References: FATE#316071
Subject: [RFC] qemu-upstream: add discard support for xen_disk

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

The discard support is enabled unconditionally. But it would be worth to
have a knob to disable it in case the backing file was intentionally
created non-sparse to avoid fragmentation.
How could this be knob be passed from domU.cfg:disk=[] to the actual
qemu process? Perhaps introduce an option discard=on|off|ignore|unmap.

Signed-off-by: Olaf Hering <olaf@aepfle.de>

---
 tools/qemu-xen-dir-remote/hw/block/xen_blkif.h |   12 ++++++++
 tools/qemu-xen-dir-remote/hw/block/xen_disk.c  |   36 +++++++++++++++++++++++++
 2 files changed, 48 insertions(+)

Index: xen-4.4.0-testing/tools/qemu-xen-dir-remote/hw/block/xen_blkif.h
===================================================================
--- xen-4.4.0-testing.orig/tools/qemu-xen-dir-remote/hw/block/xen_blkif.h
+++ xen-4.4.0-testing/tools/qemu-xen-dir-remote/hw/block/xen_blkif.h
@@ -79,6 +79,12 @@ static inline void blkif_get_x86_32_req(
 	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(
 	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++)
Index: xen-4.4.0-testing/tools/qemu-xen-dir-remote/hw/block/xen_disk.c
===================================================================
--- xen-4.4.0-testing.orig/tools/qemu-xen-dir-remote/hw/block/xen_disk.c
+++ xen-4.4.0-testing/tools/qemu-xen-dir-remote/hw/block/xen_disk.c
@@ -68,6 +68,8 @@ struct ioreq {
     int                 presync;
     int                 postsync;
     uint8_t             mapped;
+    int64_t             sector_num;
+    int                 nb_sectors;
 
     /* grant mapping */
     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
@@ -114,6 +116,7 @@ struct XenBlkDev {
     int                 requests_finished;
 
     /* Persistent grants extension */
+    gboolean            feature_discard;
     gboolean            feature_persistent;
     GTree               *persistent_gnts;
     unsigned int        persistent_gnt_count;
@@ -232,6 +235,7 @@ static void ioreq_release(struct ioreq *
 static int ioreq_parse(struct ioreq *ioreq)
 {
     struct XenBlkDev *blkdev = ioreq->blkdev;
+    struct blkif_request_discard *discard_req = (void *)&ioreq->req;
     uintptr_t mem;
     size_t len;
     int i;
@@ -244,6 +248,10 @@ static int ioreq_parse(struct ioreq *ior
     case BLKIF_OP_READ:
         ioreq->prot = PROT_WRITE; /* to memory */
         break;
+    case BLKIF_OP_DISCARD:
+        ioreq->sector_num = discard_req->sector_number;
+        ioreq->nb_sectors = discard_req->nr_sectors;
+        return 0;
     case BLKIF_OP_FLUSH_DISKCACHE:
         ioreq->presync = 1;
         if (!ioreq->req.nr_segments) {
@@ -521,6 +529,13 @@ static int ioreq_runio_qemu_aio(struct i
                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
                         qemu_aio_complete, ioreq);
         break;
+    case BLKIF_OP_DISCARD:
+        bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->nb_sectors * BLOCK_SIZE, BDRV_ACCT_WRITE);
+        ioreq->aio_inflight++;
+        bdrv_aio_discard(blkdev->bs,
+                        ioreq->sector_num, ioreq->nb_sectors,
+                        qemu_aio_complete, ioreq);
+        break;
     default:
         /* unknown operation (shouldn't happen -- parse catches this) */
         goto err;
@@ -699,6 +714,23 @@ static void blk_alloc(struct XenDevice *
     }
 }
 
+static void blk_parse_discard(struct XenBlkDev *blkdev)
+{
+    int flags = BDRV_O_UNMAP;
+#if 0
+    char *s = xenstore_read_be_str(&blkdev->xendev, "toolstack-options-discard");
+
+    if (s && bdrv_parse_discard_flags(s, &flags) < 0) {
+        xen_be_printf(xendev, 0, "property toolstack-discard-options failed to parse: %s\n", s);
+        return;
+    }
+#endif
+    blkdev->feature_discard = !!(flags & BDRV_O_UNMAP);
+    if (!blkdev->feature_discard)
+        return;
+    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);
@@ -766,6 +798,8 @@ static int blk_init(struct XenDevice *xe
     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;
 
@@ -801,6 +835,8 @@ static int blk_connect(struct XenDevice
         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;