[Qemu-devel] [PATCH v4 0/3] sheepdog: implement .bdrv_co_is_allocated

[Qemu-devel] [PATCH v4 0/3] sheepdog: implement .bdrv_co_is_allocated

[Liu Yuan]

From: Liu Yuan <tailai.ly@taobao.com>

v4:
 - correct idx iteration range

v3:
 - include discard patch for easy review and merge
 - update the comment typo
 - get variable 'idx' definition more clear

v2:
 - correct 'end' calculation
 - get the longest unallocated area
 - replace SECTOR_SIZE with BDRV_SECTOR_SIZE

PATCH 1/3 is the previous distard patch which implement .bdrv_co_discard
PATCH 2/3 is ia prepare patch
PATCH 3/3 implement .bdrv_co_is_allocated

Thanks,
Yuan
-- 
1.7.9.5

[Qemu-devel] [PATCH v4 1/3] sheepdog: add discard/trim support for sheepdog

[Liu Yuan]

From: Liu Yuan <tailai.ly@taobao.com>

The 'TRIM' command from VM that is to release underlying data storage for
better thin-provision is already supported by the Sheepdog.

This patch adds the TRIM support at QEMU part.

For older Sheepdog that doesn't support it, we return 0(success) to upper layer.

Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Liu Yuan <tailai.ly@taobao.com>
---
 block/sheepdog.c |   64 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 63 insertions(+), 1 deletion(-)

diff --git a/block/sheepdog.c b/block/sheepdog.c
index 1c5b532..c3b7770 100644
--- a/block/sheepdog.c
+++ b/block/sheepdog.c
@@ -27,6 +27,8 @@
 #define SD_OP_CREATE_AND_WRITE_OBJ  0x01
 #define SD_OP_READ_OBJ       0x02
 #define SD_OP_WRITE_OBJ      0x03
+/* 0x04 is used internally by Sheepdog */
+#define SD_OP_DISCARD_OBJ    0x05
 
 #define SD_OP_NEW_VDI        0x11
 #define SD_OP_LOCK_VDI       0x12
@@ -269,6 +271,7 @@ enum AIOCBState {
     AIOCB_WRITE_UDATA,
     AIOCB_READ_UDATA,
     AIOCB_FLUSH_CACHE,
+    AIOCB_DISCARD_OBJ,
 };
 
 struct SheepdogAIOCB {
@@ -298,6 +301,7 @@ typedef struct BDRVSheepdogState {
     char name[SD_MAX_VDI_LEN];
     bool is_snapshot;
     uint32_t cache_flags;
+    bool discard_supported;
 
     char *host_spec;
     bool is_unix;
@@ -656,7 +660,7 @@ static void coroutine_fn aio_read_response(void *opaque)
     int ret;
     AIOReq *aio_req = NULL;
     SheepdogAIOCB *acb;
-    unsigned long idx;
+    uint64_t idx;
 
     if (QLIST_EMPTY(&s->inflight_aio_head)) {
         goto out;
@@ -727,6 +731,21 @@ static void coroutine_fn aio_read_response(void *opaque)
             rsp.result = SD_RES_SUCCESS;
         }
         break;
+    case AIOCB_DISCARD_OBJ:
+        switch (rsp.result) {
+        case SD_RES_INVALID_PARMS:
+            error_report("sheep(%s) doesn't support discard command",
+                         s->host_spec);
+            rsp.result = SD_RES_SUCCESS;
+            s->discard_supported = false;
+            break;
+        case SD_RES_SUCCESS:
+            idx = data_oid_to_idx(aio_req->oid);
+            s->inode.data_vdi_id[idx] = 0;
+            break;
+        default:
+            break;
+        }
     }
 
     if (rsp.result != SD_RES_SUCCESS) {
@@ -1016,6 +1035,9 @@ static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
         wlen = datalen;
         hdr.flags = SD_FLAG_CMD_WRITE | flags;
         break;
+    case AIOCB_DISCARD_OBJ:
+        hdr.opcode = SD_OP_DISCARD_OBJ;
+        break;
     }
 
     if (s->cache_flags) {
@@ -1170,6 +1192,7 @@ static int sd_open(BlockDriverState *bs, const char *filename,
     if (flags & BDRV_O_NOCACHE) {
         s->cache_flags = SD_FLAG_CMD_DIRECT;
     }
+    s->discard_supported = true;
 
     if (snapid || tag[0] != '\0') {
         dprintf("%" PRIx32 " snapshot inode was open.\n", vid);
@@ -1633,6 +1656,15 @@ static int coroutine_fn sd_co_rw_vector(void *p)
                 flags = SD_FLAG_CMD_COW;
             }
             break;
+        case AIOCB_DISCARD_OBJ:
+            /*
+             * We discard the object only when the whole object is
+             * 1) allocated 2) trimmed. Otherwise, simply skip it.
+             */
+            if (len != SD_DATA_OBJ_SIZE || inode->data_vdi_id[idx] == 0) {
+                goto done;
+            }
+            break;
         default:
             break;
         }
@@ -2078,6 +2110,33 @@ static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data,
 }
 
 
+static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num,
+                                      int nb_sectors)
+{
+    SheepdogAIOCB *acb;
+    QEMUIOVector dummy;
+    BDRVSheepdogState *s = bs->opaque;
+    int ret;
+
+    if (!s->discard_supported) {
+            return 0;
+    }
+
+    acb = sd_aio_setup(bs, &dummy, sector_num, nb_sectors);
+    acb->aiocb_type = AIOCB_DISCARD_OBJ;
+    acb->aio_done_func = sd_finish_aiocb;
+
+    ret = sd_co_rw_vector(acb);
+    if (ret <= 0) {
+        qemu_aio_release(acb);
+        return ret;
+    }
+
+    qemu_coroutine_yield();
+
+    return acb->ret;
+}
+
 static QEMUOptionParameter sd_create_options[] = {
     {
         .name = BLOCK_OPT_SIZE,
@@ -2110,6 +2169,7 @@ static BlockDriver bdrv_sheepdog = {
     .bdrv_co_readv  = sd_co_readv,
     .bdrv_co_writev = sd_co_writev,
     .bdrv_co_flush_to_disk  = sd_co_flush_to_disk,
+    .bdrv_co_discard = sd_co_discard,
 
     .bdrv_snapshot_create   = sd_snapshot_create,
     .bdrv_snapshot_goto     = sd_snapshot_goto,
@@ -2135,6 +2195,7 @@ static BlockDriver bdrv_sheepdog_tcp = {
     .bdrv_co_readv  = sd_co_readv,
     .bdrv_co_writev = sd_co_writev,
     .bdrv_co_flush_to_disk  = sd_co_flush_to_disk,
+    .bdrv_co_discard = sd_co_discard,
 
     .bdrv_snapshot_create   = sd_snapshot_create,
     .bdrv_snapshot_goto     = sd_snapshot_goto,
@@ -2160,6 +2221,7 @@ static BlockDriver bdrv_sheepdog_unix = {
     .bdrv_co_readv  = sd_co_readv,
     .bdrv_co_writev = sd_co_writev,
     .bdrv_co_flush_to_disk  = sd_co_flush_to_disk,
+    .bdrv_co_discard = sd_co_discard,
 
     .bdrv_snapshot_create   = sd_snapshot_create,
     .bdrv_snapshot_goto     = sd_snapshot_goto,
-- 
1.7.9.5

[Qemu-devel] [PATCH v4 2/3] sheepdog: use BDRV_SECTOR_SIZE

[Liu Yuan]

From: Liu Yuan <tailai.ly@taobao.com>

Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Liu Yuan <tailai.ly@taobao.com>
---
 block/sheepdog.c |   11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/block/sheepdog.c b/block/sheepdog.c
index c3b7770..916f360 100644
--- a/block/sheepdog.c
+++ b/block/sheepdog.c
@@ -89,7 +89,6 @@
 #define SD_NR_VDIS   (1U << 24)
 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
-#define SECTOR_SIZE 512
 
 #define SD_INODE_SIZE (sizeof(SheepdogInode))
 #define CURRENT_VDI_ID 0
@@ -1219,7 +1218,7 @@ static int sd_open(BlockDriverState *bs, const char *filename,
     s->min_dirty_data_idx = UINT32_MAX;
     s->max_dirty_data_idx = 0;
 
-    bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE;
+    bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE;
     pstrcpy(s->name, sizeof(s->name), vdi);
     qemu_co_mutex_init(&s->lock);
     g_free(buf);
@@ -1604,10 +1603,10 @@ static int coroutine_fn sd_co_rw_vector(void *p)
 {
     SheepdogAIOCB *acb = p;
     int ret = 0;
-    unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE;
-    unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE;
+    unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE;
+    unsigned long idx = acb->sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE;
     uint64_t oid;
-    uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE;
+    uint64_t offset = (acb->sector_num * BDRV_SECTOR_SIZE) % SD_DATA_OBJ_SIZE;
     BDRVSheepdogState *s = acb->common.bs->opaque;
     SheepdogInode *inode = &s->inode;
     AIOReq *aio_req;
@@ -1726,7 +1725,7 @@ static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
     int ret;
 
     if (bs->growable && sector_num + nb_sectors > bs->total_sectors) {
-        ret = sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE);
+        ret = sd_truncate(bs, (sector_num + nb_sectors) * BDRV_SECTOR_SIZE);
         if (ret < 0) {
             return ret;
         }
-- 
1.7.9.5

[Qemu-devel] [PATCH v4 3/3] sheepdog: implement .bdrv_co_is_allocated()

[Liu Yuan]

From: Liu Yuan <tailai.ly@taobao.com>

Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Liu Yuan <tailai.ly@taobao.com>
---
 block/sheepdog.c |   37 +++++++++++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)

diff --git a/block/sheepdog.c b/block/sheepdog.c
index 916f360..2fe0783 100644
--- a/block/sheepdog.c
+++ b/block/sheepdog.c
@@ -2136,6 +2136,40 @@ static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num,
     return acb->ret;
 }
 
+static coroutine_fn int
+sd_co_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
+                   int *pnum)
+{
+    BDRVSheepdogState *s = bs->opaque;
+    SheepdogInode *inode = &s->inode;
+    unsigned long start = sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE,
+                  end = DIV_ROUND_UP((sector_num + nb_sectors) *
+                                     BDRV_SECTOR_SIZE, SD_DATA_OBJ_SIZE);
+    unsigned long idx;
+    int ret = 1;
+
+    for (idx = start; idx < end; idx++) {
+        if (inode->data_vdi_id[idx] == 0) {
+            break;
+        }
+    }
+    if (idx == start) {
+        /* Get the longest length of unallocated sectors */
+        ret = 0;
+        for (idx = start + 1; idx < end; idx++) {
+            if (inode->data_vdi_id[idx] != 0) {
+                break;
+            }
+        }
+    }
+
+    *pnum = (idx - start) * SD_DATA_OBJ_SIZE / BDRV_SECTOR_SIZE;
+    if (*pnum > nb_sectors) {
+        *pnum = nb_sectors;
+    }
+    return ret;
+}
+
 static QEMUOptionParameter sd_create_options[] = {
     {
         .name = BLOCK_OPT_SIZE,
@@ -2169,6 +2203,7 @@ static BlockDriver bdrv_sheepdog = {
     .bdrv_co_writev = sd_co_writev,
     .bdrv_co_flush_to_disk  = sd_co_flush_to_disk,
     .bdrv_co_discard = sd_co_discard,
+    .bdrv_co_is_allocated = sd_co_is_allocated,
 
     .bdrv_snapshot_create   = sd_snapshot_create,
     .bdrv_snapshot_goto     = sd_snapshot_goto,
@@ -2195,6 +2230,7 @@ static BlockDriver bdrv_sheepdog_tcp = {
     .bdrv_co_writev = sd_co_writev,
     .bdrv_co_flush_to_disk  = sd_co_flush_to_disk,
     .bdrv_co_discard = sd_co_discard,
+    .bdrv_co_is_allocated = sd_co_is_allocated,
 
     .bdrv_snapshot_create   = sd_snapshot_create,
     .bdrv_snapshot_goto     = sd_snapshot_goto,
@@ -2221,6 +2257,7 @@ static BlockDriver bdrv_sheepdog_unix = {
     .bdrv_co_writev = sd_co_writev,
     .bdrv_co_flush_to_disk  = sd_co_flush_to_disk,
     .bdrv_co_discard = sd_co_discard,
+    .bdrv_co_is_allocated = sd_co_is_allocated,
 
     .bdrv_snapshot_create   = sd_snapshot_create,
     .bdrv_snapshot_goto     = sd_snapshot_goto,
-- 
1.7.9.5

Re: [Qemu-devel] [PATCH v4 1/3] sheepdog: add discard/trim support for sheepdog

[Stefan Hajnoczi]

On Tue, Apr 23, 2013 at 02:03:33PM +0800, Liu Yuan wrote:
> From: Liu Yuan <tailai.ly@taobao.com>
> 
> The 'TRIM' command from VM that is to release underlying data storage for
> better thin-provision is already supported by the Sheepdog.
> 
> This patch adds the TRIM support at QEMU part.
> 
> For older Sheepdog that doesn't support it, we return 0(success) to upper layer.
> 
> Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
> Cc: Kevin Wolf <kwolf@redhat.com>
> Cc: Stefan Hajnoczi <stefanha@redhat.com>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Signed-off-by: Liu Yuan <tailai.ly@taobao.com>
> ---
>  block/sheepdog.c |   64 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 63 insertions(+), 1 deletion(-)

Thanks, applied to my block tree:
https://github.com/stefanha/qemu/commits/block

Stefan