xen_disk: add discard support

Implement discard support for xen_disk. It makes use of the existing
discard code in qemu.

The discard support is enabled unconditionally. The tool stack may
provide a property "discard-enable" in the backend node to optionally
disable discard support.  This is helpful in case the backing file was
intentionally created non-sparse to avoid fragmentation.

Signed-off-by: Olaf Hering <olaf@aepfle.de>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>

3 files changed, 52 insertions, 0 deletions

diff --git a/hw/block/xen_blkif.h b/hw/block/xen_blkif.h
index c0f4136228..711b692742 100644
--- a/hw/block/xen_blkif.h
+++ b/hw/block/xen_blkif.h
@@ -79,6 +79,12 @@ static inline void blkif_get_x86_32_req(blkif_request_t *dst, blkif_x86_32_reque
 	dst->handle = src->handle;
 	dst->id = src->id;
 	dst->sector_number = src->sector_number;
+	if (src->operation == BLKIF_OP_DISCARD) {
+		struct blkif_request_discard *s = (void *)src;
+		struct blkif_request_discard *d = (void *)dst;
+		d->nr_sectors = s->nr_sectors;
+		return;
+	}
 	if (n > src->nr_segments)
 		n = src->nr_segments;
 	for (i = 0; i < n; i++)
@@ -94,6 +100,12 @@ static inline void blkif_get_x86_64_req(blkif_request_t *dst, blkif_x86_64_reque
 	dst->handle = src->handle;
 	dst->id = src->id;
 	dst->sector_number = src->sector_number;
+	if (src->operation == BLKIF_OP_DISCARD) {
+		struct blkif_request_discard *s = (void *)src;
+		struct blkif_request_discard *d = (void *)dst;
+		d->nr_sectors = s->nr_sectors;
+		return;
+	}
 	if (n > src->nr_segments)
 		n = src->nr_segments;
 	for (i = 0; i < n; i++)
diff --git a/hw/block/xen_disk.c b/hw/block/xen_disk.c
index a8fea72edf..aed5b5b3e9 100644
--- a/hw/block/xen_disk.c
+++ b/hw/block/xen_disk.c
@@ -114,6 +114,7 @@ struct XenBlkDev {
     int                 requests_finished;
 
     /* Persistent grants extension */
+    gboolean            feature_discard;
     gboolean            feature_persistent;
     GTree               *persistent_gnts;
     unsigned int        persistent_gnt_count;
@@ -253,6 +254,8 @@ static int ioreq_parse(struct ioreq *ioreq)
     case BLKIF_OP_WRITE:
         ioreq->prot = PROT_READ; /* from memory */
         break;
+    case BLKIF_OP_DISCARD:
+        return 0;
     default:
         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
                       ioreq->req.operation);
@@ -492,6 +495,7 @@ static void qemu_aio_complete(void *opaque, int ret)
     case BLKIF_OP_READ:
         bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
         break;
+    case BLKIF_OP_DISCARD:
     default:
         break;
     }
@@ -532,6 +536,15 @@ static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
                         qemu_aio_complete, ioreq);
         break;
+    case BLKIF_OP_DISCARD:
+    {
+        struct blkif_request_discard *discard_req = (void *)&ioreq->req;
+        ioreq->aio_inflight++;
+        bdrv_aio_discard(blkdev->bs,
+                        discard_req->sector_number, discard_req->nr_sectors,
+                        qemu_aio_complete, ioreq);
+        break;
+    }
     default:
         /* unknown operation (shouldn't happen -- parse catches this) */
         goto err;
@@ -710,6 +723,21 @@ static void blk_alloc(struct XenDevice *xendev)
     }
 }
 
+static void blk_parse_discard(struct XenBlkDev *blkdev)
+{
+    int enable;
+
+    blkdev->feature_discard = true;
+
+    if (xenstore_read_be_int(&blkdev->xendev, "discard-enable", &enable) == 0) {
+        blkdev->feature_discard = !!enable;
+    }
+
+    if (blkdev->feature_discard) {
+        xenstore_write_be_int(&blkdev->xendev, "feature-discard", 1);
+    }
+}
+
 static int blk_init(struct XenDevice *xendev)
 {
     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
@@ -777,6 +805,8 @@ static int blk_init(struct XenDevice *xendev)
     xenstore_write_be_int(&blkdev->xendev, "feature-persistent", 1);
     xenstore_write_be_int(&blkdev->xendev, "info", info);
 
+    blk_parse_discard(blkdev);
+
     g_free(directiosafe);
     return 0;
 
@@ -812,6 +842,9 @@ static int blk_connect(struct XenDevice *xendev)
         qflags |= BDRV_O_RDWR;
         readonly = false;
     }
+    if (blkdev->feature_discard) {
+        qflags |= BDRV_O_UNMAP;
+    }
 
     /* init qemu block driver */
     index = (blkdev->xendev.dev - 202 * 256) / 16;
diff --git a/include/hw/xen/xen_common.h b/include/hw/xen/xen_common.h
index 2d5a25bf40..07731b9289 100644
--- a/include/hw/xen/xen_common.h
+++ b/include/hw/xen/xen_common.h
@@ -144,6 +144,13 @@ static inline int xen_xc_hvm_inject_msi(XenXC xen_xc, domid_t dom,
 {
     return -ENOSYS;
 }
+/* The followings are only to compile op_discard related code on older
+ * Xen releases. */
+#define BLKIF_OP_DISCARD 5
+struct blkif_request_discard {
+    uint64_t nr_sectors;
+    uint64_t sector_number;
+};
 #else
 static inline int xen_xc_hvm_inject_msi(XenXC xen_xc, domid_t dom,
         uint64_t addr, uint32_t data)