[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH v5 09/15] argo: implement the sendv op; evtchn: expose send_guest_global_virq



sendv operation is invoked to perform a synchronous send of buffers
contained in iovs to a remote domain's registered ring.

It takes:
 * A destination address (domid, port) for the ring to send to.
   It performs a most-specific match lookup, to allow for wildcard.
 * A source address, used to inform the destination of where to reply.
 * The address of an array of iovs containing the data to send
 * .. and the length of that array of iovs
 * and a 32-bit message type, available to communicate message context
   data (eg. kernel-to-kernel, separate from the application data).

If insufficient space exists in the destination ring, it will return
-EAGAIN and Xen will notify the caller when sufficient space becomes
available.

Accesses to the ring indices are appropriately atomic. The rings are
mapped into Xen's private address space to write as needed and the
mappings are retained for later use.

Notifications are sent to guests via VIRQ and send_guest_global_virq is
exposed in the change to enable argo to call it. VIRQ_ARGO_MESSAGE is
claimed from the VIRQ previously reserved for this purpose (#11).

The VIRQ notification method is used rather than sending events using
evtchn functions directly because:

* no current event channel type is an exact fit for the intended
  behaviour. ECS_IPI is closest, but it disallows migration to
  other VCPUs which is not necessarily a requirement for Argo.

* at the point of argo_init, allocation of an event channel is
  complicated by none of the guest VCPUs being initialized yet
  and the event channel logic expects that a valid event channel
  has a present VCPU.

* at the point of signalling a notification, the VIRQ logic is already
  defensive: if d->vcpu[0] is NULL, the notification is just silently
  dropped, whereas the evtchn_send logic is not so defensive: vcpu[0]
  must not be NULL, otherwise a null pointer dereference occurs.

Using a VIRQ removes the need for the guest to query to determine which
event channel notifications will be delivered on. This is also likely to
simplify establishing future L0/L1 nested hypervisor argo communication.

Signed-off-by: Christopher Clark <christopher.clark6@xxxxxxxxxxxxxx>
---
v4 Jan: remove use of fixed-width types from iov_count, ringbuf_insert
v4 #07 Jan: shrink critical sections in sendv
v3 #07 Jan: header: note 32-bitness of hypercall message tuype arg
v4 : use standard data structures as per common code
v4 self: bugfix memcpy_to_guest_ring: head_len must check (offset + len)
v4 #09 Roger: drop MESSAGE from VIRQ_ARGO_MESSAGE

v3 #07 Jan: rename ring_find_info* to find_ring_info*
v3 #07 Jan: fix numeric entries in printk format strings
v3 #10 Roger: move find functions to top of file and drop prototypes
v3 #04 Jan: meld compat struct checking for hypercall args
v3 #04 Roger/Jan: make lock names clearer and assert their state
v3 #04 Jan: port -> aport with type; distinguish argo port from evtchn
v3 feedback #09 Eric: fix len & offset sanity check in memcpy_to_guest_ring
v3 feedback #04 Roger: newline fix in wildcard_pending_list_insert
v3 feedback #04 Roger: drop npages struct member, calculate from len
v3 #09 Roger: simplify EFAULT return in memcpy_to_guest_ring
v3 #09 Roger: add newline before return in get_sanitized_ring
v3 #09 Roger: replace while with for loop in iov_count
v3 #09 Roger: drop 0 in struct init in ringbuf_insert
v3 #09 Roger: comment for XEN_ARGO_MAXIOV: warn of stack overflow risk
v3 #09 Roger: simplify while loop: for instead in ringbuf_insert
v3 #09 Roger: drop out label for returns in ringbuf_insert
v3 #09 Roger: drop newline in pending_queue
v3 #09 Roger: replace second goto label with error path unlock in sendv
v3 #09 Jason: check iov_len vs MAX_ARGO_MESSAGE_SIZE in iov_count
v3 #09 Jason: check padding is zeroed in sendv op
v3 #09 Jason: memcpy_to_guest_ring: simpler code with better loop

v2 self: use ring_info backpointer in pending_ent to maintain npending
v2 feedback Jan: drop cookie, implement teardown
v2 self: pending_queue: reap stale ents when in need of space
v2 self: pending_requeue: reclaim ents for stale domains
v2.feedback Jan: only override sender domid if DOMID_ANY
v2 feedback Jan: drop message from argo_message_op
v2 self: check npending vs maximum limit
v2 self: get_sanitized_ring instead of get_rx_ptr
v2 feedback v1#13 Jan: remove double read from ringbuf insert, lower MAX_IOV
v2 self: make iov_count const
v2 self: iov_count : return EMSGSIZE for message too big
v2 self: OVERHAUL
v2 self: s/argo_pending_ent/pending_ent/g
v2 feedback v1#13 Roger: use OS-supplied roundup; drop from public header
v1,2 feedback Jan/Roger/Paul: drop errno returning guest access functions
v1 feedback Roger, Jan: drop argo prefix on static functions
v1 feedback #13 Jan: drop guest_handle_okay when using copy_from_guest
    - reorder do_argo_op logic
v2 self: add _hnd suffix to iovs variable name to indicate guest handle type
v2 self: replace use of XEN_GUEST_HANDLE_NULL with two existing macros

v1 #15 feedback, Jan: sendv op : s/ECONNREFUSED/ESRCH/
v1 #5 (#15) feedback Paul: sendv: use currd in do_argo_message_op
v1 #13 (#15) feedback Paul: sendv op: do/while reindent only
v1 #13 (#15) feedback Paul: sendv op: do/while: argo_ringbuf_insert to goto 
style
v1 #13 (#15) feedback Paul: sendv op: do/while: reindent only again
v1 #13 (#15) feedback Paul: sendv op: do/while : goto
v1 #15 feedback Paul: sendv op: make page var: unsigned
v1 #15 feedback Paul: sendv op: new local var for PAGE_SIZE - offset
v1 #8 feedback Jan: XEN_GUEST_HANDLE : C89 compliance
v1 rebase after switching register op from pfns to page descriptors
v1 self: move iov DEFINE_XEN_GUEST_HANDLE out of public header into argo.c
v1 #13 (#15) feedback Paul: fix loglevel for guest-triggered messages
v1 : add compat xlat.lst entries
v1 self: switched notification to send_guest_global_virq instead of event
v1: fix gprintk use for ARM as its defn dislikes split format strings
v1: init len variable to satisfy ARM compiler initialized checking
v1 #13 feedback Jan: rename page var
v1:#14 feedback Jan: uint8_t* -> void*
v1: #13 feedback Jan: public namespace: prefix with xen
v1: #13 feedback Jan: blank line after case op in do_argo_message_op
v1: #15 feedback Jan: add comments explaining why the writes don't overrun
v1: self: add ASSERT to support comment that overrun cannot happen
v1: self: fail on short writes where guest manipulated the iov_lens
v1: self: rename ent id to domain_id
v1: self: add moan for iov rewrite
v1. feedback #15 Jan: require the pad bits are zero
v1. feedback #15 Jan: drop NULL check in argo_signal_domain as now using VIRQ
v1. self: store domain_cookie in pending ent
v1. feedback #15 Jan: use unsigned where possible
v1. feedback Jan: use handle type for iov_base in public iov interface
v1. self: log whenever visible error occurs
v1 feedback #15, Jan: drop unnecessary mb
v1 self: only update internal tx_ptr if able to return success
         and update the visible tx_ptr
v1 self: log on failure to map ring to update visible tx_ptr
v1 feedback #15 Jan: add comment re: notification size policy
v1 self/Roger? remove errant space after sizeof
v1. feedback #15 Jan: require iov pad be zero
v1. self: rename iov_base to iov_hnd for handle in public iov interface
v1: feedback #15 Jan: handle upper-halves of hypercall args; changes some
    types in function signatures to match.
v1: self: add dprintk to sendv
v1: self: add debug output to argo_iov_count
v1. feedback #14 Jan: blank line before return in argo_iov_count
v1 feedback #15 Jan: verify src id, not override

 xen/common/argo.c          | 636 +++++++++++++++++++++++++++++++++++++++++++++
 xen/common/compat/argo.c   |  19 ++
 xen/common/event_channel.c |   2 +-
 xen/include/public/argo.h  |  60 +++++
 xen/include/public/xen.h   |   2 +-
 xen/include/xen/event.h    |   7 +
 xen/include/xlat.lst       |   2 +
 7 files changed, 726 insertions(+), 2 deletions(-)

diff --git a/xen/common/argo.c b/xen/common/argo.c
index e4cd446..518aff7 100644
--- a/xen/common/argo.c
+++ b/xen/common/argo.c
@@ -30,10 +30,15 @@
 #include <public/argo.h>
 
 #define MAX_RINGS_PER_DOMAIN            128U
+#define MAX_PENDING_PER_RING             32U
 
 /* All messages on the ring are padded to a multiple of the slot size. */
 #define ROUNDUP_MESSAGE(a) ROUNDUP((a), XEN_ARGO_MSG_SLOT_SIZE)
 
+/* The maximum size of a message that may be sent on the largest Argo ring. */
+#define MAX_ARGO_MESSAGE_SIZE ((XEN_ARGO_MAX_RING_SIZE) - \
+        (sizeof(struct xen_argo_ring_message_header)) - ROUNDUP_MESSAGE(1))
+
 /* Number of PAGEs needed to hold a ring of a given size in bytes */
 #define NPAGES_RING(ring_len) \
     (ROUNDUP((ROUNDUP_MESSAGE(ring_len) + sizeof(xen_argo_ring_t)), PAGE_SIZE) 
\
@@ -41,8 +46,10 @@
 
 DEFINE_XEN_GUEST_HANDLE(xen_argo_addr_t);
 DEFINE_XEN_GUEST_HANDLE(xen_argo_gfn_t);
+DEFINE_XEN_GUEST_HANDLE(xen_argo_iov_t);
 DEFINE_XEN_GUEST_HANDLE(xen_argo_register_ring_t);
 DEFINE_XEN_GUEST_HANDLE(xen_argo_ring_t);
+DEFINE_XEN_GUEST_HANDLE(xen_argo_send_addr_t);
 DEFINE_XEN_GUEST_HANDLE(xen_argo_unregister_ring_t);
 
 static bool __read_mostly opt_argo;
@@ -352,6 +359,28 @@ find_ring_info(const struct domain *d, const struct 
argo_ring_id *id)
     return NULL;
 }
 
+static struct argo_ring_info *
+find_ring_info_by_match(const struct domain *d, xen_argo_port_t aport,
+                        domid_t partner_id)
+{
+    struct argo_ring_id id;
+    struct argo_ring_info *ring_info;
+
+    ASSERT(LOCKING_Read_rings_L2(d));
+
+    id.aport = aport;
+    id.domain_id = d->domain_id;
+    id.partner_id = partner_id;
+
+    ring_info = find_ring_info(d, &id);
+    if ( ring_info )
+        return ring_info;
+
+    id.partner_id = XEN_ARGO_DOMID_ANY;
+
+    return find_ring_info(d, &id);
+}
+
 static struct argo_send_info *
 find_send_info(const struct domain *d, const struct argo_ring_id *id)
 {
@@ -384,6 +413,14 @@ find_send_info(const struct domain *d, const struct 
argo_ring_id *id)
 }
 
 static void
+signal_domain(struct domain *d)
+{
+    argo_dprintk("signalling domid:%u\n", d->domain_id);
+
+    send_guest_global_virq(d, VIRQ_ARGO);
+}
+
+static void
 ring_unmap(const struct domain *d, struct argo_ring_info *ring_info)
 {
     unsigned int i;
@@ -477,6 +514,390 @@ update_tx_ptr(const struct domain *d, struct 
argo_ring_info *ring_info,
     smp_wmb();
 }
 
+static int
+memcpy_to_guest_ring(const struct domain *d, struct argo_ring_info *ring_info,
+                     unsigned int offset,
+                     const void *src, XEN_GUEST_HANDLE(uint8_t) src_hnd,
+                     unsigned int len)
+{
+    unsigned int mfns_index = offset >> PAGE_SHIFT;
+    void *dst;
+    int ret;
+    unsigned int src_offset = 0;
+
+    ASSERT(LOCKING_L3(d, ring_info));
+
+    offset &= ~PAGE_MASK;
+
+    if ( len + offset > XEN_ARGO_MAX_RING_SIZE )
+        return -EFAULT;
+
+    while ( len )
+    {
+        unsigned int head_len = (offset + len) > PAGE_SIZE ? PAGE_SIZE - offset
+                                                           : len;
+
+        ret = ring_map_page(d, ring_info, mfns_index, &dst);
+        if ( ret )
+            return ret;
+
+        if ( src )
+        {
+            memcpy(dst + offset, src + src_offset, head_len);
+            src_offset += head_len;
+        }
+        else
+        {
+            if ( copy_from_guest(dst + offset, src_hnd, head_len) )
+                return -EFAULT;
+
+            guest_handle_add_offset(src_hnd, head_len);
+        }
+
+        mfns_index++;
+        len -= head_len;
+        offset = 0;
+    }
+
+    return 0;
+}
+
+/*
+ * Use this with caution: rx_ptr is under guest control and may be bogus.
+ * See get_sanitized_ring for a safer alternative.
+ */
+static int
+get_rx_ptr(const struct domain *d, struct argo_ring_info *ring_info,
+           uint32_t *rx_ptr)
+{
+    void *src;
+    xen_argo_ring_t *ringp;
+    int ret;
+
+    ASSERT(LOCKING_L3(d, ring_info));
+
+    if ( !ring_info->nmfns || ring_info->nmfns < NPAGES_RING(ring_info->len) )
+        return -EINVAL;
+
+    ret = ring_map_page(d, ring_info, 0, &src);
+    if ( ret )
+        return ret;
+
+    ringp = (xen_argo_ring_t *)src;
+
+    *rx_ptr = read_atomic(&ringp->rx_ptr);
+
+    return 0;
+}
+
+/*
+ * get_sanitized_ring creates a modified copy of the ring pointers where
+ * the rx_ptr is rounded up to ensure it is aligned, and then ring
+ * wrap is handled. Simplifies safe use of the rx_ptr for available
+ * space calculation.
+ */
+static int
+get_sanitized_ring(const struct domain *d, xen_argo_ring_t *ring,
+                   struct argo_ring_info *ring_info)
+{
+    uint32_t rx_ptr;
+    int ret;
+
+    ASSERT(LOCKING_L3(d, ring_info));
+
+    ret = get_rx_ptr(d, ring_info, &rx_ptr);
+    if ( ret )
+        return ret;
+
+    ring->tx_ptr = ring_info->tx_ptr;
+
+    rx_ptr = ROUNDUP_MESSAGE(rx_ptr);
+    if ( rx_ptr >= ring_info->len )
+        rx_ptr = 0;
+
+    ring->rx_ptr = rx_ptr;
+
+    return 0;
+}
+
+/*
+ * iov_count returns its count on success via an out variable to avoid
+ * potential for a negative return value to be used incorrectly
+ * (eg. coerced into an unsigned variable resulting in a large incorrect value)
+ */
+static int
+iov_count(const xen_argo_iov_t *piov, unsigned long niov,
+          unsigned int *count)
+{
+    unsigned int sum_iov_lens = 0;
+
+    if ( niov > XEN_ARGO_MAXIOV )
+        return -EINVAL;
+
+    for ( ; niov--; piov++ )
+    {
+        /* valid iovs must have the padding field set to zero */
+        if ( piov->pad )
+        {
+            argo_dprintk("invalid iov: padding is not zero\n");
+            return -EINVAL;
+        }
+
+        /* check each to protect sum against integer overflow */
+        if ( piov->iov_len > MAX_ARGO_MESSAGE_SIZE )
+        {
+            argo_dprintk("invalid iov_len: too big (%u)>%llu\n",
+                         piov->iov_len, MAX_ARGO_MESSAGE_SIZE);
+            return -EINVAL;
+        }
+
+        sum_iov_lens += piov->iov_len;
+
+        /*
+         * Again protect sum from integer overflow
+         * and ensure total msg size will be within bounds.
+         */
+        if ( sum_iov_lens > MAX_ARGO_MESSAGE_SIZE )
+        {
+            argo_dprintk("invalid iov series: total message too big\n");
+            return -EMSGSIZE;
+        }
+    }
+
+    *count = sum_iov_lens;
+
+    return 0;
+}
+
+static int
+ringbuf_insert(const struct domain *d, struct argo_ring_info *ring_info,
+               const struct argo_ring_id *src_id,
+               XEN_GUEST_HANDLE_PARAM(xen_argo_iov_t) iovs_hnd,
+               unsigned long niov, uint32_t message_type,
+               unsigned long *out_len)
+{
+    xen_argo_ring_t ring;
+    struct xen_argo_ring_message_header mh = { };
+    int sp, ret;
+    unsigned int len = 0;
+    xen_argo_iov_t iovs[XEN_ARGO_MAXIOV];
+    xen_argo_iov_t *piov;
+    XEN_GUEST_HANDLE(uint8_t) NULL_hnd =
+       guest_handle_from_param(guest_handle_from_ptr(NULL, uint8_t), uint8_t);
+
+    ASSERT(LOCKING_L3(d, ring_info));
+
+    ret = __copy_from_guest(iovs, iovs_hnd, niov) ? -EFAULT : 0;
+    if ( ret )
+        return ret;
+
+    /*
+     * Obtain the total size of data to transmit -- sets the 'len' variable
+     * -- and sanity check that the iovs conform to size and number limits.
+     * Enforced below: no more than 'len' bytes of guest data
+     * (plus the message header) will be sent in this operation.
+     */
+    ret = iov_count(iovs, niov, &len);
+    if ( ret )
+        return ret;
+
+    /*
+     * Size bounds check against ring size and static maximum message limit.
+     * The message must not fill the ring; there must be at least one slot
+     * remaining so we can distinguish a full ring from an empty one.
+     */
+    if ( ((ROUNDUP_MESSAGE(len) +
+            sizeof(struct xen_argo_ring_message_header)) >= ring_info->len) ||
+         (len > MAX_ARGO_MESSAGE_SIZE) )
+        return -EMSGSIZE;
+
+    ret = get_sanitized_ring(d, &ring, ring_info);
+    if ( ret )
+        return ret;
+
+    argo_dprintk("ring.tx_ptr=%u ring.rx_ptr=%u ring len=%u"
+                 " ring_info->tx_ptr=%u\n",
+                 ring.tx_ptr, ring.rx_ptr, ring_info->len, ring_info->tx_ptr);
+
+    if ( ring.rx_ptr == ring.tx_ptr )
+        sp = ring_info->len;
+    else
+    {
+        sp = ring.rx_ptr - ring.tx_ptr;
+        if ( sp < 0 )
+            sp += ring_info->len;
+    }
+
+    /*
+     * Size bounds check against currently available space in the ring.
+     * Again: the message must not fill the ring leaving no space remaining.
+     */
+    if ( (ROUNDUP_MESSAGE(len) +
+            sizeof(struct xen_argo_ring_message_header)) >= sp )
+    {
+        argo_dprintk("EAGAIN\n");
+        return -EAGAIN;
+    }
+
+    mh.len = len + sizeof(struct xen_argo_ring_message_header);
+    mh.source.aport = src_id->aport;
+    mh.source.domain_id = src_id->domain_id;
+    mh.message_type = message_type;
+
+    /*
+     * For this copy to the guest ring, tx_ptr is always 16-byte aligned
+     * and the message header is 16 bytes long.
+     */
+    BUILD_BUG_ON(
+        sizeof(struct xen_argo_ring_message_header) != ROUNDUP_MESSAGE(1));
+
+    /*
+     * First data write into the destination ring: fixed size, message header.
+     * This cannot overrun because the available free space (value in 'sp')
+     * is checked above and must be at least this size.
+     */
+    ret = memcpy_to_guest_ring(d, ring_info,
+                               ring.tx_ptr + sizeof(xen_argo_ring_t),
+                               &mh, NULL_hnd, sizeof(mh));
+    if ( ret )
+    {
+        gprintk(XENLOG_ERR,
+                "argo: failed to write message header to ring (vm%u:%x 
vm%u)\n",
+                ring_info->id.domain_id, ring_info->id.aport,
+                ring_info->id.partner_id);
+
+        return ret;
+    }
+
+    ring.tx_ptr += sizeof(mh);
+    if ( ring.tx_ptr == ring_info->len )
+        ring.tx_ptr = 0;
+
+    for ( piov = iovs; niov--; piov++ )
+    {
+        XEN_GUEST_HANDLE_64(uint8_t) buf_hnd = piov->iov_hnd;
+        unsigned int iov_len = piov->iov_len;
+
+        /* If no data is provided in this iov, moan and skip on to the next */
+        if ( !iov_len )
+        {
+            gprintk(XENLOG_ERR,
+                    "argo: no data iov_len=0 iov_hnd=%p ring (vm%u:%x vm%u)\n",
+                    buf_hnd.p, ring_info->id.domain_id, ring_info->id.aport,
+                    ring_info->id.partner_id);
+
+            continue;
+        }
+
+        if ( unlikely(!guest_handle_okay(buf_hnd, iov_len)) )
+        {
+            gprintk(XENLOG_ERR,
+                    "argo: bad iov handle [%p, %u] (vm%u:%x vm%u)\n",
+                    buf_hnd.p, iov_len,
+                    ring_info->id.domain_id, ring_info->id.aport,
+                    ring_info->id.partner_id);
+
+            return -EFAULT;
+        }
+
+        sp = ring_info->len - ring.tx_ptr;
+
+        /* Check: iov data size versus free space at the tail of the ring */
+        if ( iov_len > sp )
+        {
+            /*
+             * Second possible data write: ring-tail-wrap-write.
+             * Populate the ring tail and update the internal tx_ptr to handle
+             * wrapping at the end of ring.
+             * Size of data written here: sp
+             * which is the exact full amount of free space available at the
+             * tail of the ring, so this cannot overrun.
+             */
+            ret = memcpy_to_guest_ring(d, ring_info,
+                                       ring.tx_ptr + sizeof(xen_argo_ring_t),
+                                       NULL, buf_hnd, sp);
+            if ( ret )
+            {
+                gprintk(XENLOG_ERR,
+                        "argo: failed to copy {%p, %d} (vm%u:%x vm%u)\n",
+                        buf_hnd.p, sp,
+                        ring_info->id.domain_id, ring_info->id.aport,
+                        ring_info->id.partner_id);
+
+                return ret;
+            }
+
+            ring.tx_ptr = 0;
+            iov_len -= sp;
+            guest_handle_add_offset(buf_hnd, sp);
+
+            ASSERT(iov_len <= ring_info->len);
+        }
+
+        /*
+         * Third possible data write: all data remaining for this iov.
+         * Size of data written here: iov_len
+         *
+         * Case 1: if the ring-tail-wrap-write above was performed, then
+         *         iov_len has been decreased by 'sp' and ring.tx_ptr is zero.
+         *
+         *    We know from checking the result of iov_count:
+         *      len + sizeof(message_header) <= ring_info->len
+         *    We also know that len is the total of summing all iov_lens, so:
+         *       iov_len <= len
+         *    so by transitivity:
+         *       iov_len <= len <= (ring_info->len - sizeof(msgheader))
+         *    and therefore:
+         *       (iov_len + sizeof(msgheader) <= ring_info->len) &&
+         *       (ring.tx_ptr == 0)
+         *    so this write cannot overrun here.
+         *
+         * Case 2: ring-tail-wrap-write above was not performed
+         *    -> so iov_len is the guest-supplied value and: (iov_len <= sp)
+         *    ie. less than available space at the tail of the ring:
+         *        so this write cannot overrun.
+         */
+        ret = memcpy_to_guest_ring(d, ring_info,
+                                   ring.tx_ptr + sizeof(xen_argo_ring_t),
+                                   NULL, buf_hnd, iov_len);
+        if ( ret )
+        {
+            gprintk(XENLOG_ERR,
+                    "argo: failed to copy [%p, %u] (vm%u:%x vm%u)\n",
+                    buf_hnd.p, iov_len, ring_info->id.domain_id,
+                    ring_info->id.aport, ring_info->id.partner_id);
+
+            return ret;
+        }
+
+        ring.tx_ptr += iov_len;
+
+        if ( ring.tx_ptr == ring_info->len )
+            ring.tx_ptr = 0;
+    }
+
+    ring.tx_ptr = ROUNDUP_MESSAGE(ring.tx_ptr);
+
+    if ( ring.tx_ptr >= ring_info->len )
+        ring.tx_ptr -= ring_info->len;
+
+    update_tx_ptr(d, ring_info, ring.tx_ptr);
+
+    /*
+     * At this point (and also on an error exit paths from this function) it is
+     * possible to unmap the ring_info, ie:
+     *   ring_unmap(d, ring_info);
+     * but performance should be improved by not doing so, and retaining
+     * the mapping.
+     * An XSM policy control over level of confidentiality required
+     * versus performance cost could be added to decide that here.
+     */
+
+    *out_len = len;
+
+    return ret;
+}
+
 static void
 wildcard_pending_list_remove(domid_t domain_id, struct pending_ent *ent)
 {
@@ -497,6 +918,25 @@ wildcard_pending_list_remove(domid_t domain_id, struct 
pending_ent *ent)
 }
 
 static void
+wildcard_pending_list_insert(domid_t domain_id, struct pending_ent *ent)
+{
+    struct domain *d = get_domain_by_id(domain_id);
+
+    if ( !d )
+        return;
+
+    ASSERT(LOCKING_Read_L1);
+
+    if ( d->argo )
+    {
+        spin_lock(&d->argo->wildcard_L2_lock);
+        list_add(&ent->wildcard_node, &d->argo->wildcard_pend_list);
+        spin_unlock(&d->argo->wildcard_L2_lock);
+    }
+    put_domain(d);
+}
+
+static void
 pending_remove_all(const struct domain *d, struct argo_ring_info *ring_info)
 {
     struct list_head *ring_pending = &ring_info->pending;
@@ -518,6 +958,70 @@ pending_remove_all(const struct domain *d, struct 
argo_ring_info *ring_info)
     ring_info->npending = 0;
 }
 
+static int
+pending_queue(const struct domain *d, struct argo_ring_info *ring_info,
+              domid_t src_id, unsigned int len)
+{
+    struct pending_ent *ent;
+
+    ASSERT(LOCKING_L3(d, ring_info));
+
+    if ( ring_info->npending >= MAX_PENDING_PER_RING )
+        return -ENOSPC;
+
+    ent = xmalloc(struct pending_ent);
+    if ( !ent )
+        return -ENOMEM;
+
+    ent->len = len;
+    ent->domain_id = src_id;
+    ent->ring_info = ring_info;
+
+    if ( ring_info->id.partner_id == XEN_ARGO_DOMID_ANY )
+        wildcard_pending_list_insert(src_id, ent);
+    list_add(&ent->node, &ring_info->pending);
+    ring_info->npending++;
+
+    return 0;
+}
+
+static int
+pending_requeue(const struct domain *d, struct argo_ring_info *ring_info,
+                domid_t src_id, unsigned int len)
+{
+    struct list_head *cursor, *head;
+
+    ASSERT(LOCKING_L3(d, ring_info));
+
+    /* List structure is not modified here. Update len in a match if found. */
+    head = &ring_info->pending;
+
+    for ( cursor = head->next; cursor != head; cursor = cursor->next )
+    {
+        struct pending_ent *ent = list_entry(cursor, struct pending_ent, node);
+
+        if ( ent->domain_id == src_id )
+        {
+            /*
+             * Reuse an existing queue entry for a notification rather than add
+             * another. If the existing entry is waiting for a smaller size 
than
+             * the current message then adjust the record to wait for the
+             * current (larger) size to be available before triggering a
+             * notification.
+             * This assists the waiting sender by ensuring that whenever a
+             * notification is triggered, there is sufficient space available
+             * for (at least) any one of the messages awaiting transmission.
+             */
+            if ( ent->len < len )
+                ent->len = len;
+
+            return 0;
+        }
+    }
+
+    return pending_queue(d, ring_info, src_id, len);
+}
+
 static void
 wildcard_rings_pending_remove(struct domain *d)
 {
@@ -1082,6 +1586,91 @@ register_ring(struct domain *currd,
     return ret;
 }
 
+static long
+sendv(struct domain *src_d, const xen_argo_addr_t *src_addr,
+      const xen_argo_addr_t *dst_addr,
+      XEN_GUEST_HANDLE_PARAM(xen_argo_iov_t) iovs_hnd, unsigned long niov,
+      uint32_t message_type)
+{
+    struct domain *dst_d = NULL;
+    struct argo_ring_id src_id;
+    struct argo_ring_info *ring_info;
+    int ret = 0;
+    unsigned long len = 0;
+
+    ASSERT(src_d->domain_id == src_addr->domain_id);
+
+    argo_dprintk("sendv: (%u:%x)->(%u:%x) niov:%lu iov:%p type:%u\n",
+                 src_addr->domain_id, src_addr->aport,
+                 dst_addr->domain_id, dst_addr->aport,
+                 niov, iovs_hnd.p, message_type);
+
+    src_id.aport = src_addr->aport;
+    src_id.domain_id = src_d->domain_id;
+    src_id.partner_id = dst_addr->domain_id;
+
+    dst_d = get_domain_by_id(dst_addr->domain_id);
+    if ( !dst_d )
+        return -ESRCH;
+
+    read_lock(&L1_global_argo_rwlock);
+
+    if ( !src_d->argo )
+    {
+        ret = -ENODEV;
+        goto out_unlock;
+    }
+
+    if ( !dst_d->argo )
+    {
+        argo_dprintk("!dst_d->argo, ECONNREFUSED\n");
+        ret = -ECONNREFUSED;
+        goto out_unlock;
+    }
+
+    read_lock(&dst_d->argo->rings_L2_rwlock);
+
+    ring_info = find_ring_info_by_match(dst_d, dst_addr->aport,
+                                        src_id.domain_id);
+    if ( !ring_info )
+    {
+        gprintk(XENLOG_ERR,
+                "argo: vm%u connection refused, src (vm%u:%x) dst (vm%u:%x)\n",
+                current->domain->domain_id, src_id.domain_id, src_id.aport,
+                dst_addr->domain_id, dst_addr->aport);
+
+        ret = -ECONNREFUSED;
+    }
+    else
+    {
+        spin_lock(&ring_info->L3_lock);
+
+        ret = ringbuf_insert(dst_d, ring_info, &src_id, iovs_hnd, niov,
+                             message_type, &len);
+        if ( ret == -EAGAIN )
+        {
+            argo_dprintk("argo_ringbuf_sendv failed, EAGAIN\n");
+            /* requeue to issue a notification when space is there */
+            ret = pending_requeue(dst_d, ring_info, src_id.domain_id, len);
+        }
+
+        spin_unlock(&ring_info->L3_lock);
+    }
+
+    read_unlock(&dst_d->argo->rings_L2_rwlock);
+
+ out_unlock:
+    read_unlock(&L1_global_argo_rwlock);
+
+    if ( ret >= 0 )
+        signal_domain(dst_d);
+
+    if ( dst_d )
+        put_domain(dst_d);
+
+    return ( ret < 0 ) ? ret : len;
+}
+
 long
 do_argo_op(unsigned int cmd, XEN_GUEST_HANDLE_PARAM(void) arg1,
            XEN_GUEST_HANDLE_PARAM(void) arg2, unsigned long arg3,
@@ -1145,6 +1734,53 @@ do_argo_op(unsigned int cmd, 
XEN_GUEST_HANDLE_PARAM(void) arg1,
         break;
     }
 
+    case XEN_ARGO_OP_sendv:
+    {
+        xen_argo_send_addr_t send_addr;
+
+        XEN_GUEST_HANDLE_PARAM(xen_argo_send_addr_t) send_addr_hnd =
+            guest_handle_cast(arg1, xen_argo_send_addr_t);
+        XEN_GUEST_HANDLE_PARAM(xen_argo_iov_t) iovs_hnd =
+            guest_handle_cast(arg2, xen_argo_iov_t);
+        /* arg3 is niov */
+        /* arg4 is message_type. Must be a 32-bit value. */
+
+        rc = copy_from_guest(&send_addr, send_addr_hnd, 1) ? -EFAULT : 0;
+        if ( rc )
+            break;
+
+        /*
+         * Check padding is zeroed. Reject niov above limit or message_types
+         * that are outside 32 bit range.
+         */
+        if ( unlikely(send_addr.src.pad || send_addr.dst.pad ||
+                      (arg3 > XEN_ARGO_MAXIOV) || (arg4 & ~0xffffffffUL)) )
+        {
+            rc = -EINVAL;
+            break;
+        }
+
+        if ( send_addr.src.domain_id == XEN_ARGO_DOMID_ANY )
+            send_addr.src.domain_id = currd->domain_id;
+
+        /* No domain is currently authorized to send on behalf of another */
+        if ( unlikely(send_addr.src.domain_id != currd->domain_id) )
+        {
+            rc = -EPERM;
+            break;
+        }
+
+        /*
+         * Check access to the whole array here so we can use the faster __copy
+         * operations to read each element later.
+         */
+        if ( unlikely(!guest_handle_okay(iovs_hnd, arg3)) )
+            break;
+
+        rc = sendv(currd, &send_addr.src, &send_addr.dst, iovs_hnd, arg3, 
arg4);
+        break;
+    }
+
     default:
         rc = -EOPNOTSUPP;
         break;
diff --git a/xen/common/compat/argo.c b/xen/common/compat/argo.c
index 6a1671c..6290ed6 100644
--- a/xen/common/compat/argo.c
+++ b/xen/common/compat/argo.c
@@ -23,3 +23,22 @@ CHECK_argo_addr;
 CHECK_argo_register_ring;
 CHECK_argo_ring;
 CHECK_argo_unregister_ring;
+
+/*
+ * Disable strict type checking in this compat validation macro for the
+ * following struct checks because it cannot handle fields within structs that
+ * have types that differ in the compat versus non-compat structs.
+ * Replace it with a field size check which is sufficient here.
+ */
+
+#undef CHECK_FIELD_COMMON_
+#define CHECK_FIELD_COMMON_(k, name, n, f) \
+static inline int __maybe_unused name(k xen_ ## n *x, k compat_ ## n *c) \
+{ \
+    BUILD_BUG_ON(offsetof(k xen_ ## n, f) != \
+                 offsetof(k compat_ ## n, f)); \
+    return sizeof(x->f) == sizeof(c->f); \
+}
+
+CHECK_argo_send_addr;
+CHECK_argo_iov;
diff --git a/xen/common/event_channel.c b/xen/common/event_channel.c
index f34d4f0..6fbe346 100644
--- a/xen/common/event_channel.c
+++ b/xen/common/event_channel.c
@@ -746,7 +746,7 @@ void send_guest_vcpu_virq(struct vcpu *v, uint32_t virq)
     spin_unlock_irqrestore(&v->virq_lock, flags);
 }
 
-static void send_guest_global_virq(struct domain *d, uint32_t virq)
+void send_guest_global_virq(struct domain *d, uint32_t virq)
 {
     unsigned long flags;
     int port;
diff --git a/xen/include/public/argo.h b/xen/include/public/argo.h
index 2371510..a28454a 100644
--- a/xen/include/public/argo.h
+++ b/xen/include/public/argo.h
@@ -46,6 +46,34 @@ typedef uint32_t xen_argo_port_t;
 /* gfn type: 64-bit on all architectures to aid avoiding a compat ABI */
 typedef uint64_t xen_argo_gfn_t;
 
+/*
+ * XEN_ARGO_MAXIOV : maximum number of iovs accepted in a single sendv.
+ * Caution is required if this value is increased: this determines the size of
+ * an array of xen_argo_iov_t structs on the hypervisor stack, so could cause
+ * stack overflow if the value is too large.
+ * The Linux Argo driver never passes more than two iovs.
+ *
+ * This value should also not exceed 128 to ensure that the total amount of 
data
+ * posted in a single Argo sendv operation cannot exceed 2^31 bytes, to reduce
+ * risk of integer overflow defects:
+ * Each argo iov can hold ~ 2^24 bytes, so XEN_ARGO_MAXIOV <= 2^(31-24),
+ * ie. keep XEN_ARGO_MAXIOV <= 128.
+*/
+#define XEN_ARGO_MAXIOV          8U
+
+DEFINE_XEN_GUEST_HANDLE(uint8_t);
+
+typedef struct xen_argo_iov
+{
+#ifdef XEN_GUEST_HANDLE_64
+    XEN_GUEST_HANDLE_64(uint8_t) iov_hnd;
+#else
+    uint64_t iov_hnd;
+#endif
+    uint32_t iov_len;
+    uint32_t pad;
+} xen_argo_iov_t;
+
 typedef struct xen_argo_addr
 {
     xen_argo_port_t aport;
@@ -53,6 +81,12 @@ typedef struct xen_argo_addr
     uint16_t pad;
 } xen_argo_addr_t;
 
+typedef struct xen_argo_send_addr
+{
+    xen_argo_addr_t src;
+    xen_argo_addr_t dst;
+} xen_argo_send_addr_t;
+
 typedef struct xen_argo_ring
 {
     /* Guests should use atomic operations to access rx_ptr */
@@ -153,4 +187,30 @@ struct xen_argo_ring_message_header
  */
 #define XEN_ARGO_OP_unregister_ring     2
 
+/*
+ * XEN_ARGO_OP_sendv
+ *
+ * Send a list of buffers contained in iovs.
+ *
+ * The send address struct specifies the source and destination addresses
+ * for the message being sent, which are used to find the destination ring:
+ * Xen first looks for a most-specific match with a registered ring with
+ *  (id.addr == dst) and (id.partner == sending_domain) ;
+ * if that fails, it then looks for a wildcard match (aka multicast receiver)
+ * where (id.addr == dst) and (id.partner == DOMID_ANY).
+ *
+ * For each iov entry, send iov_len bytes from iov_base to the destination 
ring.
+ * If insufficient space exists in the destination ring, it will return -EAGAIN
+ * and Xen will notify the caller when sufficient space becomes available.
+ *
+ * The message type is a 32-bit data field available to communicate message
+ * context data (eg. kernel-to-kernel, rather than application layer).
+ *
+ * arg1: XEN_GUEST_HANDLE(xen_argo_send_addr_t) source and dest addresses
+ * arg2: XEN_GUEST_HANDLE(xen_argo_iov_t) iovs
+ * arg3: unsigned long niov
+ * arg4: unsigned long message type (32-bit value)
+ */
+#define XEN_ARGO_OP_sendv               3
+
 #endif
diff --git a/xen/include/public/xen.h b/xen/include/public/xen.h
index b3f6491..ccdffc0 100644
--- a/xen/include/public/xen.h
+++ b/xen/include/public/xen.h
@@ -178,7 +178,7 @@ DEFINE_XEN_GUEST_HANDLE(xen_ulong_t);
 #define VIRQ_CON_RING   8  /* G. (DOM0) Bytes received on console            */
 #define VIRQ_PCPU_STATE 9  /* G. (DOM0) PCPU state changed                   */
 #define VIRQ_MEM_EVENT  10 /* G. (DOM0) A memory event has occurred          */
-#define VIRQ_XC_RESERVED 11 /* G. Reserved for XenClient                     */
+#define VIRQ_ARGO       11 /* G. Argo interdomain message notification       */
 #define VIRQ_ENOMEM     12 /* G. (DOM0) Low on heap memory       */
 #define VIRQ_XENPMU     13 /* V.  PMC interrupt                              */
 
diff --git a/xen/include/xen/event.h b/xen/include/xen/event.h
index ebb879e..4650887 100644
--- a/xen/include/xen/event.h
+++ b/xen/include/xen/event.h
@@ -29,6 +29,13 @@ void send_guest_vcpu_virq(struct vcpu *v, uint32_t virq);
 void send_global_virq(uint32_t virq);
 
 /*
+ * send_guest_global_virq:
+ *  @d:        Domain to which VIRQ should be sent
+ *  @virq:     Virtual IRQ number (VIRQ_*), must be global
+ */
+void send_guest_global_virq(struct domain *d, uint32_t virq);
+
+/*
  * sent_global_virq_handler: Set a global VIRQ handler.
  *  @d:        New target domain for this VIRQ
  *  @virq:     Virtual IRQ number (VIRQ_*), must be global
diff --git a/xen/include/xlat.lst b/xen/include/xlat.lst
index 411c661..3723980 100644
--- a/xen/include/xlat.lst
+++ b/xen/include/xlat.lst
@@ -152,3 +152,5 @@
 ?      argo_ring                       argo.h
 ?      argo_register_ring              argo.h
 ?      argo_unregister_ring            argo.h
+?      argo_iov                        argo.h
+?      argo_send_addr                  argo.h
-- 
2.7.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.