[Qemu-devel] [PATCH v5 00/10] qemu-img convert with copy offloading

[Qemu-devel] [PATCH v5 00/10] qemu-img convert with copy offloading

[Fam Zheng]

v5: - Fix raw offset/bytes check for read. [Eric]
    - Fix qcow2_handle_l2meta. [Stefan]
    - Add coroutine_fn whereever appropriate. [Stefan]

v4: - Fix raw offset and size. [Eric]
    - iscsi: Drop unnecessary return values and variables in favor of
      constants. [Stefan]
    - qcow2: Handle small backing case. [Stefan]
    - file-posix: Translate ENOSYS to ENOTSUP. [Stefan]
    - API documentation and commit message. [Stefan]
    - Add rev-by to patches 3, 5 - 10. [Stefan, Eric]

This series introduces block layer API for copy offloading and makes use of it
in qemu-img convert.

For now we implemented the operation in local file protocol with
copy_file_range(2).  Besides that it's possible to add similar to iscsi, nfs
and potentially more.

As far as its usage goes, in addition to qemu-img convert, we can emulate
offloading in scsi-disk (handle EXTENDED COPY command), and use the API in
block jobs too.

Fam Zheng (10):
  block: Introduce API for copy offloading
  raw: Check byte range uniformly
  raw: Implement copy offloading
  qcow2: Implement copy offloading
  file-posix: Implement bdrv_co_copy_range
  iscsi: Query and save device designator when opening
  iscsi: Create and use iscsi_co_wait_for_task
  iscsi: Implement copy offloading
  block-backend: Add blk_co_copy_range
  qemu-img: Convert with copy offloading

 block/block-backend.c          |  18 +++
 block/file-posix.c             |  96 ++++++++++++-
 block/io.c                     |  97 +++++++++++++
 block/iscsi.c                  | 314 ++++++++++++++++++++++++++++++++++++-----
 block/qcow2.c                  | 226 +++++++++++++++++++++++++----
 block/raw-format.c             |  96 +++++++++----
 include/block/block.h          |  32 +++++
 include/block/block_int.h      |  38 +++++
 include/block/raw-aio.h        |  10 +-
 include/scsi/constants.h       |   5 +
 include/sysemu/block-backend.h |   4 +
 qemu-img.c                     |  50 ++++++-
 12 files changed, 887 insertions(+), 99 deletions(-)

-- 
2.14.3

[Qemu-devel] [PATCH v5 01/10] block: Introduce API for copy offloading

[Fam Zheng]

Introduce the bdrv_co_copy_range() API for copy offloading.  Block
drivers implementing this API support efficient copy operations that
avoid reading each block from the source device and writing it to the
destination devices.  Examples of copy offload primitives are SCSI
EXTENDED COPY and Linux copy_file_range(2).

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/io.c                | 97 +++++++++++++++++++++++++++++++++++++++++++++++
 include/block/block.h     | 32 ++++++++++++++++
 include/block/block_int.h | 38 +++++++++++++++++++
 3 files changed, 167 insertions(+)

diff --git a/block/io.c b/block/io.c
index ca96b487eb..b7beaeeb9f 100644
--- a/block/io.c
+++ b/block/io.c
@@ -2835,3 +2835,100 @@ void bdrv_unregister_buf(BlockDriverState *bs, void *host)
         bdrv_unregister_buf(child->bs, host);
     }
 }
+
+static int coroutine_fn bdrv_co_copy_range_internal(BdrvChild *src,
+                                                    uint64_t src_offset,
+                                                    BdrvChild *dst,
+                                                    uint64_t dst_offset,
+                                                    uint64_t bytes,
+                                                    BdrvRequestFlags flags,
+                                                    bool recurse_src)
+{
+    int ret;
+
+    if (!src || !dst || !src->bs || !dst->bs) {
+        return -ENOMEDIUM;
+    }
+    ret = bdrv_check_byte_request(src->bs, src_offset, bytes);
+    if (ret) {
+        return ret;
+    }
+
+    ret = bdrv_check_byte_request(dst->bs, dst_offset, bytes);
+    if (ret) {
+        return ret;
+    }
+    if (flags & BDRV_REQ_ZERO_WRITE) {
+        return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, flags);
+    }
+
+    if (!src->bs->drv->bdrv_co_copy_range_from
+        || !dst->bs->drv->bdrv_co_copy_range_to
+        || src->bs->encrypted || dst->bs->encrypted) {
+        return -ENOTSUP;
+    }
+    if (recurse_src) {
+        return src->bs->drv->bdrv_co_copy_range_from(src->bs,
+                                                     src, src_offset,
+                                                     dst, dst_offset,
+                                                     bytes, flags);
+    } else {
+        return dst->bs->drv->bdrv_co_copy_range_to(dst->bs,
+                                                   src, src_offset,
+                                                   dst, dst_offset,
+                                                   bytes, flags);
+    }
+}
+
+/* Copy range from @src to @dst.
+ *
+ * See the comment of bdrv_co_copy_range for the parameter and return value
+ * semantics. */
+int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset,
+                                         BdrvChild *dst, uint64_t dst_offset,
+                                         uint64_t bytes, BdrvRequestFlags flags)
+{
+    return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
+                                       bytes, flags, true);
+}
+
+/* Copy range from @src to @dst.
+ *
+ * See the comment of bdrv_co_copy_range for the parameter and return value
+ * semantics. */
+int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset,
+                                       BdrvChild *dst, uint64_t dst_offset,
+                                       uint64_t bytes, BdrvRequestFlags flags)
+{
+    return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
+                                       bytes, flags, false);
+}
+
+int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset,
+                                    BdrvChild *dst, uint64_t dst_offset,
+                                    uint64_t bytes, BdrvRequestFlags flags)
+{
+    BdrvTrackedRequest src_req, dst_req;
+    BlockDriverState *src_bs = src->bs;
+    BlockDriverState *dst_bs = dst->bs;
+    int ret;
+
+    bdrv_inc_in_flight(src_bs);
+    bdrv_inc_in_flight(dst_bs);
+    tracked_request_begin(&src_req, src_bs, src_offset,
+                          bytes, BDRV_TRACKED_READ);
+    tracked_request_begin(&dst_req, dst_bs, dst_offset,
+                          bytes, BDRV_TRACKED_WRITE);
+
+    wait_serialising_requests(&src_req);
+    wait_serialising_requests(&dst_req);
+    ret = bdrv_co_copy_range_from(src, src_offset,
+                                  dst, dst_offset,
+                                  bytes, flags);
+
+    tracked_request_end(&src_req);
+    tracked_request_end(&dst_req);
+    bdrv_dec_in_flight(src_bs);
+    bdrv_dec_in_flight(dst_bs);
+    return ret;
+}
diff --git a/include/block/block.h b/include/block/block.h
index 3894edda9d..6cc6c7e699 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -611,4 +611,36 @@ bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
  */
 void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size);
 void bdrv_unregister_buf(BlockDriverState *bs, void *host);
+
+/**
+ *
+ * bdrv_co_copy_range:
+ *
+ * Do offloaded copy between two children. If the operation is not implemented
+ * by the driver, or if the backend storage doesn't support it, a negative
+ * error code will be returned.
+ *
+ * Note: block layer doesn't emulate or fallback to a bounce buffer approach
+ * because usually the caller shouldn't attempt offloaded copy any more (e.g.
+ * calling copy_file_range(2)) after the first error, thus it should fall back
+ * to a read+write path in the caller level.
+ *
+ * @src: Source child to copy data from
+ * @src_offset: offset in @src image to read data
+ * @dst: Destination child to copy data to
+ * @dst_offset: offset in @dst image to write data
+ * @bytes: number of bytes to copy
+ * @flags: request flags. Must be one of:
+ *         0 - actually read data from src;
+ *         BDRV_REQ_ZERO_WRITE - treat the @src range as zero data and do zero
+ *                               write on @dst as if bdrv_co_pwrite_zeroes is
+ *                               called. Used to simplify caller code, or
+ *                               during BlockDriver.bdrv_co_copy_range_from()
+ *                               recursion.
+ *
+ * Returns: 0 if succeeded; negative error code if failed.
+ **/
+int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset,
+                                    BdrvChild *dst, uint64_t dst_offset,
+                                    uint64_t bytes, BdrvRequestFlags flags);
 #endif
diff --git a/include/block/block_int.h b/include/block/block_int.h
index 76b589da57..c4d3d84901 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -204,6 +204,37 @@ struct BlockDriver {
     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
         int64_t offset, int bytes);
 
+    /* Map [offset, offset + nbytes) range onto a child of @bs to copy from,
+     * and invoke bdrv_co_copy_range_from(child, ...), or invoke
+     * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
+     *
+     * See the comment of bdrv_co_copy_range for the parameter and return value
+     * semantics.
+     */
+    int coroutine_fn (*bdrv_co_copy_range_from)(BlockDriverState *bs,
+                                                BdrvChild *src,
+                                                uint64_t offset,
+                                                BdrvChild *dst,
+                                                uint64_t dst_offset,
+                                                uint64_t bytes,
+                                                BdrvRequestFlags flags);
+
+    /* Map [offset, offset + nbytes) range onto a child of bs to copy data to,
+     * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
+     * operation if @bs is the leaf and @src has the same BlockDriver.  Return
+     * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
+     *
+     * See the comment of bdrv_co_copy_range for the parameter and return value
+     * semantics.
+     */
+    int coroutine_fn (*bdrv_co_copy_range_to)(BlockDriverState *bs,
+                                              BdrvChild *src,
+                                              uint64_t src_offset,
+                                              BdrvChild *dst,
+                                              uint64_t dst_offset,
+                                              uint64_t bytes,
+                                              BdrvRequestFlags flags);
+
     /*
      * Building block for bdrv_block_status[_above] and
      * bdrv_is_allocated[_above].  The driver should answer only
@@ -1102,4 +1133,11 @@ void bdrv_dec_in_flight(BlockDriverState *bs);
 
 void blockdev_close_all_bdrv_states(void);
 
+int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset,
+                                         BdrvChild *dst, uint64_t dst_offset,
+                                         uint64_t bytes, BdrvRequestFlags flags);
+int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset,
+                                       BdrvChild *dst, uint64_t dst_offset,
+                                       uint64_t bytes, BdrvRequestFlags flags);
+
 #endif /* BLOCK_INT_H */
-- 
2.14.3

[Qemu-devel] [PATCH v5 02/10] raw: Check byte range uniformly

[Fam Zheng]

We don't verify the request range against s->size in the I/O callbacks
except for raw_co_pwritev. This is wrong (especially for
raw_co_pwrite_zeroes and raw_co_pdiscard), so fix them.

Signed-off-by: Fam Zheng <famz@redhat.com>
---
 block/raw-format.c | 64 +++++++++++++++++++++++++++++++++---------------------
 1 file changed, 39 insertions(+), 25 deletions(-)

diff --git a/block/raw-format.c b/block/raw-format.c
index fe33693a2d..b69a0674b3 100644
--- a/block/raw-format.c
+++ b/block/raw-format.c
@@ -167,16 +167,37 @@ static void raw_reopen_abort(BDRVReopenState *state)
     state->opaque = NULL;
 }
 
+/* Check and adjust the offset, against 'offset' and 'size' options. */
+static inline int raw_adjust_offset(BlockDriverState *bs, uint64_t *offset,
+                                    uint64_t bytes, bool is_write)
+{
+    BDRVRawState *s = bs->opaque;
+
+    if (s->has_size && (*offset > s->size || bytes > (s->size - *offset))) {
+        /* There's not enough space for the write, or the read request is
+         * out-of-range. Don't read/write anything to prevent leaking out of
+         * the size specified in options. */
+        return is_write ? -ENOSPC : -EINVAL;;
+    }
+
+    if (*offset > INT64_MAX - s->offset) {
+        return -EINVAL;
+    }
+    *offset += s->offset;
+
+    return 0;
+}
+
 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
                                       uint64_t bytes, QEMUIOVector *qiov,
                                       int flags)
 {
-    BDRVRawState *s = bs->opaque;
+    int ret;
 
-    if (offset > UINT64_MAX - s->offset) {
-        return -EINVAL;
+    ret = raw_adjust_offset(bs, &offset, bytes, false);
+    if (ret) {
+        return ret;
     }
-    offset += s->offset;
 
     BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
@@ -186,23 +207,11 @@ static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
                                        uint64_t bytes, QEMUIOVector *qiov,
                                        int flags)
 {
-    BDRVRawState *s = bs->opaque;
     void *buf = NULL;
     BlockDriver *drv;
     QEMUIOVector local_qiov;
     int ret;
 
-    if (s->has_size && (offset > s->size || bytes > (s->size - offset))) {
-        /* There's not enough space for the data. Don't write anything and just
-         * fail to prevent leaking out of the size specified in options. */
-        return -ENOSPC;
-    }
-
-    if (offset > UINT64_MAX - s->offset) {
-        ret = -EINVAL;
-        goto fail;
-    }
-
     if (bs->probed && offset < BLOCK_PROBE_BUF_SIZE && bytes) {
         /* Handling partial writes would be a pain - so we just
          * require that guests have 512-byte request alignment if
@@ -237,7 +246,10 @@ static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
         qiov = &local_qiov;
     }
 
-    offset += s->offset;
+    ret = raw_adjust_offset(bs, &offset, bytes, true);
+    if (ret) {
+        goto fail;
+    }
 
     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
     ret = bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
@@ -267,22 +279,24 @@ static int coroutine_fn raw_co_pwrite_zeroes(BlockDriverState *bs,
                                              int64_t offset, int bytes,
                                              BdrvRequestFlags flags)
 {
-    BDRVRawState *s = bs->opaque;
-    if (offset > UINT64_MAX - s->offset) {
-        return -EINVAL;
+    int ret;
+
+    ret = raw_adjust_offset(bs, (uint64_t *)&offset, bytes, true);
+    if (ret) {
+        return ret;
     }
-    offset += s->offset;
     return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
 }
 
 static int coroutine_fn raw_co_pdiscard(BlockDriverState *bs,
                                         int64_t offset, int bytes)
 {
-    BDRVRawState *s = bs->opaque;
-    if (offset > UINT64_MAX - s->offset) {
-        return -EINVAL;
+    int ret;
+
+    ret = raw_adjust_offset(bs, (uint64_t *)&offset, bytes, true);
+    if (ret) {
+        return ret;
     }
-    offset += s->offset;
     return bdrv_co_pdiscard(bs->file->bs, offset, bytes);
 }
 
-- 
2.14.3

[Qemu-devel] [PATCH v5 03/10] raw: Implement copy offloading

[Fam Zheng]

Just pass down to ->file.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/raw-format.c | 32 ++++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)

diff --git a/block/raw-format.c b/block/raw-format.c
index b69a0674b3..f2e468df6f 100644
--- a/block/raw-format.c
+++ b/block/raw-format.c
@@ -497,6 +497,36 @@ static int raw_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
     return bdrv_probe_geometry(bs->file->bs, geo);
 }
 
+static int coroutine_fn raw_co_copy_range_from(BlockDriverState *bs,
+                                               BdrvChild *src, uint64_t src_offset,
+                                               BdrvChild *dst, uint64_t dst_offset,
+                                               uint64_t bytes, BdrvRequestFlags flags)
+{
+    int ret;
+
+    ret = raw_adjust_offset(bs, &src_offset, bytes, false);
+    if (ret) {
+        return ret;
+    }
+    return bdrv_co_copy_range_from(bs->file, src_offset, dst, dst_offset,
+                                   bytes, flags);
+}
+
+static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs,
+                                             BdrvChild *src, uint64_t src_offset,
+                                             BdrvChild *dst, uint64_t dst_offset,
+                                             uint64_t bytes, BdrvRequestFlags flags)
+{
+    int ret;
+
+    ret = raw_adjust_offset(bs, &dst_offset, bytes, true);
+    if (ret) {
+        return ret;
+    }
+    return bdrv_co_copy_range_to(src, src_offset, bs->file, dst_offset, bytes,
+                                 flags);
+}
+
 BlockDriver bdrv_raw = {
     .format_name          = "raw",
     .instance_size        = sizeof(BDRVRawState),
@@ -513,6 +543,8 @@ BlockDriver bdrv_raw = {
     .bdrv_co_pwrite_zeroes = &raw_co_pwrite_zeroes,
     .bdrv_co_pdiscard     = &raw_co_pdiscard,
     .bdrv_co_block_status = &raw_co_block_status,
+    .bdrv_co_copy_range_from = &raw_co_copy_range_from,
+    .bdrv_co_copy_range_to  = &raw_co_copy_range_to,
     .bdrv_truncate        = &raw_truncate,
     .bdrv_getlength       = &raw_getlength,
     .has_variable_length  = true,
-- 
2.14.3

[Qemu-devel] [PATCH v5 04/10] qcow2: Implement copy offloading

[Fam Zheng]

The two callbacks are implemented quite similarly to the read/write
functions: bdrv_co_copy_range_from maps for read and calls into bs->file
or bs->backing depending on the allocation status; bdrv_co_copy_range_to
maps for write and calls into bs->file.

Signed-off-by: Fam Zheng <famz@redhat.com>
---
 block/qcow2.c | 226 ++++++++++++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 196 insertions(+), 30 deletions(-)

diff --git a/block/qcow2.c b/block/qcow2.c
index 6d532470a8..e32a3c1518 100644
--- a/block/qcow2.c
+++ b/block/qcow2.c
@@ -1762,6 +1762,39 @@ static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
     return status;
 }
 
+static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
+                                            QCowL2Meta **pl2meta,
+                                            bool link_l2)
+{
+    int ret = 0;
+    QCowL2Meta *l2meta = *pl2meta;
+
+    while (l2meta != NULL) {
+        QCowL2Meta *next;
+
+        if (!ret && link_l2) {
+            ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
+            if (ret) {
+                goto out;
+            }
+        }
+
+        /* Take the request off the list of running requests */
+        if (l2meta->nb_clusters != 0) {
+            QLIST_REMOVE(l2meta, next_in_flight);
+        }
+
+        qemu_co_queue_restart_all(&l2meta->dependent_requests);
+
+        next = l2meta->next;
+        g_free(l2meta);
+        l2meta = next;
+    }
+out:
+    *pl2meta = l2meta;
+    return ret;
+}
+
 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
                                         uint64_t bytes, QEMUIOVector *qiov,
                                         int flags)
@@ -2048,24 +2081,9 @@ static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
             }
         }
 
-        while (l2meta != NULL) {
-            QCowL2Meta *next;
-
-            ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
-            if (ret < 0) {
-                goto fail;
-            }
-
-            /* Take the request off the list of running requests */
-            if (l2meta->nb_clusters != 0) {
-                QLIST_REMOVE(l2meta, next_in_flight);
-            }
-
-            qemu_co_queue_restart_all(&l2meta->dependent_requests);
-
-            next = l2meta->next;
-            g_free(l2meta);
-            l2meta = next;
+        ret = qcow2_handle_l2meta(bs, &l2meta, true);
+        if (ret) {
+            goto fail;
         }
 
         bytes -= cur_bytes;
@@ -2076,18 +2094,7 @@ static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
     ret = 0;
 
 fail:
-    while (l2meta != NULL) {
-        QCowL2Meta *next;
-
-        if (l2meta->nb_clusters != 0) {
-            QLIST_REMOVE(l2meta, next_in_flight);
-        }
-        qemu_co_queue_restart_all(&l2meta->dependent_requests);
-
-        next = l2meta->next;
-        g_free(l2meta);
-        l2meta = next;
-    }
+    qcow2_handle_l2meta(bs, &l2meta, false);
 
     qemu_co_mutex_unlock(&s->lock);
 
@@ -3274,6 +3281,163 @@ static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
     return ret;
 }
 
+static int coroutine_fn
+qcow2_co_copy_range_from(BlockDriverState *bs,
+                         BdrvChild *src, uint64_t src_offset,
+                         BdrvChild *dst, uint64_t dst_offset,
+                         uint64_t bytes, BdrvRequestFlags flags)
+{
+    BDRVQcow2State *s = bs->opaque;
+    int ret;
+    unsigned int cur_bytes; /* number of bytes in current iteration */
+    BdrvChild *child = NULL;
+
+    assert(!bs->encrypted);
+    qemu_co_mutex_lock(&s->lock);
+
+    while (bytes != 0) {
+        uint64_t copy_offset = 0;
+        /* prepare next request */
+        cur_bytes = MIN(bytes, INT_MAX);
+
+        ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, &copy_offset);
+        if (ret < 0) {
+            goto out;
+        }
+
+        switch (ret) {
+        case QCOW2_CLUSTER_UNALLOCATED:
+            if (bs->backing && bs->backing->bs) {
+                int64_t backing_length = bdrv_getlength(bs->backing->bs);
+                if (src_offset >= backing_length) {
+                    flags |= BDRV_REQ_ZERO_WRITE;
+                } else {
+                    child = bs->backing;
+                    cur_bytes = MIN(cur_bytes, backing_length - src_offset);
+                    copy_offset = src_offset;
+                }
+            } else {
+                flags |= BDRV_REQ_ZERO_WRITE;
+            }
+            break;
+
+        case QCOW2_CLUSTER_ZERO_PLAIN:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+            flags |= BDRV_REQ_ZERO_WRITE;
+            break;
+
+        case QCOW2_CLUSTER_COMPRESSED:
+            ret = -ENOTSUP;
+            goto out;
+            break;
+
+        case QCOW2_CLUSTER_NORMAL:
+            child = bs->file;
+            copy_offset += offset_into_cluster(s, src_offset);
+            if ((copy_offset & 511) != 0) {
+                ret = -EIO;
+                goto out;
+            }
+            break;
+
+        default:
+            abort();
+        }
+        qemu_co_mutex_unlock(&s->lock);
+        ret = bdrv_co_copy_range_from(child,
+                                      copy_offset,
+                                      dst, dst_offset,
+                                      cur_bytes, flags);
+        qemu_co_mutex_lock(&s->lock);
+        if (ret < 0) {
+            goto out;
+        }
+
+        bytes -= cur_bytes;
+        src_offset += cur_bytes;
+    }
+    ret = 0;
+
+out:
+    qemu_co_mutex_unlock(&s->lock);
+    return ret;
+}
+
+static int coroutine_fn
+qcow2_co_copy_range_to(BlockDriverState *bs,
+                       BdrvChild *src, uint64_t src_offset,
+                       BdrvChild *dst, uint64_t dst_offset,
+                       uint64_t bytes, BdrvRequestFlags flags)
+{
+    BDRVQcow2State *s = bs->opaque;
+    int offset_in_cluster;
+    int ret;
+    unsigned int cur_bytes; /* number of sectors in current iteration */
+    uint64_t cluster_offset;
+    uint8_t *cluster_data = NULL;
+    QCowL2Meta *l2meta = NULL;
+
+    assert(!bs->encrypted);
+    s->cluster_cache_offset = -1; /* disable compressed cache */
+
+    qemu_co_mutex_lock(&s->lock);
+
+    while (bytes != 0) {
+
+        l2meta = NULL;
+
+        offset_in_cluster = offset_into_cluster(s, dst_offset);
+        cur_bytes = MIN(bytes, INT_MAX);
+
+        /* TODO:
+         * If src->bs == dst->bs, we could simply copy by incrementing
+         * the refcnt, without copying user data.
+         * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
+        ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes,
+                                         &cluster_offset, &l2meta);
+        if (ret < 0) {
+            goto fail;
+        }
+
+        assert((cluster_offset & 511) == 0);
+
+        ret = qcow2_pre_write_overlap_check(bs, 0,
+                cluster_offset + offset_in_cluster, cur_bytes);
+        if (ret < 0) {
+            goto fail;
+        }
+
+        qemu_co_mutex_unlock(&s->lock);
+        ret = bdrv_co_copy_range_to(src, src_offset,
+                                    bs->file,
+                                    cluster_offset + offset_in_cluster,
+                                    cur_bytes, flags);
+        qemu_co_mutex_lock(&s->lock);
+        if (ret < 0) {
+            goto fail;
+        }
+
+        ret = qcow2_handle_l2meta(bs, &l2meta, true);
+        if (ret) {
+            goto fail;
+        }
+
+        bytes -= cur_bytes;
+        dst_offset += cur_bytes;
+    }
+    ret = 0;
+
+fail:
+    qcow2_handle_l2meta(bs, &l2meta, false);
+
+    qemu_co_mutex_unlock(&s->lock);
+
+    qemu_vfree(cluster_data);
+    trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
+
+    return ret;
+}
+
 static int qcow2_truncate(BlockDriverState *bs, int64_t offset,
                           PreallocMode prealloc, Error **errp)
 {
@@ -4522,6 +4686,8 @@ BlockDriver bdrv_qcow2 = {
 
     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
+    .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
+    .bdrv_co_copy_range_to  = qcow2_co_copy_range_to,
     .bdrv_truncate          = qcow2_truncate,
     .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed,
     .bdrv_make_empty        = qcow2_make_empty,
-- 
2.14.3

[Qemu-devel] [PATCH v5 05/10] file-posix: Implement bdrv_co_copy_range

[Fam Zheng]

With copy_file_range(2), we can implement the bdrv_co_copy_range
semantics.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/file-posix.c      | 96 +++++++++++++++++++++++++++++++++++++++++++++++--
 include/block/raw-aio.h | 10 ++++--
 2 files changed, 101 insertions(+), 5 deletions(-)

diff --git a/block/file-posix.c b/block/file-posix.c
index 5a602cfe37..550a201750 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -59,6 +59,7 @@
 #ifdef __linux__
 #include <sys/ioctl.h>
 #include <sys/param.h>
+#include <sys/syscall.h>
 #include <linux/cdrom.h>
 #include <linux/fd.h>
 #include <linux/fs.h>
@@ -187,6 +188,8 @@ typedef struct RawPosixAIOData {
 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
     off_t aio_offset;
     int aio_type;
+    int aio_fd2;
+    off_t aio_offset2;
 } RawPosixAIOData;
 
 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
@@ -1446,6 +1449,47 @@ static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
     return -ENOTSUP;
 }
 
+static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd,
+                             off_t *out_off, size_t len, unsigned int flags)
+{
+#ifdef __NR_copy_file_range
+    return syscall(__NR_copy_file_range, in_fd, in_off, out_fd,
+                   out_off, len, flags);
+#else
+    errno = ENOSYS;
+    return -1;
+#endif
+}
+
+static ssize_t handle_aiocb_copy_range(RawPosixAIOData *aiocb)
+{
+    uint64_t bytes = aiocb->aio_nbytes;
+    off_t in_off = aiocb->aio_offset;
+    off_t out_off = aiocb->aio_offset2;
+
+    while (bytes) {
+        ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off,
+                                      aiocb->aio_fd2, &out_off,
+                                      bytes, 0);
+        if (ret == -EINTR) {
+            continue;
+        }
+        if (ret < 0) {
+            if (errno == ENOSYS) {
+                return -ENOTSUP;
+            } else {
+                return -errno;
+            }
+        }
+        if (!ret) {
+            /* No progress (e.g. when beyond EOF), fall back to buffer I/O. */
+            return -ENOTSUP;
+        }
+        bytes -= ret;
+    }
+    return 0;
+}
+
 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 {
     int ret = -EOPNOTSUPP;
@@ -1526,6 +1570,9 @@ static int aio_worker(void *arg)
     case QEMU_AIO_WRITE_ZEROES:
         ret = handle_aiocb_write_zeroes(aiocb);
         break;
+    case QEMU_AIO_COPY_RANGE:
+        ret = handle_aiocb_copy_range(aiocb);
+        break;
     default:
         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
         ret = -EINVAL;
@@ -1536,9 +1583,10 @@ static int aio_worker(void *arg)
     return ret;
 }
 
-static int paio_submit_co(BlockDriverState *bs, int fd,
-                          int64_t offset, QEMUIOVector *qiov,
-                          int bytes, int type)
+static int paio_submit_co_full(BlockDriverState *bs, int fd,
+                               int64_t offset, int fd2, int64_t offset2,
+                               QEMUIOVector *qiov,
+                               int bytes, int type)
 {
     RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
     ThreadPool *pool;
@@ -1546,6 +1594,8 @@ static int paio_submit_co(BlockDriverState *bs, int fd,
     acb->bs = bs;
     acb->aio_type = type;
     acb->aio_fildes = fd;
+    acb->aio_fd2 = fd2;
+    acb->aio_offset2 = offset2;
 
     acb->aio_nbytes = bytes;
     acb->aio_offset = offset;
@@ -1561,6 +1611,13 @@ static int paio_submit_co(BlockDriverState *bs, int fd,
     return thread_pool_submit_co(pool, aio_worker, acb);
 }
 
+static inline int paio_submit_co(BlockDriverState *bs, int fd,
+                                 int64_t offset, QEMUIOVector *qiov,
+                                 int bytes, int type)
+{
+    return paio_submit_co_full(bs, fd, offset, -1, 0, qiov, bytes, type);
+}
+
 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
         int64_t offset, QEMUIOVector *qiov, int bytes,
         BlockCompletionFunc *cb, void *opaque, int type)
@@ -2451,6 +2508,35 @@ static void raw_abort_perm_update(BlockDriverState *bs)
     raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
 }
 
+static int coroutine_fn raw_co_copy_range_from(BlockDriverState *bs,
+                                               BdrvChild *src, uint64_t src_offset,
+                                               BdrvChild *dst, uint64_t dst_offset,
+                                               uint64_t bytes, BdrvRequestFlags flags)
+{
+    return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, flags);
+}
+
+static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs,
+                                             BdrvChild *src, uint64_t src_offset,
+                                             BdrvChild *dst, uint64_t dst_offset,
+                                             uint64_t bytes, BdrvRequestFlags flags)
+{
+    BDRVRawState *s = bs->opaque;
+    BDRVRawState *src_s;
+
+    assert(dst->bs == bs);
+    if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) {
+        return -ENOTSUP;
+    }
+
+    src_s = src->bs->opaque;
+    if (fd_open(bs) < 0 || fd_open(bs) < 0) {
+        return -EIO;
+    }
+    return paio_submit_co_full(bs, src_s->fd, src_offset, s->fd, dst_offset,
+                               NULL, bytes, QEMU_AIO_COPY_RANGE);
+}
+
 BlockDriver bdrv_file = {
     .format_name = "file",
     .protocol_name = "file",
@@ -2474,6 +2560,8 @@ BlockDriver bdrv_file = {
     .bdrv_co_pwritev        = raw_co_pwritev,
     .bdrv_aio_flush = raw_aio_flush,
     .bdrv_aio_pdiscard = raw_aio_pdiscard,
+    .bdrv_co_copy_range_from = raw_co_copy_range_from,
+    .bdrv_co_copy_range_to  = raw_co_copy_range_to,
     .bdrv_refresh_limits = raw_refresh_limits,
     .bdrv_io_plug = raw_aio_plug,
     .bdrv_io_unplug = raw_aio_unplug,
@@ -2952,6 +3040,8 @@ static BlockDriver bdrv_host_device = {
     .bdrv_co_pwritev        = raw_co_pwritev,
     .bdrv_aio_flush	= raw_aio_flush,
     .bdrv_aio_pdiscard   = hdev_aio_pdiscard,
+    .bdrv_co_copy_range_from = raw_co_copy_range_from,
+    .bdrv_co_copy_range_to  = raw_co_copy_range_to,
     .bdrv_refresh_limits = raw_refresh_limits,
     .bdrv_io_plug = raw_aio_plug,
     .bdrv_io_unplug = raw_aio_unplug,
diff --git a/include/block/raw-aio.h b/include/block/raw-aio.h
index 9e47b8a629..0e717fd475 100644
--- a/include/block/raw-aio.h
+++ b/include/block/raw-aio.h
@@ -25,9 +25,15 @@
 #define QEMU_AIO_FLUSH        0x0008
 #define QEMU_AIO_DISCARD      0x0010
 #define QEMU_AIO_WRITE_ZEROES 0x0020
+#define QEMU_AIO_COPY_RANGE   0x0040
 #define QEMU_AIO_TYPE_MASK \
-        (QEMU_AIO_READ|QEMU_AIO_WRITE|QEMU_AIO_IOCTL|QEMU_AIO_FLUSH| \
-         QEMU_AIO_DISCARD|QEMU_AIO_WRITE_ZEROES)
+        (QEMU_AIO_READ | \
+         QEMU_AIO_WRITE | \
+         QEMU_AIO_IOCTL | \
+         QEMU_AIO_FLUSH | \
+         QEMU_AIO_DISCARD | \
+         QEMU_AIO_WRITE_ZEROES | \
+         QEMU_AIO_COPY_RANGE)
 
 /* AIO flags */
 #define QEMU_AIO_MISALIGNED   0x1000
-- 
2.14.3

[Qemu-devel] [PATCH v5 06/10] iscsi: Query and save device designator when opening

[Fam Zheng]

The device designator data returned in INQUIRY command will be useful to
fill in source/target fields during copy offloading. Do this when
connecting to the target and save the data for later use.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/iscsi.c            | 41 +++++++++++++++++++++++++++++++++++++++++
 include/scsi/constants.h |  2 ++
 2 files changed, 43 insertions(+)

diff --git a/block/iscsi.c b/block/iscsi.c
index 3fd7203916..6d0035d4b9 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -68,6 +68,7 @@ typedef struct IscsiLun {
     QemuMutex mutex;
     struct scsi_inquiry_logical_block_provisioning lbp;
     struct scsi_inquiry_block_limits bl;
+    struct scsi_inquiry_device_designator *dd;
     unsigned char *zeroblock;
     /* The allocmap tracks which clusters (pages) on the iSCSI target are
      * allocated and which are not. In case a target returns zeros for
@@ -1740,6 +1741,30 @@ static QemuOptsList runtime_opts = {
     },
 };
 
+static void iscsi_save_designator(IscsiLun *lun,
+                                  struct scsi_inquiry_device_identification *inq_di)
+{
+    struct scsi_inquiry_device_designator *desig, *copy = NULL;
+
+    for (desig = inq_di->designators; desig; desig = desig->next) {
+        if (desig->association ||
+            desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) {
+            continue;
+        }
+        /* NAA works better than T10 vendor ID based designator. */
+        if (!copy || copy->designator_type < desig->designator_type) {
+            copy = desig;
+        }
+    }
+    if (copy) {
+        lun->dd = g_new(struct scsi_inquiry_device_designator, 1);
+        *lun->dd = *copy;
+        lun->dd->next = NULL;
+        lun->dd->designator = g_malloc(copy->designator_length);
+        memcpy(lun->dd->designator, copy->designator, copy->designator_length);
+    }
+}
+
 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
                       Error **errp)
 {
@@ -1922,6 +1947,7 @@ static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
         struct scsi_task *inq_task;
         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
         struct scsi_inquiry_block_limits *inq_bl;
+        struct scsi_inquiry_device_identification *inq_di;
         switch (inq_vpd->pages[i]) {
         case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
@@ -1947,6 +1973,17 @@ static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
                    sizeof(struct scsi_inquiry_block_limits));
             scsi_free_scsi_task(inq_task);
             break;
+        case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION:
+            inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
+                                    SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION,
+                                    (void **) &inq_di, errp);
+            if (inq_task == NULL) {
+                ret = -EINVAL;
+                goto out;
+            }
+            iscsi_save_designator(iscsilun, inq_di);
+            scsi_free_scsi_task(inq_task);
+            break;
         default:
             break;
         }
@@ -2003,6 +2040,10 @@ static void iscsi_close(BlockDriverState *bs)
         iscsi_logout_sync(iscsi);
     }
     iscsi_destroy_context(iscsi);
+    if (iscsilun->dd) {
+        g_free(iscsilun->dd->designator);
+        g_free(iscsilun->dd);
+    }
     g_free(iscsilun->zeroblock);
     iscsi_allocmap_free(iscsilun);
     qemu_mutex_destroy(&iscsilun->mutex);
diff --git a/include/scsi/constants.h b/include/scsi/constants.h
index a141dd71f8..54733b7110 100644
--- a/include/scsi/constants.h
+++ b/include/scsi/constants.h
@@ -311,4 +311,6 @@
 #define MMC_PROFILE_HDDVD_RW_DL         0x005A
 #define MMC_PROFILE_INVALID             0xFFFF
 
+#define IDENT_DESCR_TGT_DESCR 0xE4
+
 #endif
-- 
2.14.3

[Qemu-devel] [PATCH v5 07/10] iscsi: Create and use iscsi_co_wait_for_task

[Fam Zheng]

This loop is repeated a growing number times. Make a helper.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
---
 block/iscsi.c | 54 +++++++++++++++++-------------------------------------
 1 file changed, 17 insertions(+), 37 deletions(-)

diff --git a/block/iscsi.c b/block/iscsi.c
index 6d0035d4b9..6a365cb07b 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -556,6 +556,17 @@ static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
                                offset / iscsilun->cluster_size) == size);
 }
 
+static void coroutine_fn iscsi_co_wait_for_task(IscsiTask *iTask,
+                                                IscsiLun *iscsilun)
+{
+    while (!iTask->complete) {
+        iscsi_set_events(iscsilun);
+        qemu_mutex_unlock(&iscsilun->mutex);
+        qemu_coroutine_yield();
+        qemu_mutex_lock(&iscsilun->mutex);
+    }
+}
+
 static int coroutine_fn
 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
                 QEMUIOVector *iov, int flags)
@@ -617,12 +628,7 @@ retry:
     scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
                           iov->niov);
 #endif
-    while (!iTask.complete) {
-        iscsi_set_events(iscsilun);
-        qemu_mutex_unlock(&iscsilun->mutex);
-        qemu_coroutine_yield();
-        qemu_mutex_lock(&iscsilun->mutex);
-    }
+    iscsi_co_wait_for_task(&iTask, iscsilun);
 
     if (iTask.task != NULL) {
         scsi_free_scsi_task(iTask.task);
@@ -693,13 +699,7 @@ retry:
         ret = -ENOMEM;
         goto out_unlock;
     }
-
-    while (!iTask.complete) {
-        iscsi_set_events(iscsilun);
-        qemu_mutex_unlock(&iscsilun->mutex);
-        qemu_coroutine_yield();
-        qemu_mutex_lock(&iscsilun->mutex);
-    }
+    iscsi_co_wait_for_task(&iTask, iscsilun);
 
     if (iTask.do_retry) {
         if (iTask.task != NULL) {
@@ -863,13 +863,8 @@ retry:
 #if LIBISCSI_API_VERSION < (20160603)
     scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
 #endif
-    while (!iTask.complete) {
-        iscsi_set_events(iscsilun);
-        qemu_mutex_unlock(&iscsilun->mutex);
-        qemu_coroutine_yield();
-        qemu_mutex_lock(&iscsilun->mutex);
-    }
 
+    iscsi_co_wait_for_task(&iTask, iscsilun);
     if (iTask.task != NULL) {
         scsi_free_scsi_task(iTask.task);
         iTask.task = NULL;
@@ -906,12 +901,7 @@ retry:
         return -ENOMEM;
     }
 
-    while (!iTask.complete) {
-        iscsi_set_events(iscsilun);
-        qemu_mutex_unlock(&iscsilun->mutex);
-        qemu_coroutine_yield();
-        qemu_mutex_lock(&iscsilun->mutex);
-    }
+    iscsi_co_wait_for_task(&iTask, iscsilun);
 
     if (iTask.task != NULL) {
         scsi_free_scsi_task(iTask.task);
@@ -1143,12 +1133,7 @@ retry:
         goto out_unlock;
     }
 
-    while (!iTask.complete) {
-        iscsi_set_events(iscsilun);
-        qemu_mutex_unlock(&iscsilun->mutex);
-        qemu_coroutine_yield();
-        qemu_mutex_lock(&iscsilun->mutex);
-    }
+    iscsi_co_wait_for_task(&iTask, iscsilun);
 
     if (iTask.task != NULL) {
         scsi_free_scsi_task(iTask.task);
@@ -1244,12 +1229,7 @@ retry:
         return -ENOMEM;
     }
 
-    while (!iTask.complete) {
-        iscsi_set_events(iscsilun);
-        qemu_mutex_unlock(&iscsilun->mutex);
-        qemu_coroutine_yield();
-        qemu_mutex_lock(&iscsilun->mutex);
-    }
+    iscsi_co_wait_for_task(&iTask, iscsilun);
 
     if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
         iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
-- 
2.14.3

[Qemu-devel] [PATCH v5 08/10] iscsi: Implement copy offloading

[Fam Zheng]

Issue EXTENDED COPY (LID1) command to implement the copy_range API.

The parameter data construction code is modified from libiscsi's
iscsi-dd.c.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/iscsi.c            | 219 +++++++++++++++++++++++++++++++++++++++++++++++
 include/scsi/constants.h |   3 +
 2 files changed, 222 insertions(+)

diff --git a/block/iscsi.c b/block/iscsi.c
index 6a365cb07b..5ea75646d9 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -2205,6 +2205,221 @@ static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs,
     iscsi_allocmap_invalidate(iscsilun);
 }
 
+static int coroutine_fn iscsi_co_copy_range_from(BlockDriverState *bs,
+                                                 BdrvChild *src,
+                                                 uint64_t src_offset,
+                                                 BdrvChild *dst,
+                                                 uint64_t dst_offset,
+                                                 uint64_t bytes,
+                                                 BdrvRequestFlags flags)
+{
+    return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, flags);
+}
+
+static struct scsi_task *iscsi_xcopy_task(int param_len)
+{
+    struct scsi_task *task;
+
+    task = g_new0(struct scsi_task, 1);
+
+    task->cdb[0]     = EXTENDED_COPY;
+    task->cdb[10]    = (param_len >> 24) & 0xFF;
+    task->cdb[11]    = (param_len >> 16) & 0xFF;
+    task->cdb[12]    = (param_len >> 8) & 0xFF;
+    task->cdb[13]    = param_len & 0xFF;
+    task->cdb_size   = 16;
+    task->xfer_dir   = SCSI_XFER_WRITE;
+    task->expxferlen = param_len;
+
+    return task;
+}
+
+static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun)
+{
+    struct scsi_inquiry_device_designator *dd = lun->dd;
+
+    memset(desc, 0, 32);
+    desc[0] = IDENT_DESCR_TGT_DESCR;
+    desc[4] = dd->code_set;
+    desc[5] = (dd->designator_type & 0xF)
+        | ((dd->association & 3) << 4);
+    desc[7] = dd->designator_length;
+    memcpy(desc + 8, dd->designator, dd->designator_length);
+
+    desc[28] = 0;
+    desc[29] = (lun->block_size >> 16) & 0xFF;
+    desc[30] = (lun->block_size >> 8) & 0xFF;
+    desc[31] = lun->block_size & 0xFF;
+}
+
+static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index,
+                                 int dst_index)
+{
+    hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */
+    hdr[1] = ((dc << 1) | cat) & 0xFF;
+    hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF;
+    /* don't account for the first 4 bytes in descriptor header*/
+    hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF;
+    hdr[4] = (src_index >> 8) & 0xFF;
+    hdr[5] = src_index & 0xFF;
+    hdr[6] = (dst_index >> 8) & 0xFF;
+    hdr[7] = dst_index & 0xFF;
+}
+
+static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat,
+                                      int src_index, int dst_index, int num_blks,
+                                      uint64_t src_lba, uint64_t dst_lba)
+{
+    iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index);
+
+    /* The caller should verify the request size */
+    assert(num_blks < 65536);
+    desc[10] = (num_blks >> 8) & 0xFF;
+    desc[11] = num_blks & 0xFF;
+    desc[12] = (src_lba >> 56) & 0xFF;
+    desc[13] = (src_lba >> 48) & 0xFF;
+    desc[14] = (src_lba >> 40) & 0xFF;
+    desc[15] = (src_lba >> 32) & 0xFF;
+    desc[16] = (src_lba >> 24) & 0xFF;
+    desc[17] = (src_lba >> 16) & 0xFF;
+    desc[18] = (src_lba >> 8) & 0xFF;
+    desc[19] = src_lba & 0xFF;
+    desc[20] = (dst_lba >> 56) & 0xFF;
+    desc[21] = (dst_lba >> 48) & 0xFF;
+    desc[22] = (dst_lba >> 40) & 0xFF;
+    desc[23] = (dst_lba >> 32) & 0xFF;
+    desc[24] = (dst_lba >> 24) & 0xFF;
+    desc[25] = (dst_lba >> 16) & 0xFF;
+    desc[26] = (dst_lba >> 8) & 0xFF;
+    desc[27] = dst_lba & 0xFF;
+}
+
+static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str,
+                                        int list_id_usage, int prio,
+                                        int tgt_desc_len,
+                                        int seg_desc_len, int inline_data_len)
+{
+    buf[0] = list_id;
+    buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7);
+    buf[2] = (tgt_desc_len >> 8) & 0xFF;
+    buf[3] = tgt_desc_len & 0xFF;
+    buf[8] = (seg_desc_len >> 24) & 0xFF;
+    buf[9] = (seg_desc_len >> 16) & 0xFF;
+    buf[10] = (seg_desc_len >> 8) & 0xFF;
+    buf[11] = seg_desc_len & 0xFF;
+    buf[12] = (inline_data_len >> 24) & 0xFF;
+    buf[13] = (inline_data_len >> 16) & 0xFF;
+    buf[14] = (inline_data_len >> 8) & 0xFF;
+    buf[15] = inline_data_len & 0xFF;
+}
+
+static void iscsi_xcopy_data(struct iscsi_data *data,
+                             IscsiLun *src, int64_t src_lba,
+                             IscsiLun *dst, int64_t dst_lba,
+                             uint16_t num_blocks)
+{
+    uint8_t *buf;
+    const int src_offset = XCOPY_DESC_OFFSET;
+    const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE;
+    const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE;
+
+    data->size = XCOPY_DESC_OFFSET +
+                 IDENT_DESCR_TGT_DESCR_SIZE * 2 +
+                 XCOPY_BLK2BLK_SEG_DESC_SIZE;
+    data->data = g_malloc0(data->size);
+    buf = data->data;
+
+    /* Initialise the parameter list header */
+    iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */,
+                                0, 2 * IDENT_DESCR_TGT_DESCR_SIZE,
+                                XCOPY_BLK2BLK_SEG_DESC_SIZE,
+                                0);
+
+    /* Initialise CSCD list with one src + one dst descriptor */
+    iscsi_populate_target_desc(&buf[src_offset], src);
+    iscsi_populate_target_desc(&buf[dst_offset], dst);
+
+    /* Initialise one segment descriptor */
+    iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks,
+                              src_lba, dst_lba);
+}
+
+static int coroutine_fn iscsi_co_copy_range_to(BlockDriverState *bs,
+                                               BdrvChild *src,
+                                               uint64_t src_offset,
+                                               BdrvChild *dst,
+                                               uint64_t dst_offset,
+                                               uint64_t bytes,
+                                               BdrvRequestFlags flags)
+{
+    IscsiLun *dst_lun = dst->bs->opaque;
+    IscsiLun *src_lun;
+    struct IscsiTask iscsi_task;
+    struct iscsi_data data;
+    int r = 0;
+    int block_size;
+
+    if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) {
+        return -ENOTSUP;
+    }
+    src_lun = src->bs->opaque;
+
+    if (!src_lun->dd || !dst_lun->dd) {
+        return -ENOTSUP;
+    }
+    if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) {
+        return -ENOTSUP;
+    }
+    if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) {
+        return -ENOTSUP;
+    }
+    if (dst_lun->block_size != src_lun->block_size ||
+        !dst_lun->block_size) {
+        return -ENOTSUP;
+    }
+
+    block_size = dst_lun->block_size;
+    if (bytes / block_size > 65535) {
+        return -ENOTSUP;
+    }
+
+    iscsi_xcopy_data(&data,
+                     src_lun, src_offset / block_size,
+                     dst_lun, dst_offset / block_size,
+                     bytes / block_size);
+
+    iscsi_co_init_iscsitask(dst_lun, &iscsi_task);
+
+    qemu_mutex_lock(&dst_lun->mutex);
+    iscsi_task.task = iscsi_xcopy_task(data.size);
+retry:
+    if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun,
+                                 iscsi_task.task, iscsi_co_generic_cb,
+                                 &data,
+                                 &iscsi_task) != 0) {
+        r = -EIO;
+        goto out_unlock;
+    }
+
+    iscsi_co_wait_for_task(&iscsi_task, dst_lun);
+
+    if (iscsi_task.do_retry) {
+        iscsi_task.complete = 0;
+        goto retry;
+    }
+
+    if (iscsi_task.status != SCSI_STATUS_GOOD) {
+        r = iscsi_task.err_code;
+        goto out_unlock;
+    }
+
+out_unlock:
+    g_free(iscsi_task.task);
+    qemu_mutex_unlock(&dst_lun->mutex);
+    g_free(iscsi_task.err_str);
+    return r;
+}
+
 static QemuOptsList iscsi_create_opts = {
     .name = "iscsi-create-opts",
     .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
@@ -2239,6 +2454,8 @@ static BlockDriver bdrv_iscsi = {
 
     .bdrv_co_block_status  = iscsi_co_block_status,
     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
+    .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
+    .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
     .bdrv_co_readv         = iscsi_co_readv,
     .bdrv_co_writev        = iscsi_co_writev,
@@ -2274,6 +2491,8 @@ static BlockDriver bdrv_iser = {
 
     .bdrv_co_block_status  = iscsi_co_block_status,
     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
+    .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
+    .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
     .bdrv_co_readv         = iscsi_co_readv,
     .bdrv_co_writev        = iscsi_co_writev,
diff --git a/include/scsi/constants.h b/include/scsi/constants.h
index 54733b7110..b53ff8adcb 100644
--- a/include/scsi/constants.h
+++ b/include/scsi/constants.h
@@ -311,6 +311,9 @@
 #define MMC_PROFILE_HDDVD_RW_DL         0x005A
 #define MMC_PROFILE_INVALID             0xFFFF
 
+#define XCOPY_DESC_OFFSET 16
 #define IDENT_DESCR_TGT_DESCR 0xE4
+#define IDENT_DESCR_TGT_DESCR_SIZE 32
+#define XCOPY_BLK2BLK_SEG_DESC_SIZE 28
 
 #endif
-- 
2.14.3

[Qemu-devel] [PATCH v5 09/10] block-backend: Add blk_co_copy_range

[Fam Zheng]

It's a BlockBackend wrapper of the BDS interface.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/block-backend.c          | 18 ++++++++++++++++++
 include/sysemu/block-backend.h |  4 ++++
 2 files changed, 22 insertions(+)

diff --git a/block/block-backend.c b/block/block-backend.c
index 89f47b00ea..d55c328736 100644
--- a/block/block-backend.c
+++ b/block/block-backend.c
@@ -2211,3 +2211,21 @@ void blk_unregister_buf(BlockBackend *blk, void *host)
 {
     bdrv_unregister_buf(blk_bs(blk), host);
 }
+
+int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
+                                   BlockBackend *blk_out, int64_t off_out,
+                                   int bytes, BdrvRequestFlags flags)
+{
+    int r;
+    r = blk_check_byte_request(blk_in, off_in, bytes);
+    if (r) {
+        return r;
+    }
+    r = blk_check_byte_request(blk_out, off_out, bytes);
+    if (r) {
+        return r;
+    }
+    return bdrv_co_copy_range(blk_in->root, off_in,
+                              blk_out->root, off_out,
+                              bytes, flags);
+}
diff --git a/include/sysemu/block-backend.h b/include/sysemu/block-backend.h
index 92ab624fac..8d03d493c2 100644
--- a/include/sysemu/block-backend.h
+++ b/include/sysemu/block-backend.h
@@ -232,4 +232,8 @@ void blk_set_force_allow_inactivate(BlockBackend *blk);
 void blk_register_buf(BlockBackend *blk, void *host, size_t size);
 void blk_unregister_buf(BlockBackend *blk, void *host);
 
+int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
+                                   BlockBackend *blk_out, int64_t off_out,
+                                   int bytes, BdrvRequestFlags flags);
+
 #endif
-- 
2.14.3

[Qemu-devel] [PATCH v5 10/10] qemu-img: Convert with copy offloading

[Fam Zheng]

The new blk_co_copy_range interface offers a more efficient way in the
case of network based storage. Make use of it to allow faster convert
operation.

Since copy offloading cannot do zero detection ('-S') and compression
(-c), only try it when these options are not used.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 qemu-img.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 48 insertions(+), 2 deletions(-)

diff --git a/qemu-img.c b/qemu-img.c
index 2b5a5706b6..67f3631a81 100644
--- a/qemu-img.c
+++ b/qemu-img.c
@@ -1543,6 +1543,7 @@ typedef struct ImgConvertState {
     bool compressed;
     bool target_has_backing;
     bool wr_in_order;
+    bool copy_range;
     int min_sparse;
     size_t cluster_sectors;
     size_t buf_sectors;
@@ -1736,6 +1737,37 @@ static int coroutine_fn convert_co_write(ImgConvertState *s, int64_t sector_num,
     return 0;
 }
 
+static int coroutine_fn convert_co_copy_range(ImgConvertState *s, int64_t sector_num,
+                                              int nb_sectors)
+{
+    int n, ret;
+
+    while (nb_sectors > 0) {
+        BlockBackend *blk;
+        int src_cur;
+        int64_t bs_sectors, src_cur_offset;
+        int64_t offset;
+
+        convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
+        offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
+        blk = s->src[src_cur];
+        bs_sectors = s->src_sectors[src_cur];
+
+        n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
+
+        ret = blk_co_copy_range(blk, offset, s->target,
+                                sector_num << BDRV_SECTOR_BITS,
+                                n << BDRV_SECTOR_BITS, 0);
+        if (ret < 0) {
+            return ret;
+        }
+
+        sector_num += n;
+        nb_sectors -= n;
+    }
+    return 0;
+}
+
 static void coroutine_fn convert_co_do_copy(void *opaque)
 {
     ImgConvertState *s = opaque;
@@ -1758,6 +1790,7 @@ static void coroutine_fn convert_co_do_copy(void *opaque)
         int n;
         int64_t sector_num;
         enum ImgConvertBlockStatus status;
+        bool copy_range;
 
         qemu_co_mutex_lock(&s->lock);
         if (s->ret != -EINPROGRESS || s->sector_num >= s->total_sectors) {
@@ -1787,7 +1820,9 @@ static void coroutine_fn convert_co_do_copy(void *opaque)
                                         s->allocated_sectors, 0);
         }
 
-        if (status == BLK_DATA) {
+retry:
+        copy_range = s->copy_range && s->status == BLK_DATA;
+        if (status == BLK_DATA && !copy_range) {
             ret = convert_co_read(s, sector_num, n, buf);
             if (ret < 0) {
                 error_report("error while reading sector %" PRId64
@@ -1809,7 +1844,15 @@ static void coroutine_fn convert_co_do_copy(void *opaque)
         }
 
         if (s->ret == -EINPROGRESS) {
-            ret = convert_co_write(s, sector_num, n, buf, status);
+            if (copy_range) {
+                ret = convert_co_copy_range(s, sector_num, n);
+                if (ret) {
+                    s->copy_range = false;
+                    goto retry;
+                }
+            } else {
+                ret = convert_co_write(s, sector_num, n, buf, status);
+            }
             if (ret < 0) {
                 error_report("error while writing sector %" PRId64
                              ": %s", sector_num, strerror(-ret));
@@ -1932,6 +1975,7 @@ static int img_convert(int argc, char **argv)
     ImgConvertState s = (ImgConvertState) {
         /* Need at least 4k of zeros for sparse detection */
         .min_sparse         = 8,
+        .copy_range         = true,
         .buf_sectors        = IO_BUF_SIZE / BDRV_SECTOR_SIZE,
         .wr_in_order        = true,
         .num_coroutines     = 8,
@@ -1972,6 +2016,7 @@ static int img_convert(int argc, char **argv)
             break;
         case 'c':
             s.compressed = true;
+            s.copy_range = false;
             break;
         case 'o':
             if (!is_valid_option_list(optarg)) {
@@ -2013,6 +2058,7 @@ static int img_convert(int argc, char **argv)
             }
 
             s.min_sparse = sval / BDRV_SECTOR_SIZE;
+            s.copy_range = false;
             break;
         }
         case 'p':
-- 
2.14.3

Re: [Qemu-devel] [PATCH v5 02/10] raw: Check byte range uniformly

[Eric Blake]

On 05/22/2018 10:04 PM, Fam Zheng wrote:
> We don't verify the request range against s->size in the I/O callbacks
> except for raw_co_pwritev. This is wrong (especially for
> raw_co_pwrite_zeroes and raw_co_pdiscard), so fix them.

I'd also mention ...

> 
> Signed-off-by: Fam Zheng <famz@redhat.com>
> ---
>   block/raw-format.c | 64 +++++++++++++++++++++++++++++++++---------------------
>   1 file changed, 39 insertions(+), 25 deletions(-)

Should this cc: qemu-stable?

Do we have iotests coverage of this?

> 
> diff --git a/block/raw-format.c b/block/raw-format.c
> index fe33693a2d..b69a0674b3 100644
> --- a/block/raw-format.c
> +++ b/block/raw-format.c
> @@ -167,16 +167,37 @@ static void raw_reopen_abort(BDRVReopenState *state)
>       state->opaque = NULL;
>   }
>   
> +/* Check and adjust the offset, against 'offset' and 'size' options. */
> +static inline int raw_adjust_offset(BlockDriverState *bs, uint64_t *offset,
> +                                    uint64_t bytes, bool is_write)
> +{
> +    BDRVRawState *s = bs->opaque;
> +
> +    if (s->has_size && (*offset > s->size || bytes > (s->size - *offset))) {
> +        /* There's not enough space for the write, or the read request is
> +         * out-of-range. Don't read/write anything to prevent leaking out of
> +         * the size specified in options. */
> +        return is_write ? -ENOSPC : -EINVAL;;
> +    }
> +
> +    if (*offset > INT64_MAX - s->offset) {
> +        return -EINVAL;

...that this change to a 63-bit check...

> @@ -186,23 +207,11 @@ static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,

> -    if (offset > UINT64_MAX - s->offset) {
> -        ret = -EINVAL;
> -        goto fail;
> -    }

...from a previous 64-bit check is intentional.

With improved commit message, and ideally with followup commits that add 
iotest coverage,

Reviewed-by: Eric Blake <eblake@redhat.com>

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

Re: [Qemu-devel] [PATCH v5 02/10] raw: Check byte range uniformly

[Fam Zheng]

On Wed, 05/23 13:28, Eric Blake wrote:
> On 05/22/2018 10:04 PM, Fam Zheng wrote:
> > We don't verify the request range against s->size in the I/O callbacks
> > except for raw_co_pwritev. This is wrong (especially for
> > raw_co_pwrite_zeroes and raw_co_pdiscard), so fix them.
> 
> I'd also mention ...
> 
> > 
> > Signed-off-by: Fam Zheng <famz@redhat.com>
> > ---
> >   block/raw-format.c | 64 +++++++++++++++++++++++++++++++++---------------------
> >   1 file changed, 39 insertions(+), 25 deletions(-)
> 
> Should this cc: qemu-stable?
> 
> Do we have iotests coverage of this?

No for both questions.

This is not reachable from the user interface. In that sense "wrong" is not
accurate in the commit message, "inconsistent" or "confusing" might be more
appropriate.

Out of range requests can only reach us if blk_set_allow_write_beyond_eof(blk,
true) was called, which can only happen when creating image, where the I/O are
all generated internally.

So there is no behavior change in this patch. But verifying and returning error
is still better than asserting, IMO, since we don't want to assert how block
layer uses the driver callbacks. I'll tweak the commit message.

> 
> > 
> > diff --git a/block/raw-format.c b/block/raw-format.c
> > index fe33693a2d..b69a0674b3 100644
> > --- a/block/raw-format.c
> > +++ b/block/raw-format.c
> > @@ -167,16 +167,37 @@ static void raw_reopen_abort(BDRVReopenState *state)
> >       state->opaque = NULL;
> >   }
> > +/* Check and adjust the offset, against 'offset' and 'size' options. */
> > +static inline int raw_adjust_offset(BlockDriverState *bs, uint64_t *offset,
> > +                                    uint64_t bytes, bool is_write)
> > +{
> > +    BDRVRawState *s = bs->opaque;
> > +
> > +    if (s->has_size && (*offset > s->size || bytes > (s->size - *offset))) {
> > +        /* There's not enough space for the write, or the read request is
> > +         * out-of-range. Don't read/write anything to prevent leaking out of
> > +         * the size specified in options. */
> > +        return is_write ? -ENOSPC : -EINVAL;;
> > +    }
> > +
> > +    if (*offset > INT64_MAX - s->offset) {
> > +        return -EINVAL;
> 
> ...that this change to a 63-bit check...
> 
> > @@ -186,23 +207,11 @@ static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
> 
> > -    if (offset > UINT64_MAX - s->offset) {
> > -        ret = -EINVAL;
> > -        goto fail;
> > -    }
> 
> ...from a previous 64-bit check is intentional.

OK!

> 
> With improved commit message, and ideally with followup commits that add
> iotest coverage,
> 
> Reviewed-by: Eric Blake <eblake@redhat.com>
> 
> -- 
> Eric Blake, Principal Software Engineer
> Red Hat, Inc.           +1-919-301-3266
> Virtualization:  qemu.org | libvirt.org

Fam

Re: [Qemu-devel] [PATCH v5 04/10] qcow2: Implement copy offloading

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 575 bytes --]

On Wed, May 23, 2018 at 11:04:30AM +0800, Fam Zheng wrote:
> The two callbacks are implemented quite similarly to the read/write
> functions: bdrv_co_copy_range_from maps for read and calls into bs->file
> or bs->backing depending on the allocation status; bdrv_co_copy_range_to
> maps for write and calls into bs->file.
> 
> Signed-off-by: Fam Zheng <famz@redhat.com>
> ---
>  block/qcow2.c | 226 ++++++++++++++++++++++++++++++++++++++++++++++++++--------
>  1 file changed, 196 insertions(+), 30 deletions(-)

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]

Re: [Qemu-devel] [PATCH v5 02/10] raw: Check byte range uniformly

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 591 bytes --]

On Wed, May 23, 2018 at 11:04:28AM +0800, Fam Zheng wrote:
> We don't verify the request range against s->size in the I/O callbacks
> except for raw_co_pwritev. This is wrong (especially for
> raw_co_pwrite_zeroes and raw_co_pdiscard), so fix them.
> 
> Signed-off-by: Fam Zheng <famz@redhat.com>
> ---
>  block/raw-format.c | 64 +++++++++++++++++++++++++++++++++---------------------
>  1 file changed, 39 insertions(+), 25 deletions(-)

The uint64_t vs int64_t inconsistency is ugly but pre-existing due to
the interfaces:

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]