| /* 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. |
| * |
| * Utility for ChromeOS-specific GPT partitions, Please see corresponding .c |
| * files for more details. |
| */ |
| |
| #include "cgpt.h" |
| |
| #include <errno.h> |
| #include <fcntl.h> |
| #include <getopt.h> |
| #include <stdint.h> |
| #include <stdio.h> |
| #include <stdlib.h> |
| #include <string.h> |
| #include <sys/ioctl.h> |
| #include <sys/mount.h> |
| #include <sys/stat.h> |
| #include <sys/types.h> |
| #include <unistd.h> |
| #include <assert.h> |
| #include <stdarg.h> |
| |
| #include "cgptlib_internal.h" |
| #include "crc32.h" |
| |
| |
| void Error(const char *format, ...) { |
| va_list ap; |
| va_start(ap, format); |
| fprintf(stderr, "ERROR: %s %s: ", progname, command); |
| vfprintf(stderr, format, ap); |
| va_end(ap); |
| } |
| |
| |
| int CheckValid(const struct drive *drive) { |
| if ((drive->gpt.valid_headers != MASK_BOTH) || |
| (drive->gpt.valid_entries != MASK_BOTH)) { |
| fprintf(stderr, "\nWARNING: one of the GPT header/entries is invalid, " |
| "please run '%s repair'\n", progname); |
| return CGPT_FAILED; |
| } |
| return CGPT_OK; |
| } |
| |
| /* Loads sectors from 'fd'. |
| * *buf is pointed to an allocated memory when returned, and should be |
| * freed by cgpt_close(). |
| * |
| * fd -- file descriptot. |
| * buf -- pointer to buffer pointer |
| * sector -- offset of starting sector (in sectors) |
| * sector_bytes -- bytes per sector |
| * sector_count -- number of sectors to load |
| * |
| * Returns CGPT_OK for successful. Aborts if any error occurs. |
| */ |
| static int Load(const int fd, uint8_t **buf, |
| const uint64_t sector, |
| const uint64_t sector_bytes, |
| const uint64_t sector_count) { |
| int count; /* byte count to read */ |
| int nread; |
| |
| assert(buf); |
| count = sector_bytes * sector_count; |
| *buf = malloc(count); |
| assert(*buf); |
| |
| if (-1 == lseek(fd, sector * sector_bytes, SEEK_SET)) |
| goto error_free; |
| |
| nread = read(fd, *buf, count); |
| if (nread < count) |
| goto error_free; |
| |
| return CGPT_OK; |
| |
| error_free: |
| free(*buf); |
| *buf = 0; |
| return CGPT_FAILED; |
| } |
| |
| |
| int ReadPMBR(struct drive *drive) { |
| if (-1 == lseek(drive->fd, 0, SEEK_SET)) |
| return CGPT_FAILED; |
| |
| int nread = read(drive->fd, &drive->pmbr, sizeof(struct pmbr)); |
| if (nread != sizeof(struct pmbr)) |
| return CGPT_FAILED; |
| |
| return CGPT_OK; |
| } |
| |
| int WritePMBR(struct drive *drive) { |
| if (-1 == lseek(drive->fd, 0, SEEK_SET)) |
| return CGPT_FAILED; |
| |
| int nwrote = write(drive->fd, &drive->pmbr, sizeof(struct pmbr)); |
| if (nwrote != sizeof(struct pmbr)) |
| return CGPT_FAILED; |
| |
| return CGPT_OK; |
| } |
| |
| /* Saves sectors to 'fd'. |
| * |
| * fd -- file descriptot. |
| * buf -- pointer to buffer |
| * sector -- starting sector offset |
| * sector_bytes -- bytes per sector |
| * sector_count -- number of sector to save |
| * |
| * Returns CGPT_OK for successful, CGPT_FAILED for failed. |
| */ |
| static int Save(const int fd, const uint8_t *buf, |
| const uint64_t sector, |
| const uint64_t sector_bytes, |
| const uint64_t sector_count) { |
| int count; /* byte count to write */ |
| int nwrote; |
| |
| assert(buf); |
| count = sector_bytes * sector_count; |
| |
| if (-1 == lseek(fd, sector * sector_bytes, SEEK_SET)) |
| return CGPT_FAILED; |
| |
| nwrote = write(fd, buf, count); |
| if (nwrote < count) |
| return CGPT_FAILED; |
| |
| return CGPT_OK; |
| } |
| |
| |
| // Opens a block device or file, loads raw GPT data from it. |
| // |
| // Returns CGPT_FAILED if any error happens. |
| // Returns CGPT_OK if success and information are stored in 'drive'. */ |
| int DriveOpen(const char *drive_path, struct drive *drive) { |
| struct stat stat; |
| |
| assert(drive_path); |
| assert(drive); |
| |
| // Clear struct for proper error handling. |
| memset(drive, 0, sizeof(struct drive)); |
| |
| drive->fd = open(drive_path, O_RDWR | O_LARGEFILE); |
| if (drive->fd == -1) { |
| Error("Can't open %s: %s\n", drive_path, strerror(errno)); |
| return CGPT_FAILED; |
| } |
| |
| if (fstat(drive->fd, &stat) == -1) { |
| goto error_close; |
| } |
| if ((stat.st_mode & S_IFMT) != S_IFREG) { |
| if (ioctl(drive->fd, BLKGETSIZE64, &drive->size) < 0) { |
| Error("Can't read drive size from %s: %s\n", drive_path, strerror(errno)); |
| goto error_close; |
| } |
| if (ioctl(drive->fd, BLKSSZGET, &drive->gpt.sector_bytes) < 0) { |
| Error("Can't read sector size from %s: %s\n", |
| drive_path, strerror(errno)); |
| goto error_close; |
| } |
| } else { |
| drive->gpt.sector_bytes = 512; /* bytes */ |
| drive->size = stat.st_size; |
| } |
| if (drive->size % drive->gpt.sector_bytes) { |
| Error("Media size (%llu) is not a multiple of sector size(%d)\n", |
| (long long unsigned int)drive->size, drive->gpt.sector_bytes); |
| goto error_close; |
| } |
| drive->gpt.drive_sectors = drive->size / drive->gpt.sector_bytes; |
| |
| // Read the data. |
| if (CGPT_OK != Load(drive->fd, &drive->gpt.primary_header, |
| GPT_PMBR_SECTOR, |
| drive->gpt.sector_bytes, GPT_HEADER_SECTOR)) { |
| goto error_close; |
| } |
| if (CGPT_OK != Load(drive->fd, &drive->gpt.secondary_header, |
| drive->gpt.drive_sectors - GPT_PMBR_SECTOR, |
| drive->gpt.sector_bytes, GPT_HEADER_SECTOR)) { |
| goto error_close; |
| } |
| if (CGPT_OK != Load(drive->fd, &drive->gpt.primary_entries, |
| GPT_PMBR_SECTOR + GPT_HEADER_SECTOR, |
| drive->gpt.sector_bytes, GPT_ENTRIES_SECTORS)) { |
| goto error_close; |
| } |
| if (CGPT_OK != Load(drive->fd, &drive->gpt.secondary_entries, |
| drive->gpt.drive_sectors - GPT_HEADER_SECTOR |
| - GPT_ENTRIES_SECTORS, |
| drive->gpt.sector_bytes, GPT_ENTRIES_SECTORS)) { |
| goto error_close; |
| } |
| |
| // We just load the data. Caller must validate it. |
| return CGPT_OK; |
| |
| error_close: |
| (void) DriveClose(drive, 0); |
| return CGPT_FAILED; |
| } |
| |
| |
| int DriveClose(struct drive *drive, int update_as_needed) { |
| int errors = 0; |
| |
| if (update_as_needed) { |
| if (drive->gpt.modified & GPT_MODIFIED_HEADER1) { |
| if (CGPT_OK != Save(drive->fd, drive->gpt.primary_header, |
| GPT_PMBR_SECTOR, |
| drive->gpt.sector_bytes, GPT_HEADER_SECTOR)) { |
| errors++; |
| Error("Cannot write primary header: %s\n", strerror(errno)); |
| } |
| } |
| |
| if (drive->gpt.modified & GPT_MODIFIED_HEADER2) { |
| if(CGPT_OK != Save(drive->fd, drive->gpt.secondary_header, |
| drive->gpt.drive_sectors - GPT_PMBR_SECTOR, |
| drive->gpt.sector_bytes, GPT_HEADER_SECTOR)) { |
| errors++; |
| Error("Cannot write secondary header: %s\n", strerror(errno)); |
| } |
| } |
| if (drive->gpt.modified & GPT_MODIFIED_ENTRIES1) { |
| if (CGPT_OK != Save(drive->fd, drive->gpt.primary_entries, |
| GPT_PMBR_SECTOR + GPT_HEADER_SECTOR, |
| drive->gpt.sector_bytes, GPT_ENTRIES_SECTORS)) { |
| errors++; |
| Error("Cannot write primary entries: %s\n", strerror(errno)); |
| } |
| } |
| if (drive->gpt.modified & GPT_MODIFIED_ENTRIES2) { |
| if (CGPT_OK != Save(drive->fd, drive->gpt.secondary_entries, |
| drive->gpt.drive_sectors - GPT_HEADER_SECTOR |
| - GPT_ENTRIES_SECTORS, |
| drive->gpt.sector_bytes, GPT_ENTRIES_SECTORS)) { |
| errors++; |
| Error("Cannot write secondary entries: %s\n", strerror(errno)); |
| } |
| } |
| } |
| |
| close(drive->fd); |
| |
| if (drive->gpt.primary_header) |
| free(drive->gpt.primary_header); |
| drive->gpt.primary_header = 0; |
| if (drive->gpt.primary_entries) |
| free(drive->gpt.primary_entries); |
| drive->gpt.primary_entries = 0; |
| if (drive->gpt.secondary_header) |
| free(drive->gpt.secondary_header); |
| drive->gpt.secondary_header = 0; |
| if (drive->gpt.secondary_entries) |
| free(drive->gpt.secondary_entries); |
| drive->gpt.secondary_entries = 0; |
| |
| return errors ? CGPT_FAILED : CGPT_OK; |
| } |
| |
| |
| |
| /* GUID conversion functions. Accepted format: |
| * |
| * "C12A7328-F81F-11D2-BA4B-00A0C93EC93B" |
| * |
| * Returns CGPT_OK if parsing is successful; otherwise CGPT_FAILED. |
| */ |
| int StrToGuid(const char *str, Guid *guid) { |
| uint32_t time_low; |
| uint16_t time_mid; |
| uint16_t time_high_and_version; |
| unsigned int chunk[11]; |
| |
| if (11 != sscanf(str, "%08X-%04X-%04X-%02X%02X-%02X%02X%02X%02X%02X%02X", |
| chunk+0, |
| chunk+1, |
| chunk+2, |
| chunk+3, |
| chunk+4, |
| chunk+5, |
| chunk+6, |
| chunk+7, |
| chunk+8, |
| chunk+9, |
| chunk+10)) { |
| printf("FAILED\n"); |
| return CGPT_FAILED; |
| } |
| |
| time_low = chunk[0] & 0xffffffff; |
| time_mid = chunk[1] & 0xffff; |
| time_high_and_version = chunk[2] & 0xffff; |
| |
| guid->u.Uuid.time_low = htole32(time_low); |
| guid->u.Uuid.time_mid = htole16(time_mid); |
| guid->u.Uuid.time_high_and_version = htole16(time_high_and_version); |
| |
| guid->u.Uuid.clock_seq_high_and_reserved = chunk[3] & 0xff; |
| guid->u.Uuid.clock_seq_low = chunk[4] & 0xff; |
| guid->u.Uuid.node[0] = chunk[5] & 0xff; |
| guid->u.Uuid.node[1] = chunk[6] & 0xff; |
| guid->u.Uuid.node[2] = chunk[7] & 0xff; |
| guid->u.Uuid.node[3] = chunk[8] & 0xff; |
| guid->u.Uuid.node[4] = chunk[9] & 0xff; |
| guid->u.Uuid.node[5] = chunk[10] & 0xff; |
| |
| return CGPT_OK; |
| } |
| void GuidToStr(const Guid *guid, char *str) { |
| sprintf(str, "%08X-%04X-%04X-%02X%02X-%02X%02X%02X%02X%02X%02X", |
| le32toh(guid->u.Uuid.time_low), le16toh(guid->u.Uuid.time_mid), |
| le16toh(guid->u.Uuid.time_high_and_version), |
| guid->u.Uuid.clock_seq_high_and_reserved, guid->u.Uuid.clock_seq_low, |
| guid->u.Uuid.node[0], guid->u.Uuid.node[1], guid->u.Uuid.node[2], |
| guid->u.Uuid.node[3], guid->u.Uuid.node[4], guid->u.Uuid.node[5]); |
| } |
| |
| /* Convert UTF16 string to UTF8. Rewritten from gpt utility. |
| * Caller must prepare enough space for UTF8. The rough estimation is: |
| * |
| * utf8 length = bytecount(utf16) * 1.5 |
| */ |
| #define SIZEOF_GPTENTRY_NAME 36 /* sizeof(GptEntry.name[]) */ |
| void UTF16ToUTF8(const uint16_t *utf16, uint8_t *utf8) |
| { |
| size_t s8idx, s16idx, s16len; |
| uint32_t utfchar; |
| unsigned int next_utf16; |
| |
| for (s16len = 0; s16len < SIZEOF_GPTENTRY_NAME && utf16[s16len]; ++s16len); |
| |
| *utf8 = s8idx = s16idx = 0; |
| while (s16idx < s16len) { |
| utfchar = le16toh(utf16[s16idx++]); |
| if ((utfchar & 0xf800) == 0xd800) { |
| next_utf16 = le16toh(utf16[s16idx]); |
| if ((utfchar & 0x400) != 0 || (next_utf16 & 0xfc00) != 0xdc00) |
| utfchar = 0xfffd; |
| else |
| s16idx++; |
| } |
| if (utfchar < 0x80) { |
| utf8[s8idx++] = utfchar; |
| } else if (utfchar < 0x800) { |
| utf8[s8idx++] = 0xc0 | (utfchar >> 6); |
| utf8[s8idx++] = 0x80 | (utfchar & 0x3f); |
| } else if (utfchar < 0x10000) { |
| utf8[s8idx++] = 0xe0 | (utfchar >> 12); |
| utf8[s8idx++] = 0x80 | ((utfchar >> 6) & 0x3f); |
| utf8[s8idx++] = 0x80 | (utfchar & 0x3f); |
| } else if (utfchar < 0x200000) { |
| utf8[s8idx++] = 0xf0 | (utfchar >> 18); |
| utf8[s8idx++] = 0x80 | ((utfchar >> 12) & 0x3f); |
| utf8[s8idx++] = 0x80 | ((utfchar >> 6) & 0x3f); |
| utf8[s8idx++] = 0x80 | (utfchar & 0x3f); |
| } |
| } |
| utf8[s8idx++] = 0; |
| } |
| |
| /* Convert UTF8 string to UTF16. Rewritten from gpt utility. |
| * Caller must prepare enough space for UTF16. The conservative estimation is: |
| * |
| * utf16 bytecount = bytecount(utf8) / 3 * 4 |
| */ |
| void UTF8ToUTF16(const uint8_t *utf8, uint16_t *utf16) |
| { |
| size_t s16idx, s8idx, s8len; |
| uint32_t utfchar; |
| unsigned int c, utfbytes; |
| |
| for (s8len = 0; utf8[s8len]; ++s8len); |
| |
| s8idx = s16idx = 0; |
| utfbytes = 0; |
| do { |
| c = utf8[s8idx++]; |
| if ((c & 0xc0) != 0x80) { |
| /* Initial characters. */ |
| if (utfbytes != 0) { |
| /* Incomplete encoding. */ |
| utf16[s16idx++] = 0xfffd; |
| } |
| if ((c & 0xf8) == 0xf0) { |
| utfchar = c & 0x07; |
| utfbytes = 3; |
| } else if ((c & 0xf0) == 0xe0) { |
| utfchar = c & 0x0f; |
| utfbytes = 2; |
| } else if ((c & 0xe0) == 0xc0) { |
| utfchar = c & 0x1f; |
| utfbytes = 1; |
| } else { |
| utfchar = c & 0x7f; |
| utfbytes = 0; |
| } |
| } else { |
| /* Followup characters. */ |
| if (utfbytes > 0) { |
| utfchar = (utfchar << 6) + (c & 0x3f); |
| utfbytes--; |
| } else if (utfbytes == 0) |
| utfbytes = -1; |
| utfchar = 0xfffd; |
| } |
| if (utfbytes == 0) { |
| if (utfchar >= 0x10000) { |
| utf16[s16idx++] = htole16(0xd800 | ((utfchar>>10)-0x40)); |
| if (s16idx >= SIZEOF_GPTENTRY_NAME) break; |
| utf16[s16idx++] = htole16(0xdc00 | (utfchar & 0x3ff)); |
| } else { |
| utf16[s16idx++] = htole16(utfchar); |
| } |
| } |
| } while (c != 0 && s16idx < SIZEOF_GPTENTRY_NAME); |
| if (s16idx < SIZEOF_GPTENTRY_NAME) |
| utf16[s16idx++] = 0; |
| } |
| |
| struct { |
| Guid type; |
| char *name; |
| char *description; |
| } supported_types[] = { |
| {GPT_ENT_TYPE_CHROMEOS_KERNEL, "kernel", "ChromeOS kernel"}, |
| {GPT_ENT_TYPE_CHROMEOS_ROOTFS, "rootfs", "ChromeOS rootfs"}, |
| {GPT_ENT_TYPE_LINUX_DATA, "data", "Linux data"}, |
| {GPT_ENT_TYPE_CHROMEOS_RESERVED, "reserved", "ChromeOS reserved"}, |
| {GPT_ENT_TYPE_EFI, "efi", "EFI System Partition"}, |
| {GPT_ENT_TYPE_UNUSED, "unused", "Unused (nonexistent) partition"}, |
| }; |
| |
| /* Resolves human-readable GPT type. |
| * Returns CGPT_OK if found. |
| * Returns CGPT_FAILED if no known type found. */ |
| int ResolveType(const Guid *type, char *buf) { |
| int i; |
| for (i = 0; i < ARRAY_COUNT(supported_types); ++i) { |
| if (!memcmp(type, &supported_types[i].type, sizeof(Guid))) { |
| strcpy(buf, supported_types[i].description); |
| return CGPT_OK; |
| } |
| } |
| return CGPT_FAILED; |
| } |
| |
| int SupportedType(const char *name, Guid *type) { |
| int i; |
| for (i = 0; i < ARRAY_COUNT(supported_types); ++i) { |
| if (!strcmp(name, supported_types[i].name)) { |
| memcpy(type, &supported_types[i].type, sizeof(Guid)); |
| return CGPT_OK; |
| } |
| } |
| return CGPT_FAILED; |
| } |
| |
| void PrintTypes(void) { |
| int i; |
| printf("The partition type may also be given as one of these aliases:\n\n"); |
| for (i = 0; i < ARRAY_COUNT(supported_types); ++i) { |
| printf(" %-10s %s\n", supported_types[i].name, |
| supported_types[i].description); |
| } |
| printf("\n"); |
| } |
| |
| uint32_t GetNumberOfEntries(const GptData *gpt) { |
| GptHeader *header = 0; |
| if (gpt->valid_headers & MASK_PRIMARY) |
| header = (GptHeader*)gpt->primary_header; |
| else if (gpt->valid_headers & MASK_SECONDARY) |
| header = (GptHeader*)gpt->secondary_header; |
| else |
| return 0; |
| return header->number_of_entries; |
| } |
| |
| static uint32_t GetSizeOfEntries(const GptData *gpt) { |
| GptHeader *header = 0; |
| if (gpt->valid_headers & MASK_PRIMARY) |
| header = (GptHeader*)gpt->primary_header; |
| else if (gpt->valid_headers & MASK_SECONDARY) |
| header = (GptHeader*)gpt->secondary_header; |
| else |
| return 0; |
| return header->number_of_entries; |
| } |
| |
| GptEntry *GetEntry(GptData *gpt, int secondary, int entry_index) { |
| uint8_t *entries; |
| int stride = GetSizeOfEntries(gpt); |
| if (!stride) |
| return 0; |
| |
| if (secondary == PRIMARY) { |
| entries = gpt->primary_entries; |
| } else { |
| entries = gpt->secondary_entries; |
| } |
| |
| return (GptEntry*)(&entries[stride * entry_index]); |
| } |
| |
| void SetPriority(GptData *gpt, int secondary, int entry_index, int priority) { |
| GptEntry *entry; |
| entry = GetEntry(gpt, secondary, entry_index); |
| |
| assert(priority >= 0 && priority <= CGPT_ATTRIBUTE_MAX_PRIORITY); |
| entry->attrs.fields.gpt_att &= ~CGPT_ATTRIBUTE_PRIORITY_MASK; |
| entry->attrs.fields.gpt_att |= priority << CGPT_ATTRIBUTE_PRIORITY_OFFSET; |
| } |
| |
| int GetPriority(GptData *gpt, int secondary, int entry_index) { |
| GptEntry *entry; |
| entry = GetEntry(gpt, secondary, entry_index); |
| return (entry->attrs.fields.gpt_att & CGPT_ATTRIBUTE_PRIORITY_MASK) >> |
| CGPT_ATTRIBUTE_PRIORITY_OFFSET; |
| } |
| |
| void SetTries(GptData *gpt, int secondary, int entry_index, int tries) { |
| GptEntry *entry; |
| entry = GetEntry(gpt, secondary, entry_index); |
| |
| assert(tries >= 0 && tries <= CGPT_ATTRIBUTE_MAX_TRIES); |
| entry->attrs.fields.gpt_att &= ~CGPT_ATTRIBUTE_TRIES_MASK; |
| entry->attrs.fields.gpt_att |= tries << CGPT_ATTRIBUTE_TRIES_OFFSET; |
| } |
| |
| int GetTries(GptData *gpt, int secondary, int entry_index) { |
| GptEntry *entry; |
| entry = GetEntry(gpt, secondary, entry_index); |
| return (entry->attrs.fields.gpt_att & CGPT_ATTRIBUTE_TRIES_MASK) >> |
| CGPT_ATTRIBUTE_TRIES_OFFSET; |
| } |
| |
| void SetSuccessful(GptData *gpt, int secondary, int entry_index, int success) { |
| GptEntry *entry; |
| entry = GetEntry(gpt, secondary, entry_index); |
| |
| assert(success >= 0 && success <= CGPT_ATTRIBUTE_MAX_SUCCESSFUL); |
| entry->attrs.fields.gpt_att &= ~CGPT_ATTRIBUTE_SUCCESSFUL_MASK; |
| entry->attrs.fields.gpt_att |= success << CGPT_ATTRIBUTE_SUCCESSFUL_OFFSET; |
| } |
| |
| int GetSuccessful(GptData *gpt, int secondary, int entry_index) { |
| GptEntry *entry; |
| entry = GetEntry(gpt, secondary, entry_index); |
| return (entry->attrs.fields.gpt_att & CGPT_ATTRIBUTE_SUCCESSFUL_MASK) >> |
| CGPT_ATTRIBUTE_SUCCESSFUL_OFFSET; |
| } |
| |
| |
| #define TOSTRING(A) #A |
| const char *GptError(int errnum) { |
| const char *error_string[] = { |
| TOSTRING(GPT_SUCCESS), |
| TOSTRING(GPT_ERROR_NO_VALID_KERNEL), |
| TOSTRING(GPT_ERROR_INVALID_HEADERS), |
| TOSTRING(GPT_ERROR_INVALID_ENTRIES), |
| TOSTRING(GPT_ERROR_INVALID_SECTOR_SIZE), |
| TOSTRING(GPT_ERROR_INVALID_SECTOR_NUMBER), |
| TOSTRING(GPT_ERROR_INVALID_UPDATE_TYPE) |
| }; |
| if (errnum < 0 || errnum >= ARRAY_COUNT(error_string)) |
| return "<illegal value>"; |
| return error_string[errnum]; |
| } |
| |
| /* Update CRC value if necessary. */ |
| void UpdateCrc(GptData *gpt) { |
| GptHeader *primary_header, *secondary_header; |
| |
| primary_header = (GptHeader*)gpt->primary_header; |
| secondary_header = (GptHeader*)gpt->secondary_header; |
| |
| if (gpt->modified & GPT_MODIFIED_ENTRIES1) { |
| primary_header->entries_crc32 = |
| Crc32(gpt->primary_entries, TOTAL_ENTRIES_SIZE); |
| } |
| if (gpt->modified & GPT_MODIFIED_ENTRIES2) { |
| secondary_header->entries_crc32 = |
| Crc32(gpt->secondary_entries, TOTAL_ENTRIES_SIZE); |
| } |
| if (gpt->modified & GPT_MODIFIED_HEADER1) { |
| primary_header->header_crc32 = 0; |
| primary_header->header_crc32 = Crc32( |
| (const uint8_t *)primary_header, primary_header->size); |
| } |
| if (gpt->modified & GPT_MODIFIED_HEADER2) { |
| secondary_header->header_crc32 = 0; |
| secondary_header->header_crc32 = Crc32( |
| (const uint8_t *)secondary_header, secondary_header->size); |
| } |
| } |
| /* Two headers are NOT bitwise identical. For example, my_lba pointers to header |
| * itself so that my_lba in primary and secondary is definitely different. |
| * Only the following fields should be identical. |
| * |
| * first_usable_lba |
| * last_usable_lba |
| * number_of_entries |
| * size_of_entry |
| * disk_uuid |
| * |
| * If any of above field are not matched, overwrite secondary with primary since |
| * we always trust primary. |
| * If any one of header is invalid, copy from another. */ |
| int IsSynonymous(const GptHeader* a, const GptHeader* b) { |
| if ((a->first_usable_lba == b->first_usable_lba) && |
| (a->last_usable_lba == b->last_usable_lba) && |
| (a->number_of_entries == b->number_of_entries) && |
| (a->size_of_entry == b->size_of_entry) && |
| (!memcmp(&a->disk_uuid, &b->disk_uuid, sizeof(Guid)))) |
| return 1; |
| return 0; |
| } |
| |
| /* Primary entries and secondary entries should be bitwise identical. |
| * If two entries tables are valid, compare them. If not the same, |
| * overwrites secondary with primary (primary always has higher priority), |
| * and marks secondary as modified. |
| * If only one is valid, overwrites invalid one. |
| * If all are invalid, does nothing. |
| * This function returns bit masks for GptData.modified field. |
| * Note that CRC is NOT re-computed in this function. |
| */ |
| uint8_t RepairEntries(GptData *gpt, const uint32_t valid_entries) { |
| if (valid_entries == MASK_BOTH) { |
| if (memcmp(gpt->primary_entries, gpt->secondary_entries, |
| TOTAL_ENTRIES_SIZE)) { |
| memcpy(gpt->secondary_entries, gpt->primary_entries, TOTAL_ENTRIES_SIZE); |
| return GPT_MODIFIED_ENTRIES2; |
| } |
| } else if (valid_entries == MASK_PRIMARY) { |
| memcpy(gpt->secondary_entries, gpt->primary_entries, TOTAL_ENTRIES_SIZE); |
| return GPT_MODIFIED_ENTRIES2; |
| } else if (valid_entries == MASK_SECONDARY) { |
| memcpy(gpt->primary_entries, gpt->secondary_entries, TOTAL_ENTRIES_SIZE); |
| return GPT_MODIFIED_ENTRIES1; |
| } |
| |
| return 0; |
| } |
| |
| /* The above five fields are shared between primary and secondary headers. |
| * We can recover one header from another through copying those fields. */ |
| void CopySynonymousParts(GptHeader* target, const GptHeader* source) { |
| target->first_usable_lba = source->first_usable_lba; |
| target->last_usable_lba = source->last_usable_lba; |
| target->number_of_entries = source->number_of_entries; |
| target->size_of_entry = source->size_of_entry; |
| memcpy(&target->disk_uuid, &source->disk_uuid, sizeof(Guid)); |
| } |
| |
| /* This function repairs primary and secondary headers if possible. |
| * If both headers are valid (CRC32 is correct) but |
| * a) indicate inconsistent usable LBA ranges, |
| * b) inconsistent partition entry size and number, |
| * c) inconsistent disk_uuid, |
| * we will use the primary header to overwrite secondary header. |
| * If primary is invalid (CRC32 is wrong), then we repair it from secondary. |
| * If secondary is invalid (CRC32 is wrong), then we repair it from primary. |
| * This function returns the bitmasks for modified header. |
| * Note that CRC value is NOT re-computed in this function. UpdateCrc() will |
| * do it later. |
| */ |
| uint8_t RepairHeader(GptData *gpt, const uint32_t valid_headers) { |
| GptHeader *primary_header, *secondary_header; |
| |
| primary_header = (GptHeader*)gpt->primary_header; |
| secondary_header = (GptHeader*)gpt->secondary_header; |
| |
| if (valid_headers == MASK_BOTH) { |
| if (!IsSynonymous(primary_header, secondary_header)) { |
| CopySynonymousParts(secondary_header, primary_header); |
| return GPT_MODIFIED_HEADER2; |
| } |
| } else if (valid_headers == MASK_PRIMARY) { |
| memcpy(secondary_header, primary_header, primary_header->size); |
| secondary_header->my_lba = gpt->drive_sectors - 1; /* the last sector */ |
| secondary_header->alternate_lba = primary_header->my_lba; |
| secondary_header->entries_lba = secondary_header->my_lba - |
| GPT_ENTRIES_SECTORS; |
| return GPT_MODIFIED_HEADER2; |
| } else if (valid_headers == MASK_SECONDARY) { |
| memcpy(primary_header, secondary_header, secondary_header->size); |
| primary_header->my_lba = GPT_PMBR_SECTOR; /* the second sector on drive */ |
| primary_header->alternate_lba = secondary_header->my_lba; |
| primary_header->entries_lba = primary_header->my_lba + GPT_HEADER_SECTOR; |
| return GPT_MODIFIED_HEADER1; |
| } |
| |
| return 0; |
| } |
| |
| |
| int IsZero(const Guid *gp) { |
| return (0 == memcmp(gp, &guid_unused, sizeof(Guid))); |
| } |
| |
| void PMBRToStr(struct pmbr *pmbr, char *str) { |
| char buf[256]; |
| if (IsZero(&pmbr->boot_guid)) { |
| sprintf(str, "PMBR"); |
| } else { |
| GuidToStr(&pmbr->boot_guid, buf); |
| sprintf(str, "PMBR (Boot GUID: %s)", buf); |
| } |
| } |
| |