[Qemu-devel] [PATCH 0/2] bugfixes of sheepdog driver for a case of live snapshot

[Qemu-devel] [PATCH 0/2] bugfixes of sheepdog driver for a case of live snapshot

[Hitoshi Mitake]

Current sheepdog driver has two problems in a mechanism of inode
object reloading for live snapshot. It causes inconsistent state of
snapshot volumes. A new GC algorithm implemented in sheepdog exposes
the problems. This patchset contains bugfixes for them.

Hitoshi Mitake (2):
  sheepdog: fix vdi object update after live snapshot
  sheepdog: reload only header in a case of live snapshot

 block/sheepdog.c | 49 +++++++++++++++++++++++++++++--------------------
 1 file changed, 29 insertions(+), 20 deletions(-)

-- 
1.9.1

[Qemu-devel] [PATCH 1/2] sheepdog: fix vdi object update after live snapshot

[Hitoshi Mitake]

sheepdog driver should decide a write request is COW or not based on
inode object which is active when the write request is issued.

Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Liu Yuan <namei.unix@gmail.com>
Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Hitoshi Mitake <mitake.hitoshi@lab.ntt.co.jp>
---
 block/sheepdog.c | 40 +++++++++++++++++++++++-----------------
 1 file changed, 23 insertions(+), 17 deletions(-)

diff --git a/block/sheepdog.c b/block/sheepdog.c
index 2c3fb01..8266382 100644
--- a/block/sheepdog.c
+++ b/block/sheepdog.c
@@ -282,6 +282,7 @@ typedef struct AIOReq {
     unsigned int data_len;
     uint8_t flags;
     uint32_t id;
+    bool create;
 
     QLIST_ENTRY(AIOReq) aio_siblings;
 } AIOReq;
@@ -404,7 +405,7 @@ static const char * sd_strerror(int err)
 
 static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
                                     uint64_t oid, unsigned int data_len,
-                                    uint64_t offset, uint8_t flags,
+                                    uint64_t offset, uint8_t flags, bool create,
                                     uint64_t base_oid, unsigned int iov_offset)
 {
     AIOReq *aio_req;
@@ -418,6 +419,7 @@ static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
     aio_req->data_len = data_len;
     aio_req->flags = flags;
     aio_req->id = s->aioreq_seq_num++;
+    aio_req->create = create;
 
     acb->nr_pending++;
     return aio_req;
@@ -668,8 +670,8 @@ static int do_req(int sockfd, SheepdogReq *hdr, void *data,
 }
 
 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
-                           struct iovec *iov, int niov, bool create,
-                           enum AIOCBState aiocb_type);
+                                         struct iovec *iov, int niov,
+                                         enum AIOCBState aiocb_type);
 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req);
 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag);
 static int get_sheep_fd(BDRVSheepdogState *s);
@@ -702,7 +704,7 @@ static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid)
         /* move aio_req from pending list to inflight one */
         QLIST_REMOVE(aio_req, aio_siblings);
         QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
-        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, false,
+        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
                         acb->aiocb_type);
     }
 }
@@ -798,7 +800,7 @@ static void coroutine_fn aio_read_response(void *opaque)
         }
         idx = data_oid_to_idx(aio_req->oid);
 
-        if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) {
+        if (aio_req->create) {
             /*
              * If the object is newly created one, we need to update
              * the vdi object (metadata object).  min_dirty_data_idx
@@ -1117,8 +1119,8 @@ out:
 }
 
 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
-                           struct iovec *iov, int niov, bool create,
-                           enum AIOCBState aiocb_type)
+                                         struct iovec *iov, int niov,
+                                         enum AIOCBState aiocb_type)
 {
     int nr_copies = s->inode.nr_copies;
     SheepdogObjReq hdr;
@@ -1129,6 +1131,7 @@ static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
     uint64_t offset = aio_req->offset;
     uint8_t flags = aio_req->flags;
     uint64_t old_oid = aio_req->base_oid;
+    bool create = aio_req->create;
 
     if (!nr_copies) {
         error_report("bug");
@@ -1310,6 +1313,7 @@ static bool check_simultaneous_create(BDRVSheepdogState *s, AIOReq *aio_req)
             DPRINTF("simultaneous create to %" PRIx64 "\n", aio_req->oid);
             aio_req->flags = 0;
             aio_req->base_oid = 0;
+            aio_req->create = false;
             QLIST_REMOVE(aio_req, aio_siblings);
             QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, aio_siblings);
             return true;
@@ -1322,7 +1326,8 @@ static bool check_simultaneous_create(BDRVSheepdogState *s, AIOReq *aio_req)
 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req)
 {
     SheepdogAIOCB *acb = aio_req->aiocb;
-    bool create = false;
+
+    aio_req->create = false;
 
     /* check whether this request becomes a CoW one */
     if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) {
@@ -1340,17 +1345,17 @@ static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req)
             aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx);
             aio_req->flags |= SD_FLAG_CMD_COW;
         }
-        create = true;
+        aio_req->create = true;
     }
 out:
     if (is_data_obj(aio_req->oid)) {
-        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, create,
+        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
                         acb->aiocb_type);
     } else {
         struct iovec iov;
         iov.iov_base = &s->inode;
         iov.iov_len = sizeof(s->inode);
-        add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA);
+        add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
     }
 }
 
@@ -1831,9 +1836,9 @@ static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
         iov.iov_base = &s->inode;
         iov.iov_len = sizeof(s->inode);
         aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
-                                data_len, offset, 0, 0, offset);
+                                data_len, offset, 0, false, 0, offset);
         QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
-        add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA);
+        add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
 
         acb->aio_done_func = sd_finish_aiocb;
         acb->aiocb_type = AIOCB_WRITE_UDATA;
@@ -2023,7 +2028,8 @@ static int coroutine_fn sd_co_rw_vector(void *p)
             DPRINTF("new oid %" PRIx64 "\n", oid);
         }
 
-        aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done);
+        aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create,
+                                old_oid, done);
         QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
 
         if (create) {
@@ -2032,7 +2038,7 @@ static int coroutine_fn sd_co_rw_vector(void *p)
             }
         }
 
-        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, create,
+        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
                         acb->aiocb_type);
     done:
         offset = 0;
@@ -2112,9 +2118,9 @@ static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
     acb->aio_done_func = sd_finish_aiocb;
 
     aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
-                            0, 0, 0, 0, 0);
+                            0, 0, 0, false, 0, 0);
     QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
-    add_aio_request(s, aio_req, NULL, 0, false, acb->aiocb_type);
+    add_aio_request(s, aio_req, NULL, 0, acb->aiocb_type);
 
     qemu_coroutine_yield();
     return acb->ret;
-- 
1.9.1

[Qemu-devel] [PATCH 2/2] sheepdog: reload only header in a case of live snapshot

[Hitoshi Mitake]

sheepdog dirver doesn't need to read data_vdi_id[] when a live
snapshot is created.

Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Liu Yuan <namei.unix@gmail.com>
Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Hitoshi Mitake <mitake.hitoshi@lab.ntt.co.jp>
---
 block/sheepdog.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/block/sheepdog.c b/block/sheepdog.c
index 8266382..06c4180 100644
--- a/block/sheepdog.c
+++ b/block/sheepdog.c
@@ -200,6 +200,8 @@ typedef struct SheepdogInode {
     uint32_t data_vdi_id[MAX_DATA_OBJS];
 } SheepdogInode;
 
+#define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
+
 /*
  * 64 bit FNV-1a non-zero initial basis
  */
@@ -1275,7 +1277,7 @@ static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag)
         return -EIO;
     }
 
-    inode = g_malloc(sizeof(s->inode));
+    inode = g_malloc(SD_INODE_HEADER_SIZE);
 
     ret = find_vdi_name(s, s->name, snapid, tag, &vid, false);
     if (ret) {
@@ -1283,13 +1285,14 @@ static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag)
     }
 
     ret = read_object(fd, (char *)inode, vid_to_vdi_oid(vid),
-                      s->inode.nr_copies, sizeof(*inode), 0, s->cache_flags);
+                      s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0,
+                      s->cache_flags);
     if (ret < 0) {
         goto out;
     }
 
     if (inode->vdi_id != s->inode.vdi_id) {
-        memcpy(&s->inode, inode, sizeof(s->inode));
+        memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE);
     }
 
 out:
-- 
1.9.1

Re: [Qemu-devel] [PATCH 2/2] sheepdog: reload only header in a case of live snapshot

[Eric Blake]

[-- Attachment #1: Type: text/plain, Size: 274 bytes --]

On 06/02/2014 09:06 AM, Hitoshi Mitake wrote:
> sheepdog dirver doesn't need to read data_vdi_id[] when a live

s/dirver/driver/

> snapshot is created.
> 

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]

Re: [Qemu-devel] [PATCH 2/2] sheepdog: reload only header in a case of live snapshot

[Hitoshi Mitake]

On Tue, Jun 3, 2014 at 12:48 AM, Eric Blake <eblake@redhat.com> wrote:
> On 06/02/2014 09:06 AM, Hitoshi Mitake wrote:
>> sheepdog dirver doesn't need to read data_vdi_id[] when a live
>
> s/dirver/driver/

Thanks for your comment. I'll send v2 later.

Thanks,
Hitoshi

>
>> snapshot is created.
>>
>
> --
> Eric Blake   eblake redhat com    +1-919-301-3266
> Libvirt virtualization library http://libvirt.org
>