diff mbox

[3/8] qed: add support for Copy-on-Read

Message ID 1303910855-28999-4-git-send-email-stefanha@linux.vnet.ibm.com
State New
Headers show

Commit Message

Stefan Hajnoczi April 27, 2011, 1:27 p.m. UTC 
From: Anthony Liguori <aliguori@us.ibm.com>

When creating an image using qemu-img, just pass '-o copy_on_read' and then
whenever QED reads from a backing file, it will write the block to the QED
file after the read completes ensuring that you only fetch from the backing
device once.

This is very useful for streaming images over a slow connection.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
---
 block/qed.c |   51 ++++++++++++++++++++++++++++++++++++++++++++++++---
 block/qed.h |   15 +++++++++++----
 2 files changed, 59 insertions(+), 7 deletions(-)

Comments

Paolo Bonzini April 27, 2011, 2:29 p.m. UTC | #1 
On 04/27/2011 03:27 PM, Stefan Hajnoczi wrote:
> From: Anthony Liguori<aliguori@us.ibm.com>
>
> When creating an image using qemu-img, just pass '-o copy_on_read' and then
> whenever QED reads from a backing file, it will write the block to the QED
> file after the read completes ensuring that you only fetch from the backing
> device once.
>
> This is very useful for streaming images over a slow connection.

While having the default in the file is sane, it seems to me that it 
should be overridable at runtime, too.

Paolo
Kevin Wolf April 29, 2011, 12:14 p.m. UTC | #2 
Am 27.04.2011 15:27, schrieb Stefan Hajnoczi:
> From: Anthony Liguori <aliguori@us.ibm.com>
> 
> When creating an image using qemu-img, just pass '-o copy_on_read' and then
> whenever QED reads from a backing file, it will write the block to the QED
> file after the read completes ensuring that you only fetch from the backing
> device once.

As discussed previously, this is not the right primary interface.

Also, the patch still seems to be broken with respect to concurrent writes.

Kevin
Stefan Hajnoczi May 6, 2011, 1:24 p.m. UTC | #3 
On Fri, Apr 29, 2011 at 1:14 PM, Kevin Wolf <kwolf@redhat.com> wrote:
> Am 27.04.2011 15:27, schrieb Stefan Hajnoczi:
>> From: Anthony Liguori <aliguori@us.ibm.com>
>>
>> When creating an image using qemu-img, just pass '-o copy_on_read' and then
>> whenever QED reads from a backing file, it will write the block to the QED
>> file after the read completes ensuring that you only fetch from the backing
>> device once.
>
> As discussed previously, this is not the right primary interface.
>
> Also, the patch still seems to be broken with respect to concurrent writes.

Both good points, I really wanted to give a full RFC patch series to
show all the code that has been written today.  But your points need
to be addressed for the next version.

Stefan
diff mbox

Patch

diff --git a/block/qed.c b/block/qed.c
index c8c5930..7487683 100644
--- a/block/qed.c
+++ b/block/qed.c
@@ -448,7 +448,8 @@  static int bdrv_qed_flush(BlockDriverState *bs)
 
 static int qed_create(const char *filename, uint32_t cluster_size,
                       uint64_t image_size, uint32_t table_size,
-                      const char *backing_file, const char *backing_fmt)
+                      const char *backing_file, const char *backing_fmt,
+                      bool copy_on_read)
 {
     QEDHeader header = {
         .magic = QED_MAGIC,
@@ -490,6 +491,9 @@  static int qed_create(const char *filename, uint32_t cluster_size,
         if (qed_fmt_is_raw(backing_fmt)) {
             header.features |= QED_F_BACKING_FORMAT_NO_PROBE;
         }
+        if (copy_on_read) {
+            header.compat_features |= QED_CF_COPY_ON_READ;
+        }
     }
 
     qed_header_cpu_to_le(&header, &le_header);
@@ -523,6 +527,7 @@  static int bdrv_qed_create(const char *filename, QEMUOptionParameter *options)
     uint32_t table_size = QED_DEFAULT_TABLE_SIZE;
     const char *backing_file = NULL;
     const char *backing_fmt = NULL;
+    bool copy_on_read = false;
 
     while (options && options->name) {
         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
@@ -539,6 +544,10 @@  static int bdrv_qed_create(const char *filename, QEMUOptionParameter *options)
             if (options->value.n) {
                 table_size = options->value.n;
             }
+        } else if (!strcmp(options->name, "copy_on_read")) {
+            if (options->value.n) {
+                copy_on_read = true;
+            }
         }
         options++;
     }
@@ -559,9 +568,14 @@  static int bdrv_qed_create(const char *filename, QEMUOptionParameter *options)
                 qed_max_image_size(cluster_size, table_size));
         return -EINVAL;
     }
+    if (copy_on_read && !backing_file) {
+        fprintf(stderr,
+                "QED only supports Copy-on-Read with a backing file\n");
+        return -EINVAL;
+    }
 
     return qed_create(filename, cluster_size, image_size, table_size,
-                      backing_file, backing_fmt);
+                      backing_file, backing_fmt, copy_on_read);
 }
 
 typedef struct {
@@ -1092,6 +1106,27 @@  static void qed_aio_write_data(void *opaque, int ret,
 }
 
 /**
+ * Copy on read callback
+ *
+ * Write data from backing file to QED that's been read if CoR is enabled.
+ */
+static void qed_copy_on_read_cb(void *opaque, int ret)
+{
+    QEDAIOCB *acb = opaque;
+    BDRVQEDState *s = acb_to_s(acb);
+    BlockDriverAIOCB *cor_acb;
+
+    cor_acb = bdrv_aio_writev(s->bs,
+                              acb->cur_pos / BDRV_SECTOR_SIZE,
+                              &acb->cur_qiov,
+                              acb->cur_qiov.size / BDRV_SECTOR_SIZE,
+                              qed_aio_next_io, acb);
+    if (!cor_acb) {
+        qed_aio_complete(acb, -EIO);
+    }
+}
+
+/**
  * Read data cluster
  *
  * @opaque:     Read request
@@ -1127,8 +1162,14 @@  static void qed_aio_read_data(void *opaque, int ret,
         qed_aio_next_io(acb, 0);
         return;
     } else if (ret != QED_CLUSTER_FOUND) {
+        BlockDriverCompletionFunc *cb = qed_aio_next_io;
+
+        if (bs->backing_hd &&
+            (s->header.compat_features & QED_CF_COPY_ON_READ)) {
+            cb = qed_copy_on_read_cb;
+        }
         qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov,
-                              qed_aio_next_io, acb);
+                              cb, acb);
         return;
     }
 
@@ -1349,6 +1390,10 @@  static QEMUOptionParameter qed_create_options[] = {
         .name = BLOCK_OPT_TABLE_SIZE,
         .type = OPT_SIZE,
         .help = "L1/L2 table size (in clusters)"
+    }, {
+        .name = "copy_on_read",
+        .type = OPT_FLAG,
+        .help = "Copy blocks from base image on read"
     },
     { /* end of list */ }
 };
diff --git a/block/qed.h b/block/qed.h
index 3e1ab84..845a80e 100644
--- a/block/qed.h
+++ b/block/qed.h
@@ -56,12 +56,19 @@  enum {
     /* The backing file format must not be probed, treat as raw image */
     QED_F_BACKING_FORMAT_NO_PROBE = 0x04,
 
-    /* Feature bits must be used when the on-disk format changes */
-    QED_FEATURE_MASK = QED_F_BACKING_FILE | /* supported feature bits */
+    /* Reads to the backing file should populate the image file */
+    QED_CF_COPY_ON_READ = 0x01,
+
+    /* Supported feature bits */
+    QED_FEATURE_MASK = QED_F_BACKING_FILE |
                        QED_F_NEED_CHECK |
                        QED_F_BACKING_FORMAT_NO_PROBE,
-    QED_COMPAT_FEATURE_MASK = 0,            /* supported compat feature bits */
-    QED_AUTOCLEAR_FEATURE_MASK = 0,         /* supported autoclear feature bits */
+
+    /* Supported compat feature bits */
+    QED_COMPAT_FEATURE_MASK = QED_CF_COPY_ON_READ,
+
+    /* Supported autoclear feature bits */
+    QED_AUTOCLEAR_FEATURE_MASK = 0,
 
     /* Data is stored in groups of sectors called clusters.  Cluster size must
      * be large to avoid keeping too much metadata.  I/O requests that have