2007-12-24 17:11:51 +01:00
|
|
|
/*
|
|
|
|
* Generic SCSI Device support
|
|
|
|
*
|
|
|
|
* Copyright (c) 2007 Bull S.A.S.
|
|
|
|
* Based on code by Paul Brook
|
|
|
|
* Based on code by Fabrice Bellard
|
|
|
|
*
|
|
|
|
* Written by Laurent Vivier <Laurent.Vivier@bull.net>
|
|
|
|
*
|
2011-06-26 04:21:35 +02:00
|
|
|
* This code is licensed under the LGPL.
|
2007-12-24 17:11:51 +01:00
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "qemu-common.h"
|
2010-02-18 16:24:31 +01:00
|
|
|
#include "qemu-error.h"
|
2009-10-30 09:54:00 +01:00
|
|
|
#include "scsi.h"
|
2010-08-24 17:22:24 +02:00
|
|
|
#include "blockdev.h"
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2009-08-31 14:24:04 +02:00
|
|
|
#ifdef __linux__
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
//#define DEBUG_SCSI
|
|
|
|
|
|
|
|
#ifdef DEBUG_SCSI
|
2009-05-13 19:53:17 +02:00
|
|
|
#define DPRINTF(fmt, ...) \
|
|
|
|
do { printf("scsi-generic: " fmt , ## __VA_ARGS__); } while (0)
|
2007-12-24 17:11:51 +01:00
|
|
|
#else
|
2009-05-13 19:53:17 +02:00
|
|
|
#define DPRINTF(fmt, ...) do {} while(0)
|
2007-12-24 17:11:51 +01:00
|
|
|
#endif
|
|
|
|
|
2009-05-13 19:53:17 +02:00
|
|
|
#define BADF(fmt, ...) \
|
|
|
|
do { fprintf(stderr, "scsi-generic: " fmt , ## __VA_ARGS__); } while (0)
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <sys/types.h>
|
|
|
|
#include <sys/stat.h>
|
|
|
|
#include <unistd.h>
|
|
|
|
#include <scsi/sg.h>
|
2009-11-26 15:33:53 +01:00
|
|
|
#include "scsi-defs.h"
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2008-09-23 15:38:44 +02:00
|
|
|
#define SCSI_SENSE_BUF_SIZE 96
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
#define SG_ERR_DRIVER_TIMEOUT 0x06
|
|
|
|
#define SG_ERR_DRIVER_SENSE 0x08
|
|
|
|
|
|
|
|
#ifndef MAX_UINT
|
|
|
|
#define MAX_UINT ((unsigned int)-1)
|
|
|
|
#endif
|
|
|
|
|
2009-08-31 14:24:04 +02:00
|
|
|
typedef struct SCSIGenericState SCSIGenericState;
|
|
|
|
|
2009-11-26 15:33:48 +01:00
|
|
|
typedef struct SCSIGenericReq {
|
|
|
|
SCSIRequest req;
|
2007-12-24 17:11:51 +01:00
|
|
|
uint8_t *buf;
|
|
|
|
int buflen;
|
|
|
|
int len;
|
|
|
|
sg_io_hdr_t io_header;
|
2009-11-26 15:33:48 +01:00
|
|
|
} SCSIGenericReq;
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2009-08-31 14:24:04 +02:00
|
|
|
struct SCSIGenericState
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2009-08-31 14:24:04 +02:00
|
|
|
SCSIDevice qdev;
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
BlockDriverState *bs;
|
2007-12-24 17:11:51 +01:00
|
|
|
int lun;
|
|
|
|
int driver_status;
|
|
|
|
uint8_t sensebuf[SCSI_SENSE_BUF_SIZE];
|
2008-10-17 10:08:56 +02:00
|
|
|
uint8_t senselen;
|
2007-12-24 17:11:51 +01:00
|
|
|
};
|
|
|
|
|
2011-04-18 12:53:14 +02:00
|
|
|
static void scsi_set_sense(SCSIGenericState *s, SCSISense sense)
|
|
|
|
{
|
|
|
|
s->senselen = scsi_build_sense(sense, s->sensebuf, SCSI_SENSE_BUF_SIZE, 0);
|
|
|
|
s->driver_status = SG_ERR_DRIVER_SENSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void scsi_clear_sense(SCSIGenericState *s)
|
|
|
|
{
|
|
|
|
memset(s->sensebuf, 0, SCSI_SENSE_BUF_SIZE);
|
|
|
|
s->senselen = 0;
|
|
|
|
s->driver_status = 0;
|
|
|
|
}
|
|
|
|
|
2011-04-18 13:36:02 +02:00
|
|
|
static int scsi_get_sense(SCSIRequest *req, uint8_t *outbuf, int len)
|
|
|
|
{
|
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, req->dev);
|
|
|
|
int size = SCSI_SENSE_BUF_SIZE;
|
|
|
|
|
|
|
|
if (!(s->driver_status & SG_ERR_DRIVER_SENSE)) {
|
|
|
|
size = scsi_build_sense(SENSE_CODE(NO_SENSE), s->sensebuf,
|
|
|
|
SCSI_SENSE_BUF_SIZE, 0);
|
|
|
|
}
|
|
|
|
if (size > len) {
|
|
|
|
size = len;
|
|
|
|
}
|
|
|
|
memcpy(outbuf, s->sensebuf, size);
|
|
|
|
|
|
|
|
return size;
|
|
|
|
}
|
|
|
|
|
2011-07-11 15:02:24 +02:00
|
|
|
static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
|
|
|
|
void *hba_private)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2009-11-26 15:33:50 +01:00
|
|
|
SCSIRequest *req;
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2011-07-11 15:02:24 +02:00
|
|
|
req = scsi_req_alloc(sizeof(SCSIGenericReq), d, tag, lun, hba_private);
|
2011-04-18 12:35:39 +02:00
|
|
|
return req;
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
2011-04-18 16:01:56 +02:00
|
|
|
static void scsi_free_request(SCSIRequest *req)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2011-04-18 16:01:56 +02:00
|
|
|
SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
|
|
|
|
|
2009-11-26 15:33:49 +01:00
|
|
|
qemu_free(r->buf);
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Helper function for command completion. */
|
|
|
|
static void scsi_command_complete(void *opaque, int ret)
|
|
|
|
{
|
2011-08-03 10:49:06 +02:00
|
|
|
int status;
|
2009-11-26 15:33:48 +01:00
|
|
|
SCSIGenericReq *r = (SCSIGenericReq *)opaque;
|
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, r->req.dev);
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2011-05-25 16:53:46 +02:00
|
|
|
r->req.aiocb = NULL;
|
2007-12-24 17:11:51 +01:00
|
|
|
s->driver_status = r->io_header.driver_status;
|
2008-10-17 10:08:56 +02:00
|
|
|
if (s->driver_status & SG_ERR_DRIVER_SENSE)
|
|
|
|
s->senselen = r->io_header.sb_len_wr;
|
|
|
|
|
2011-04-18 12:53:14 +02:00
|
|
|
if (ret != 0) {
|
|
|
|
switch (ret) {
|
2011-04-18 14:29:16 +02:00
|
|
|
case -EDOM:
|
2011-08-03 10:49:06 +02:00
|
|
|
status = TASK_SET_FULL;
|
2011-04-18 14:29:16 +02:00
|
|
|
break;
|
2011-04-18 12:53:14 +02:00
|
|
|
case -EINVAL:
|
2011-08-03 10:49:06 +02:00
|
|
|
status = CHECK_CONDITION;
|
2011-04-18 12:53:14 +02:00
|
|
|
scsi_set_sense(s, SENSE_CODE(INVALID_FIELD));
|
|
|
|
break;
|
|
|
|
case -ENOMEM:
|
2011-08-03 10:49:06 +02:00
|
|
|
status = CHECK_CONDITION;
|
2011-04-18 12:53:14 +02:00
|
|
|
scsi_set_sense(s, SENSE_CODE(TARGET_FAILURE));
|
|
|
|
break;
|
|
|
|
default:
|
2011-08-03 10:49:06 +02:00
|
|
|
status = CHECK_CONDITION;
|
2011-04-18 12:53:14 +02:00
|
|
|
scsi_set_sense(s, SENSE_CODE(IO_ERROR));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
} else {
|
2007-12-24 17:11:51 +01:00
|
|
|
if (s->driver_status & SG_ERR_DRIVER_TIMEOUT) {
|
2011-08-03 10:49:06 +02:00
|
|
|
status = BUSY;
|
2007-12-24 17:11:51 +01:00
|
|
|
BADF("Driver Timeout\n");
|
2011-08-03 10:49:06 +02:00
|
|
|
} else if (r->io_header.status) {
|
|
|
|
status = r->io_header.status;
|
|
|
|
} else if (s->driver_status & SG_ERR_DRIVER_SENSE) {
|
|
|
|
status = CHECK_CONDITION;
|
|
|
|
} else {
|
|
|
|
status = GOOD;
|
|
|
|
}
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
2008-10-17 10:08:56 +02:00
|
|
|
DPRINTF("Command complete 0x%p tag=0x%x status=%d\n",
|
2011-08-03 10:49:06 +02:00
|
|
|
r, r->req.tag, status);
|
2009-11-26 15:34:00 +01:00
|
|
|
|
2011-08-03 10:49:06 +02:00
|
|
|
scsi_req_complete(&r->req, status);
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Cancel a pending data transfer. */
|
2011-04-18 12:35:39 +02:00
|
|
|
static void scsi_cancel_io(SCSIRequest *req)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2011-04-18 12:35:39 +02:00
|
|
|
SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
|
|
|
|
|
|
|
|
DPRINTF("Cancel tag=0x%x\n", req->tag);
|
|
|
|
if (r->req.aiocb) {
|
|
|
|
bdrv_aio_cancel(r->req.aiocb);
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
2011-04-18 12:35:39 +02:00
|
|
|
r->req.aiocb = NULL;
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static int execute_command(BlockDriverState *bdrv,
|
2009-11-26 15:33:48 +01:00
|
|
|
SCSIGenericReq *r, int direction,
|
2007-12-24 17:11:51 +01:00
|
|
|
BlockDriverCompletionFunc *complete)
|
|
|
|
{
|
2009-11-26 15:33:48 +01:00
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, r->req.dev);
|
|
|
|
|
2007-12-24 17:11:51 +01:00
|
|
|
r->io_header.interface_id = 'S';
|
|
|
|
r->io_header.dxfer_direction = direction;
|
|
|
|
r->io_header.dxferp = r->buf;
|
|
|
|
r->io_header.dxfer_len = r->buflen;
|
2009-11-26 15:33:51 +01:00
|
|
|
r->io_header.cmdp = r->req.cmd.buf;
|
|
|
|
r->io_header.cmd_len = r->req.cmd.len;
|
2009-11-26 15:33:48 +01:00
|
|
|
r->io_header.mx_sb_len = sizeof(s->sensebuf);
|
|
|
|
r->io_header.sbp = s->sensebuf;
|
2007-12-24 17:11:51 +01:00
|
|
|
r->io_header.timeout = MAX_UINT;
|
|
|
|
r->io_header.usr_ptr = r;
|
|
|
|
r->io_header.flags |= SG_FLAG_DIRECT_IO;
|
|
|
|
|
2009-11-26 15:33:48 +01:00
|
|
|
r->req.aiocb = bdrv_aio_ioctl(bdrv, SG_IO, &r->io_header, complete, r);
|
|
|
|
if (r->req.aiocb == NULL) {
|
2007-12-24 17:11:51 +01:00
|
|
|
BADF("execute_command: read failed !\n");
|
2011-04-18 12:53:14 +02:00
|
|
|
return -ENOMEM;
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void scsi_read_complete(void * opaque, int ret)
|
|
|
|
{
|
2009-11-26 15:33:48 +01:00
|
|
|
SCSIGenericReq *r = (SCSIGenericReq *)opaque;
|
2007-12-24 17:11:51 +01:00
|
|
|
int len;
|
|
|
|
|
2011-05-25 16:53:46 +02:00
|
|
|
r->req.aiocb = NULL;
|
2007-12-24 17:11:51 +01:00
|
|
|
if (ret) {
|
2010-09-01 16:33:21 +02:00
|
|
|
DPRINTF("IO error ret %d\n", ret);
|
2007-12-24 17:11:51 +01:00
|
|
|
scsi_command_complete(r, ret);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
len = r->io_header.dxfer_len - r->io_header.resid;
|
2009-11-26 15:33:48 +01:00
|
|
|
DPRINTF("Data ready tag=0x%x len=%d\n", r->req.tag, len);
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
r->len = -1;
|
2011-05-03 14:15:59 +02:00
|
|
|
if (len == 0) {
|
2008-10-17 10:08:56 +02:00
|
|
|
scsi_command_complete(r, 0);
|
2011-05-03 14:15:59 +02:00
|
|
|
} else {
|
2011-04-18 14:59:13 +02:00
|
|
|
scsi_req_data(&r->req, len);
|
2011-05-03 14:15:59 +02:00
|
|
|
}
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Read more data from scsi device into buffer. */
|
2011-04-18 12:35:39 +02:00
|
|
|
static void scsi_read_data(SCSIRequest *req)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2011-04-18 12:35:39 +02:00
|
|
|
SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
|
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, r->req.dev);
|
2007-12-24 17:11:51 +01:00
|
|
|
int ret;
|
|
|
|
|
2011-04-18 12:35:39 +02:00
|
|
|
DPRINTF("scsi_read_data 0x%x\n", req->tag);
|
2007-12-24 17:11:51 +01:00
|
|
|
if (r->len == -1) {
|
|
|
|
scsi_command_complete(r, 0);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2009-11-26 15:33:51 +01:00
|
|
|
if (r->req.cmd.buf[0] == REQUEST_SENSE && s->driver_status & SG_ERR_DRIVER_SENSE)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2008-10-17 10:08:56 +02:00
|
|
|
s->senselen = MIN(r->len, s->senselen);
|
|
|
|
memcpy(r->buf, s->sensebuf, s->senselen);
|
2007-12-24 17:11:51 +01:00
|
|
|
r->io_header.driver_status = 0;
|
2008-10-17 10:08:56 +02:00
|
|
|
r->io_header.status = 0;
|
|
|
|
r->io_header.dxfer_len = s->senselen;
|
2007-12-24 17:11:51 +01:00
|
|
|
r->len = -1;
|
2009-11-26 15:33:48 +01:00
|
|
|
DPRINTF("Data ready tag=0x%x len=%d\n", r->req.tag, s->senselen);
|
2008-09-23 15:38:44 +02:00
|
|
|
DPRINTF("Sense: %d %d %d %d %d %d %d %d\n",
|
|
|
|
r->buf[0], r->buf[1], r->buf[2], r->buf[3],
|
|
|
|
r->buf[4], r->buf[5], r->buf[6], r->buf[7]);
|
2011-04-18 14:59:13 +02:00
|
|
|
scsi_req_data(&r->req, s->senselen);
|
2011-04-18 12:53:14 +02:00
|
|
|
/* Clear sensebuf after REQUEST_SENSE */
|
|
|
|
scsi_clear_sense(s);
|
2007-12-24 17:11:51 +01:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
ret = execute_command(s->bs, r, SG_DXFER_FROM_DEV, scsi_read_complete);
|
2011-04-18 12:53:14 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
scsi_command_complete(r, ret);
|
2007-12-24 17:11:51 +01:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void scsi_write_complete(void * opaque, int ret)
|
|
|
|
{
|
2009-11-26 15:33:48 +01:00
|
|
|
SCSIGenericReq *r = (SCSIGenericReq *)opaque;
|
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, r->req.dev);
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
DPRINTF("scsi_write_complete() ret = %d\n", ret);
|
2011-05-25 16:53:46 +02:00
|
|
|
r->req.aiocb = NULL;
|
2007-12-24 17:11:51 +01:00
|
|
|
if (ret) {
|
|
|
|
DPRINTF("IO error\n");
|
|
|
|
scsi_command_complete(r, ret);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2009-11-26 15:33:51 +01:00
|
|
|
if (r->req.cmd.buf[0] == MODE_SELECT && r->req.cmd.buf[4] == 12 &&
|
2009-11-26 15:33:54 +01:00
|
|
|
s->qdev.type == TYPE_TAPE) {
|
2009-11-26 15:33:52 +01:00
|
|
|
s->qdev.blocksize = (r->buf[9] << 16) | (r->buf[10] << 8) | r->buf[11];
|
2010-09-01 16:33:21 +02:00
|
|
|
DPRINTF("block size %d\n", s->qdev.blocksize);
|
2008-10-17 10:08:56 +02:00
|
|
|
}
|
|
|
|
|
2007-12-24 17:11:51 +01:00
|
|
|
scsi_command_complete(r, ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Write data to a scsi device. Returns nonzero on failure.
|
|
|
|
The transfer may complete asynchronously. */
|
2011-04-22 09:39:16 +02:00
|
|
|
static void scsi_write_data(SCSIRequest *req)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2011-04-18 12:35:39 +02:00
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, req->dev);
|
|
|
|
SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
|
2007-12-24 17:11:51 +01:00
|
|
|
int ret;
|
|
|
|
|
2011-04-18 12:35:39 +02:00
|
|
|
DPRINTF("scsi_write_data 0x%x\n", req->tag);
|
2007-12-24 17:11:51 +01:00
|
|
|
if (r->len == 0) {
|
|
|
|
r->len = r->buflen;
|
2011-04-18 14:59:13 +02:00
|
|
|
scsi_req_data(&r->req, r->len);
|
2011-04-22 09:39:16 +02:00
|
|
|
return;
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
ret = execute_command(s->bs, r, SG_DXFER_TO_DEV, scsi_write_complete);
|
2011-04-18 12:53:14 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
scsi_command_complete(r, ret);
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return a pointer to the data buffer. */
|
2011-04-18 12:35:39 +02:00
|
|
|
static uint8_t *scsi_get_buf(SCSIRequest *req)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2011-04-18 12:35:39 +02:00
|
|
|
SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
|
|
|
|
|
2007-12-24 17:11:51 +01:00
|
|
|
return r->buf;
|
|
|
|
}
|
|
|
|
|
2009-11-26 15:33:55 +01:00
|
|
|
static void scsi_req_fixup(SCSIRequest *req)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2009-11-26 15:33:55 +01:00
|
|
|
switch(req->cmd.buf[0]) {
|
2007-12-24 17:11:51 +01:00
|
|
|
case WRITE_10:
|
2009-11-26 15:33:55 +01:00
|
|
|
req->cmd.buf[1] &= ~0x08; /* disable FUA */
|
2007-12-24 17:11:51 +01:00
|
|
|
break;
|
|
|
|
case READ_10:
|
2009-11-26 15:33:55 +01:00
|
|
|
req->cmd.buf[1] &= ~0x08; /* disable FUA */
|
2008-09-23 15:38:44 +02:00
|
|
|
break;
|
|
|
|
case REWIND:
|
|
|
|
case START_STOP:
|
2009-11-26 15:33:55 +01:00
|
|
|
if (req->dev->type == TYPE_TAPE) {
|
|
|
|
/* force IMMED, otherwise qemu waits end of command */
|
|
|
|
req->cmd.buf[1] = 0x01;
|
|
|
|
}
|
2008-09-23 15:38:44 +02:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-12-24 17:11:51 +01:00
|
|
|
/* Execute a scsi command. Returns the length of the data expected by the
|
|
|
|
command. This will be Positive for data transfers from the device
|
|
|
|
(eg. disk reads), negative for transfers to the device (eg. disk writes),
|
|
|
|
and zero if the command does not transfer any data. */
|
|
|
|
|
2011-04-18 12:35:39 +02:00
|
|
|
static int32_t scsi_send_command(SCSIRequest *req, uint8_t *cmd)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2011-04-18 12:35:39 +02:00
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, req->dev);
|
|
|
|
SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
|
2007-12-24 17:11:51 +01:00
|
|
|
int ret;
|
|
|
|
|
2011-05-19 16:47:28 +02:00
|
|
|
if (cmd[0] != REQUEST_SENSE && req->lun != s->lun) {
|
|
|
|
DPRINTF("Unimplemented LUN %d\n", req->lun);
|
2011-04-18 12:53:14 +02:00
|
|
|
scsi_set_sense(s, SENSE_CODE(LUN_NOT_SUPPORTED));
|
2011-08-03 10:49:06 +02:00
|
|
|
scsi_req_complete(&r->req, CHECK_CONDITION);
|
2008-10-17 10:08:56 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-11-26 15:33:55 +01:00
|
|
|
if (-1 == scsi_req_parse(&r->req, cmd)) {
|
|
|
|
BADF("Unsupported command length, command %x\n", cmd[0]);
|
2011-04-18 12:53:14 +02:00
|
|
|
scsi_command_complete(r, -EINVAL);
|
2009-11-26 15:33:55 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
scsi_req_fixup(&r->req);
|
|
|
|
|
2010-09-01 16:33:21 +02:00
|
|
|
DPRINTF("Command: lun=%d tag=0x%x len %zd data=0x%02x", lun, tag,
|
|
|
|
r->req.cmd.xfer, cmd[0]);
|
|
|
|
|
|
|
|
#ifdef DEBUG_SCSI
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
for (i = 1; i < r->req.cmd.len; i++) {
|
|
|
|
printf(" 0x%02x", cmd[i]);
|
|
|
|
}
|
|
|
|
printf("\n");
|
|
|
|
}
|
|
|
|
#endif
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2009-11-26 15:33:55 +01:00
|
|
|
if (r->req.cmd.xfer == 0) {
|
2007-12-24 17:11:51 +01:00
|
|
|
if (r->buf != NULL)
|
2009-11-15 19:18:18 +01:00
|
|
|
qemu_free(r->buf);
|
2007-12-24 17:11:51 +01:00
|
|
|
r->buflen = 0;
|
|
|
|
r->buf = NULL;
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
ret = execute_command(s->bs, r, SG_DXFER_NONE, scsi_command_complete);
|
2011-04-18 12:53:14 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
scsi_command_complete(r, ret);
|
|
|
|
return 0;
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-11-26 15:33:55 +01:00
|
|
|
if (r->buflen != r->req.cmd.xfer) {
|
2007-12-24 17:11:51 +01:00
|
|
|
if (r->buf != NULL)
|
2009-11-15 19:18:18 +01:00
|
|
|
qemu_free(r->buf);
|
2009-11-26 15:33:55 +01:00
|
|
|
r->buf = qemu_malloc(r->req.cmd.xfer);
|
|
|
|
r->buflen = r->req.cmd.xfer;
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
memset(r->buf, 0, r->buflen);
|
2009-11-26 15:33:55 +01:00
|
|
|
r->len = r->req.cmd.xfer;
|
2009-11-26 15:33:57 +01:00
|
|
|
if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
|
2007-12-24 17:11:51 +01:00
|
|
|
r->len = 0;
|
2011-04-18 12:35:39 +02:00
|
|
|
return -r->req.cmd.xfer;
|
2011-04-18 16:01:56 +02:00
|
|
|
} else {
|
2011-04-18 12:35:39 +02:00
|
|
|
return r->req.cmd.xfer;
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static int get_blocksize(BlockDriverState *bdrv)
|
|
|
|
{
|
|
|
|
uint8_t cmd[10];
|
|
|
|
uint8_t buf[8];
|
|
|
|
uint8_t sensebuf[8];
|
|
|
|
sg_io_hdr_t io_header;
|
|
|
|
int ret;
|
|
|
|
|
2008-09-22 17:27:54 +02:00
|
|
|
memset(cmd, 0, sizeof(cmd));
|
|
|
|
memset(buf, 0, sizeof(buf));
|
scsi: Sanitize command definitions
Sanitize SCSI command definitions.
Add _10 suffix to READ_CAPACITY, WRITE_VERIFY, VERIFY, READ_LONG,
WRITE_LONG, and WRITE_SAME.
Add new command definitions for LOCATE_10, UNMAP, VARLENGTH_CDB,
WRITE_FILEMARKS_16, EXTENDED_COPY, ATA_PASSTHROUGH, ACCESS_CONTROL_IN,
ACCESS_CONTROL_OUT, COMPARE_AND_WRITE, VERIFY_16, SYNCHRONIZE_CACHE_16,
LOCATE_16, ERASE_16, WRITE_LONG_16, LOAD_UNLOAD, VERIFY_12.
Remove invalid definition of WRITE_LONG_2.
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2011-07-22 16:51:15 +02:00
|
|
|
cmd[0] = READ_CAPACITY_10;
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
memset(&io_header, 0, sizeof(io_header));
|
|
|
|
io_header.interface_id = 'S';
|
|
|
|
io_header.dxfer_direction = SG_DXFER_FROM_DEV;
|
|
|
|
io_header.dxfer_len = sizeof(buf);
|
|
|
|
io_header.dxferp = buf;
|
|
|
|
io_header.cmdp = cmd;
|
|
|
|
io_header.cmd_len = sizeof(cmd);
|
|
|
|
io_header.mx_sb_len = sizeof(sensebuf);
|
|
|
|
io_header.sbp = sensebuf;
|
|
|
|
io_header.timeout = 6000; /* XXX */
|
|
|
|
|
2009-03-28 18:28:41 +01:00
|
|
|
ret = bdrv_ioctl(bdrv, SG_IO, &io_header);
|
2009-03-12 20:57:08 +01:00
|
|
|
if (ret < 0)
|
2007-12-24 17:11:51 +01:00
|
|
|
return -1;
|
|
|
|
|
|
|
|
return (buf[4] << 24) | (buf[5] << 16) | (buf[6] << 8) | buf[7];
|
|
|
|
}
|
|
|
|
|
2008-10-17 10:08:56 +02:00
|
|
|
static int get_stream_blocksize(BlockDriverState *bdrv)
|
|
|
|
{
|
|
|
|
uint8_t cmd[6];
|
|
|
|
uint8_t buf[12];
|
|
|
|
uint8_t sensebuf[8];
|
|
|
|
sg_io_hdr_t io_header;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
memset(cmd, 0, sizeof(cmd));
|
|
|
|
memset(buf, 0, sizeof(buf));
|
|
|
|
cmd[0] = MODE_SENSE;
|
|
|
|
cmd[4] = sizeof(buf);
|
|
|
|
|
|
|
|
memset(&io_header, 0, sizeof(io_header));
|
|
|
|
io_header.interface_id = 'S';
|
|
|
|
io_header.dxfer_direction = SG_DXFER_FROM_DEV;
|
|
|
|
io_header.dxfer_len = sizeof(buf);
|
|
|
|
io_header.dxferp = buf;
|
|
|
|
io_header.cmdp = cmd;
|
|
|
|
io_header.cmd_len = sizeof(cmd);
|
|
|
|
io_header.mx_sb_len = sizeof(sensebuf);
|
|
|
|
io_header.sbp = sensebuf;
|
|
|
|
io_header.timeout = 6000; /* XXX */
|
|
|
|
|
2009-03-28 18:28:41 +01:00
|
|
|
ret = bdrv_ioctl(bdrv, SG_IO, &io_header);
|
2009-03-12 20:57:08 +01:00
|
|
|
if (ret < 0)
|
2008-10-17 10:08:56 +02:00
|
|
|
return -1;
|
|
|
|
|
|
|
|
return (buf[9] << 16) | (buf[10] << 8) | buf[11];
|
|
|
|
}
|
|
|
|
|
2010-09-06 16:07:33 +02:00
|
|
|
static void scsi_generic_reset(DeviceState *dev)
|
|
|
|
{
|
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev.qdev, dev);
|
|
|
|
|
2011-04-18 18:47:12 +02:00
|
|
|
scsi_device_purge_requests(&s->qdev);
|
2010-09-06 16:07:33 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static void scsi_destroy(SCSIDevice *d)
|
|
|
|
{
|
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, d);
|
|
|
|
|
2011-04-18 18:47:12 +02:00
|
|
|
scsi_device_purge_requests(&s->qdev);
|
2010-05-05 16:36:52 +02:00
|
|
|
blockdev_mark_auto_del(s->qdev.conf.bs);
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
|
|
|
|
2009-08-31 14:24:04 +02:00
|
|
|
static int scsi_generic_initfn(SCSIDevice *dev)
|
2007-12-24 17:11:51 +01:00
|
|
|
{
|
2009-08-31 14:24:04 +02:00
|
|
|
SCSIGenericState *s = DO_UPCAST(SCSIGenericState, qdev, dev);
|
2007-12-24 17:11:51 +01:00
|
|
|
int sg_version;
|
|
|
|
struct sg_scsi_id scsiid;
|
|
|
|
|
2010-05-05 16:36:52 +02:00
|
|
|
if (!s->qdev.conf.bs) {
|
2010-02-18 17:25:24 +01:00
|
|
|
error_report("scsi-generic: drive property not set");
|
2009-08-31 14:24:04 +02:00
|
|
|
return -1;
|
|
|
|
}
|
2010-05-05 16:36:52 +02:00
|
|
|
s->bs = s->qdev.conf.bs;
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2009-08-31 14:24:04 +02:00
|
|
|
/* check we are really using a /dev/sg* file */
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
if (!bdrv_is_sg(s->bs)) {
|
2010-02-18 17:25:24 +01:00
|
|
|
error_report("scsi-generic: not /dev/sg*");
|
2009-08-31 14:24:04 +02:00
|
|
|
return -1;
|
|
|
|
}
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2010-05-27 20:02:28 +02:00
|
|
|
if (bdrv_get_on_error(s->bs, 0) != BLOCK_ERR_STOP_ENOSPC) {
|
|
|
|
error_report("Device doesn't support drive option werror");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
if (bdrv_get_on_error(s->bs, 1) != BLOCK_ERR_REPORT) {
|
|
|
|
error_report("Device doesn't support drive option rerror");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2007-12-24 17:11:51 +01:00
|
|
|
/* check we are using a driver managing SG_IO (version 3 and after */
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
if (bdrv_ioctl(s->bs, SG_GET_VERSION_NUM, &sg_version) < 0 ||
|
2009-08-31 14:24:04 +02:00
|
|
|
sg_version < 30000) {
|
2010-02-18 17:25:24 +01:00
|
|
|
error_report("scsi-generic: scsi generic interface too old");
|
2009-08-31 14:24:04 +02:00
|
|
|
return -1;
|
|
|
|
}
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
/* get LUN of the /dev/sg? */
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
if (bdrv_ioctl(s->bs, SG_GET_SCSI_ID, &scsiid)) {
|
2010-02-18 17:25:24 +01:00
|
|
|
error_report("scsi-generic: SG_GET_SCSI_ID ioctl failed");
|
2009-08-31 14:24:04 +02:00
|
|
|
return -1;
|
|
|
|
}
|
2007-12-24 17:11:51 +01:00
|
|
|
|
|
|
|
/* define device state */
|
|
|
|
s->lun = scsiid.lun;
|
2008-10-17 10:08:56 +02:00
|
|
|
DPRINTF("LUN %d\n", s->lun);
|
2009-11-26 15:33:54 +01:00
|
|
|
s->qdev.type = scsiid.scsi_type;
|
|
|
|
DPRINTF("device type %d\n", s->qdev.type);
|
|
|
|
if (s->qdev.type == TYPE_TAPE) {
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
s->qdev.blocksize = get_stream_blocksize(s->bs);
|
2009-11-26 15:33:52 +01:00
|
|
|
if (s->qdev.blocksize == -1)
|
|
|
|
s->qdev.blocksize = 0;
|
2008-10-17 10:08:56 +02:00
|
|
|
} else {
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
s->qdev.blocksize = get_blocksize(s->bs);
|
2008-10-17 10:08:56 +02:00
|
|
|
/* removable media returns 0 if not present */
|
2009-11-26 15:33:52 +01:00
|
|
|
if (s->qdev.blocksize <= 0) {
|
2009-11-26 15:33:54 +01:00
|
|
|
if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM)
|
2009-11-26 15:33:52 +01:00
|
|
|
s->qdev.blocksize = 2048;
|
2008-10-17 10:08:56 +02:00
|
|
|
else
|
2009-11-26 15:33:52 +01:00
|
|
|
s->qdev.blocksize = 512;
|
2008-10-17 10:08:56 +02:00
|
|
|
}
|
|
|
|
}
|
2009-11-26 15:33:52 +01:00
|
|
|
DPRINTF("block size %d\n", s->qdev.blocksize);
|
2007-12-24 17:11:51 +01:00
|
|
|
s->driver_status = 0;
|
|
|
|
memset(s->sensebuf, 0, sizeof(s->sensebuf));
|
block: Fix virtual media change for if=none
BlockDriverState member removable controls whether virtual media
change (monitor commands change, eject) is allowed. It is set when
the "type hint" is BDRV_TYPE_CDROM or BDRV_TYPE_FLOPPY.
The type hint is only set by drive_init(). It sets BDRV_TYPE_FLOPPY
for if=floppy. It sets BDRV_TYPE_CDROM for media=cdrom and if=ide,
scsi, xen, or none.
if=ide and if=scsi work, because the type hint makes it a CD-ROM.
if=xen likewise, I think.
For the same reason, if=none works when it's used by ide-drive or
scsi-disk. For other guest devices, there are problems:
* fdc: you can't change virtual media
$ qemu [...] -drive if=none,id=foo,... -global isa-fdc.driveA=foo
QEMU 0.12.50 monitor - type 'help' for more information
(qemu) eject foo
Device 'foo' is not removable
unless you add media=cdrom, but that makes it readonly.
* virtio: if you add media=cdrom, you can change virtual media. If
you eject, the guest gets I/O errors. If you change, the guest sees
the drive's contents suddenly change.
* scsi-generic: if you add media=cdrom, you can change virtual media.
I didn't test what that does to the guest or the physical device,
but it can't be pretty.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2010-06-25 13:42:14 +02:00
|
|
|
bdrv_set_removable(s->bs, 0);
|
2009-08-31 14:24:04 +02:00
|
|
|
return 0;
|
|
|
|
}
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2009-08-31 14:24:04 +02:00
|
|
|
static SCSIDeviceInfo scsi_generic_info = {
|
|
|
|
.qdev.name = "scsi-generic",
|
|
|
|
.qdev.desc = "pass through generic scsi device (/dev/sg*)",
|
|
|
|
.qdev.size = sizeof(SCSIGenericState),
|
2010-09-06 16:07:33 +02:00
|
|
|
.qdev.reset = scsi_generic_reset,
|
2009-08-31 14:24:04 +02:00
|
|
|
.init = scsi_generic_initfn,
|
|
|
|
.destroy = scsi_destroy,
|
2011-04-18 12:35:39 +02:00
|
|
|
.alloc_req = scsi_new_request,
|
2011-04-18 16:01:56 +02:00
|
|
|
.free_req = scsi_free_request,
|
2009-08-31 14:24:04 +02:00
|
|
|
.send_command = scsi_send_command,
|
|
|
|
.read_data = scsi_read_data,
|
|
|
|
.write_data = scsi_write_data,
|
|
|
|
.cancel_io = scsi_cancel_io,
|
|
|
|
.get_buf = scsi_get_buf,
|
2011-04-18 13:36:02 +02:00
|
|
|
.get_sense = scsi_get_sense,
|
2009-08-31 14:24:04 +02:00
|
|
|
.qdev.props = (Property[]) {
|
block: add topology qdev properties
Add three new qdev properties to export block topology information to
the guest. This is needed to get optimal I/O alignment for RAID arrays
or SSDs.
The options are:
- physical_block_size to specify the physical block size of the device,
this is going to increase from 512 bytes to 4096 kilobytes for many
modern storage devices
- min_io_size to specify the minimal I/O size without performance impact,
this is typically set to the RAID chunk size for arrays.
- opt_io_size to specify the optimal sustained I/O size, this is
typically the RAID stripe width for arrays.
I decided to not auto-probe these values from blkid which might easily
be possible as I don't know how to deal with these issues on migration.
Note that we specificly only set the physical_block_size, and not the
logial one which is the unit all I/O is described in. The reason for
that is that IDE does not support increasing the logical block size and
at last for now I want to stick to one meachnisms in queue and allow
for easy switching of transports for a given backing image which would
not be possible if scsi and virtio use real 4k sectors, while ide only
uses the physical block exponent.
To make this more common for the different block drivers introduce a
new BlockConf structure holding all common block properties and a
DEFINE_BLOCK_PROPERTIES macro to add them all together, mirroring
what is done for network drivers. Also switch over all block drivers
to use it, except for the floppy driver which has weird driveA/driveB
properties and probably won't require any advanced block options ever.
Example usage for a virtio device with 4k physical block size and
8k optimal I/O size:
-drive file=scratch.img,media=disk,cache=none,id=scratch \
-device virtio-blk-pci,drive=scratch,physical_block_size=4096,opt_io_size=8192
aliguori: updated patch to take into account BLOCK events
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
2010-02-10 23:37:09 +01:00
|
|
|
DEFINE_BLOCK_PROPERTIES(SCSIGenericState, qdev.conf),
|
2009-08-31 14:24:04 +02:00
|
|
|
DEFINE_PROP_END_OF_LIST(),
|
|
|
|
},
|
|
|
|
};
|
2007-12-24 17:11:51 +01:00
|
|
|
|
2009-08-31 14:24:04 +02:00
|
|
|
static void scsi_generic_register_devices(void)
|
|
|
|
{
|
|
|
|
scsi_qdev_register(&scsi_generic_info);
|
2007-12-24 17:11:51 +01:00
|
|
|
}
|
2009-08-31 14:24:04 +02:00
|
|
|
device_init(scsi_generic_register_devices)
|
|
|
|
|
2007-12-24 17:11:51 +01:00
|
|
|
#endif /* __linux__ */
|