[Qemu-devel] [PATCH v12 00/10] qcow2: cluster space preallocation

[Qemu-devel] [PATCH v12 00/10] qcow2: cluster space preallocation

[Anton Nefedov]

new in v12:
   - patch 9: pre-write overlap check added

v11: http://lists.nongnu.org/archive/html/qemu-devel/2018-12/msg04342.html
    - patch 4, 9: fixed commentary format
    - patch 4: removed one hunk with a dead check
    - patch 5: added commentary to BDRV_REQ_ALLOCATE definition
    - new auxiliary patch 6 for the following patch-7 change:
    - patch 7: reset BDRV_REQ_ALLOCATE from supported flag if CONFIG_FALLOCATE
               is false
    - patch 9: add commentary about missing qcow2_pre_write_overlap_check().
               Omit redundant changes in the test 060.

v10: http://lists.nongnu.org/archive/html/qemu-devel/2018-12/msg00121.html
    - patches 1-3,6,7: rebase after REQ_WRITE_UNCHANGED
    - patch 3: drop supported_zero_flags. My bad, no write_zeroes in quorum.
    - patch 4: almost trivial rebase. RB-tags not stripped.
               Choose another constant for BDRV_REQ_ALLOCATE
    - patch 5: rebase. Instead of marking REQ_ALLOCATE serialising, accompany
               it with REQ_SERIALISING.
    - patch 7: add symmetric copy-on-read change
    - patch 8: trivial rebase. RB-tags not stripped.

----

This pull request is to start to improve a few performance points of
qcow2 format:

  1. non cluster-aligned write requests (to unallocated clusters) explicitly
     pad data with zeroes if there is no backing data.
     Resulting increase in ops number and potential cluster fragmentation
     (on the host file) is already solved by:
       ee22a9d qcow2: Merge the writing of the COW regions with the guest data
     However, in case of zero COW regions, that can be avoided at all
     but the whole clusters are preallocated and zeroed in a single
     efficient write_zeroes() operation

  2. moreover, efficient write_zeroes() operation can be used to preallocate
     space megabytes (*configurable number) ahead which gives noticeable
     improvement on some storage types (e.g. distributed storage)
     where the space allocation operation might be expensive)
     (Not included in this patchset since v6).

  3. this will also allow to enable simultaneous writes to the same unallocated
     cluster after the space has been allocated & zeroed but before
     the first data is written and the cluster is linked to L2.
     (Not included in this patchset).

Efficient write_zeroes usually implies that the blocks are not actually
written to but only reserved and marked as zeroed by the storage.
In this patchset, file-posix driver is marked as supporting this operation
if it supports (/configured to support) fallocate() operation.

Existing bdrv_write_zeroes() falls back to writing zero buffers if
write_zeroes is not supported by the driver.
A new flag (BDRV_REQ_ALLOCATE) is introduced to avoid that but return ENOTSUP.
Such allocate requests are also implemented to possibly overlap with the
other requests. No wait is performed but an error returned in such case as well.
So the operation should be considered advisory and a fallback scenario still
handled by the caller (in this case, qcow2 driver).

simple perf test:

  qemu-img create -f qcow2 test.img 4G && \
  qemu-img bench -c $((1024*1024)) -f qcow2 -n -s 4k -t none -w test.img

test results (seconds):

    +-----------+-------+------+-------+------+------+
    |   file    |    before    |     after    | gain |
    +-----------+-------+------+-------+------+------+
    |    ssd    |      61.153  |      36.313  |  41% |
    |    hdd    |     112.676  |     122.056  |  -8% |
    +-----------+--------------+--------------+------+

Anton Nefedov (10):
  mirror: inherit supported write/zero flags
  blkverify: set supported write/zero flags
  quorum: set supported write flags
  block: introduce BDRV_REQ_ALLOCATE flag
  block: treat BDRV_REQ_ALLOCATE as serialising
  file-posix: reset fallocate-related flags without CONFIG_FALLOCATE*
  file-posix: support BDRV_REQ_ALLOCATE
  block: support BDRV_REQ_ALLOCATE in passthrough drivers
  qcow2: skip writing zero buffers to empty COW areas
  iotest 134: test cluster-misaligned encrypted write

 qapi/block-core.json       |  4 +-
 block/qcow2.h              |  6 +++
 include/block/block.h      | 13 +++++-
 include/block/block_int.h  |  3 +-
 block/blkdebug.c           |  2 +-
 block/blkverify.c          | 10 ++++-
 block/copy-on-read.c       |  4 +-
 block/file-posix.c         | 16 +++++--
 block/io.c                 | 45 +++++++++++++++----
 block/mirror.c             |  8 +++-
 block/qcow2-cluster.c      |  2 +-
 block/qcow2.c              | 91 +++++++++++++++++++++++++++++++++++++-
 block/quorum.c             | 19 +++++++-
 block/raw-format.c         |  2 +-
 block/trace-events         |  1 +
 tests/qemu-iotests/060     |  7 ++-
 tests/qemu-iotests/060.out |  5 ++-
 tests/qemu-iotests/134     |  9 ++++
 tests/qemu-iotests/134.out | 10 +++++
 19 files changed, 229 insertions(+), 28 deletions(-)

-- 
2.17.1

[Qemu-devel] [PATCH v12 02/10] blkverify: set supported write/zero flags

[Anton Nefedov]

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block/blkverify.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/block/blkverify.c b/block/blkverify.c
index 89bf4386e3..bb52596cbb 100644
--- a/block/blkverify.c
+++ b/block/blkverify.c
@@ -141,8 +141,14 @@ static int blkverify_open(BlockDriverState *bs, QDict *options, int flags,
         goto fail;
     }
 
-    bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED;
-    bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED;
+    bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
+        (BDRV_REQ_FUA &
+         bs->file->bs->supported_write_flags &
+         s->test_file->bs->supported_write_flags);
+    bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
+        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP) &
+         bs->file->bs->supported_zero_flags &
+         s->test_file->bs->supported_zero_flags);
 
     ret = 0;
 fail:
-- 
2.17.1

[Qemu-devel] [PATCH v12 01/10] mirror: inherit supported write/zero flags

[Anton Nefedov]

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block/mirror.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/block/mirror.c b/block/mirror.c
index f0b211a9c8..7b5a5f13a2 100644
--- a/block/mirror.c
+++ b/block/mirror.c
@@ -1529,8 +1529,12 @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs,
         mirror_top_bs->implicit = true;
     }
     mirror_top_bs->total_sectors = bs->total_sectors;
-    mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED;
-    mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED;
+    mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
+        (BDRV_REQ_FUA & bs->supported_write_flags);
+    mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
+        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP)
+         & bs->supported_zero_flags);
+
     bs_opaque = g_new0(MirrorBDSOpaque, 1);
     mirror_top_bs->opaque = bs_opaque;
     bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs));
-- 
2.17.1

[Qemu-devel] [PATCH v12 03/10] quorum: set supported write flags

[Anton Nefedov]

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block/quorum.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/block/quorum.c b/block/quorum.c
index 16b3c8067c..d21a6a3b8e 100644
--- a/block/quorum.c
+++ b/block/quorum.c
@@ -857,6 +857,19 @@ static QemuOptsList quorum_runtime_opts = {
     },
 };
 
+static void quorum_set_supported_flags(BlockDriverState *bs)
+{
+    BDRVQuorumState *s = bs->opaque;
+    int i;
+
+    bs->supported_write_flags = BDRV_REQ_FUA;
+    for (i = 0; i < s->num_children; i++) {
+        bs->supported_write_flags &= s->children[i]->bs->supported_write_flags;
+    }
+
+    bs->supported_write_flags |= BDRV_REQ_WRITE_UNCHANGED;
+}
+
 static int quorum_open(BlockDriverState *bs, QDict *options, int flags,
                        Error **errp)
 {
@@ -950,7 +963,7 @@ static int quorum_open(BlockDriverState *bs, QDict *options, int flags,
     }
     s->next_child_index = s->num_children;
 
-    bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED;
+    quorum_set_supported_flags(bs);
 
     g_free(opened);
     goto exit;
@@ -1025,6 +1038,8 @@ static void quorum_add_child(BlockDriverState *bs, BlockDriverState *child_bs,
     s->children = g_renew(BdrvChild *, s->children, s->num_children + 1);
     s->children[s->num_children++] = child;
 
+    quorum_set_supported_flags(bs);
+
 out:
     bdrv_drained_end(bs);
 }
@@ -1063,6 +1078,8 @@ static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
     bdrv_unref_child(bs, child);
 
     bdrv_drained_end(bs);
+
+    quorum_set_supported_flags(bs);
 }
 
 static void quorum_refresh_filename(BlockDriverState *bs, QDict *options)
-- 
2.17.1

[Qemu-devel] [PATCH v12 04/10] block: introduce BDRV_REQ_ALLOCATE flag

[Anton Nefedov]

The flag is supposed to indicate that the region of the disk image has
to be sufficiently allocated so it reads as zeroes.

The call with the flag set must return -ENOTSUP if allocation cannot
be done efficiently.
This has to be made sure of by both
  - the drivers that support the flag
  - and the common block layer (so it will not fall back to any slowpath
    (like writing zero buffers) in case the driver does not support
    the flag).

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
---
 include/block/block.h     | 10 +++++++++-
 include/block/block_int.h |  3 ++-
 block/io.c                | 14 +++++++++++++-
 3 files changed, 24 insertions(+), 3 deletions(-)

diff --git a/include/block/block.h b/include/block/block.h
index f70a843b72..643d32f4b8 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -83,8 +83,16 @@ typedef enum {
      */
     BDRV_REQ_SERIALISING        = 0x80,
 
+    /*
+     * The BDRV_REQ_ALLOCATE flag is used to indicate that the driver has to
+     * efficiently allocate the space so it reads as zeroes, or return an error.
+     * If this flag is set then BDRV_REQ_ZERO_WRITE must also be set.
+     * This flag cannot be set together with BDRV_REQ_MAY_UNMAP.
+     */
+    BDRV_REQ_ALLOCATE           = 0x100,
+
     /* Mask of valid flags */
-    BDRV_REQ_MASK               = 0xff,
+    BDRV_REQ_MASK               = 0x1ff,
 } BdrvRequestFlags;
 
 typedef struct BlockSizes {
diff --git a/include/block/block_int.h b/include/block/block_int.h
index f605622216..833129d912 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -724,7 +724,8 @@ struct BlockDriverState {
      * their children. */
     unsigned int supported_write_flags;
     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
-     * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED) */
+     * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED, BDRV_REQ_ALLOCATE)
+     */
     unsigned int supported_zero_flags;
 
     /* the following member gives a name to every node on the bs graph. */
diff --git a/block/io.c b/block/io.c
index bd9d688f8b..66006a089d 100644
--- a/block/io.c
+++ b/block/io.c
@@ -1534,7 +1534,7 @@ static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
             assert(!bs->supported_zero_flags);
         }
 
-        if (ret == -ENOTSUP) {
+        if (ret == -ENOTSUP && !(flags & BDRV_REQ_ALLOCATE)) {
             /* Fall back to bounce buffer if write zeroes is unsupported */
             BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE;
 
@@ -1773,6 +1773,9 @@ static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child,
 
     assert(flags & BDRV_REQ_ZERO_WRITE);
     if (head_padding_bytes || tail_padding_bytes) {
+        if (flags & BDRV_REQ_ALLOCATE) {
+            return -ENOTSUP;
+        }
         buf = qemu_blockalign(bs, align);
         iov = (struct iovec) {
             .iov_base   = buf,
@@ -1858,6 +1861,9 @@ int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
     bool use_local_qiov = false;
     int ret;
 
+    assert(!((flags & BDRV_REQ_ALLOCATE) && (flags & BDRV_REQ_MAY_UNMAP)));
+    assert(!((flags & BDRV_REQ_ALLOCATE) && !(flags & BDRV_REQ_ZERO_WRITE)));
+
     trace_bdrv_co_pwritev(child->bs, offset, bytes, flags);
 
     if (!bs->drv) {
@@ -1980,6 +1986,12 @@ int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset,
 {
     trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags);
 
+    if ((flags & BDRV_REQ_ALLOCATE) &&
+        !(child->bs->supported_zero_flags & BDRV_REQ_ALLOCATE))
+    {
+        return -ENOTSUP;
+    }
+
     if (!(child->bs->open_flags & BDRV_O_UNMAP)) {
         flags &= ~BDRV_REQ_MAY_UNMAP;
     }
-- 
2.17.1

[Qemu-devel] [PATCH v12 05/10] block: treat BDRV_REQ_ALLOCATE as serialising

[Anton Nefedov]

The idea is that ALLOCATE requests may overlap with other requests.
Reuse the existing block layer infrastructure for serialising requests.
Use the following approach:
  - mark ALLOCATE also SERIALISING, so subsequent requests to the area wait
  - ALLOCATE request itself must never wait if another request is in flight
    already. Return EAGAIN, let the caller reconsider.

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
---
 include/block/block.h |  3 +++
 block/io.c            | 31 ++++++++++++++++++++++++-------
 2 files changed, 27 insertions(+), 7 deletions(-)

diff --git a/include/block/block.h b/include/block/block.h
index 643d32f4b8..dfc0fc1b8f 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -88,6 +88,9 @@ typedef enum {
      * efficiently allocate the space so it reads as zeroes, or return an error.
      * If this flag is set then BDRV_REQ_ZERO_WRITE must also be set.
      * This flag cannot be set together with BDRV_REQ_MAY_UNMAP.
+     * This flag implicitly sets BDRV_REQ_SERIALISING meaning it is protected
+     * from conflicts with overlapping requests. If such conflict is detected,
+     * -EAGAIN is returned.
      */
     BDRV_REQ_ALLOCATE           = 0x100,
 
diff --git a/block/io.c b/block/io.c
index 66006a089d..4451714a60 100644
--- a/block/io.c
+++ b/block/io.c
@@ -720,12 +720,13 @@ void bdrv_dec_in_flight(BlockDriverState *bs)
     bdrv_wakeup(bs);
 }
 
-static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
+static bool coroutine_fn find_or_wait_serialising_requests(
+    BdrvTrackedRequest *self, bool wait)
 {
     BlockDriverState *bs = self->bs;
     BdrvTrackedRequest *req;
     bool retry;
-    bool waited = false;
+    bool found = false;
 
     if (!atomic_read(&bs->serialising_in_flight)) {
         return false;
@@ -751,11 +752,14 @@ static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
                  * will wait for us as soon as it wakes up, then just go on
                  * (instead of producing a deadlock in the former case). */
                 if (!req->waiting_for) {
+                    found = true;
+                    if (!wait) {
+                        break;
+                    }
                     self->waiting_for = req;
                     qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock);
                     self->waiting_for = NULL;
                     retry = true;
-                    waited = true;
                     break;
                 }
             }
@@ -763,7 +767,12 @@ static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
         qemu_co_mutex_unlock(&bs->reqs_lock);
     } while (retry);
 
-    return waited;
+    return found;
+}
+
+static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
+{
+    return find_or_wait_serialising_requests(self, true);
 }
 
 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
@@ -1585,7 +1594,7 @@ bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes,
                           BdrvTrackedRequest *req, int flags)
 {
     BlockDriverState *bs = child->bs;
-    bool waited;
+    bool found;
     int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
 
     if (bs->read_only) {
@@ -1602,9 +1611,13 @@ bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes,
         mark_request_serialising(req, bdrv_get_cluster_size(bs));
     }
 
-    waited = wait_serialising_requests(req);
+    found = find_or_wait_serialising_requests(req,
+                                              !(flags & BDRV_REQ_ALLOCATE));
+    if (found && (flags & BDRV_REQ_ALLOCATE)) {
+        return -EAGAIN;
+    }
 
-    assert(!waited || !req->serialising ||
+    assert(!found || !req->serialising ||
            is_request_serialising_and_aligned(req));
     assert(req->overlap_offset <= offset);
     assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
@@ -1864,6 +1877,10 @@ int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
     assert(!((flags & BDRV_REQ_ALLOCATE) && (flags & BDRV_REQ_MAY_UNMAP)));
     assert(!((flags & BDRV_REQ_ALLOCATE) && !(flags & BDRV_REQ_ZERO_WRITE)));
 
+    if (flags & BDRV_REQ_ALLOCATE) {
+        flags |= BDRV_REQ_SERIALISING;
+    }
+
     trace_bdrv_co_pwritev(child->bs, offset, bytes, flags);
 
     if (!bs->drv) {
-- 
2.17.1

[Qemu-devel] [PATCH v12 06/10] file-posix: reset fallocate-related flags without CONFIG_FALLOCATE*

[Anton Nefedov]

these flags currently affect nothing without CONFIG_FALLOCATE*, so it's
not a bug, but fixing it makes possible to adjust supported zero flag
BDRV_REQ_ALLOCATE regardless of configuration.

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
---
 block/file-posix.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/block/file-posix.c b/block/file-posix.c
index 8aee7a3fb8..8d3ec96627 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -1488,9 +1488,7 @@ static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
 static int handle_aiocb_write_zeroes(void *opaque)
 {
     RawPosixAIOData *aiocb = opaque;
-#if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
     BDRVRawState *s = aiocb->bs->opaque;
-#endif
 #ifdef CONFIG_FALLOCATE
     int64_t len;
 #endif
@@ -1514,6 +1512,8 @@ static int handle_aiocb_write_zeroes(void *opaque)
         }
         s->has_write_zeroes = false;
     }
+#else
+    s->has_write_zeroes = false;
 #endif
 
 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
@@ -1533,6 +1533,8 @@ static int handle_aiocb_write_zeroes(void *opaque)
             s->has_discard = false;
         }
     }
+#else
+    s->has_discard = false;
 #endif
 
 #ifdef CONFIG_FALLOCATE
@@ -1546,6 +1548,8 @@ static int handle_aiocb_write_zeroes(void *opaque)
         }
         s->has_fallocate = false;
     }
+#else
+    s->has_fallocate = false;
 #endif
 
     return -ENOTSUP;
-- 
2.17.1

[Qemu-devel] [PATCH v12 07/10] file-posix: support BDRV_REQ_ALLOCATE

[Anton Nefedov]

Current write_zeroes implementation is good enough to satisfy this flag too

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
---
 block/file-posix.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/block/file-posix.c b/block/file-posix.c
index 8d3ec96627..dac218ec7f 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -607,6 +607,7 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,
         } else {
             s->discard_zeroes = true;
             s->has_fallocate = true;
+            bs->supported_zero_flags = BDRV_REQ_ALLOCATE;
         }
     } else {
         if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
@@ -650,10 +651,11 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,
 #ifdef CONFIG_XFS
     if (platform_test_xfs_fd(s->fd)) {
         s->is_xfs = true;
+        bs->supported_zero_flags = BDRV_REQ_ALLOCATE;
     }
 #endif
 
-    bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
+    bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
     ret = 0;
 fail:
     if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
@@ -1552,6 +1554,10 @@ static int handle_aiocb_write_zeroes(void *opaque)
     s->has_fallocate = false;
 #endif
 
+    if (!s->has_fallocate) {
+        aiocb->bs->supported_zero_flags &= ~BDRV_REQ_ALLOCATE;
+    }
+
     return -ENOTSUP;
 }
 
-- 
2.17.1

[Qemu-devel] [PATCH v12 08/10] block: support BDRV_REQ_ALLOCATE in passthrough drivers

[Anton Nefedov]

Support the flag if the underlying BDS supports it

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
---
 block/blkdebug.c     | 2 +-
 block/blkverify.c    | 2 +-
 block/copy-on-read.c | 4 ++--
 block/mirror.c       | 2 +-
 block/raw-format.c   | 2 +-
 5 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/block/blkdebug.c b/block/blkdebug.c
index 0759452925..f0fc2ec276 100644
--- a/block/blkdebug.c
+++ b/block/blkdebug.c
@@ -401,7 +401,7 @@ static int blkdebug_open(BlockDriverState *bs, QDict *options, int flags,
     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
         (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
-        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP) &
+        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_ALLOCATE) &
             bs->file->bs->supported_zero_flags);
     ret = -EINVAL;
 
diff --git a/block/blkverify.c b/block/blkverify.c
index bb52596cbb..9cb4f94b68 100644
--- a/block/blkverify.c
+++ b/block/blkverify.c
@@ -146,7 +146,7 @@ static int blkverify_open(BlockDriverState *bs, QDict *options, int flags,
          bs->file->bs->supported_write_flags &
          s->test_file->bs->supported_write_flags);
     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
-        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP) &
+        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_ALLOCATE) &
          bs->file->bs->supported_zero_flags &
          s->test_file->bs->supported_zero_flags);
 
diff --git a/block/copy-on-read.c b/block/copy-on-read.c
index 64dcc424b5..1eb993699a 100644
--- a/block/copy-on-read.c
+++ b/block/copy-on-read.c
@@ -38,8 +38,8 @@ static int cor_open(BlockDriverState *bs, QDict *options, int flags,
                                     bs->file->bs->supported_write_flags);
 
     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
-                               ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP) &
-                                    bs->file->bs->supported_zero_flags);
+        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_ALLOCATE) &
+         bs->file->bs->supported_zero_flags);
 
     return 0;
 }
diff --git a/block/mirror.c b/block/mirror.c
index 7b5a5f13a2..057516acb9 100644
--- a/block/mirror.c
+++ b/block/mirror.c
@@ -1532,7 +1532,7 @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs,
     mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
         (BDRV_REQ_FUA & bs->supported_write_flags);
     mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
-        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP)
+        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_ALLOCATE)
          & bs->supported_zero_flags);
 
     bs_opaque = g_new0(MirrorBDSOpaque, 1);
diff --git a/block/raw-format.c b/block/raw-format.c
index 6f6dc99b2c..ad7453dc83 100644
--- a/block/raw-format.c
+++ b/block/raw-format.c
@@ -432,7 +432,7 @@ static int raw_open(BlockDriverState *bs, QDict *options, int flags,
     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
         (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
-        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP) &
+        ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_ALLOCATE) &
             bs->file->bs->supported_zero_flags);
 
     if (bs->probed && !bdrv_is_read_only(bs)) {
-- 
2.17.1

[Qemu-devel] [PATCH v12 09/10] qcow2: skip writing zero buffers to empty COW areas

[Anton Nefedov]

If COW areas of the newly allocated clusters are zeroes on the backing image,
efficient bdrv_write_zeroes(flags=BDRV_REQ_ALLOCATE) can be used on the whole
cluster instead of writing explicit zero buffers later in perform_cow().

iotest 060:
write to the discarded cluster does not trigger COW anymore.
Use a backing image instead.

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 qapi/block-core.json       |  4 +-
 block/qcow2.h              |  6 +++
 block/qcow2-cluster.c      |  2 +-
 block/qcow2.c              | 91 +++++++++++++++++++++++++++++++++++++-
 block/trace-events         |  1 +
 tests/qemu-iotests/060     |  7 ++-
 tests/qemu-iotests/060.out |  5 ++-
 7 files changed, 110 insertions(+), 6 deletions(-)

diff --git a/qapi/block-core.json b/qapi/block-core.json
index 762000f31f..204528b3f6 100644
--- a/qapi/block-core.json
+++ b/qapi/block-core.json
@@ -3009,6 +3009,8 @@
 #
 # @cor_write: a write due to copy-on-read (since 2.11)
 #
+# @cluster_alloc_space: an allocation of file space for a cluster (since 4.0)
+#
 # Since: 2.9
 ##
 { 'enum': 'BlkdebugEvent', 'prefix': 'BLKDBG',
@@ -3027,7 +3029,7 @@
             'pwritev_rmw_tail', 'pwritev_rmw_after_tail', 'pwritev',
             'pwritev_zero', 'pwritev_done', 'empty_image_prepare',
             'l1_shrink_write_table', 'l1_shrink_free_l2_clusters',
-            'cor_write'] }
+            'cor_write', 'cluster_alloc_space'] }
 
 ##
 # @BlkdebugInjectErrorOptions:
diff --git a/block/qcow2.h b/block/qcow2.h
index 438a1dee9e..dad4b1c7ca 100644
--- a/block/qcow2.h
+++ b/block/qcow2.h
@@ -389,6 +389,12 @@ typedef struct QCowL2Meta
      */
     Qcow2COWRegion cow_end;
 
+    /*
+     * Indicates that COW regions are already handled and do not require
+     * any more processing.
+     */
+    bool skip_cow;
+
     /**
      * The I/O vector with the data from the actual guest write request.
      * If non-NULL, this is meant to be merged together with the data
diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
index 30eca26c47..e5f936a82c 100644
--- a/block/qcow2-cluster.c
+++ b/block/qcow2-cluster.c
@@ -806,7 +806,7 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m)
     assert(start->offset + start->nb_bytes <= end->offset);
     assert(!m->data_qiov || m->data_qiov->size == data_bytes);
 
-    if (start->nb_bytes == 0 && end->nb_bytes == 0) {
+    if ((start->nb_bytes == 0 && end->nb_bytes == 0) || m->skip_cow) {
         return 0;
     }
 
diff --git a/block/qcow2.c b/block/qcow2.c
index 4897abae5e..05a7cbebbd 100644
--- a/block/qcow2.c
+++ b/block/qcow2.c
@@ -2021,6 +2021,11 @@ static bool merge_cow(uint64_t offset, unsigned bytes,
             continue;
         }
 
+        /* If COW regions are handled already, skip this too */
+        if (m->skip_cow) {
+            continue;
+        }
+
         /* The data (middle) region must be immediately after the
          * start region */
         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
@@ -2046,6 +2051,79 @@ static bool merge_cow(uint64_t offset, unsigned bytes,
     return false;
 }
 
+static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes)
+{
+    int64_t nr;
+    return !bytes ||
+        (!bdrv_is_allocated_above(bs, NULL, offset, bytes, &nr) && nr == bytes);
+}
+
+static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
+{
+    /*
+     * This check is designed for optimization shortcut so it must be
+     * efficient.
+     * Instead of is_zero(), use is_unallocated() as it is faster (but not
+     * as accurate and can result in false negatives).
+     */
+    return is_unallocated(bs, m->offset + m->cow_start.offset,
+                          m->cow_start.nb_bytes) &&
+           is_unallocated(bs, m->offset + m->cow_end.offset,
+                          m->cow_end.nb_bytes);
+}
+
+static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
+{
+    BDRVQcow2State *s = bs->opaque;
+    QCowL2Meta *m;
+
+    if (!(bs->file->bs->supported_zero_flags & BDRV_REQ_ALLOCATE)) {
+        return 0;
+    }
+
+    if (bs->encrypted) {
+        return 0;
+    }
+
+    for (m = l2meta; m != NULL; m = m->next) {
+        int ret;
+
+        if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
+            continue;
+        }
+
+        if (!is_zero_cow(bs, m)) {
+            continue;
+        }
+
+        /*
+         * instead of writing zero COW buffers,
+         * efficiently zero out the whole clusters
+         */
+
+        ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset,
+                                            m->nb_clusters * s->cluster_size);
+        if (ret < 0) {
+            return ret;
+        }
+
+        BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
+        ret = bdrv_co_pwrite_zeroes(bs->file, m->alloc_offset,
+                                    m->nb_clusters * s->cluster_size,
+                                    BDRV_REQ_ALLOCATE);
+        if (ret < 0) {
+            if (ret != -ENOTSUP && ret != -EAGAIN) {
+                return ret;
+            }
+            continue;
+        }
+
+        trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
+        m->skip_cow = true;
+    }
+    return 0;
+}
+
 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
                                          uint64_t bytes, QEMUIOVector *qiov,
                                          int flags)
@@ -2126,24 +2204,33 @@ static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
             goto fail;
         }
 
+        qemu_co_mutex_unlock(&s->lock);
+
+        ret = handle_alloc_space(bs, l2meta);
+        if (ret < 0) {
+            qemu_co_mutex_lock(&s->lock);
+            goto fail;
+        }
+
         /* If we need to do COW, check if it's possible to merge the
          * writing of the guest data together with that of the COW regions.
          * If it's not possible (or not necessary) then write the
          * guest data now. */
         if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) {
-            qemu_co_mutex_unlock(&s->lock);
             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
             trace_qcow2_writev_data(qemu_coroutine_self(),
                                     cluster_offset + offset_in_cluster);
             ret = bdrv_co_pwritev(bs->file,
                                   cluster_offset + offset_in_cluster,
                                   cur_bytes, &hd_qiov, 0);
-            qemu_co_mutex_lock(&s->lock);
             if (ret < 0) {
+                qemu_co_mutex_lock(&s->lock);
                 goto fail;
             }
         }
 
+        qemu_co_mutex_lock(&s->lock);
+
         ret = qcow2_handle_l2meta(bs, &l2meta, true);
         if (ret) {
             goto fail;
diff --git a/block/trace-events b/block/trace-events
index 693c14c443..4243c3469a 100644
--- a/block/trace-events
+++ b/block/trace-events
@@ -69,6 +69,7 @@ qcow2_writev_done_part(void *co, int cur_bytes) "co %p cur_bytes %d"
 qcow2_writev_data(void *co, uint64_t offset) "co %p offset 0x%" PRIx64
 qcow2_pwrite_zeroes_start_req(void *co, int64_t offset, int count) "co %p offset 0x%" PRIx64 " count %d"
 qcow2_pwrite_zeroes(void *co, int64_t offset, int count) "co %p offset 0x%" PRIx64 " count %d"
+qcow2_skip_cow(void *co, uint64_t offset, int nb_clusters) "co %p offset 0x%" PRIx64 " nb_clusters %d"
 
 # block/qcow2-cluster.c
 qcow2_alloc_clusters_offset(void *co, uint64_t offset, int bytes) "co %p offset 0x%" PRIx64 " bytes %d"
diff --git a/tests/qemu-iotests/060 b/tests/qemu-iotests/060
index af0588ae9a..163fb075ea 100755
--- a/tests/qemu-iotests/060
+++ b/tests/qemu-iotests/060
@@ -150,10 +150,15 @@ $QEMU_IO -c "$OPEN_RO" -c "read -P 1 0 512" | _filter_qemu_io
 echo
 echo "=== Testing overlap while COW is in flight ==="
 echo
+BACKING_IMG=$TEST_IMG.base
+TEST_IMG=$BACKING_IMG _make_test_img 1G
+
+$QEMU_IO -c 'write 0k 64k' "$BACKING_IMG" | _filter_qemu_io
+
 # compat=0.10 is required in order to make the following discard actually
 # unallocate the sector rather than make it a zero sector - we want COW, after
 # all.
-IMGOPTS='compat=0.10' _make_test_img 1G
+IMGOPTS='compat=0.10' _make_test_img -b "$BACKING_IMG" 1G
 # Write two clusters, the second one enforces creation of an L2 table after
 # the first data cluster.
 $QEMU_IO -c 'write 0k 64k' -c 'write 512M 64k' "$TEST_IMG" | _filter_qemu_io
diff --git a/tests/qemu-iotests/060.out b/tests/qemu-iotests/060.out
index af623cfd86..afd9212c07 100644
--- a/tests/qemu-iotests/060.out
+++ b/tests/qemu-iotests/060.out
@@ -97,7 +97,10 @@ read 512/512 bytes at offset 0
 
 === Testing overlap while COW is in flight ===
 
-Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1073741824
+Formatting 'TEST_DIR/t.IMGFMT.base', fmt=IMGFMT size=1073741824
+wrote 65536/65536 bytes at offset 0
+64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1073741824 backing_file=TEST_DIR/t.IMGFMT.base
 wrote 65536/65536 bytes at offset 0
 64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
 wrote 65536/65536 bytes at offset 536870912
-- 
2.17.1

[Qemu-devel] [PATCH v12 10/10] iotest 134: test cluster-misaligned encrypted write

[Anton Nefedov]

COW (even empty/zero) areas require encryption too

Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
---
 tests/qemu-iotests/134     |  9 +++++++++
 tests/qemu-iotests/134.out | 10 ++++++++++
 2 files changed, 19 insertions(+)

diff --git a/tests/qemu-iotests/134 b/tests/qemu-iotests/134
index cacabcd28b..792c8ca12f 100755
--- a/tests/qemu-iotests/134
+++ b/tests/qemu-iotests/134
@@ -57,6 +57,15 @@ echo
 echo "== reading whole image =="
 $QEMU_IO --object $SECRET -c "read 0 $size" --image-opts $IMGSPEC | _filter_qemu_io | _filter_testdir
 
+echo
+echo "== rewriting cluster part =="
+$QEMU_IO --object $SECRET -c "write -P 0xb 512 512" --image-opts $IMGSPEC | _filter_qemu_io | _filter_testdir
+
+echo
+echo "== verify pattern =="
+$QEMU_IO --object $SECRET -c "read -P 0 0 512"  --image-opts $IMGSPEC | _filter_qemu_io | _filter_testdir
+$QEMU_IO --object $SECRET -c "read -P 0xb 512 512"  --image-opts $IMGSPEC | _filter_qemu_io | _filter_testdir
+
 echo
 echo "== rewriting whole image =="
 $QEMU_IO --object $SECRET -c "write -P 0xa 0 $size" --image-opts $IMGSPEC | _filter_qemu_io | _filter_testdir
diff --git a/tests/qemu-iotests/134.out b/tests/qemu-iotests/134.out
index 972be49d91..09d46f6b17 100644
--- a/tests/qemu-iotests/134.out
+++ b/tests/qemu-iotests/134.out
@@ -5,6 +5,16 @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=134217728 encryption=on encrypt.
 read 134217728/134217728 bytes at offset 0
 128 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
 
+== rewriting cluster part ==
+wrote 512/512 bytes at offset 512
+512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+
+== verify pattern ==
+read 512/512 bytes at offset 0
+512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+read 512/512 bytes at offset 512
+512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+
 == rewriting whole image ==
 wrote 134217728/134217728 bytes at offset 0
 128 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-- 
2.17.1

Re: [Qemu-devel] [PATCH v12 09/10] qcow2: skip writing zero buffers to empty COW areas

[Alberto Garcia]

On Mon 14 Jan 2019 12:18:30 PM CET, Anton Nefedov wrote:
> If COW areas of the newly allocated clusters are zeroes on the backing image,
> efficient bdrv_write_zeroes(flags=BDRV_REQ_ALLOCATE) can be used on the whole
> cluster instead of writing explicit zero buffers later in perform_cow().
>
> iotest 060:
> write to the discarded cluster does not trigger COW anymore.
> Use a backing image instead.
>
> Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>

Reviewed-by: Alberto Garcia <berto@igalia.com>

> +        ret = handle_alloc_space(bs, l2meta);

I insist that it would be nice to have a short comment explaining what
this does.

Berto

Re: [Qemu-devel] [PATCH v12 09/10] qcow2: skip writing zero buffers to empty COW areas

[Anton Nefedov]

On 15/1/2019 6:27 PM, Alberto Garcia wrote:
> On Mon 14 Jan 2019 12:18:30 PM CET, Anton Nefedov wrote:
>> If COW areas of the newly allocated clusters are zeroes on the backing image,
>> efficient bdrv_write_zeroes(flags=BDRV_REQ_ALLOCATE) can be used on the whole
>> cluster instead of writing explicit zero buffers later in perform_cow().
>>
>> iotest 060:
>> write to the discarded cluster does not trigger COW anymore.
>> Use a backing image instead.
>>
>> Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
>> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> 
> Reviewed-by: Alberto Garcia <berto@igalia.com>
> 
>> +        ret = handle_alloc_space(bs, l2meta);
> 
> I insist that it would be nice to have a short comment explaining what
> this does.
> 

Right sorry forgot your comment.
I'd go with:

+        /* Try to efficiently initialize the physical space with zeroes */
          ret = handle_alloc_space(bs, l2meta);
          if (ret < 0) {
              qemu_co_mutex_lock(&s->lock);

Re: [Qemu-devel] [PATCH v12 09/10] qcow2: skip writing zero buffers to empty COW areas

[Alberto Garcia]

On Wed 16 Jan 2019 10:32:33 AM CET, Anton Nefedov wrote:
>>> +        ret = handle_alloc_space(bs, l2meta);
>> 
>> I insist that it would be nice to have a short comment explaining
>> what this does.
>
> Right sorry forgot your comment.
> I'd go with:
>
> +        /* Try to efficiently initialize the physical space with zeroes */
>           ret = handle_alloc_space(bs, l2meta);
>           if (ret < 0) {
>               qemu_co_mutex_lock(&s->lock);

That looks good, thanks!

Berto

Re: [Qemu-devel] [PATCH v12 00/10] qcow2: cluster space preallocation

[Anton Nefedov]

On 14/1/2019 2:18 PM, Anton Nefedov wrote:
> This pull request is to start to improve a few performance points of
> qcow2 format:
> 
>    1. non cluster-aligned write requests (to unallocated clusters) explicitly
>       pad data with zeroes if there is no backing data.
>       Resulting increase in ops number and potential cluster fragmentation
>       (on the host file) is already solved by:
>         ee22a9d qcow2: Merge the writing of the COW regions with the guest data
>       However, in case of zero COW regions, that can be avoided at all
>       but the whole clusters are preallocated and zeroed in a single
>       efficient write_zeroes() operation
> 
>    2. moreover, efficient write_zeroes() operation can be used to preallocate
>       space megabytes (*configurable number) ahead which gives noticeable
>       improvement on some storage types (e.g. distributed storage)
>       where the space allocation operation might be expensive)
>       (Not included in this patchset since v6).
> 
>    3. this will also allow to enable simultaneous writes to the same unallocated
>       cluster after the space has been allocated & zeroed but before
>       the first data is written and the cluster is linked to L2.
>       (Not included in this patchset).
> 
> Efficient write_zeroes usually implies that the blocks are not actually
> written to but only reserved and marked as zeroed by the storage.
> In this patchset, file-posix driver is marked as supporting this operation
> if it supports (/configured to support) fallocate() operation.
> 
> Existing bdrv_write_zeroes() falls back to writing zero buffers if
> write_zeroes is not supported by the driver.
> A new flag (BDRV_REQ_ALLOCATE) is introduced to avoid that but return ENOTSUP.
> Such allocate requests are also implemented to possibly overlap with the
> other requests. No wait is performed but an error returned in such case as well.
> So the operation should be considered advisory and a fallback scenario still
> handled by the caller (in this case, qcow2 driver).
> 
> simple perf test:
> 
>    qemu-img create -f qcow2 test.img 4G && \
>    qemu-img bench -c $((1024*1024)) -f qcow2 -n -s 4k -t none -w test.img
> 
> test results (seconds):
> 
>      +-----------+-------+------+-------+------+------+
>      |   file    |    before    |     after    | gain |
>      +-----------+-------+------+-------+------+------+
>      |    ssd    |      61.153  |      36.313  |  41% |
>      |    hdd    |     112.676  |     122.056  |  -8% |
>      +-----------+--------------+--------------+------+
> 

ping