[RFC PATCH v1 12/15] hw/xen: Add backend implementation of grant table operations

[David Woodhouse]

From: David Woodhouse <dwmw@amazon.co.uk>

This is limited to mapping a single grant at a time, because under Xen the
pages are mapped *contiguously* into qemu's address space, and that's very
hard to do when those pages actually come from anonymous mappings in qemu
in the first place.

Eventually perhaps we can look at using shared mappings of actual objects
for system RAM, and then we can make new mappings of the same backing
store (be it deleted files, shmem, whatever). But for now let's stuck to
a page at a time.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
---
 hw/i386/kvm/xen_gnttab.c | 294 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 291 insertions(+), 3 deletions(-)

diff --git a/hw/i386/kvm/xen_gnttab.c b/hw/i386/kvm/xen_gnttab.c
index 5e8fc0184e..8008a391bd 100644
--- a/hw/i386/kvm/xen_gnttab.c
+++ b/hw/i386/kvm/xen_gnttab.c
@@ -21,6 +21,7 @@
 
 #include "hw/sysbus.h"
 #include "hw/xen/xen.h"
+#include "hw/xen/xen_backend_ops.h"
 #include "xen_overlay.h"
 #include "xen_gnttab.h"
 
@@ -33,12 +34,11 @@
 #define TYPE_XEN_GNTTAB "xen-gnttab"
 OBJECT_DECLARE_SIMPLE_TYPE(XenGnttabState, XEN_GNTTAB)
 
-#define XEN_PAGE_SHIFT 12
-#define XEN_PAGE_SIZE (1ULL << XEN_PAGE_SHIFT)
-
 #define ENTRIES_PER_FRAME_V1 (XEN_PAGE_SIZE / sizeof(grant_entry_v1_t))
 #define ENTRIES_PER_FRAME_V2 (XEN_PAGE_SIZE / sizeof(grant_entry_v2_t))
 
+static struct gnttab_backend_ops emu_gnttab_backend_ops;
+
 struct XenGnttabState {
     /*< private >*/
     SysBusDevice busdev;
@@ -57,6 +57,8 @@ struct XenGnttabState {
     MemoryRegion gnt_frames;
     MemoryRegion *gnt_aliases;
     uint64_t *gnt_frame_gpas;
+
+    uint8_t *map_track;
 };
 
 struct XenGnttabState *xen_gnttab_singleton;
@@ -88,9 +90,15 @@ static void xen_gnttab_realize(DeviceState *dev, Error 
**errp)
         s->gnt_frame_gpas[i] = INVALID_GPA;
     }
 
+    s->entries.v1[GNTTAB_RESERVED_XENSTORE].flags = GTF_permit_access;
+    s->entries.v1[GNTTAB_RESERVED_XENSTORE].frame = XEN_SPECIAL_PFN(XENSTORE);
     qemu_mutex_init(&s->gnt_lock);
 
     xen_gnttab_singleton = s;
+
+    s->map_track = g_new0(uint8_t, s->max_frames * ENTRIES_PER_FRAME_V1);
+
+    xen_gnttab_ops = &emu_gnttab_backend_ops;
 }
 
 static int xen_gnttab_post_load(void *opaque, int version_id)
@@ -236,3 +244,283 @@ int xen_gnttab_query_size_op(struct gnttab_query_size 
*size)
     size->max_nr_frames = s->max_frames;
     return 0;
 }
+
+/* Track per-open refs, to allow close() to clean up. */
+struct active_ref {
+    MemoryRegionSection mrs;
+    void *virtaddr;
+    uint32_t refcnt;
+    int prot;
+};
+
+static void gnt_unref(XenGnttabState *s, grant_ref_t ref,
+                      MemoryRegionSection *mrs, int prot)
+{
+    if (mrs && mrs->mr) {
+       if (prot & PROT_WRITE) {
+            memory_region_set_dirty(mrs->mr, mrs->offset_within_region,
+                                    XEN_PAGE_SIZE);
+        }
+        memory_region_unref(mrs->mr);
+        mrs->mr = NULL;
+    }
+    assert(s->map_track[ref] != 0);
+
+    if (--s->map_track[ref] == 0) {
+        grant_entry_v1_t *gnt_p = &s->entries.v1[ref];
+        qatomic_and(&gnt_p->flags, ~(GTF_reading | GTF_writing));
+    }
+}
+
+static uint64_t gnt_ref(XenGnttabState *s, grant_ref_t ref, int prot)
+{
+    uint16_t mask = GTF_type_mask | GTF_sub_page;
+    volatile grant_entry_v1_t *gnt_p;
+    grant_entry_v1_t gnt;
+    int retries = 0;
+
+    if (ref >= s->max_frames * ENTRIES_PER_FRAME_V1 ||
+        s->map_track[ref] == UINT8_MAX) {
+        return INVALID_GPA;
+    }
+
+    if (prot & PROT_WRITE) {
+        mask |= GTF_readonly;
+    }
+
+    gnt_p = &s->entries.v1[ref];
+
+    /*
+     * The guest can legitimately be changing the GTF_readonly flag. Allow
+     * that, but don't let a malicious guest cause a livelock.
+     */
+    for (retries = 0; retries < 5; retries++) {
+        uint16_t new_flags;
+        gnt = *gnt_p;
+
+        if ((gnt.flags & mask) != GTF_permit_access ||
+            gnt.domid != DOMID_QEMU) {
+            return INVALID_GPA;
+        }
+
+        new_flags = gnt.flags | GTF_reading;
+        if (prot & PROT_WRITE) {
+            new_flags |= GTF_writing;
+        }
+
+        if (qatomic_cmpxchg(&gnt_p->flags, gnt.flags, new_flags) == gnt.flags) 
{
+            return (uint64_t)gnt.frame << XEN_PAGE_SHIFT;
+        }
+    }
+
+    return INVALID_GPA;
+}
+
+struct xengntdev_handle {
+    GHashTable *active_maps;
+};
+
+static int xen_be_gnttab_set_max_grants(struct xengntdev_handle *xgt,
+                                        uint32_t nr_grants)
+{
+    return 0;
+}
+
+static void *xen_be_gnttab_map_refs(struct xengntdev_handle *xgt, uint32_t 
count,
+                                    uint32_t domid, uint32_t *refs, int prot)
+{
+    XenGnttabState *s = xen_gnttab_singleton;
+    struct active_ref *act;
+
+    if (!s) {
+        errno = ENOTSUP;
+        return NULL;
+    }
+
+    if (domid != xen_domid) {
+        errno = EINVAL;
+        return NULL;
+    }
+
+    if (!count || count > 4096) {
+        errno = EINVAL;
+        return NULL;
+    }
+
+    /*
+     * Making a contiguous mapping from potentially discontiguous grant
+     * references would be... distinctly non-trivial. We don't support it.
+     * Even changing the API to return an array of pointers, one per page,
+     * wouldn't be simple to use in PV backends because some structures
+     * actually cross page boundaries (e.g. 32-bit blkif_response ring
+     * entries are 12 bytes).
+     */
+    if (count != 1) {
+        errno = EINVAL;
+        return NULL;
+    }
+
+    QEMU_LOCK_GUARD(&s->gnt_lock);
+
+    act = g_hash_table_lookup(xgt->active_maps, GINT_TO_POINTER(refs[0]));
+    if (act) {
+        if ((prot & PROT_WRITE) && !(act->prot & PROT_WRITE)) {
+            if (gnt_ref(s, refs[0], prot) == INVALID_GPA) {
+                return NULL;
+            }
+            act->prot |= PROT_WRITE;
+        }
+        act->refcnt++;
+    } else {
+        uint64_t gpa = gnt_ref(s, refs[0], prot);
+        if (gpa == INVALID_GPA) {
+            errno = EINVAL;
+            return NULL;
+        }
+
+        act = g_new0(struct active_ref, 1);
+        act->prot = prot;
+        act->refcnt = 1;
+        act->mrs = memory_region_find(get_system_memory(), gpa, XEN_PAGE_SIZE);
+
+        if (!act->mrs.mr || act->mrs.size < XEN_PAGE_SIZE ||
+            memory_region_get_ram_addr(act->mrs.mr) == RAM_ADDR_INVALID ||
+            !(act->virtaddr = qemu_map_ram_ptr(act->mrs.mr->ram_block,
+                                               
act->mrs.offset_within_region))) {
+            gnt_unref(s, refs[0], &act->mrs, 0);
+            g_free(act);
+            errno = EINVAL;
+            return NULL;
+        }
+
+        s->map_track[refs[0]]++;
+        g_hash_table_insert(xgt->active_maps, GINT_TO_POINTER(refs[0]), act);
+    }
+
+    return act->virtaddr;
+}
+
+static gboolean do_unmap(gpointer key, gpointer value, gpointer user_data)
+{
+    XenGnttabState *s = user_data;
+    grant_ref_t gref = GPOINTER_TO_INT(key);
+    struct active_ref *act = value;
+
+    gnt_unref(s, gref, &act->mrs, act->prot);
+    g_free(act);
+    return true;
+}
+
+static int xen_be_gnttab_unmap(struct xengntdev_handle *xgt,
+                               void *start_address, uint32_t *refs,
+                               uint32_t count)
+{
+    XenGnttabState *s = xen_gnttab_singleton;
+    struct active_ref *act;
+
+    if (!s) {
+        return -ENOTSUP;
+    }
+
+    if (count != 1) {
+        return -EINVAL;
+    }
+
+    QEMU_LOCK_GUARD(&s->gnt_lock);
+
+    act = g_hash_table_lookup(xgt->active_maps, GINT_TO_POINTER(refs[0]));
+    if (!act) {
+        return -ENOENT;
+    }
+
+    if (act->virtaddr != start_address) {
+        return -EINVAL;
+    }
+
+    if (!--act->refcnt) {
+        do_unmap(GINT_TO_POINTER(refs[0]), act, s);
+        g_hash_table_remove(xgt->active_maps, GINT_TO_POINTER(refs[0]));
+    }
+
+    return 0;
+}
+
+/*
+ * This looks a bit like the one for true Xen in xen-operations.c but
+ * in emulation we don't support multi-page mappings. And under Xen we
+ * *want* the multi-page mappings so we have fewer bounces through the
+ * kernel and the hypervisor. So the code paths end up being similar,
+ * but different.
+ */
+static int xen_be_gnttab_copy(struct xengntdev_handle *xgt, bool to_domain,
+                              uint32_t domid, XenGrantCopySegment *segs,
+                              uint32_t nr_segs, Error **errp)
+{
+    int prot = to_domain ? PROT_WRITE : PROT_READ;
+    unsigned int i;
+
+    for (i = 0; i < nr_segs; i++) {
+        XenGrantCopySegment *seg = &segs[i];
+        void *page;
+        uint32_t ref = to_domain ? seg->dest.foreign.ref :
+            seg->source.foreign.ref;
+
+        page = xen_be_gnttab_map_refs(xgt, 1, domid, &ref, prot);
+        if (!page) {
+            if (errp) {
+                error_setg_errno(errp, errno,
+                                 "xen_be_gnttab_map_refs failed");
+            }
+            return -errno;
+        }
+
+        if (to_domain) {
+            memcpy(page + seg->dest.foreign.offset, seg->source.virt,
+                   seg->len);
+        } else {
+            memcpy(seg->dest.virt, page + seg->source.foreign.offset,
+                   seg->len);
+        }
+
+        if (xen_be_gnttab_unmap(xgt, page, &ref, 1)) {
+            if (errp) {
+                error_setg_errno(errp, errno, "xen_be_gnttab_unmap failed");
+            }
+            return -errno;
+        }
+    }
+
+    return 0;
+}
+
+static struct xengntdev_handle *xen_be_gnttab_open(void)
+{
+    struct xengntdev_handle *xgt = g_new0(struct xengntdev_handle, 1);
+
+    xgt->active_maps = g_hash_table_new(g_direct_hash, g_direct_equal);
+    return xgt;
+}
+
+static int xen_be_gnttab_close(struct xengntdev_handle *xgt)
+{
+    XenGnttabState *s = xen_gnttab_singleton;
+
+    if (!s) {
+        return -ENOTSUP;
+    }
+
+    g_hash_table_foreach_remove(xgt->active_maps, do_unmap, s);
+    g_hash_table_destroy(xgt->active_maps);
+    g_free(xgt);
+    return 0;
+}
+
+static struct gnttab_backend_ops emu_gnttab_backend_ops = {
+    .open = xen_be_gnttab_open,
+    .close = xen_be_gnttab_close,
+    .grant_copy = xen_be_gnttab_copy,
+    .set_max_grants = xen_be_gnttab_set_max_grants,
+    .map_refs = xen_be_gnttab_map_refs,
+    .unmap = xen_be_gnttab_unmap,
+};
+
-- 
2.35.3