@@ -2038,6 +2038,11 @@ int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
goto error;
}
}
+ if (!reopen_state->bs->blk && !(reopen_state->flags & BDRV_O_CACHE_WB)) {
+ error_setg(errp, "Cannot disable cache.writeback: No BlockBackend");
+ ret = -EINVAL;
+ goto error;
+ }
/* node-name and driver must be unchanged. Put them back into the QDict, so
* that they are checked at the end of this function. */
@@ -2138,10 +2143,10 @@ void bdrv_reopen_commit(BDRVReopenState *reopen_state)
reopen_state->bs->explicit_options = reopen_state->explicit_options;
reopen_state->bs->open_flags = reopen_state->flags;
- reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
- BDRV_O_CACHE_WB);
reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
+ bdrv_set_enable_write_cache(reopen_state->bs,
+ !!(reopen_state->flags & BDRV_O_CACHE_WB));
bdrv_refresh_limits(reopen_state->bs, NULL);
}
@@ -2271,9 +2276,6 @@ static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
BlockDriverState *bs_src)
{
/* move some fields that need to stay attached to the device */
-
- /* dev info */
- bs_dest->enable_write_cache = bs_src->enable_write_cache;
}
static void change_parent_backing_link(BlockDriverState *from,
@@ -2753,12 +2755,18 @@ int bdrv_is_sg(BlockDriverState *bs)
int bdrv_enable_write_cache(BlockDriverState *bs)
{
- return bs->enable_write_cache;
+ if (bs->blk) {
+ return blk_enable_write_cache(bs->blk);
+ } else {
+ return true;
+ }
}
void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
{
- bs->enable_write_cache = wce;
+ if (bs->blk) {
+ blk_set_enable_write_cache(bs->blk, wce);
+ }
/* so a reopen() will preserve wce */
if (wce) {
@@ -3618,8 +3626,8 @@ void bdrv_img_create(const char *filename, const char *fmt,
}
/* backing files always opened read-only */
- back_flags =
- flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
+ back_flags = flags | BDRV_O_CACHE_WB;
+ back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
if (backing_fmt) {
backing_options = qdict_new();
@@ -46,6 +46,8 @@ struct BlockBackend {
* can be used to restore those options in the new BDS on insert) */
BlockBackendRootState root_state;
+ bool enable_write_cache;
+
/* I/O stats (display with "info blockstats"). */
BlockAcctStats stats;
@@ -715,11 +717,17 @@ static int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
unsigned int bytes, QEMUIOVector *qiov,
BdrvRequestFlags flags)
{
- int ret = blk_check_byte_request(blk, offset, bytes);
+ int ret;
+
+ ret = blk_check_byte_request(blk, offset, bytes);
if (ret < 0) {
return ret;
}
+ if (!blk->enable_write_cache) {
+ flags |= BDRV_REQ_FUA;
+ }
+
return bdrv_co_do_pwritev(blk_bs(blk), offset, bytes, qiov, flags);
}
@@ -1226,26 +1234,19 @@ int blk_is_sg(BlockBackend *blk)
int blk_enable_write_cache(BlockBackend *blk)
{
- BlockDriverState *bs = blk_bs(blk);
-
- if (bs) {
- return bdrv_enable_write_cache(bs);
- } else {
- return !!(blk->root_state.open_flags & BDRV_O_CACHE_WB);
- }
+ return blk->enable_write_cache;
}
void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
{
- BlockDriverState *bs = blk_bs(blk);
+ blk->enable_write_cache = wce;
- if (bs) {
- bdrv_set_enable_write_cache(bs, wce);
- } else {
+ /* TODO Remove this when BDRV_O_CACHE_WB isn't used any more */
+ if (blk->root) {
if (wce) {
- blk->root_state.open_flags |= BDRV_O_CACHE_WB;
+ blk->root->bs->open_flags |= BDRV_O_CACHE_WB;
} else {
- blk->root_state.open_flags &= ~BDRV_O_CACHE_WB;
+ blk->root->bs->open_flags &= ~BDRV_O_CACHE_WB;
}
}
}
@@ -1508,11 +1509,22 @@ int blk_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
int64_t pos, int size)
{
+ int ret;
+
if (!blk_is_available(blk)) {
return -ENOMEDIUM;
}
- return bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
+ ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
+ if (ret < 0) {
+ return ret;
+ }
+
+ if (ret == size && !blk->enable_write_cache) {
+ ret = bdrv_flush(blk_bs(blk));
+ }
+
+ return ret < 0 ? ret : size;
}
int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
@@ -1156,7 +1156,7 @@ static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
}
bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
- if (ret == 0 && !bs->enable_write_cache) {
+ if (ret == 0 && (flags & BDRV_REQ_FUA)) {
ret = bdrv_co_flush(bs);
}
@@ -476,7 +476,7 @@ static int coroutine_fn iscsi_co_writev(BlockDriverState *bs,
num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
iscsi_co_init_iscsitask(iscsilun, &iTask);
retry:
- fua = iscsilun->dpofua && !bs->enable_write_cache;
+ fua = iscsilun->dpofua && !bdrv_enable_write_cache(bs);
iTask.force_next_flush = !fua;
if (iscsilun->use_16_for_rw) {
iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
@@ -64,6 +64,7 @@ typedef enum {
*/
BDRV_REQ_MAY_UNMAP = 0x4,
BDRV_REQ_NO_SERIALISING = 0x8,
+ BDRV_REQ_FUA = 0x10,
} BdrvRequestFlags;
typedef struct BlockSizes {
@@ -435,9 +435,6 @@ struct BlockDriverState {
/* Alignment requirement for offset/length of I/O requests */
unsigned int request_alignment;
- /* do we need to tell the quest if we have a volatile write cache? */
- int enable_write_cache;
-
/* the following member gives a name to every node on the bs graph. */
char node_name[32];
/* element of the list of named nodes building the graph */
@@ -338,8 +338,8 @@ echo
# TODO Implement node-name support for 'qemu-io' HMP command for -c
# Can use only -o to access child node options for now
-hmp_cmds="qemu-io none0 \"reopen -o file.cache.writeback=off,file.cache.direct=off,file.cache.no-flush=off\"
-qemu-io none0 \"reopen -o backing.file.cache.writeback=on,backing.file.cache.direct=off,backing.file.cache.no-flush=on\"
+hmp_cmds="qemu-io none0 \"reopen -o file.cache.direct=off,file.cache.no-flush=off\"
+qemu-io none0 \"reopen -o backing.file.cache.direct=off,backing.file.cache.no-flush=on\"
qemu-io none0 \"reopen -c none\"
info block image
info block file
@@ -132,7 +132,7 @@ cache.direct=on on backing-file
cache.writeback=off on none0
Cache mode: writethrough
- Cache mode: writethrough
+ Cache mode: writeback
Cache mode: writeback
Cache mode: writeback
Cache mode: writeback
@@ -342,7 +342,7 @@ cache.direct=on on backing-file
cache.writeback=off on none0
Cache mode: writeback, direct
- Cache mode: writethrough
+ Cache mode: writeback
Cache mode: writeback
Cache mode: writeback
Cache mode: writeback
@@ -503,7 +503,7 @@ cache.direct=on on backing-file
cache.writeback=off on blk
Cache mode: writethrough
- Cache mode: writethrough
+ Cache mode: writeback
Cache mode: writeback
Cache mode: writeback
Cache mode: writeback
@@ -707,7 +707,7 @@ cache.no-flush=on on backing-file
--- Change cache mode after reopening child ---
Cache mode: writeback, direct
- Cache mode: writethrough
+ Cache mode: writeback
Cache mode: writeback, direct
Cache mode: writeback, ignore flushes
*** done
Whether a write cache is used or not is a decision that concerns the user (e.g. the guest device) rather than the backend. It was already logically part of the BB level as bdrv_move_feature_fields() always kept it on top of the BDS tree; with this patch, the core of it (the actual flag and the additional flushes) is also implemented there. Direct callers of bdrv_open() must pass BDRV_O_CACHE_WB now if bs doesn't have a BlockBackend attached. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block.c | 26 +++++++++++++++++--------- block/block-backend.c | 42 +++++++++++++++++++++++++++--------------- block/io.c | 2 +- block/iscsi.c | 2 +- include/block/block.h | 1 + include/block/block_int.h | 3 --- tests/qemu-iotests/142 | 4 ++-- tests/qemu-iotests/142.out | 8 ++++---- 8 files changed, 53 insertions(+), 35 deletions(-)