提交 0e7106d8 编写于 作者: L Liu Yuan 提交者: Kevin Wolf

sheepdog: implement direct write semantics

Sheepdog supports both writeback/writethrough write but has not yet supported
DIRECTIO semantics which bypass the cache completely even if Sheepdog daemon is
set up with cache enabled.

Suppose cache is enabled on Sheepdog daemon size, the new cache control is

cache=writeback # enable the writeback semantics for write
cache=writethrough # enable the emulated writethrough semantics for write
cache=directsync # disable cache competely

Guest WCE toggling on the run time to toggle writeback/writethrough is also
supported.

Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@gmail.com>
Signed-off-by: NLiu Yuan <tailai.ly@taobao.com>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: NMORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: NKevin Wolf <kwolf@redhat.com>
上级 8e895599
...@@ -36,7 +36,8 @@ ...@@ -36,7 +36,8 @@
#define SD_FLAG_CMD_WRITE 0x01 #define SD_FLAG_CMD_WRITE 0x01
#define SD_FLAG_CMD_COW 0x02 #define SD_FLAG_CMD_COW 0x02
#define SD_FLAG_CMD_CACHE 0x04 #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */
#define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */
#define SD_RES_SUCCESS 0x00 /* Success */ #define SD_RES_SUCCESS 0x00 /* Success */
#define SD_RES_UNKNOWN 0x01 /* Unknown error */ #define SD_RES_UNKNOWN 0x01 /* Unknown error */
...@@ -293,7 +294,7 @@ typedef struct BDRVSheepdogState { ...@@ -293,7 +294,7 @@ typedef struct BDRVSheepdogState {
char name[SD_MAX_VDI_LEN]; char name[SD_MAX_VDI_LEN];
bool is_snapshot; bool is_snapshot;
bool cache_enabled; uint32_t cache_flags;
char *addr; char *addr;
char *port; char *port;
...@@ -977,8 +978,8 @@ static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, ...@@ -977,8 +978,8 @@ static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
hdr.flags = SD_FLAG_CMD_WRITE | flags; hdr.flags = SD_FLAG_CMD_WRITE | flags;
} }
if (s->cache_enabled) { if (s->cache_flags) {
hdr.flags |= SD_FLAG_CMD_CACHE; hdr.flags |= s->cache_flags;
} }
hdr.oid = oid; hdr.oid = oid;
...@@ -1023,7 +1024,7 @@ static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, ...@@ -1023,7 +1024,7 @@ static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
static int read_write_object(int fd, char *buf, uint64_t oid, int copies, static int read_write_object(int fd, char *buf, uint64_t oid, int copies,
unsigned int datalen, uint64_t offset, unsigned int datalen, uint64_t offset,
bool write, bool create, bool cache) bool write, bool create, uint32_t cache_flags)
{ {
SheepdogObjReq hdr; SheepdogObjReq hdr;
SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
...@@ -1047,9 +1048,7 @@ static int read_write_object(int fd, char *buf, uint64_t oid, int copies, ...@@ -1047,9 +1048,7 @@ static int read_write_object(int fd, char *buf, uint64_t oid, int copies,
hdr.opcode = SD_OP_READ_OBJ; hdr.opcode = SD_OP_READ_OBJ;
} }
if (cache) { hdr.flags |= cache_flags;
hdr.flags |= SD_FLAG_CMD_CACHE;
}
hdr.oid = oid; hdr.oid = oid;
hdr.data_length = datalen; hdr.data_length = datalen;
...@@ -1072,18 +1071,19 @@ static int read_write_object(int fd, char *buf, uint64_t oid, int copies, ...@@ -1072,18 +1071,19 @@ static int read_write_object(int fd, char *buf, uint64_t oid, int copies,
} }
static int read_object(int fd, char *buf, uint64_t oid, int copies, static int read_object(int fd, char *buf, uint64_t oid, int copies,
unsigned int datalen, uint64_t offset, bool cache) unsigned int datalen, uint64_t offset,
uint32_t cache_flags)
{ {
return read_write_object(fd, buf, oid, copies, datalen, offset, false, return read_write_object(fd, buf, oid, copies, datalen, offset, false,
false, cache); false, cache_flags);
} }
static int write_object(int fd, char *buf, uint64_t oid, int copies, static int write_object(int fd, char *buf, uint64_t oid, int copies,
unsigned int datalen, uint64_t offset, bool create, unsigned int datalen, uint64_t offset, bool create,
bool cache) uint32_t cache_flags)
{ {
return read_write_object(fd, buf, oid, copies, datalen, offset, true, return read_write_object(fd, buf, oid, copies, datalen, offset, true,
create, cache); create, cache_flags);
} }
static int sd_open(BlockDriverState *bs, const char *filename, int flags) static int sd_open(BlockDriverState *bs, const char *filename, int flags)
...@@ -1118,12 +1118,22 @@ static int sd_open(BlockDriverState *bs, const char *filename, int flags) ...@@ -1118,12 +1118,22 @@ static int sd_open(BlockDriverState *bs, const char *filename, int flags)
goto out; goto out;
} }
s->cache_enabled = true; /*
s->flush_fd = connect_to_sdog(s->addr, s->port); * QEMU block layer emulates writethrough cache as 'writeback + flush', so
if (s->flush_fd < 0) { * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
error_report("failed to connect"); */
ret = s->flush_fd; s->cache_flags = SD_FLAG_CMD_CACHE;
goto out; if (flags & BDRV_O_NOCACHE) {
s->cache_flags = SD_FLAG_CMD_DIRECT;
}
if (s->cache_flags == SD_FLAG_CMD_CACHE) {
s->flush_fd = connect_to_sdog(s->addr, s->port);
if (s->flush_fd < 0) {
error_report("failed to connect");
ret = s->flush_fd;
goto out;
}
} }
if (snapid || tag[0] != '\0') { if (snapid || tag[0] != '\0') {
...@@ -1140,7 +1150,7 @@ static int sd_open(BlockDriverState *bs, const char *filename, int flags) ...@@ -1140,7 +1150,7 @@ static int sd_open(BlockDriverState *bs, const char *filename, int flags)
buf = g_malloc(SD_INODE_SIZE); buf = g_malloc(SD_INODE_SIZE);
ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0,
s->cache_enabled); s->cache_flags);
closesocket(fd); closesocket(fd);
...@@ -1387,7 +1397,7 @@ static void sd_close(BlockDriverState *bs) ...@@ -1387,7 +1397,7 @@ static void sd_close(BlockDriverState *bs)
qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL);
closesocket(s->fd); closesocket(s->fd);
if (s->cache_enabled) { if (s->cache_flags) {
closesocket(s->flush_fd); closesocket(s->flush_fd);
} }
g_free(s->addr); g_free(s->addr);
...@@ -1423,7 +1433,7 @@ static int sd_truncate(BlockDriverState *bs, int64_t offset) ...@@ -1423,7 +1433,7 @@ static int sd_truncate(BlockDriverState *bs, int64_t offset)
datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
s->inode.vdi_size = offset; s->inode.vdi_size = offset;
ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
s->inode.nr_copies, datalen, 0, false, s->cache_enabled); s->inode.nr_copies, datalen, 0, false, s->cache_flags);
close(fd); close(fd);
if (ret < 0) { if (ret < 0) {
...@@ -1506,7 +1516,7 @@ static int sd_create_branch(BDRVSheepdogState *s) ...@@ -1506,7 +1516,7 @@ static int sd_create_branch(BDRVSheepdogState *s)
} }
ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies,
SD_INODE_SIZE, 0, s->cache_enabled); SD_INODE_SIZE, 0, s->cache_flags);
closesocket(fd); closesocket(fd);
...@@ -1707,7 +1717,7 @@ static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) ...@@ -1707,7 +1717,7 @@ static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
int ret; int ret;
unsigned int wlen = 0, rlen = 0; unsigned int wlen = 0, rlen = 0;
if (!s->cache_enabled) { if (s->cache_flags != SD_FLAG_CMD_CACHE) {
return 0; return 0;
} }
...@@ -1723,7 +1733,7 @@ static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) ...@@ -1723,7 +1733,7 @@ static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
if (rsp->result == SD_RES_INVALID_PARMS) { if (rsp->result == SD_RES_INVALID_PARMS) {
dprintf("disable write cache since the server doesn't support it\n"); dprintf("disable write cache since the server doesn't support it\n");
s->cache_enabled = false; s->cache_flags = SD_FLAG_CMD_DIRECT;
closesocket(s->flush_fd); closesocket(s->flush_fd);
return 0; return 0;
} }
...@@ -1774,7 +1784,7 @@ static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) ...@@ -1774,7 +1784,7 @@ static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
} }
ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
s->inode.nr_copies, datalen, 0, false, s->cache_enabled); s->inode.nr_copies, datalen, 0, false, s->cache_flags);
if (ret < 0) { if (ret < 0) {
error_report("failed to write snapshot's inode."); error_report("failed to write snapshot's inode.");
goto cleanup; goto cleanup;
...@@ -1791,7 +1801,7 @@ static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) ...@@ -1791,7 +1801,7 @@ static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
inode = (SheepdogInode *)g_malloc(datalen); inode = (SheepdogInode *)g_malloc(datalen);
ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid), ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid),
s->inode.nr_copies, datalen, 0, s->cache_enabled); s->inode.nr_copies, datalen, 0, s->cache_flags);
if (ret < 0) { if (ret < 0) {
error_report("failed to read new inode info. %s", strerror(errno)); error_report("failed to read new inode info. %s", strerror(errno));
...@@ -1845,7 +1855,7 @@ static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) ...@@ -1845,7 +1855,7 @@ static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
buf = g_malloc(SD_INODE_SIZE); buf = g_malloc(SD_INODE_SIZE);
ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies,
SD_INODE_SIZE, 0, s->cache_enabled); SD_INODE_SIZE, 0, s->cache_flags);
closesocket(fd); closesocket(fd);
...@@ -1942,7 +1952,7 @@ static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) ...@@ -1942,7 +1952,7 @@ static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
/* we don't need to read entire object */ /* we don't need to read entire object */
ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid), ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid),
0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0,
s->cache_enabled); s->cache_flags);
if (ret) { if (ret) {
continue; continue;
...@@ -2003,11 +2013,11 @@ static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, ...@@ -2003,11 +2013,11 @@ static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
if (load) { if (load) {
ret = read_object(fd, (char *)data, vmstate_oid, ret = read_object(fd, (char *)data, vmstate_oid,
s->inode.nr_copies, data_len, offset, s->inode.nr_copies, data_len, offset,
s->cache_enabled); s->cache_flags);
} else { } else {
ret = write_object(fd, (char *)data, vmstate_oid, ret = write_object(fd, (char *)data, vmstate_oid,
s->inode.nr_copies, data_len, offset, create, s->inode.nr_copies, data_len, offset, create,
s->cache_enabled); s->cache_flags);
} }
if (ret < 0) { if (ret < 0) {
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册