VMCIHashTable *
VMCIHashTable_Create(int size)
{
VMCIHashTable *table = VMCI_AllocKernelMem(sizeof *table,
VMCI_MEMORY_NONPAGED);
if (table == NULL) {
return NULL;
}
table->entries = VMCI_AllocKernelMem(sizeof *table->entries * size,
VMCI_MEMORY_NONPAGED);
if (table->entries == NULL) {
VMCI_FreeKernelMem(table, sizeof *table);
return NULL;
}
memset(table->entries, 0, sizeof *table->entries * size);
table->size = size;
if (VMCIHashTableInitLock(&table->lock, "VMCIHashTableLock") < VMCI_SUCCESS) {
VMCI_FreeKernelMem(table->entries, sizeof *table->entries * size);
VMCI_FreeKernelMem(table, sizeof *table);
return NULL;
}
return table;
}
void
DsListDestroy(DsList *list) // IN:
{
if (list == NULL) {
return;
}
if (list->elements != NULL) {
VMCI_FreeKernelMem(list->elements, list->capacity * sizeof *list->elements);
list->elements = NULL;
}
VMCI_FreeKernelMem(list, sizeof *list);
}
void
VMCI_FreePPNSet(PPNSet *ppnSet) // IN:
{
ASSERT(ppnSet);
if (ppnSet->initialized) {
/* Do not call these functions on NULL inputs. */
ASSERT(ppnSet->producePPNs && ppnSet->consumePPNs);
VMCI_FreeKernelMem(ppnSet->producePPNs,
ppnSet->numProducePages * sizeof *ppnSet->producePPNs);
VMCI_FreeKernelMem(ppnSet->consumePPNs,
ppnSet->numConsumePages * sizeof *ppnSet->consumePPNs);
}
memset(ppnSet, 0, sizeof *ppnSet);
}
static Bool
VMCIUtilCheckHostCapabilities(void)
{
int result;
VMCIResourcesQueryMsg *msg;
uint32 msgSize = sizeof(VMCIResourcesQueryHdr) +
VMCI_UTIL_NUM_RESOURCES * sizeof(VMCI_Resource);
VMCIDatagram *checkMsg = VMCI_AllocKernelMem(msgSize, VMCI_MEMORY_NONPAGED);
if (checkMsg == NULL) {
VMCI_WARNING((LGPFX"Check host: Insufficient memory.\n"));
return FALSE;
}
checkMsg->dst = VMCI_MAKE_HANDLE(VMCI_HYPERVISOR_CONTEXT_ID,
VMCI_RESOURCES_QUERY);
checkMsg->src = VMCI_ANON_SRC_HANDLE;
checkMsg->payloadSize = msgSize - VMCI_DG_HEADERSIZE;
msg = (VMCIResourcesQueryMsg *)VMCI_DG_PAYLOAD(checkMsg);
msg->numResources = VMCI_UTIL_NUM_RESOURCES;
msg->resources[0] = VMCI_GET_CONTEXT_ID;
result = VMCI_SendDatagram(checkMsg);
VMCI_FreeKernelMem(checkMsg, msgSize);
/* We need the vector. There are no fallbacks. */
return (result == 0x1);
}
int
vmci_datagram_destroy_handle(VMCIHandle handle) // IN
{
DatagramEntry *entry;
VMCIResource *resource = VMCIResource_Get(handle,
VMCI_RESOURCE_TYPE_DATAGRAM);
if (resource == NULL) {
VMCI_DEBUG_LOG(4, (LGPFX"Failed to destroy datagram (handle=0x%x:0x%x).\n",
handle.context, handle.resource));
return VMCI_ERROR_NOT_FOUND;
}
entry = RESOURCE_CONTAINER(resource, DatagramEntry, resource);
VMCIResource_Remove(handle, VMCI_RESOURCE_TYPE_DATAGRAM);
/*
* We now wait on the destroyEvent and release the reference we got
* above.
*/
VMCI_WaitOnEvent(&entry->destroyEvent, DatagramReleaseCB, entry);
/*
* We know that we are now the only reference to the above entry so
* can safely free it.
*/
VMCI_DestroyEvent(&entry->destroyEvent);
VMCI_FreeKernelMem(entry, sizeof *entry);
return VMCI_SUCCESS;
}
static Bool
DsListInit(DsList **list, // OUT:
int capacity) // IN:
{
DsList *l = VMCI_AllocKernelMem(sizeof(DsList),
VMCI_MEMORY_NONPAGED | VMCI_MEMORY_ATOMIC);
ASSERT(list);
ASSERT(capacity >= 1);
if (l == NULL) {
return FALSE;
}
l->size = 0;
l->capacity = capacity;
l->elements = VMCI_AllocKernelMem(sizeof(DsListElement) * capacity,
VMCI_MEMORY_NONPAGED | VMCI_MEMORY_ATOMIC);
if (l->elements == NULL) {
VMCI_FreeKernelMem(l, sizeof *l);
return FALSE;
}
*list = l;
return TRUE;
}
int
VMCIDatagram_InvokeGuestHandler(VMCIDatagram *dg) // IN
{
#if defined(VMKERNEL)
VMCI_WARNING((LGPFX"Cannot dispatch within guest in VMKERNEL.\n"));
return VMCI_ERROR_DST_UNREACHABLE;
#else // VMKERNEL
int retval;
VMCIResource *resource;
DatagramEntry *dstEntry;
ASSERT(dg);
resource = VMCIResource_Get(dg->dst, VMCI_RESOURCE_TYPE_DATAGRAM);
if (NULL == resource) {
VMCI_DEBUG_LOG(4, (LGPFX"destination (handle=0x%x:0x%x) doesn't exist.\n",
dg->dst.context, dg->dst.resource));
return VMCI_ERROR_NO_HANDLE;
}
dstEntry = RESOURCE_CONTAINER(resource, DatagramEntry, resource);
if (dstEntry->runDelayed) {
VMCIDelayedDatagramInfo *dgInfo;
dgInfo = VMCI_AllocKernelMem(sizeof *dgInfo + (size_t)dg->payloadSize,
(VMCI_MEMORY_ATOMIC | VMCI_MEMORY_NONPAGED));
if (NULL == dgInfo) {
VMCIResource_Release(resource);
retval = VMCI_ERROR_NO_MEM;
goto exit;
}
dgInfo->inDGHostQueue = FALSE;
dgInfo->entry = dstEntry;
memcpy(&dgInfo->msg, dg, VMCI_DG_SIZE(dg));
retval = VMCI_ScheduleDelayedWork(VMCIDatagramDelayedDispatchCB, dgInfo);
if (retval < VMCI_SUCCESS) {
VMCI_WARNING((LGPFX"Failed to schedule delayed work for datagram "
"(result=%d).\n", retval));
VMCI_FreeKernelMem(dgInfo, sizeof *dgInfo + (size_t)dg->payloadSize);
VMCIResource_Release(resource);
dgInfo = NULL;
goto exit;
}
} else {
dstEntry->recvCB(dstEntry->clientData, dg);
VMCIResource_Release(resource);
retval = VMCI_SUCCESS;
}
exit:
return retval;
#endif // VMKERNEL
}
static int
DsListInsert(DsList *list, // IN:
const char *name, // IN:
VMCIHandle handle, // IN:
VMCIId contextID) // IN:
{
int nameLen;
char *nameMem;
if (!list || !name || VMCI_HANDLE_EQUAL(handle, VMCI_INVALID_HANDLE) ||
contextID == VMCI_INVALID_ID) {
return VMCI_ERROR_INVALID_ARGS;
}
/* Check for duplicates */
if (DsListLookupIndex(list, name) >= 0) {
return VMCI_ERROR_ALREADY_EXISTS;
}
if (list->capacity == list->size) {
/* We need to expand the list */
int newCapacity = list->capacity * 2;
DsListElement *elms = VMCI_AllocKernelMem(sizeof(DsListElement) *
newCapacity,
VMCI_MEMORY_NONPAGED |
VMCI_MEMORY_ATOMIC);
if (elms == NULL) {
return VMCI_ERROR_NO_MEM;
}
memcpy(elms, list->elements, sizeof(DsListElement) * list->capacity);
VMCI_FreeKernelMem(list->elements,
sizeof *list->elements * list->capacity);
list->elements = elms;
list->capacity = newCapacity;
}
ASSERT(list->capacity > list->size);
nameLen = strlen(name) + 1;
nameMem = VMCI_AllocKernelMem(nameLen,
VMCI_MEMORY_NONPAGED | VMCI_MEMORY_ATOMIC);
if (nameMem == NULL) {
return VMCI_ERROR_NO_MEM;
}
memcpy(nameMem, name, nameLen);
list->elements[list->size].name = nameMem;
list->elements[list->size].handle = handle;
list->elements[list->size].contextID = contextID;
list->size = list->size + 1;
return VMCI_SUCCESS;
}
static void
VMCIDatagramDelayedDispatchCB(void *data) // IN
{
Bool inDGHostQueue;
VMCIDelayedDatagramInfo *dgInfo = (VMCIDelayedDatagramInfo *)data;
ASSERT(data);
dgInfo->entry->recvCB(dgInfo->entry->clientData, &dgInfo->msg);
VMCIResource_Release(&dgInfo->entry->resource);
inDGHostQueue = dgInfo->inDGHostQueue;
VMCI_FreeKernelMem(dgInfo, sizeof *dgInfo + (size_t)dgInfo->msg.payloadSize);
if (inDGHostQueue) {
Atomic_Dec(&delayedDGHostQueueSize);
}
}
void
VPageChannel_Destroy(VPageChannel *channel) // IN/OUT
{
ASSERT(channel);
VPageChannelDestroyQueuePair(channel);
if (!VMCI_HANDLE_INVALID(channel->doorbellHandle)) {
vmci_doorbell_destroy(channel->doorbellHandle);
}
if (!VMCI_HANDLE_INVALID(channel->dgHandle)) {
vmci_datagram_destroy_handle(channel->dgHandle);
}
channel->state = VPCState_Free;
VMCI_FreeKernelMem(channel, sizeof *channel);
VMCI_DEBUG_LOG(10,
(LGPFX"Destroyed (channel=%p).\n",
channel));
}
static int
DsListRemoveElement(DsList *list, // IN:
int index) // IN: index of the element to remove
{
if (!list || index < 0) {
return VMCI_ERROR_INVALID_ARGS;
}
if (index > list->size - 1) {
return VMCI_ERROR_NOT_FOUND;
}
/* Free name. */
VMCI_FreeKernelMem(list->elements[index].name,
strlen(list->elements[index].name) + 1);
/* Move elements one spot up. */
memmove(&list->elements[index],
&list->elements[index + 1],
(list->size - index - 1) * sizeof list->elements[index]);
list->size--;
/* Zero out the last element. */
memset(&list->elements[list->size], 0, sizeof list->elements[list->size]);
return VMCI_SUCCESS;
}
static void
VPageChannelDoDoorbellCallback(VPageChannel *channel) // IN/OUT
{
Bool inUse;
unsigned long flags;
VPageChannelPacket packetHeader;
ASSERT(channel);
if (VPCState_Connected != channel->state) {
VMCI_WARNING((LGPFX"Not connected (channel=%p).\n",
channel));
return;
}
VPageChannelAcquireRecvLock(channel, &flags);
inUse = channel->inPoll;
channel->inPoll = TRUE;
VPageChannelReleaseRecvLock(channel, flags);
if (inUse) {
return;
}
retry:
while (vmci_qpair_consume_buf_ready(channel->qpair) >= sizeof packetHeader) {
ssize_t retSize, totalSize;
VPageChannelPacket *packet;
retSize = vmci_qpair_peek(channel->qpair, &packetHeader,
sizeof packetHeader,
/* XXX, UTIL_VMKERNEL_BUFFER for VMKernel. */
0);
if (retSize < sizeof packetHeader) {
/*
* XXX, deal with partial read.
*/
VMCI_WARNING((LGPFX"Failed to peek (channel=%p) "
"(required=%"FMTSZ"d) (err=%"FMTSZ"d).\n",
channel,
sizeof packetHeader,
retSize));
break;
}
totalSize = sizeof packetHeader + packetHeader.msgLen +
packetHeader.numElems * sizeof(VPageChannelElem);
retSize = vmci_qpair_consume_buf_ready(channel->qpair);
if (retSize < totalSize) {
/*
* XXX, deal with partial read.
*/
VMCI_WARNING((LGPFX"Received partial packet (channel=%p) "
"(type=%d) (len=%d) (num elems=%d) (avail=%"FMTSZ"d) "
"(requested=%"FMTSZ"d).\n",
channel,
packetHeader.type,
packetHeader.msgLen,
packetHeader.numElems,
retSize,
totalSize));
break;
}
packet = (VPageChannelPacket *)
VMCI_AllocKernelMem(totalSize, VMCI_MEMORY_ATOMIC);
if (!packet) {
VMCI_WARNING((LGPFX"Failed to allocate packet (channel=%p) "
"(size=%"FMTSZ"d).\n",
channel,
totalSize));
break;
}
retSize = vmci_qpair_dequeue(channel->qpair, packet,
totalSize,
/* XXX, UTIL_VMKERNEL_BUFFER for VMKernel. */
0);
if (retSize < totalSize) {
/*
* XXX, deal with partial read.
*/
VMCI_WARNING((LGPFX"Failed to dequeue (channel=%p) "
"(required=%"FMTSZ"d) (err=%"FMTSZ"d).\n",
channel,
totalSize,
retSize));
VMCI_FreeKernelMem(packet, totalSize);
break;
}
VPageChannelRecvPacket(channel, packet);
VMCI_FreeKernelMem(packet, totalSize);
}
VPageChannelAcquireRecvLock(channel, &flags);
/*
* The doorbell may have been notified between when we we finished reading
* data and when we grabbed the lock. If that happens, then there may be
* data, but we bailed out of that second notification because inPoll was
* already set. So that we don't miss anything, do a final check here under
* the lock for any data that might have arrived.
*/
if (vmci_qpair_consume_buf_ready(channel->qpair) >= sizeof packetHeader) {
VPageChannelReleaseRecvLock(channel, flags);
goto retry;
}
channel->inPoll = FALSE;
VPageChannelReleaseRecvLock(channel, flags);
}
static int
VMCIDatagramDispatchAsHost(VMCIId contextID, // IN:
VMCIDatagram *dg) // IN:
{
int retval;
size_t dgSize;
VMCIPrivilegeFlags srcPrivFlags;
ASSERT(dg);
ASSERT(VMCI_HostPersonalityActive());
dgSize = VMCI_DG_SIZE(dg);
if (contextID == VMCI_HOST_CONTEXT_ID &&
dg->dst.context == VMCI_HYPERVISOR_CONTEXT_ID) {
VMCI_DEBUG_LOG(4, (LGPFX"Host cannot talk to hypervisor\n"));
return VMCI_ERROR_DST_UNREACHABLE;
}
ASSERT(dg->dst.context != VMCI_HYPERVISOR_CONTEXT_ID);
/* Chatty. */
// VMCI_DEBUG_LOG(10, (LGPFX"Sending from (handle=0x%x:0x%x) to "
// "(handle=0x%x:0x%x) (size=%u bytes).\n",
// dg->src.context, dg->src.resource,
// dg->dst.context, dg->dst.resource, (uint32)dgSize));
/*
* Check that source handle matches sending context.
*/
if (dg->src.context != contextID) {
VMCI_DEBUG_LOG(4, (LGPFX"Sender context (ID=0x%x) is not owner of src "
"datagram entry (handle=0x%x:0x%x).\n",
contextID, dg->src.context, dg->src.resource));
return VMCI_ERROR_NO_ACCESS;
}
/*
* Get hold of privileges of sending endpoint.
*/
retval = VMCIDatagramGetPrivFlagsInt(contextID, dg->src, &srcPrivFlags);
if (retval != VMCI_SUCCESS) {
VMCI_WARNING((LGPFX"Couldn't get privileges (handle=0x%x:0x%x).\n",
dg->src.context, dg->src.resource));
return retval;
}
/* Determine if we should route to host or guest destination. */
if (dg->dst.context == VMCI_HOST_CONTEXT_ID) {
/* Route to host datagram entry. */
DatagramEntry *dstEntry;
VMCIResource *resource;
if (dg->src.context == VMCI_HYPERVISOR_CONTEXT_ID &&
dg->dst.resource == VMCI_EVENT_HANDLER) {
return VMCIEvent_Dispatch(dg);
}
resource = VMCIResource_Get(dg->dst, VMCI_RESOURCE_TYPE_DATAGRAM);
if (resource == NULL) {
VMCI_DEBUG_LOG(4, (LGPFX"Sending to invalid destination "
"(handle=0x%x:0x%x).\n",
dg->dst.context, dg->dst.resource));
return VMCI_ERROR_INVALID_RESOURCE;
}
dstEntry = RESOURCE_CONTAINER(resource, DatagramEntry, resource);
if (VMCIDenyInteraction(srcPrivFlags, dstEntry->privFlags)) {
VMCIResource_Release(resource);
return VMCI_ERROR_NO_ACCESS;
}
ASSERT(dstEntry->recvCB);
/*
* If a VMCI datagram destined for the host is also sent by the
* host, we always run it delayed. This ensures that no locks
* are held when the datagram callback runs.
*/
if (dstEntry->runDelayed ||
(dg->src.context == VMCI_HOST_CONTEXT_ID &&
VMCI_CanScheduleDelayedWork())) {
VMCIDelayedDatagramInfo *dgInfo;
if (Atomic_FetchAndAdd(&delayedDGHostQueueSize, 1) ==
VMCI_MAX_DELAYED_DG_HOST_QUEUE_SIZE) {
Atomic_Dec(&delayedDGHostQueueSize);
VMCIResource_Release(resource);
return VMCI_ERROR_NO_MEM;
}
dgInfo = VMCI_AllocKernelMem(sizeof *dgInfo + (size_t)dg->payloadSize,
(VMCI_MEMORY_ATOMIC |
VMCI_MEMORY_NONPAGED));
if (NULL == dgInfo) {
Atomic_Dec(&delayedDGHostQueueSize);
VMCIResource_Release(resource);
return VMCI_ERROR_NO_MEM;
}
dgInfo->inDGHostQueue = TRUE;
dgInfo->entry = dstEntry;
memcpy(&dgInfo->msg, dg, dgSize);
retval = VMCI_ScheduleDelayedWork(VMCIDatagramDelayedDispatchCB, dgInfo);
if (retval < VMCI_SUCCESS) {
VMCI_WARNING((LGPFX"Failed to schedule delayed work for datagram "
"(result=%d).\n", retval));
VMCI_FreeKernelMem(dgInfo, sizeof *dgInfo + (size_t)dg->payloadSize);
VMCIResource_Release(resource);
Atomic_Dec(&delayedDGHostQueueSize);
return retval;
}
} else {
retval = dstEntry->recvCB(dstEntry->clientData, dg);
VMCIResource_Release(resource);
if (retval < VMCI_SUCCESS) {
return retval;
}
}
} else {
/*
* Route to destination VM context.
*/
VMCIDatagram *newDG;
if (contextID != dg->dst.context) {
if (VMCIDenyInteraction(srcPrivFlags,
vmci_context_get_priv_flags(dg->dst.context))) {
VMCI_DEBUG_LOG(4, (LGPFX"Interaction denied (%X/%X - %X/%X)\n",
contextID, srcPrivFlags,
dg->dst.context,
vmci_context_get_priv_flags(dg->dst.context)));
return VMCI_ERROR_NO_ACCESS;
} else if (VMCI_CONTEXT_IS_VM(contextID)) {
/*
* If the sending context is a VM, it cannot reach another VM.
*/
if (!vmkernel) {
VMCI_DEBUG_LOG(4, (LGPFX"Datagram communication between VMs not "
"supported (src=0x%x, dst=0x%x).\n",
contextID, dg->dst.context));
return VMCI_ERROR_DST_UNREACHABLE;
}
}
}
/* We make a copy to enqueue. */
newDG = VMCI_AllocKernelMem(dgSize, VMCI_MEMORY_NORMAL);
if (newDG == NULL) {
VMCI_DEBUG_LOG(4, (LGPFX"No memory for datagram\n"));
return VMCI_ERROR_NO_MEM;
}
memcpy(newDG, dg, dgSize);
retval = VMCIContext_EnqueueDatagram(dg->dst.context, newDG);
if (retval < VMCI_SUCCESS) {
VMCI_FreeKernelMem(newDG, dgSize);
VMCI_DEBUG_LOG(4, (LGPFX"Enqueue failed\n"));
return retval;
}
}
/* The datagram is freed when the context reads it. */
/* Chatty. */
// VMCI_DEBUG_LOG(10, (LGPFX"Sent datagram (size=%u bytes).\n",
// (uint32)dgSize));
/*
* We currently truncate the size to signed 32 bits. This doesn't
* matter for this handler as it only support 4Kb messages.
*/
return (int)dgSize;
}
void
VMCIHost_ReleaseUserMemory(PageStoreAttachInfo *attach, // IN/OUT
VMCIQueue *produceQ, // OUT
VMCIQueue *consumeQ) // OUT
{
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
int i;
ASSERT(attach->producePages);
ASSERT(attach->consumePages);
kunmap(attach->producePages[0]);
kunmap(attach->consumePages[0]);
for (i = 0; i < attach->numProducePages; i++) {
ASSERT(attach->producePages[i]);
set_page_dirty(attach->producePages[i]);
page_cache_release(attach->producePages[i]);
}
for (i = 0; i < attach->numConsumePages; i++) {
ASSERT(attach->consumePages[i]);
set_page_dirty(attach->consumePages[i]);
page_cache_release(attach->consumePages[i]);
}
VMCI_FreeKernelMem(attach->producePages,
attach->numProducePages *
sizeof attach->producePages[0]);
VMCI_FreeKernelMem(attach->consumePages,
attach->numConsumePages *
sizeof attach->consumePages[0]);
#else
/*
* Host queue pair support for earlier kernels temporarily
* disabled. See bug 365496.
*/
ASSERT_NOT_IMPLEMENTED(FALSE);
#if 0
kunmap(attach->produceIoBuf->maplist[0]);
kunmap(attach->consumeIoBuf->maplist[0]);
mark_dirty_kiobuf(attach->produceIoBuf,
attach->numProducePages * PAGE_SIZE);
unmap_kiobuf(attach->produceIoBuf);
mark_dirty_kiobuf(attach->consumeIoBuf,
attach->numConsumePages * PAGE_SIZE);
unmap_kiobuf(attach->consumeIoBuf);
VMCI_FreeKernelMem(attach->produceIoBuf,
sizeof *attach->produceIoBuf);
VMCI_FreeKernelMem(attach->consumeIoBuf,
sizeof *attach->consumeIoBuf);
#endif
#endif
}
static int
DatagramCreateHnd(VMCIId resourceID, // IN:
uint32 flags, // IN:
VMCIPrivilegeFlags privFlags, // IN:
VMCIDatagramRecvCB recvCB, // IN:
void *clientData, // IN:
VMCIHandle *outHandle) // OUT:
{
int result;
VMCIId contextID;
VMCIHandle handle;
DatagramEntry *entry;
ASSERT(recvCB != NULL);
ASSERT(outHandle != NULL);
ASSERT(!(privFlags & ~VMCI_PRIVILEGE_ALL_FLAGS));
if ((flags & VMCI_FLAG_WELLKNOWN_DG_HND) != 0) {
return VMCI_ERROR_INVALID_ARGS;
} else {
if ((flags & VMCI_FLAG_ANYCID_DG_HND) != 0) {
contextID = VMCI_INVALID_ID;
} else {
contextID = vmci_get_context_id();
if (contextID == VMCI_INVALID_ID) {
return VMCI_ERROR_NO_RESOURCES;
}
}
if (resourceID == VMCI_INVALID_ID) {
resourceID = VMCIResource_GetID(contextID);
if (resourceID == VMCI_INVALID_ID) {
return VMCI_ERROR_NO_HANDLE;
}
}
handle = VMCI_MAKE_HANDLE(contextID, resourceID);
}
entry = VMCI_AllocKernelMem(sizeof *entry, VMCI_MEMORY_NONPAGED);
if (entry == NULL) {
VMCI_WARNING((LGPFX"Failed allocating memory for datagram entry.\n"));
return VMCI_ERROR_NO_MEM;
}
if (!VMCI_CanScheduleDelayedWork()) {
if (flags & VMCI_FLAG_DG_DELAYED_CB) {
VMCI_FreeKernelMem(entry, sizeof *entry);
return VMCI_ERROR_INVALID_ARGS;
}
entry->runDelayed = FALSE;
} else {
entry->runDelayed = (flags & VMCI_FLAG_DG_DELAYED_CB) ? TRUE : FALSE;
}
entry->flags = flags;
entry->recvCB = recvCB;
entry->clientData = clientData;
VMCI_CreateEvent(&entry->destroyEvent);
entry->privFlags = privFlags;
/* Make datagram resource live. */
result = VMCIResource_Add(&entry->resource, VMCI_RESOURCE_TYPE_DATAGRAM,
handle, DatagramFreeCB, entry);
if (result != VMCI_SUCCESS) {
VMCI_WARNING((LGPFX"Failed to add new resource (handle=0x%x:0x%x).\n",
handle.context, handle.resource));
VMCI_DestroyEvent(&entry->destroyEvent);
VMCI_FreeKernelMem(entry, sizeof *entry);
return result;
}
*outHandle = handle;
return VMCI_SUCCESS;
}
void
VMCI_FreeBuffer(VMCIBuffer buf, // IN:
size_t size) // IN: Unused on Linux
{
VMCI_FreeKernelMem(buf, size);
}
int
VMCI_AllocPPNSet(void *produceQ, // IN:
uint64 numProducePages, // IN: for queue plus header
void *consumeQ, // IN:
uint64 numConsumePages, // IN: for queue plus header
PPNSet *ppnSet) // OUT:
{
VMCIPpnList producePPNs;
VMCIPpnList consumePPNs;
uint64 i;
if (!produceQ || !numProducePages || !consumeQ || !numConsumePages ||
!ppnSet) {
return VMCI_ERROR_INVALID_ARGS;
}
if (ppnSet->initialized) {
return VMCI_ERROR_ALREADY_EXISTS;
}
producePPNs =
VMCI_AllocKernelMem(numProducePages * sizeof *producePPNs,
VMCI_MEMORY_NORMAL);
if (!producePPNs) {
return VMCI_ERROR_NO_MEM;
}
consumePPNs =
VMCI_AllocKernelMem(numConsumePages * sizeof *consumePPNs,
VMCI_MEMORY_NORMAL);
if (!consumePPNs) {
VMCI_FreeKernelMem(producePPNs, numProducePages * sizeof *producePPNs);
return VMCI_ERROR_NO_MEM;
}
producePPNs[0] = VMCIKVaToMPN(produceQ);
for (i = 1; i < numProducePages; i++) {
unsigned long pfn;
producePPNs[i] = pfn = page_to_pfn(((VMCIQueue *)produceQ)->page[i - 1]);
/*
* Fail allocation if PFN isn't supported by hypervisor.
*/
if (sizeof pfn > sizeof *producePPNs &&
pfn != producePPNs[i]) {
goto ppnError;
}
}
consumePPNs[0] = VMCIKVaToMPN(consumeQ);
for (i = 1; i < numConsumePages; i++) {
unsigned long pfn;
consumePPNs[i] = pfn = page_to_pfn(((VMCIQueue *)consumeQ)->page[i - 1]);
/*
* Fail allocation if PFN isn't supported by hypervisor.
*/
if (sizeof pfn > sizeof *consumePPNs &&
pfn != consumePPNs[i]) {
goto ppnError;
}
}
ppnSet->numProducePages = numProducePages;
ppnSet->numConsumePages = numConsumePages;
ppnSet->producePPNs = producePPNs;
ppnSet->consumePPNs = consumePPNs;
ppnSet->initialized = TRUE;
return VMCI_SUCCESS;
ppnError:
VMCI_FreeKernelMem(producePPNs, numProducePages * sizeof *producePPNs);
VMCI_FreeKernelMem(consumePPNs, numConsumePages * sizeof *consumePPNs);
return VMCI_ERROR_INVALID_ARGS;
}
int
VMCIHost_GetUserMemory(PageStoreAttachInfo *attach, // IN/OUT
VMCIQueue *produceQ, // OUT
VMCIQueue *consumeQ) // OUT
{
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
int retval;
int err = VMCI_SUCCESS;
attach->producePages =
VMCI_AllocKernelMem(attach->numProducePages * sizeof attach->producePages[0],
VMCI_MEMORY_NORMAL);
if (attach->producePages == NULL) {
return VMCI_ERROR_NO_MEM;
}
attach->consumePages =
VMCI_AllocKernelMem(attach->numConsumePages * sizeof attach->consumePages[0],
VMCI_MEMORY_NORMAL);
if (attach->consumePages == NULL) {
err = VMCI_ERROR_NO_MEM;
goto errorDealloc;
}
down_write(¤t->mm->mmap_sem);
retval = get_user_pages(current,
current->mm,
(VA)attach->produceBuffer,
attach->numProducePages,
1, 0,
attach->producePages,
NULL);
if (retval < attach->numProducePages) {
Log("get_user_pages(produce) failed: %d\n", retval);
if (retval > 0) {
int i;
for (i = 0; i < retval; i++) {
page_cache_release(attach->producePages[i]);
}
}
err = VMCI_ERROR_NO_MEM;
goto out;
}
retval = get_user_pages(current,
current->mm,
(VA)attach->consumeBuffer,
attach->numConsumePages,
1, 0,
attach->consumePages,
NULL);
if (retval < attach->numConsumePages) {
int i;
Log("get_user_pages(consume) failed: %d\n", retval);
if (retval > 0) {
for (i = 0; i < retval; i++) {
page_cache_release(attach->consumePages[i]);
}
}
for (i = 0; i < attach->numProducePages; i++) {
page_cache_release(attach->producePages[i]);
}
err = VMCI_ERROR_NO_MEM;
}
if (err == VMCI_SUCCESS) {
produceQ->queueHeaderPtr = kmap(attach->producePages[0]);
produceQ->page = &attach->producePages[1];
consumeQ->queueHeaderPtr = kmap(attach->consumePages[0]);
consumeQ->page = &attach->consumePages[1];
}
out:
up_write(¤t->mm->mmap_sem);
errorDealloc:
if (err < VMCI_SUCCESS) {
if (attach->producePages != NULL) {
VMCI_FreeKernelMem(attach->producePages,
attach->numProducePages *
sizeof attach->producePages[0]);
}
if (attach->consumePages != NULL) {
VMCI_FreeKernelMem(attach->consumePages,
attach->numConsumePages *
sizeof attach->consumePages[0]);
}
}
return err;
#else
/*
* Host queue pair support for earlier kernels temporarily
* disabled. See bug 365496.
*/
ASSERT_NOT_IMPLEMENTED(FALSE);
#if 0
attach->produceIoBuf = VMCI_AllocKernelMem(sizeof *attach->produceIoBuf,
VMCI_MEMORY_NORMAL);
if (attach->produceIoBuf == NULL) {
return VMCI_ERROR_NO_MEM;
}
attach->consumeIoBuf = VMCI_AllocKernelMem(sizeof *attach->consumeIoBuf,
VMCI_MEMORY_NORMAL);
if (attach->consumeIoBuf == NULL) {
VMCI_FreeKernelMem(attach->produceIoBuf,
sizeof *attach->produceIoBuf);
return VMCI_ERROR_NO_MEM;
}
retval = map_user_kiobuf(WRITE, attach->produceIoBuf,
(VA)attach->produceBuffer,
attach->numProducePages * PAGE_SIZE);
if (retval < 0) {
err = VMCI_ERROR_NO_ACCESS;
goto out;
}
retval = map_user_kiobuf(WRITE, attach->consumeIoBuf,
(VA)attach->consumeBuffer,
attach->numConsumePages * PAGE_SIZE);
if (retval < 0) {
unmap_kiobuf(attach->produceIoBuf);
err = VMCI_ERROR_NO_ACCESS;
}
if (err == VMCI_SUCCESS) {
produceQ->queueHeaderPtr = kmap(attach->produceIoBuf->maplist[0]);
produceQ->page = &attach->produceIoBuf->maplist[1];
consumeQ->queueHeaderPtr = kmap(attach->consumeIoBuf->maplist[0]);
consumeQ->page = &attach->consumeIoBuf->maplist[1];
}
out:
if (err < VMCI_SUCCESS) {
if (attach->produceIoBuf != NULL) {
VMCI_FreeKernelMem(attach->produceIoBuf,
sizeof *attach->produceIoBuf);
}
if (attach->consumeIoBuf != NULL) {
VMCI_FreeKernelMem(attach->consumeIoBuf,
sizeof *attach->consumeIoBuf);
}
}
return err;
#else // 0 -- Instead just return FALSE
return FALSE;
#endif // 0
#endif // Linux version >= 2.6.0
}
static int
VPageChannelAddRecvBuffers(VPageChannel *channel, // IN
int numElems, // IN
Bool onInit) // IN
{
int n;
int sent;
int maxElems;
Bool isAtomic;
size_t size;
unsigned long flags;
VPageChannelElem *elems;
VPageChannelPacket *packet;
ASSERT(channel);
sent = 0;
size = 0;
elems = NULL;
packet = NULL;
if (onInit || (channel->flags & VPAGECHANNEL_FLAGS_RECV_DELAYED)) {
/*
* If we are initializing the channel, or we are running in a delayed
* context (recv() in this case), then we can using blocking allocation
* and we can allocate large packets. Also, no need to take the
* send lock here, we can just take it for each packet.
*/
isAtomic = FALSE;
maxElems = VMCI_PACKET_DGRAM_MAX_ELEMS;
flags = 0; /* Silence compiler. */
} else {
/*
* We're in an atomic context. We must allocate page-sized packets
* atomically and send them over the queuepair. Since this can
* cause a lot of signalling, we optimize by taking the send lock
* once for all packets, and only signalling when we are done.
*/
isAtomic = TRUE;
maxElems = VMCI_PACKET_PAGE_MAX_ELEMS;
VPageChannelAcquireSendLock(channel, &flags);
}
n = min_t(int, maxElems, numElems);
while (n > 0) {
int retval;
int allocNum;
/*
* First packet is always big enough to cover any remaining elements,
* so just allocate it once.
*/
if (NULL == packet) {
size = sizeof(VPageChannelPacket) + (n * sizeof(VPageChannelElem));
packet = (VPageChannelPacket *)
VMCI_AllocKernelMem(size,
isAtomic ? VMCI_MEMORY_ATOMIC : VMCI_MEMORY_NORMAL);
if (packet == NULL) {
VMCI_WARNING((LGPFX"Failed to allocate packet (channel=%p) "
"(size=%"FMTSZ"u).\n",
channel,
size));
goto exit;
}
packet->type = VPCPacket_SetRecvBuffer;
packet->msgLen = 0;
elems = VPAGECHANNEL_PACKET_ELEMS(packet);
}
allocNum = channel->elemAllocFn(channel->allocClientData, elems, n);
if (0 == allocNum) {
/*
* If the client failed to allocate any elements at all then just
* bail out and return whatever number we managed to send so far
* (if any).
*/
VMCI_WARNING((LGPFX"Failed to allocate receive buffer (channel=%p) "
"(expected=%d).\n",
channel,
n));
goto exit;
}
/*
* We wanted "n" elements, but we might only have "allocNum" because
* that's all the client could allocate. Pass down whatever we got.
*/
packet->numElems = allocNum;
if (onInit) {
retval = VPageChannelSendControl(channel, VPCPacket_SetRecvBuffer,
NULL, 0, allocNum, elems);
} else {
/*
* Do not ask for the lock here if we are atomic, we take care of
* that ourselves. Similarly, if we are atomic then we will do our
* own signalling, so inform the send that there is a signal already
* pending.
*/
retval = VPageChannelSendPacket(channel, packet,
isAtomic ? FALSE : TRUE, // needsLock
isAtomic ? TRUE : FALSE); // signalPending
/*
* XXX, what if this is a non-blocking queuepair and we fail to
* send because it's full and we can't wait? Is it even worth it
* to loop?
*/
}
if (retval < VMCI_SUCCESS) {
/*
* Failure to send is fatal. Release the client's elements and
* bail out.
*/
VMCI_WARNING((LGPFX"Failed to set receive buffers (channel=%p) "
"(err=%d).\n",
channel,
retval));
channel->elemFreeFn(channel->freeClientData, elems, allocNum);
goto exit;
}
Atomic_Add32(&channel->curRecvBufs, allocNum);
sent += allocNum;
numElems -= allocNum;
n = min_t(int, maxElems, numElems);
}
exit:
if (isAtomic) {
/*
* We're done sending packets, so now we can signal. Even if we only
* sent some of the requested buffers, we must signal anyway, otherwise
* the peer won't know about the ones we did send.
*/
(void)VPageChannelSignal(channel);
VPageChannelReleaseSendLock(channel, flags);
}
if (NULL != packet) {
VMCI_FreeKernelMem(packet, size);
}
return sent;
}
int
VPageChannel_Send(VPageChannel *channel, // IN/OUT
VPageChannelPacketType type, // IN
char *message, // IN
int len, // IN
VPageChannelBuffer *buffer) // IN
{
int retval;
int numElems;
ssize_t totalSize;
VPageChannelPacket *packet;
ASSERT(channel);
if (VPCState_Connected != channel->state) {
VMCI_WARNING((LGPFX"Not connected (channel=%p).\n",
channel));
return VMCI_ERROR_DST_UNREACHABLE;
}
if (buffer) {
numElems = buffer->numElems;
} else {
numElems = 0;
}
totalSize = sizeof(VPageChannelPacket) + len +
numElems * sizeof(VPageChannelElem);
packet = (VPageChannelPacket *)
VMCI_AllocKernelMem(totalSize,
channel->flags & VPAGECHANNEL_FLAGS_SEND_WHILE_ATOMIC ?
VMCI_MEMORY_ATOMIC : VMCI_MEMORY_NORMAL);
if (!packet) {
VMCI_WARNING((LGPFX"Failed to allocate packet (channel=%p) "
"(size=%"FMTSZ"d).",
channel,
totalSize));
return VMCI_ERROR_NO_MEM;
}
packet->type = type;
packet->msgLen = len;
packet->numElems = numElems;
if (len) {
ASSERT(message);
memcpy(VPAGECHANNEL_PACKET_MESSAGE(packet), message, len);
}
if (numElems) {
ASSERT(buffer);
ASSERT(buffer->elems);
memcpy(VPAGECHANNEL_PACKET_ELEMS(packet), buffer->elems,
numElems * sizeof (VPageChannelElem));
}
retval = VPageChannel_SendPacket(channel, packet);
VMCI_FreeKernelMem(packet, totalSize);
return retval;
}
static int
VPageChannelSendControl(VPageChannel *channel, // IN
VPageChannelPacketType type, // IN
char *message, // IN
int len, // IN
int numElems, // IN
VPageChannelElem *elems) // IN
{
int retval;
VPageChannelPacket *packet;
VMCIDatagram *dg;
ASSERT(channel);
ASSERT(type == VPCPacket_Data ||
type == VPCPacket_GuestConnect ||
type == VPCPacket_SetRecvBuffer ||
type == VPCPacket_GuestDisconnect);
dg = NULL;
retval = VPageChannelAllocDatagram(channel, len, numElems, &dg);
if (retval < VMCI_SUCCESS) {
return retval;
}
packet = (VPageChannelPacket *)VMCI_DG_PAYLOAD(dg);
packet->type = type;
packet->msgLen = len;
packet->numElems = numElems;
if (len) {
ASSERT(message);
memcpy(VPAGECHANNEL_PACKET_MESSAGE(packet), message, len);
}
if (numElems) {
ASSERT(elems);
memcpy(VPAGECHANNEL_PACKET_ELEMS(packet), elems,
numElems * sizeof (VPageChannelElem));
}
retval = vmci_datagram_send(dg);
if (retval < VMCI_SUCCESS) {
VMCI_WARNING((LGPFX"Failed to send packet (channel=%p) to "
"(handle=0x%x:0x%x) (err=%d).\n",
channel,
dg->dst.context,
dg->dst.resource,
retval));
} else {
/*
* We don't care about how many bytes were sent, and callers may not
* expect > 0 to mean success, so just convert to exactly success.
*/
retval = VMCI_SUCCESS;
}
VMCI_FreeKernelMem(dg, VMCI_DG_SIZE(dg));
return retval;
}