Check both headers; we need at least one valid header. */
|
|
if (0 == CheckHeader(header1, 0, gpt->drive_sectors)) {
|
|
gpt->valid_headers |= MASK_PRIMARY;
|
|
goodhdr = header1;
|
|
}
|
|
if (0 == CheckHeader(header2, 1, gpt->drive_sectors)) {
|
|
gpt->valid_headers |= MASK_SECONDARY;
|
|
if (!goodhdr)
|
|
goodhdr = header2;
|
|
}
|
|
|
|
if (!gpt->valid_headers)
|
|
return GPT_ERROR_INVALID_HEADERS;
|
|
|
|
/* Checks if entries are valid.
|
|
*
|
|
* Note that we use the same header in both checks. This way we'll
|
|
* catch the case where (header1,entries1) and (header2,entries2)
|
|
* are both valid, but (entries1 != entries2). */
|
|
if (0 == CheckEntries(entries1, goodhdr))
|
|
gpt->valid_entries |= MASK_PRIMARY;
|
|
if (0 == CheckEntries(entries2, goodhdr))
|
|
gpt->valid_entries |= MASK_SECONDARY;
|
|
|
|
/* If both headers are good but neither entries were good, check the
|
|
* entries with the secondary header. */
|
|
if (MASK_BOTH == gpt->valid_headers && !gpt->valid_entries) {
|
|
if (0 == CheckEntries(entries1, header2))
|
|
gpt->valid_entries |= MASK_PRIMARY;
|
|
if (0 == CheckEntries(entries2, header2))
|
|
gpt->valid_entries |= MASK_SECONDARY;
|
|
if (gpt->valid_entries) {
|
|
/* Sure enough, header2 had a good CRC for one of the entries. Mark
|
|
* header1 invalid, so we'll update its entries CRC. */
|
|
gpt->valid_headers &= ~MASK_PRIMARY;
|
|
goodhdr = header2;
|
|
}
|
|
}
|
|
|
|
if (!gpt->valid_entries)
|
|
return GPT_ERROR_INVALID_ENTRIES;
|
|
|
|
/* Now that we've determined which header contains a good CRC for
|
|
* the entries, make sure the headers are otherwise identical. */
|
|
if (MASK_BOTH == gpt->valid_headers &&
|
|
0 != HeaderFieldsSame(header1, header2))
|
|
gpt->valid_headers &= ~MASK_SECONDARY;
|
|
|
|
return GPT_SUCCESS;
|
|
}
|
|
|
|
|
|
void GptRepair(GptData *gpt) {
|
|
GptHeader* header1 = (GptHeader*)(gpt->primary_header);
|
|
GptHeader* header2 = (GptHeader*)(gpt->secondary_header);
|
|
GptEntry* entries1 = (GptEntry*)(gpt->primary_entries);
|
|
GptEntry* entries2 = (GptEntry*)(gpt->secondary_entries);
|
|
int entries_size;
|
|
|
|
/* Need at least one good header and one good set of entries. */
|
|
if (MASK_NONE == gpt->valid_headers || MASK_NONE == gpt->valid_entries)
|
|
return;
|
|
|
|
/* Repair headers if necessary */
|
|
if (MASK_PRIMARY == gpt->valid_headers) {
|
|
/* Primary is good, secondary is bad */
|
|
Memcpy(header2, header1, sizeof(GptHeader));
|
|
header2->my_lba = gpt->drive_sectors - 1;
|
|
header2->alternate_lba = 1;
|
|
header2->entries_lba = header2->my_lba - GPT_ENTRIES_SECTORS;
|
|
header2->header_crc32 = HeaderCrc(header2);
|
|
gpt->modified |= GPT_MODIFIED_HEADER2;
|
|
}
|
|
else if (MASK_SECONDARY == gpt->valid_headers) {
|
|
/* Secondary is good, primary is bad */
|
|
Memcpy(header1, header2, sizeof(GptHeader));
|
|
header1->my_lba = 1;
|
|
header1->alternate_lba = gpt->drive_sectors - 1;
|
|
header1->entries_lba = header1->my_lba + 1;
|
|
header1->header_crc32 = HeaderCrc(header1);
|
|
gpt->modified |= GPT_MODIFIED_HEADER1;
|
|
}
|
|
gpt->valid_headers = MASK_BOTH;
|
|
|
|
/* Repair entries if necessary */
|
|
entries_size = header1->size_of_entry * header1->number_of_entries;
|
|
if (MASK_PRIMARY == gpt->valid_entries) {
|
|
/* Primary is good, secondary is bad */
|
|
Memcpy(entries2, entries1, entries_size);
|
|
gpt->modified |= GPT_MODIFIED_ENTRIES2;
|
|
}
|
|
else if (MASK_SECONDARY == gpt->valid_entries) {
|
|
/* Secondary is good, primary is bad */
|
|
Memcpy(entries1, entries2, entries_size);
|
|
gpt->modified |= GPT_MODIFIED_ENTRIES1;
|
|
}
|
|
gpt->valid_entries = MASK_BOTH;
|
|
}
|
|
|
|
|
|
int GetEntrySuccessful(const GptEntry* e) {
|
|
return (e->attrs.fields.gpt_att & CGPT_ATTRIBUTE_SUCCESSFUL_MASK) >>
|
|
CGPT_ATTRIBUTE_SUCCESSFUL_OFFSET;
|
|
}
|
|
|
|
|
|
int GetEntryPriority(const GptEntry* e) {
|
|
return (e->attrs.fields.gpt_att & CGPT_ATTRIBUTE_PRIORITY_MASK) >>
|
|
CGPT_ATTRIBUTE_PRIORITY_OFFSET;
|
|
}
|
|
|
|
|
|
int GetEntryTries(const GptEntry* e) {
|
|
return (e->attrs.fields.gpt_att & CGPT_ATTRIBUTE_TRIES_MASK) >>
|
|
CGPT_ATTRIBUTE_TRIES_OFFSET;
|
|
}
|
|
|
|
|
|
void SetEntrySuccessful(GptEntry* e, int successful) {
|
|
if (successful)
|
|
e->attrs.fields.gpt_att |= CGPT_ATTRIBUTE_SUCCESSFUL_MASK;
|
|
else
|
|
e->attrs.fields.gpt_att &= ~CGPT_ATTRIBUTE_SUCCESSFUL_MASK;
|
|
}
|
|
|
|
|
|
void SetEntryPriority(GptEntry* e, int priority) {
|
|
e->attrs.fields.gpt_att &= ~CGPT_ATTRIBUTE_PRIORITY_MASK;
|
|
e->attrs.fields.gpt_att |= (priority << CGPT_ATTRIBUTE_PRIORITY_OFFSET) &
|
|
CGPT_ATTRIBUTE_PRIORITY_MASK;
|
|
}
|
|
|
|
|
|
void SetEntryTries(GptEntry* e, int tries) {
|
|
e->attrs.fields.gpt_att &= ~CGPT_ATTRIBUTE_TRIES_MASK;
|
|
e->attrs.fields.gpt_att |= (tries << CGPT_ATTRIBUTE_TRIES_OFFSET) &
|
|
CGPT_ATTRIBUTE_TRIES_MASK;
|
|
}
|
|
|
|
void GetCurrentKernelUniqueGuid(GptData *gpt, void *dest) {
|
|
GptEntry* entries = (GptEntry*)gpt->primary_entries;
|
|
GptEntry* e = entries + gpt->current_kernel;
|
|
Memcpy(dest, &e->unique, sizeof(Guid));
|
|
}
|