int
efi_alloc_and_init(int fd, uint32_t nparts, struct dk_gpt **vtoc)
{
diskaddr_t capacity;
uint_t lbsize;
uint_t nblocks;
size_t length;
struct dk_gpt *vptr;
struct uuid uuid;
if (read_disk_info(fd, &capacity, &lbsize) != 0) {
if (efi_debug)
(void) fprintf(stderr,
"couldn't read disk information\n");
return (-1);
}
nblocks = NBLOCKS(nparts, lbsize);
if ((nblocks * lbsize) < EFI_MIN_ARRAY_SIZE + lbsize) {
/* 16K plus one block for the GPT */
nblocks = EFI_MIN_ARRAY_SIZE / lbsize + 1;
}
if (nparts > MAX_PARTS) {
if (efi_debug) {
(void) fprintf(stderr,
"the maximum number of partitions supported is %lu\n",
MAX_PARTS);
}
return (-1);
}
length = sizeof (struct dk_gpt) +
sizeof (struct dk_part) * (nparts - 1);
if ((*vtoc = calloc(length, 1)) == NULL)
return (-1);
vptr = *vtoc;
vptr->efi_version = EFI_VERSION_CURRENT;
vptr->efi_lbasize = lbsize;
vptr->efi_nparts = nparts;
/*
* add one block here for the PMBR; on disks with a 512 byte
* block size and 128 or fewer partitions, efi_first_u_lba
* should work out to "34"
*/
vptr->efi_first_u_lba = nblocks + 1;
vptr->efi_last_lba = capacity - 1;
vptr->efi_altern_lba = capacity -1;
vptr->efi_last_u_lba = vptr->efi_last_lba - nblocks;
(void) uuid_generate((uchar_t *)&uuid);
UUID_LE_CONVERT(vptr->efi_disk_uguid, uuid);
return (0);
}
static int
efi_read(int fd, struct dk_gpt *vtoc)
{
int i, j;
int label_len;
int rval = 0;
int md_flag = 0;
int vdc_flag = 0;
struct dk_minfo disk_info;
dk_efi_t dk_ioc;
efi_gpt_t *efi;
efi_gpe_t *efi_parts;
struct dk_cinfo dki_info;
uint32_t user_length;
boolean_t legacy_label = B_FALSE;
/*
* get the partition number for this file descriptor.
*/
if (ioctl(fd, DKIOCINFO, (caddr_t)&dki_info) == -1) {
if (efi_debug) {
(void) fprintf(stderr, "DKIOCINFO errno 0x%x\n", errno);
}
switch (errno) {
case EIO:
return (VT_EIO);
case EINVAL:
return (VT_EINVAL);
default:
return (VT_ERROR);
}
}
if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
(strncmp(dki_info.dki_dname, "md", 3) == 0)) {
md_flag++;
} else if ((strncmp(dki_info.dki_cname, "vdc", 4) == 0) &&
(strncmp(dki_info.dki_dname, "vdc", 4) == 0)) {
/*
* The controller and drive name "vdc" (virtual disk client)
* indicates a LDoms virtual disk.
*/
vdc_flag++;
}
/* get the LBA size */
if (ioctl(fd, DKIOCGMEDIAINFO, (caddr_t)&disk_info) == -1) {
if (efi_debug) {
(void) fprintf(stderr,
"assuming LBA 512 bytes %d\n",
errno);
}
disk_info.dki_lbsize = DEV_BSIZE;
}
if (disk_info.dki_lbsize == 0) {
if (efi_debug) {
(void) fprintf(stderr,
"efi_read: assuming LBA 512 bytes\n");
}
disk_info.dki_lbsize = DEV_BSIZE;
}
/*
* Read the EFI GPT to figure out how many partitions we need
* to deal with.
*/
dk_ioc.dki_lba = 1;
if (NBLOCKS(vtoc->efi_nparts, disk_info.dki_lbsize) < 34) {
label_len = EFI_MIN_ARRAY_SIZE + disk_info.dki_lbsize;
} else {
label_len = vtoc->efi_nparts * (int) sizeof (efi_gpe_t) +
disk_info.dki_lbsize;
if (label_len % disk_info.dki_lbsize) {
/* pad to physical sector size */
label_len += disk_info.dki_lbsize;
label_len &= ~(disk_info.dki_lbsize - 1);
}
}
if ((dk_ioc.dki_data = calloc(label_len, 1)) == NULL)
return (VT_ERROR);
dk_ioc.dki_length = disk_info.dki_lbsize;
user_length = vtoc->efi_nparts;
efi = dk_ioc.dki_data;
if (md_flag) {
dk_ioc.dki_length = label_len;
if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
switch (errno) {
case EIO:
return (VT_EIO);
default:
return (VT_ERROR);
}
}
} else if ((rval = check_label(fd, &dk_ioc)) == VT_EINVAL) {
/*
* No valid label here; try the alternate. Note that here
* we just read GPT header and save it into dk_ioc.data,
* Later, we will read GUID partition entry array if we
* can get valid GPT header.
*/
/*
* This is a workaround for legacy systems. In the past, the
* last sector of SCSI disk was invisible on x86 platform. At
* that time, backup label was saved on the next to the last
* sector. It is possible for users to move a disk from previous
* solaris system to present system. Here, we attempt to search
* legacy backup EFI label first.
*/
dk_ioc.dki_lba = disk_info.dki_capacity - 2;
dk_ioc.dki_length = disk_info.dki_lbsize;
rval = check_label(fd, &dk_ioc);
if (rval == VT_EINVAL) {
/*
* we didn't find legacy backup EFI label, try to
* search backup EFI label in the last block.
*/
dk_ioc.dki_lba = disk_info.dki_capacity - 1;
dk_ioc.dki_length = disk_info.dki_lbsize;
rval = check_label(fd, &dk_ioc);
if (rval == 0) {
legacy_label = B_TRUE;
if (efi_debug)
(void) fprintf(stderr,
"efi_read: primary label corrupt; "
"using EFI backup label located on"
" the last block\n");
}
} else {
if ((efi_debug) && (rval == 0))
(void) fprintf(stderr, "efi_read: primary label"
" corrupt; using legacy EFI backup label "
" located on the next to last block\n");
}
if (rval == 0) {
dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
vtoc->efi_flags |= EFI_GPT_PRIMARY_CORRUPT;
vtoc->efi_nparts =
LE_32(efi->efi_gpt_NumberOfPartitionEntries);
/*
* Partition tables are between backup GPT header
* table and ParitionEntryLBA (the starting LBA of
* the GUID partition entries array). Now that we
* already got valid GPT header and saved it in
* dk_ioc.dki_data, we try to get GUID partition
* entry array here.
*/
/* LINTED */
dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
+ disk_info.dki_lbsize);
if (legacy_label)
dk_ioc.dki_length = disk_info.dki_capacity - 1 -
dk_ioc.dki_lba;
else
dk_ioc.dki_length = disk_info.dki_capacity - 2 -
dk_ioc.dki_lba;
dk_ioc.dki_length *= disk_info.dki_lbsize;
if (dk_ioc.dki_length >
((len_t)label_len - sizeof (*dk_ioc.dki_data))) {
rval = VT_EINVAL;
} else {
/*
* read GUID partition entry array
*/
rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
}
}
} else if (rval == 0) {
dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
/* LINTED */
dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
+ disk_info.dki_lbsize);
dk_ioc.dki_length = label_len - disk_info.dki_lbsize;
rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
} else if (vdc_flag && rval == VT_ERROR && errno == EINVAL) {
/*
* When the device is a LDoms virtual disk, the DKIOCGETEFI
* ioctl can fail with EINVAL if the virtual disk backend
* is a ZFS volume serviced by a domain running an old version
* of Solaris. This is because the DKIOCGETEFI ioctl was
* initially incorrectly implemented for a ZFS volume and it
* expected the GPT and GPE to be retrieved with a single ioctl.
* So we try to read the GPT and the GPE using that old style
* ioctl.
*/
dk_ioc.dki_lba = 1;
dk_ioc.dki_length = label_len;
rval = check_label(fd, &dk_ioc);
}
if (rval < 0) {
free(efi);
return (rval);
}
/* LINTED -- always longlong aligned */
efi_parts = (efi_gpe_t *)(((char *)efi) + disk_info.dki_lbsize);
/*
* Assemble this into a "dk_gpt" struct for easier
* digestibility by applications.
*/
vtoc->efi_version = LE_32(efi->efi_gpt_Revision);
vtoc->efi_nparts = LE_32(efi->efi_gpt_NumberOfPartitionEntries);
vtoc->efi_part_size = LE_32(efi->efi_gpt_SizeOfPartitionEntry);
vtoc->efi_lbasize = disk_info.dki_lbsize;
vtoc->efi_last_lba = disk_info.dki_capacity - 1;
vtoc->efi_first_u_lba = LE_64(efi->efi_gpt_FirstUsableLBA);
vtoc->efi_last_u_lba = LE_64(efi->efi_gpt_LastUsableLBA);
vtoc->efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
UUID_LE_CONVERT(vtoc->efi_disk_uguid, efi->efi_gpt_DiskGUID);
/*
* If the array the user passed in is too small, set the length
* to what it needs to be and return
*/
if (user_length < vtoc->efi_nparts) {
return (VT_EINVAL);
}
for (i = 0; i < vtoc->efi_nparts; i++) {
UUID_LE_CONVERT(vtoc->efi_parts[i].p_guid,
efi_parts[i].efi_gpe_PartitionTypeGUID);
for (j = 0;
j < sizeof (conversion_array)
/ sizeof (struct uuid_to_ptag); j++) {
if (bcmp(&vtoc->efi_parts[i].p_guid,
&conversion_array[j].uuid,
sizeof (struct uuid)) == 0) {
vtoc->efi_parts[i].p_tag = j;
break;
}
}
if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED)
continue;
vtoc->efi_parts[i].p_flag =
LE_16(efi_parts[i].efi_gpe_Attributes.PartitionAttrs);
vtoc->efi_parts[i].p_start =
LE_64(efi_parts[i].efi_gpe_StartingLBA);
vtoc->efi_parts[i].p_size =
LE_64(efi_parts[i].efi_gpe_EndingLBA) -
vtoc->efi_parts[i].p_start + 1;
for (j = 0; j < EFI_PART_NAME_LEN; j++) {
vtoc->efi_parts[i].p_name[j] =
(uchar_t)LE_16(
efi_parts[i].efi_gpe_PartitionName[j]);
}
UUID_LE_CONVERT(vtoc->efi_parts[i].p_uguid,
efi_parts[i].efi_gpe_UniquePartitionGUID);
}
free(efi);
return (dki_info.dki_partition);
}
/*ARGSUSED*/
int
zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
{
zvol_state_t *zv;
struct dk_cinfo dkc;
struct dk_minfo dkm;
dk_efi_t efi;
efi_gpt_t gpt;
efi_gpe_t gpe;
struct uuid uuid = EFI_RESERVED;
uint32_t crc;
int error = 0;
mutex_enter(&zvol_state_lock);
zv = ddi_get_soft_state(zvol_state, getminor(dev));
if (zv == NULL) {
mutex_exit(&zvol_state_lock);
return (ENXIO);
}
switch (cmd) {
case DKIOCINFO:
bzero(&dkc, sizeof (dkc));
(void) strcpy(dkc.dki_cname, "zvol");
(void) strcpy(dkc.dki_dname, "zvol");
dkc.dki_ctype = DKC_UNKNOWN;
dkc.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
mutex_exit(&zvol_state_lock);
if (ddi_copyout(&dkc, (void *)arg, sizeof (dkc), flag))
error = EFAULT;
return (error);
case DKIOCGMEDIAINFO:
bzero(&dkm, sizeof (dkm));
dkm.dki_lbsize = 1U << zv->zv_min_bs;
dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
dkm.dki_media_type = DK_UNKNOWN;
mutex_exit(&zvol_state_lock);
if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
error = EFAULT;
return (error);
case DKIOCGETEFI:
if (ddi_copyin((void *)arg, &efi, sizeof (dk_efi_t), flag)) {
mutex_exit(&zvol_state_lock);
return (EFAULT);
}
bzero(&gpt, sizeof (gpt));
bzero(&gpe, sizeof (gpe));
efi.dki_data = (void *)(uintptr_t)efi.dki_data_64;
if (efi.dki_length < sizeof (gpt) + sizeof (gpe)) {
mutex_exit(&zvol_state_lock);
return (EINVAL);
}
efi.dki_length = sizeof (gpt) + sizeof (gpe);
gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
gpt.efi_gpt_FirstUsableLBA = LE_64(0ULL);
gpt.efi_gpt_LastUsableLBA =
LE_64((zv->zv_volsize >> zv->zv_min_bs) - 1);
gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
gpt.efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (gpe));
UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
gpe.efi_gpe_StartingLBA = gpt.efi_gpt_FirstUsableLBA;
gpe.efi_gpe_EndingLBA = gpt.efi_gpt_LastUsableLBA;
CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
mutex_exit(&zvol_state_lock);
if (ddi_copyout(&gpt, efi.dki_data, sizeof (gpt), flag) ||
ddi_copyout(&gpe, efi.dki_data + 1, sizeof (gpe), flag))
error = EFAULT;
return (error);
default:
error = ENOTSUP;
break;
}
mutex_exit(&zvol_state_lock);
return (error);
}
static int
efi_read(int fd, struct dk_gpt *vtoc)
{
int i, j;
int label_len;
int rval = 0;
int md_flag = 0;
struct dk_minfo disk_info;
dk_efi_t dk_ioc;
efi_gpt_t *efi;
efi_gpe_t *efi_parts;
struct dk_cinfo dki_info;
uint32_t user_length;
/*
* get the partition number for this file descriptor.
*/
if (ioctl(fd, DKIOCINFO, (caddr_t)&dki_info) == -1) {
if (efi_debug)
(void) fprintf(stderr, "DKIOCINFO errno 0x%x\n", errno);
switch (errno) {
case EIO:
return (VT_EIO);
case EINVAL:
return (VT_EINVAL);
default:
return (VT_ERROR);
}
}
if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
(strncmp(dki_info.dki_dname, "md", 3) == 0)) {
md_flag++;
}
/* get the LBA size */
if (ioctl(fd, DKIOCGMEDIAINFO, (caddr_t)&disk_info) == -1) {
if (efi_debug) {
(void) fprintf(stderr,
"assuming LBA 512 bytes %d\n",
errno);
}
disk_info.dki_lbsize = DEV_BSIZE;
}
if (disk_info.dki_lbsize == 0) {
if (efi_debug) {
(void) fprintf(stderr,
"efi_read: assuming LBA 512 bytes\n");
}
disk_info.dki_lbsize = DEV_BSIZE;
}
/*
* Read the EFI GPT to figure out how many partitions we need
* to deal with.
*/
dk_ioc.dki_lba = 1;
if (NBLOCKS(vtoc->efi_nparts, disk_info.dki_lbsize) < 34) {
label_len = EFI_MIN_ARRAY_SIZE + disk_info.dki_lbsize;
} else {
label_len = vtoc->efi_nparts * (int) sizeof (efi_gpe_t) +
disk_info.dki_lbsize;
if (label_len % disk_info.dki_lbsize) {
/* pad to physical sector size */
label_len += disk_info.dki_lbsize;
label_len &= ~(disk_info.dki_lbsize - 1);
}
}
if ((dk_ioc.dki_data = calloc(label_len, 1)) == NULL)
return (VT_ERROR);
dk_ioc.dki_length = label_len;
user_length = vtoc->efi_nparts;
efi = dk_ioc.dki_data;
if (md_flag) {
if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
switch (errno) {
case EIO:
return (VT_EIO);
default:
return (VT_ERROR);
}
}
} else if ((rval = check_label(fd, &dk_ioc)) == VT_EINVAL) {
/* no valid label here; try the alternate */
dk_ioc.dki_lba = disk_info.dki_capacity - 1;
dk_ioc.dki_length = disk_info.dki_lbsize;
rval = check_label(fd, &dk_ioc);
if (rval != 0) {
/*
* This is a workaround for legacy systems.
*
* In the past, the last sector of SCSI disk was
* invisible on x86 platform. At that time, backup
* label was saved on the next to the last sector.
* It is possible for users to move a disk from
* previous solaris system to present system.
*/
dk_ioc.dki_lba = disk_info.dki_capacity - 2;
dk_ioc.dki_length = disk_info.dki_lbsize;
rval = check_label(fd, &dk_ioc);
if (efi_debug && (rval == 0)) {
(void) fprintf(stderr,
"efi_read: primary label corrupt; "
"using legacy EFI backup label\n");
}
}
if (rval == 0) {
if (efi_debug) {
(void) fprintf(stderr,
"efi_read: primary label corrupt; "
"using backup\n");
}
dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
vtoc->efi_flags |= EFI_GPT_PRIMARY_CORRUPT;
vtoc->efi_nparts =
LE_32(efi->efi_gpt_NumberOfPartitionEntries);
/*
* partitions are between last usable LBA and
* backup partition header
*/
dk_ioc.dki_data++;
dk_ioc.dki_length = disk_info.dki_capacity -
dk_ioc.dki_lba - 1;
dk_ioc.dki_length *= disk_info.dki_lbsize;
if (dk_ioc.dki_length > (len_t)label_len) {
rval = VT_EINVAL;
} else {
rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
}
}
}
if (rval < 0) {
free(efi);
return (rval);
}
/* partitions start in the next block */
/* LINTED -- always longlong aligned */
efi_parts = (efi_gpe_t *)(((char *)efi) + disk_info.dki_lbsize);
/*
* Assemble this into a "dk_gpt" struct for easier
* digestibility by applications.
*/
vtoc->efi_version = LE_32(efi->efi_gpt_Revision);
vtoc->efi_nparts = LE_32(efi->efi_gpt_NumberOfPartitionEntries);
vtoc->efi_part_size = LE_32(efi->efi_gpt_SizeOfPartitionEntry);
vtoc->efi_lbasize = disk_info.dki_lbsize;
vtoc->efi_last_lba = disk_info.dki_capacity - 1;
vtoc->efi_first_u_lba = LE_64(efi->efi_gpt_FirstUsableLBA);
vtoc->efi_last_u_lba = LE_64(efi->efi_gpt_LastUsableLBA);
UUID_LE_CONVERT(vtoc->efi_disk_uguid, efi->efi_gpt_DiskGUID);
/*
* If the array the user passed in is too small, set the length
* to what it needs to be and return
*/
if (user_length < vtoc->efi_nparts) {
return (VT_EINVAL);
}
for (i = 0; i < vtoc->efi_nparts; i++) {
UUID_LE_CONVERT(vtoc->efi_parts[i].p_guid,
efi_parts[i].efi_gpe_PartitionTypeGUID);
for (j = 0;
j < sizeof (conversion_array) / sizeof (struct uuid_to_ptag);
j++) {
if (bcmp(&vtoc->efi_parts[i].p_guid,
&conversion_array[j].uuid,
sizeof (struct uuid)) == 0) {
vtoc->efi_parts[i].p_tag = j;
break;
}
}
if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED)
continue;
vtoc->efi_parts[i].p_flag =
LE_16(efi_parts[i].efi_gpe_Attributes.PartitionAttrs);
vtoc->efi_parts[i].p_start =
LE_64(efi_parts[i].efi_gpe_StartingLBA);
vtoc->efi_parts[i].p_size =
LE_64(efi_parts[i].efi_gpe_EndingLBA) -
vtoc->efi_parts[i].p_start + 1;
for (j = 0; j < EFI_PART_NAME_LEN; j++) {
vtoc->efi_parts[i].p_name[j] =
(uchar_t)LE_16(efi_parts[i].efi_gpe_PartitionName[j]);
}
UUID_LE_CONVERT(vtoc->efi_parts[i].p_uguid,
efi_parts[i].efi_gpe_UniquePartitionGUID);
}
free(efi);
return (dki_info.dki_partition);
}
