/* $Id: PDMAsyncCompletionFileNormal.cpp 29450 2010-05-13 15:35:35Z vboxsync $ */ /** @file * PDM Async I/O - Transport data asynchronous in R3 using EMT. * Async File I/O manager. */ /* * Copyright (C) 2006-2008 Oracle Corporation * * This file is part of VirtualBox Open Source Edition (OSE), as * available from http://www.virtualbox.org. This file is free software; * you can redistribute it and/or modify it under the terms of the GNU * General Public License (GPL) as published by the Free Software * Foundation, in version 2 as it comes in the "COPYING" file of the * VirtualBox OSE distribution. VirtualBox OSE is distributed in the * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind. */ #define LOG_GROUP LOG_GROUP_PDM_ASYNC_COMPLETION #include #include #include #include #include #include #include #include "PDMAsyncCompletionFileInternal.h" /** The update period for the I/O load statistics in ms. */ #define PDMACEPFILEMGR_LOAD_UPDATE_PERIOD 1000 /** Maximum number of requests a manager will handle. */ #define PDMACEPFILEMGR_REQS_STEP 512 /******************************************************************************* * Internal functions * *******************************************************************************/ static int pdmacFileAioMgrNormalProcessTaskList(PPDMACTASKFILE pTaskHead, PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint); static PPDMACTASKFILE pdmacFileAioMgrNormalRangeLockFree(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACFILERANGELOCK pRangeLock); int pdmacFileAioMgrNormalInit(PPDMACEPFILEMGR pAioMgr) { int rc = VINF_SUCCESS; pAioMgr->cRequestsActiveMax = PDMACEPFILEMGR_REQS_STEP; rc = RTFileAioCtxCreate(&pAioMgr->hAioCtx, RTFILEAIO_UNLIMITED_REQS); if (rc == VERR_OUT_OF_RANGE) rc = RTFileAioCtxCreate(&pAioMgr->hAioCtx, pAioMgr->cRequestsActiveMax); if (RT_SUCCESS(rc)) { /* Initialize request handle array. */ pAioMgr->iFreeEntry = 0; pAioMgr->cReqEntries = pAioMgr->cRequestsActiveMax; pAioMgr->pahReqsFree = (RTFILEAIOREQ *)RTMemAllocZ(pAioMgr->cReqEntries * sizeof(RTFILEAIOREQ)); if (pAioMgr->pahReqsFree) { /* Create the range lock memcache. */ rc = RTMemCacheCreate(&pAioMgr->hMemCacheRangeLocks, sizeof(PDMACFILERANGELOCK), 0, UINT32_MAX, NULL, NULL, NULL, 0); if (RT_SUCCESS(rc)) return VINF_SUCCESS; RTMemFree(pAioMgr->pahReqsFree); } else { RTFileAioCtxDestroy(pAioMgr->hAioCtx); rc = VERR_NO_MEMORY; } } return rc; } void pdmacFileAioMgrNormalDestroy(PPDMACEPFILEMGR pAioMgr) { RTFileAioCtxDestroy(pAioMgr->hAioCtx); while (pAioMgr->iFreeEntry > 0) { pAioMgr->iFreeEntry--; Assert(pAioMgr->pahReqsFree[pAioMgr->iFreeEntry] != NIL_RTFILEAIOREQ); RTFileAioReqDestroy(pAioMgr->pahReqsFree[pAioMgr->iFreeEntry]); } RTMemFree(pAioMgr->pahReqsFree); RTMemCacheDestroy(pAioMgr->hMemCacheRangeLocks); } /** * Sorts the endpoint list with insertion sort. */ static void pdmacFileAioMgrNormalEndpointsSortByLoad(PPDMACEPFILEMGR pAioMgr) { PPDMASYNCCOMPLETIONENDPOINTFILE pEpPrev, pEpCurr, pEpNextToSort; pEpPrev = pAioMgr->pEndpointsHead; pEpCurr = pEpPrev->AioMgr.pEndpointNext; while (pEpCurr) { /* Remember the next element to sort because the list might change. */ pEpNextToSort = pEpCurr->AioMgr.pEndpointNext; /* Unlink the current element from the list. */ PPDMASYNCCOMPLETIONENDPOINTFILE pPrev = pEpCurr->AioMgr.pEndpointPrev; PPDMASYNCCOMPLETIONENDPOINTFILE pNext = pEpCurr->AioMgr.pEndpointNext; if (pPrev) pPrev->AioMgr.pEndpointNext = pNext; else pAioMgr->pEndpointsHead = pNext; if (pNext) pNext->AioMgr.pEndpointPrev = pPrev; /* Go back until we reached the place to insert the current endpoint into. */ while (pEpPrev && (pEpPrev->AioMgr.cReqsPerSec < pEpCurr->AioMgr.cReqsPerSec)) pEpPrev = pEpPrev->AioMgr.pEndpointPrev; /* Link the endpoint into the list. */ if (pEpPrev) pNext = pEpPrev->AioMgr.pEndpointNext; else pNext = pAioMgr->pEndpointsHead; pEpCurr->AioMgr.pEndpointNext = pNext; pEpCurr->AioMgr.pEndpointPrev = pEpPrev; if (pNext) pNext->AioMgr.pEndpointPrev = pEpCurr; if (pEpPrev) pEpPrev->AioMgr.pEndpointNext = pEpCurr; else pAioMgr->pEndpointsHead = pEpCurr; pEpCurr = pEpNextToSort; } #ifdef DEBUG /* Validate sorting alogrithm */ unsigned cEndpoints = 0; pEpCurr = pAioMgr->pEndpointsHead; AssertMsg(pEpCurr, ("No endpoint in the list?\n")); AssertMsg(!pEpCurr->AioMgr.pEndpointPrev, ("First element in the list points to previous element\n")); while (pEpCurr) { cEndpoints++; PPDMASYNCCOMPLETIONENDPOINTFILE pNext = pEpCurr->AioMgr.pEndpointNext; PPDMASYNCCOMPLETIONENDPOINTFILE pPrev = pEpCurr->AioMgr.pEndpointPrev; Assert(!pNext || pNext->AioMgr.cReqsPerSec <= pEpCurr->AioMgr.cReqsPerSec); Assert(!pPrev || pPrev->AioMgr.cReqsPerSec >= pEpCurr->AioMgr.cReqsPerSec); pEpCurr = pNext; } AssertMsg(cEndpoints == pAioMgr->cEndpoints, ("Endpoints lost during sort!\n")); #endif } /** * Removes an endpoint from the currently assigned manager. * * @returns TRUE if there are still requests pending on the current manager for this endpoint. * FALSE otherwise. * @param pEndpointRemove The endpoint to remove. */ static bool pdmacFileAioMgrNormalRemoveEndpoint(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpointRemove) { PPDMASYNCCOMPLETIONENDPOINTFILE pPrev = pEndpointRemove->AioMgr.pEndpointPrev; PPDMASYNCCOMPLETIONENDPOINTFILE pNext = pEndpointRemove->AioMgr.pEndpointNext; PPDMACEPFILEMGR pAioMgr = pEndpointRemove->pAioMgr; pAioMgr->cEndpoints--; if (pPrev) pPrev->AioMgr.pEndpointNext = pNext; else pAioMgr->pEndpointsHead = pNext; if (pNext) pNext->AioMgr.pEndpointPrev = pPrev; /* Make sure that there is no request pending on this manager for the endpoint. */ if (!pEndpointRemove->AioMgr.cRequestsActive) { Assert(!pEndpointRemove->pFlushReq); /* Reopen the file so that the new endpoint can reassociate with the file */ RTFileClose(pEndpointRemove->File); int rc = RTFileOpen(&pEndpointRemove->File, pEndpointRemove->Core.pszUri, pEndpointRemove->fFlags); AssertRC(rc); return false; } return true; } static bool pdmacFileAioMgrNormalIsBalancePossible(PPDMACEPFILEMGR pAioMgr) { /* Balancing doesn't make sense with only one endpoint. */ if (pAioMgr->cEndpoints == 1) return false; /* Doesn't make sens to move endpoints if only one produces the whole load */ unsigned cEndpointsWithLoad = 0; PPDMASYNCCOMPLETIONENDPOINTFILE pCurr = pAioMgr->pEndpointsHead; while (pCurr) { if (pCurr->AioMgr.cReqsPerSec) cEndpointsWithLoad++; pCurr = pCurr->AioMgr.pEndpointNext; } return (cEndpointsWithLoad > 1); } /** * Creates a new I/O manager and spreads the I/O load of the endpoints * between the given I/O manager and the new one. * * @returns nothing. * @param pAioMgr The I/O manager with high I/O load. */ static void pdmacFileAioMgrNormalBalanceLoad(PPDMACEPFILEMGR pAioMgr) { PPDMACEPFILEMGR pAioMgrNew = NULL; int rc = VINF_SUCCESS; /* * Check if balancing would improve the situation. */ if (pdmacFileAioMgrNormalIsBalancePossible(pAioMgr)) { PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pAioMgr->pEndpointsHead->Core.pEpClass; rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgrNew, PDMACEPFILEMGRTYPE_ASYNC); if (RT_SUCCESS(rc)) { /* We will sort the list by request count per second. */ pdmacFileAioMgrNormalEndpointsSortByLoad(pAioMgr); /* Now move some endpoints to the new manager. */ unsigned cReqsHere = pAioMgr->pEndpointsHead->AioMgr.cReqsPerSec; unsigned cReqsOther = 0; PPDMASYNCCOMPLETIONENDPOINTFILE pCurr = pAioMgr->pEndpointsHead->AioMgr.pEndpointNext; while (pCurr) { if (cReqsHere <= cReqsOther) { /* * The other manager has more requests to handle now. * We will keep the current endpoint. */ Log(("Keeping endpoint %#p{%s} with %u reqs/s\n", pCurr->Core.pszUri, pCurr->AioMgr.cReqsPerSec)); cReqsHere += pCurr->AioMgr.cReqsPerSec; pCurr = pCurr->AioMgr.pEndpointNext; } else { /* Move to other endpoint. */ Log(("Moving endpoint %#p{%s} with %u reqs/s to other manager\n", pCurr, pCurr->Core.pszUri, pCurr->AioMgr.cReqsPerSec)); cReqsOther += pCurr->AioMgr.cReqsPerSec; PPDMASYNCCOMPLETIONENDPOINTFILE pMove = pCurr; pCurr = pCurr->AioMgr.pEndpointNext; bool fReqsPending = pdmacFileAioMgrNormalRemoveEndpoint(pMove); if (fReqsPending) { pMove->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_REMOVING; pMove->AioMgr.fMoving = true; pMove->AioMgr.pAioMgrDst = pAioMgrNew; } else { pMove->AioMgr.fMoving = false; pMove->AioMgr.pAioMgrDst = NULL; pdmacFileAioMgrAddEndpoint(pAioMgrNew, pMove); } } } } else { /* Don't process further but leave a log entry about reduced performance. */ LogRel(("AIOMgr: Could not create new I/O manager (rc=%Rrc). Expect reduced performance\n", rc)); } } else Log(("AIOMgr: Load balancing would not improve anything\n")); } /** * Increase the maximum number of active requests for the given I/O manager. * * @returns VBox status code. * @param pAioMgr The I/O manager to grow. */ static int pdmacFileAioMgrNormalGrow(PPDMACEPFILEMGR pAioMgr) { int rc = VINF_SUCCESS; RTFILEAIOCTX hAioCtxNew = NIL_RTFILEAIOCTX; LogFlowFunc(("pAioMgr=%#p\n", pAioMgr)); AssertMsg( pAioMgr->enmState == PDMACEPFILEMGRSTATE_GROWING && !pAioMgr->cRequestsActive, ("Invalid state of the I/O manager\n")); #ifdef RT_OS_WINDOWS /* * Reopen the files of all assigned endpoints first so we can assign them to the new * I/O context. */ PPDMASYNCCOMPLETIONENDPOINTFILE pCurr = pAioMgr->pEndpointsHead; while (pCurr) { RTFileClose(pCurr->File); rc = RTFileOpen(&pCurr->File, pCurr->Core.pszUri, pCurr->fFlags); AssertRC(rc); pCurr = pCurr->AioMgr.pEndpointNext; } #endif /* Create the new bigger context. */ pAioMgr->cRequestsActiveMax += PDMACEPFILEMGR_REQS_STEP; rc = RTFileAioCtxCreate(&hAioCtxNew, RTFILEAIO_UNLIMITED_REQS); if (rc == VERR_OUT_OF_RANGE) rc = RTFileAioCtxCreate(&hAioCtxNew, pAioMgr->cRequestsActiveMax); if (RT_SUCCESS(rc)) { /* Close the old context. */ rc = RTFileAioCtxDestroy(pAioMgr->hAioCtx); AssertRC(rc); pAioMgr->hAioCtx = hAioCtxNew; /* Create a new I/O task handle array */ uint32_t cReqEntriesNew = pAioMgr->cRequestsActiveMax + 1; RTFILEAIOREQ *pahReqNew = (RTFILEAIOREQ *)RTMemAllocZ(cReqEntriesNew * sizeof(RTFILEAIOREQ)); if (pahReqNew) { /* Copy the cached request handles. */ for (uint32_t iReq = 0; iReq < pAioMgr->cReqEntries; iReq++) pahReqNew[iReq] = pAioMgr->pahReqsFree[iReq]; RTMemFree(pAioMgr->pahReqsFree); pAioMgr->pahReqsFree = pahReqNew; pAioMgr->cReqEntries = cReqEntriesNew; LogFlowFunc(("I/O manager increased to handle a maximum of %u requests\n", pAioMgr->cRequestsActiveMax)); } else rc = VERR_NO_MEMORY; } #ifdef RT_OS_WINDOWS /* Assign the file to the new context. */ pCurr = pAioMgr->pEndpointsHead; while (pCurr) { rc = RTFileAioCtxAssociateWithFile(pAioMgr->hAioCtx, pCurr->File); AssertRC(rc); pCurr = pCurr->AioMgr.pEndpointNext; } #endif if (RT_FAILURE(rc)) { LogFlow(("Increasing size of the I/O manager failed with rc=%Rrc\n", rc)); pAioMgr->cRequestsActiveMax -= PDMACEPFILEMGR_REQS_STEP; } pAioMgr->enmState = PDMACEPFILEMGRSTATE_RUNNING; LogFlowFunc(("returns rc=%Rrc\n", rc)); return rc; } /** * Checks if a given status code is fatal. * Non fatal errors can be fixed by migrating the endpoint to a * failsafe manager. * * @returns true If the error is fatal and migrating to a failsafe manager doesn't help * false If the error can be fixed by a migration. (image on NFS disk for example) * @param rcReq The status code to check. */ DECLINLINE(bool) pdmacFileAioMgrNormalRcIsFatal(int rcReq) { return rcReq == VERR_DEV_IO_ERROR || rcReq == VERR_FILE_IO_ERROR || rcReq == VERR_DISK_IO_ERROR || rcReq == VERR_DISK_FULL || rcReq == VERR_FILE_TOO_BIG; } /** * Error handler which will create the failsafe managers and destroy the failed I/O manager. * * @returns VBox status code * @param pAioMgr The I/O manager the error ocurred on. * @param rc The error code. */ static int pdmacFileAioMgrNormalErrorHandler(PPDMACEPFILEMGR pAioMgr, int rc, RT_SRC_POS_DECL) { LogRel(("AIOMgr: I/O manager %#p encountered a critical error (rc=%Rrc) during operation. Falling back to failsafe mode. Expect reduced performance\n", pAioMgr, rc)); LogRel(("AIOMgr: Error happened in %s:(%u){%s}\n", RT_SRC_POS_ARGS)); LogRel(("AIOMgr: Please contact the product vendor\n")); PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pAioMgr->pEndpointsHead->Core.pEpClass; pAioMgr->enmState = PDMACEPFILEMGRSTATE_FAULT; ASMAtomicWriteU32((volatile uint32_t *)&pEpClassFile->enmMgrTypeOverride, PDMACEPFILEMGRTYPE_SIMPLE); AssertMsgFailed(("Implement\n")); return VINF_SUCCESS; } /** * Put a list of tasks in the pending request list of an endpoint. */ DECLINLINE(void) pdmacFileAioMgrEpAddTaskList(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTaskHead) { /* Add the rest of the tasks to the pending list */ if (!pEndpoint->AioMgr.pReqsPendingHead) { Assert(!pEndpoint->AioMgr.pReqsPendingTail); pEndpoint->AioMgr.pReqsPendingHead = pTaskHead; } else { Assert(pEndpoint->AioMgr.pReqsPendingTail); pEndpoint->AioMgr.pReqsPendingTail->pNext = pTaskHead; } /* Update the tail. */ while (pTaskHead->pNext) pTaskHead = pTaskHead->pNext; pEndpoint->AioMgr.pReqsPendingTail = pTaskHead; pTaskHead->pNext = NULL; } /** * Put one task in the pending request list of an endpoint. */ DECLINLINE(void) pdmacFileAioMgrEpAddTask(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask) { /* Add the rest of the tasks to the pending list */ if (!pEndpoint->AioMgr.pReqsPendingHead) { Assert(!pEndpoint->AioMgr.pReqsPendingTail); pEndpoint->AioMgr.pReqsPendingHead = pTask; } else { Assert(pEndpoint->AioMgr.pReqsPendingTail); pEndpoint->AioMgr.pReqsPendingTail->pNext = pTask; } pEndpoint->AioMgr.pReqsPendingTail = pTask; pTask->pNext = NULL; } /** * Allocates a async I/O request. * * @returns Handle to the request. * @param pAioMgr The I/O manager. */ static RTFILEAIOREQ pdmacFileAioMgrNormalRequestAlloc(PPDMACEPFILEMGR pAioMgr) { RTFILEAIOREQ hReq = NIL_RTFILEAIOREQ; /* Get a request handle. */ if (pAioMgr->iFreeEntry > 0) { pAioMgr->iFreeEntry--; hReq = pAioMgr->pahReqsFree[pAioMgr->iFreeEntry]; pAioMgr->pahReqsFree[pAioMgr->iFreeEntry] = NIL_RTFILEAIOREQ; Assert(hReq != NIL_RTFILEAIOREQ); } else { int rc = RTFileAioReqCreate(&hReq); AssertRC(rc); } return hReq; } /** * Frees a async I/O request handle. * * @returns nothing. * @param pAioMgr The I/O manager. * @param hReq The I/O request handle to free. */ static void pdmacFileAioMgrNormalRequestFree(PPDMACEPFILEMGR pAioMgr, RTFILEAIOREQ hReq) { Assert(pAioMgr->iFreeEntry < pAioMgr->cReqEntries); Assert(pAioMgr->pahReqsFree[pAioMgr->iFreeEntry] == NIL_RTFILEAIOREQ); pAioMgr->pahReqsFree[pAioMgr->iFreeEntry] = hReq; pAioMgr->iFreeEntry++; } /** * Wrapper around RTFIleAioCtxSubmit() which is also doing error handling. */ static int pdmacFileAioMgrNormalReqsEnqueue(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PRTFILEAIOREQ pahReqs, unsigned cReqs) { int rc; pAioMgr->cRequestsActive += cReqs; pEndpoint->AioMgr.cRequestsActive += cReqs; LogFlow(("Enqueuing %d requests. I/O manager has a total of %d active requests now\n", cReqs, pAioMgr->cRequestsActive)); LogFlow(("Endpoint has a total of %d active requests now\n", pEndpoint->AioMgr.cRequestsActive)); rc = RTFileAioCtxSubmit(pAioMgr->hAioCtx, pahReqs, cReqs); if (RT_FAILURE(rc)) { PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass; unsigned cReqsResubmit = 0; RTFILEAIOREQ ahReqsResubmit[20]; /* * We run out of resources. * Need to check which requests got queued * and put the rest on the pending list again. */ for (size_t i = 0; i < cReqs; i++) { int rcReq = RTFileAioReqGetRC(pahReqs[i], NULL); if (rcReq != VERR_FILE_AIO_IN_PROGRESS) { PPDMACTASKFILE pTask = (PPDMACTASKFILE)RTFileAioReqGetUser(pahReqs[i]); if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH) { /* Mark as not supported. */ if (rcReq != VERR_FILE_AIO_NOT_SUBMITTED) { LogFlow(("Async flushes are not supported for this endpoint, disabling\n")); pEndpoint->fAsyncFlushSupported = false; pdmacFileAioMgrNormalRequestFree(pAioMgr, pahReqs[i]); rc = VINF_SUCCESS; } else { AssertMsg(rc == VERR_FILE_AIO_INSUFFICIENT_RESSOURCES, ("Flush wasn't submitted but we are not out of ressources\n")); /* Clear the pending flush */ pdmacFileAioMgrEpAddTask(pEndpoint, pTask); Assert(pEndpoint->pFlushReq == pTask); pEndpoint->pFlushReq = NULL; } } else { AssertMsg(rcReq == VERR_FILE_AIO_NOT_SUBMITTED, ("Request returned unexpected return code: rc=%Rrc\n", rcReq)); if (rc == VERR_FILE_AIO_INSUFFICIENT_RESSOURCES) { PPDMACTASKFILE pTasksWaiting; pdmacFileAioMgrNormalRequestFree(pAioMgr, pahReqs[i]); if (pTask->cbBounceBuffer) RTMemPageFree(pTask->pvBounceBuffer, pTask->cbBounceBuffer); pTask->fPrefetch = false; pTask->cbBounceBuffer = 0; /* Free the lock and process pending tasks if neccessary */ pTasksWaiting = pdmacFileAioMgrNormalRangeLockFree(pAioMgr, pEndpoint, pTask->pRangeLock); pdmacFileAioMgrEpAddTask(pEndpoint, pTask); if (pTasksWaiting) pdmacFileAioMgrEpAddTaskList(pEndpoint, pTasksWaiting); } else { ahReqsResubmit[cReqsResubmit] = pahReqs[i]; cReqsResubmit++; } } pEndpoint->AioMgr.cRequestsActive--; pAioMgr->cRequestsActive--; if (cReqsResubmit == RT_ELEMENTS(ahReqsResubmit)) { int rc2 = RTFileAioCtxSubmit(pAioMgr->hAioCtx, ahReqsResubmit, cReqsResubmit); AssertRC(rc2); cReqsResubmit = 0; } } /* Resubmit tasks. */ if (cReqsResubmit) { int rc2 = RTFileAioCtxSubmit(pAioMgr->hAioCtx, ahReqsResubmit, cReqsResubmit); AssertRC(rc2); cReqsResubmit = 0; } else if ( pEndpoint->pFlushReq && !pAioMgr->cRequestsActive && !pEndpoint->fAsyncFlushSupported) { /* * Complete a pending flush if we don't have requests enqueued and the host doesn't support * the async flush API. * Happens only if this we just noticed that this is not supported * and the only active request was a flush. */ PPDMACTASKFILE pFlush = pEndpoint->pFlushReq; pEndpoint->pFlushReq = NULL; pFlush->pfnCompleted(pFlush, pFlush->pvUser, VINF_SUCCESS); pdmacFileTaskFree(pEndpoint, pFlush); } } if (rc == VERR_FILE_AIO_INSUFFICIENT_RESSOURCES) { pAioMgr->cRequestsActiveMax = pAioMgr->cRequestsActive; /* Print an entry in the release log */ if (RT_UNLIKELY(!pEpClass->fOutOfResourcesWarningPrinted)) { pEpClass->fOutOfResourcesWarningPrinted = true; LogRel(("AIOMgr: Host limits number of active IO requests to %u. Expect a performance impact.\n", pAioMgr->cRequestsActive)); } } LogFlow(("Removed requests. I/O manager has a total of %u active requests now\n", pAioMgr->cRequestsActive)); LogFlow(("Endpoint has a total of %u active requests now\n", pEndpoint->AioMgr.cRequestsActive)); } return rc; } static bool pdmacFileAioMgrNormalIsRangeLocked(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, RTFOFF offStart, size_t cbRange, PPDMACTASKFILE pTask) { PPDMACFILERANGELOCK pRangeLock = NULL; /** < Range lock */ AssertMsg( pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE || pTask->enmTransferType == PDMACTASKFILETRANSFER_READ, ("Invalid task type %d\n", pTask->enmTransferType)); pRangeLock = (PPDMACFILERANGELOCK)RTAvlrFileOffsetRangeGet(pEndpoint->AioMgr.pTreeRangesLocked, offStart); if (!pRangeLock) { pRangeLock = (PPDMACFILERANGELOCK)RTAvlrFileOffsetGetBestFit(pEndpoint->AioMgr.pTreeRangesLocked, offStart, true); /* Check if we intersect with the range. */ if ( !pRangeLock || !( (pRangeLock->Core.Key) <= (offStart + (RTFOFF)cbRange - 1) && (pRangeLock->Core.KeyLast) >= offStart)) { pRangeLock = NULL; /* False alarm */ } } /* Check whether we have one of the situations explained below */ if ( pRangeLock #if 0 /** @todo: later. For now we will just block all requests if they interfere */ && ( (pRangeLock->fReadLock && pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE) || (!pRangeLock->fReadLock) #endif ) { /* Add to the list. */ pTask->pNext = NULL; if (!pRangeLock->pWaitingTasksHead) { Assert(!pRangeLock->pWaitingTasksTail); pRangeLock->pWaitingTasksHead = pTask; pRangeLock->pWaitingTasksTail = pTask; } else { AssertPtr(pRangeLock->pWaitingTasksTail); pRangeLock->pWaitingTasksTail->pNext = pTask; pRangeLock->pWaitingTasksTail = pTask; } return true; } return false; } static int pdmacFileAioMgrNormalRangeLock(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, RTFOFF offStart, size_t cbRange, PPDMACTASKFILE pTask) { AssertMsg(!pdmacFileAioMgrNormalIsRangeLocked(pEndpoint, offStart, cbRange, pTask), ("Range is already locked offStart=%RTfoff cbRange=%u\n", offStart, cbRange)); PPDMACFILERANGELOCK pRangeLock = (PPDMACFILERANGELOCK)RTMemCacheAlloc(pAioMgr->hMemCacheRangeLocks); if (!pRangeLock) return VERR_NO_MEMORY; /* Init the lock. */ pRangeLock->Core.Key = offStart; pRangeLock->Core.KeyLast = offStart + cbRange - 1; pRangeLock->cRefs = 1; pRangeLock->fReadLock = pTask->enmTransferType == PDMACTASKFILETRANSFER_READ; pRangeLock->pWaitingTasksHead = NULL; pRangeLock->pWaitingTasksTail = NULL; bool fInserted = RTAvlrFileOffsetInsert(pEndpoint->AioMgr.pTreeRangesLocked, &pRangeLock->Core); AssertMsg(fInserted, ("Range lock was not inserted!\n")); /* Let the task point to its lock. */ pTask->pRangeLock = pRangeLock; return VINF_SUCCESS; } static PPDMACTASKFILE pdmacFileAioMgrNormalRangeLockFree(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACFILERANGELOCK pRangeLock) { PPDMACTASKFILE pTasksWaitingHead; AssertPtr(pRangeLock); Assert(pRangeLock->cRefs == 1); RTAvlrFileOffsetRemove(pEndpoint->AioMgr.pTreeRangesLocked, pRangeLock->Core.Key); pTasksWaitingHead = pRangeLock->pWaitingTasksHead; pRangeLock->pWaitingTasksHead = NULL; pRangeLock->pWaitingTasksTail = NULL; RTMemCacheFree(pAioMgr->hMemCacheRangeLocks, pRangeLock); return pTasksWaitingHead; } static int pdmacFileAioMgrNormalTaskPrepareBuffered(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask, PRTFILEAIOREQ phReq) { int rc = VINF_SUCCESS; RTFILEAIOREQ hReq = NIL_RTFILEAIOREQ; PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass; void *pvBuf = pTask->DataSeg.pvSeg; AssertMsg( pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE || (uint64_t)(pTask->Off + pTask->DataSeg.cbSeg) <= pEndpoint->cbFile, ("Read exceeds file size offStart=%RTfoff cbToTransfer=%d cbFile=%llu\n", pTask->Off, pTask->DataSeg.cbSeg, pEndpoint->cbFile)); pTask->fPrefetch = false; pTask->cbBounceBuffer = 0; /* * Before we start to setup the request we have to check whether there is a task * already active which range intersects with ours. We have to defer execution * of this task in two cases: * - The pending task is a write and the current is either read or write * - The pending task is a read and the current task is a write task. * * To check whether a range is currently "locked" we use the AVL tree where every pending task * is stored by its file offset range. The current task will be added to the active task * and will be executed when the active one completes. (The method below * which checks whether a range is already used will add the task) * * This is neccessary because of the requirement to align all requests to a 512 boundary * which is enforced by the host OS (Linux and Windows atm). It is possible that * we have to process unaligned tasks and need to align them using bounce buffers. * While the data is fetched from the file another request might arrive writing to * the same range. This will result in data corruption if both are executed concurrently. */ bool fLocked = pdmacFileAioMgrNormalIsRangeLocked(pEndpoint, pTask->Off, pTask->DataSeg.cbSeg, pTask); if (!fLocked) { /* Get a request handle. */ hReq = pdmacFileAioMgrNormalRequestAlloc(pAioMgr); AssertMsg(hReq != NIL_RTFILEAIOREQ, ("Out of request handles\n")); if (pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE) { /* Grow the file if needed. */ if (RT_UNLIKELY((uint64_t)(pTask->Off + pTask->DataSeg.cbSeg) > pEndpoint->cbFile)) { ASMAtomicWriteU64(&pEndpoint->cbFile, pTask->Off + pTask->DataSeg.cbSeg); RTFileSetSize(pEndpoint->File, pTask->Off + pTask->DataSeg.cbSeg); } rc = RTFileAioReqPrepareWrite(hReq, pEndpoint->File, pTask->Off, pTask->DataSeg.pvSeg, pTask->DataSeg.cbSeg, pTask); } else rc = RTFileAioReqPrepareRead(hReq, pEndpoint->File, pTask->Off, pTask->DataSeg.pvSeg, pTask->DataSeg.cbSeg, pTask); AssertRC(rc); rc = pdmacFileAioMgrNormalRangeLock(pAioMgr, pEndpoint, pTask->Off, pTask->DataSeg.cbSeg, pTask); if (RT_SUCCESS(rc)) *phReq = hReq; } else LogFlow(("Task %#p was deferred because the access range is locked\n", pTask)); return rc; } static int pdmacFileAioMgrNormalTaskPrepareNonBuffered(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask, PRTFILEAIOREQ phReq) { int rc = VINF_SUCCESS; RTFILEAIOREQ hReq = NIL_RTFILEAIOREQ; PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass; void *pvBuf = pTask->DataSeg.pvSeg; /* * Check if the alignment requirements are met. * Offset, transfer size and buffer address * need to be on a 512 boundary. */ RTFOFF offStart = pTask->Off & ~(RTFOFF)(512-1); size_t cbToTransfer = RT_ALIGN_Z(pTask->DataSeg.cbSeg + (pTask->Off - offStart), 512); PDMACTASKFILETRANSFER enmTransferType = pTask->enmTransferType; AssertMsg( pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE || (uint64_t)(offStart + cbToTransfer) <= pEndpoint->cbFile, ("Read exceeds file size offStart=%RTfoff cbToTransfer=%d cbFile=%llu\n", offStart, cbToTransfer, pEndpoint->cbFile)); pTask->fPrefetch = false; /* * Before we start to setup the request we have to check whether there is a task * already active which range intersects with ours. We have to defer execution * of this task in two cases: * - The pending task is a write and the current is either read or write * - The pending task is a read and the current task is a write task. * * To check whether a range is currently "locked" we use the AVL tree where every pending task * is stored by its file offset range. The current task will be added to the active task * and will be executed when the active one completes. (The method below * which checks whether a range is already used will add the task) * * This is neccessary because of the requirement to align all requests to a 512 boundary * which is enforced by the host OS (Linux and Windows atm). It is possible that * we have to process unaligned tasks and need to align them using bounce buffers. * While the data is fetched from the file another request might arrive writing to * the same range. This will result in data corruption if both are executed concurrently. */ bool fLocked = pdmacFileAioMgrNormalIsRangeLocked(pEndpoint, offStart, cbToTransfer, pTask); if (!fLocked) { /* Get a request handle. */ hReq = pdmacFileAioMgrNormalRequestAlloc(pAioMgr); AssertMsg(hReq != NIL_RTFILEAIOREQ, ("Out of request handles\n")); if ( RT_UNLIKELY(cbToTransfer != pTask->DataSeg.cbSeg) || RT_UNLIKELY(offStart != pTask->Off) || ((pEpClassFile->uBitmaskAlignment & (RTR3UINTPTR)pvBuf) != (RTR3UINTPTR)pvBuf)) { LogFlow(("Using bounce buffer for task %#p cbToTransfer=%zd cbSeg=%zd offStart=%RTfoff off=%RTfoff\n", pTask, cbToTransfer, pTask->DataSeg.cbSeg, offStart, pTask->Off)); /* Create bounce buffer. */ pTask->cbBounceBuffer = cbToTransfer; AssertMsg(pTask->Off >= offStart, ("Overflow in calculation Off=%llu offStart=%llu\n", pTask->Off, offStart)); pTask->offBounceBuffer = pTask->Off - offStart; /** @todo: I think we need something like a RTMemAllocAligned method here. * Current assumption is that the maximum alignment is 4096byte * (GPT disk on Windows) * so we can use RTMemPageAlloc here. */ pTask->pvBounceBuffer = RTMemPageAlloc(cbToTransfer); if (RT_LIKELY(pTask->pvBounceBuffer)) { pvBuf = pTask->pvBounceBuffer; if (pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE) { if ( RT_UNLIKELY(cbToTransfer != pTask->DataSeg.cbSeg) || RT_UNLIKELY(offStart != pTask->Off)) { /* We have to fill the buffer first before we can update the data. */ LogFlow(("Prefetching data for task %#p\n", pTask)); pTask->fPrefetch = true; enmTransferType = PDMACTASKFILETRANSFER_READ; } else memcpy(pvBuf, pTask->DataSeg.pvSeg, pTask->DataSeg.cbSeg); } } else rc = VERR_NO_MEMORY; } else pTask->cbBounceBuffer = 0; if (RT_SUCCESS(rc)) { AssertMsg((pEpClassFile->uBitmaskAlignment & (RTR3UINTPTR)pvBuf) == (RTR3UINTPTR)pvBuf, ("AIO: Alignment restrictions not met! pvBuf=%p uBitmaskAlignment=%p\n", pvBuf, pEpClassFile->uBitmaskAlignment)); if (enmTransferType == PDMACTASKFILETRANSFER_WRITE) { /* Grow the file if needed. */ if (RT_UNLIKELY((uint64_t)(pTask->Off + pTask->DataSeg.cbSeg) > pEndpoint->cbFile)) { ASMAtomicWriteU64(&pEndpoint->cbFile, pTask->Off + pTask->DataSeg.cbSeg); RTFileSetSize(pEndpoint->File, pTask->Off + pTask->DataSeg.cbSeg); } rc = RTFileAioReqPrepareWrite(hReq, pEndpoint->File, offStart, pvBuf, cbToTransfer, pTask); } else rc = RTFileAioReqPrepareRead(hReq, pEndpoint->File, offStart, pvBuf, cbToTransfer, pTask); AssertRC(rc); rc = pdmacFileAioMgrNormalRangeLock(pAioMgr, pEndpoint, offStart, cbToTransfer, pTask); if (RT_SUCCESS(rc)) *phReq = hReq; else { /* Cleanup */ if (pTask->cbBounceBuffer) RTMemPageFree(pTask->pvBounceBuffer, pTask->cbBounceBuffer); } } } else LogFlow(("Task %#p was deferred because the access range is locked\n", pTask)); return rc; } static int pdmacFileAioMgrNormalProcessTaskList(PPDMACTASKFILE pTaskHead, PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint) { RTFILEAIOREQ apReqs[20]; unsigned cRequests = 0; unsigned cMaxRequests = pAioMgr->cRequestsActiveMax - pAioMgr->cRequestsActive; int rc = VINF_SUCCESS; AssertMsg(pEndpoint->enmState == PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE, ("Trying to process request lists of a non active endpoint!\n")); /* Go through the list and queue the requests until we get a flush request */ while ( pTaskHead && !pEndpoint->pFlushReq && (pAioMgr->cRequestsActive + cRequests < pAioMgr->cRequestsActiveMax) && RT_SUCCESS(rc)) { PPDMACTASKFILE pCurr = pTaskHead; if (!pdmacFileBwMgrIsTransferAllowed(pEndpoint->pBwMgr, (uint32_t)pCurr->DataSeg.cbSeg)) { pAioMgr->fBwLimitReached = true; break; } pTaskHead = pTaskHead->pNext; pCurr->pNext = NULL; AssertMsg(VALID_PTR(pCurr->pEndpoint) && (pCurr->pEndpoint == pEndpoint), ("Endpoints do not match\n")); switch (pCurr->enmTransferType) { case PDMACTASKFILETRANSFER_FLUSH: { /* If there is no data transfer request this flush request finished immediately. */ if (pEndpoint->fAsyncFlushSupported) { /* Issue a flush to the host. */ RTFILEAIOREQ hReq = pdmacFileAioMgrNormalRequestAlloc(pAioMgr); AssertMsg(hReq != NIL_RTFILEAIOREQ, ("Out of request handles\n")); LogFlow(("Flush request %#p\n", hReq)); rc = RTFileAioReqPrepareFlush(hReq, pEndpoint->File, pCurr); if (RT_FAILURE(rc)) { pEndpoint->fAsyncFlushSupported = false; pdmacFileAioMgrNormalRequestFree(pAioMgr, hReq); rc = VINF_SUCCESS; /* Fake success */ } else { apReqs[cRequests] = hReq; pEndpoint->AioMgr.cReqsProcessed++; cRequests++; } } if ( !pEndpoint->AioMgr.cRequestsActive && !pEndpoint->fAsyncFlushSupported) { pCurr->pfnCompleted(pCurr, pCurr->pvUser, VINF_SUCCESS); pdmacFileTaskFree(pEndpoint, pCurr); } else { Assert(!pEndpoint->pFlushReq); pEndpoint->pFlushReq = pCurr; } break; } case PDMACTASKFILETRANSFER_READ: case PDMACTASKFILETRANSFER_WRITE: { RTFILEAIOREQ hReq = NIL_RTFILEAIOREQ; if (pEndpoint->enmBackendType == PDMACFILEEPBACKEND_BUFFERED) rc = pdmacFileAioMgrNormalTaskPrepareBuffered(pAioMgr, pEndpoint, pCurr, &hReq); else if (pEndpoint->enmBackendType == PDMACFILEEPBACKEND_NON_BUFFERED) rc = pdmacFileAioMgrNormalTaskPrepareNonBuffered(pAioMgr, pEndpoint, pCurr, &hReq); else AssertMsgFailed(("Invalid backend type %d\n", pEndpoint->enmBackendType)); AssertRC(rc); LogFlow(("Read/Write request %#p\n", hReq)); if (hReq != NIL_RTFILEAIOREQ) { apReqs[cRequests] = hReq; pEndpoint->AioMgr.cReqsProcessed++; cRequests++; } break; } default: AssertMsgFailed(("Invalid transfer type %d\n", pCurr->enmTransferType)); } /* switch transfer type */ /* Queue the requests if the array is full. */ if (cRequests == RT_ELEMENTS(apReqs)) { rc = pdmacFileAioMgrNormalReqsEnqueue(pAioMgr, pEndpoint, apReqs, cRequests); cRequests = 0; AssertMsg(RT_SUCCESS(rc) || (rc == VERR_FILE_AIO_INSUFFICIENT_RESSOURCES), ("Unexpected return code\n")); } } if (cRequests) { rc = pdmacFileAioMgrNormalReqsEnqueue(pAioMgr, pEndpoint, apReqs, cRequests); AssertMsg(RT_SUCCESS(rc) || (rc == VERR_FILE_AIO_INSUFFICIENT_RESSOURCES), ("Unexpected return code rc=%Rrc\n", rc)); } if (pTaskHead) { /* Add the rest of the tasks to the pending list */ pdmacFileAioMgrEpAddTaskList(pEndpoint, pTaskHead); if (RT_UNLIKELY( pAioMgr->cRequestsActiveMax == pAioMgr->cRequestsActive && !pEndpoint->pFlushReq && !pAioMgr->fBwLimitReached)) { #if 0 /* * The I/O manager has no room left for more requests * but there are still requests to process. * Create a new I/O manager and let it handle some endpoints. */ pdmacFileAioMgrNormalBalanceLoad(pAioMgr); #else /* Grow the I/O manager */ pAioMgr->enmState = PDMACEPFILEMGRSTATE_GROWING; #endif } } /* Insufficient resources are not fatal. */ if (rc == VERR_FILE_AIO_INSUFFICIENT_RESSOURCES) rc = VINF_SUCCESS; return rc; } /** * Adds all pending requests for the given endpoint * until a flush request is encountered or there is no * request anymore. * * @returns VBox status code. * @param pAioMgr The async I/O manager for the endpoint * @param pEndpoint The endpoint to get the requests from. */ static int pdmacFileAioMgrNormalQueueReqs(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint) { int rc = VINF_SUCCESS; PPDMACTASKFILE pTasksHead = NULL; AssertMsg(pEndpoint->enmState == PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE, ("Trying to process request lists of a non active endpoint!\n")); Assert(!pEndpoint->pFlushReq); /* Check the pending list first */ if (pEndpoint->AioMgr.pReqsPendingHead) { LogFlow(("Queuing pending requests first\n")); pTasksHead = pEndpoint->AioMgr.pReqsPendingHead; /* * Clear the list as the processing routine will insert them into the list * again if it gets a flush request. */ pEndpoint->AioMgr.pReqsPendingHead = NULL; pEndpoint->AioMgr.pReqsPendingTail = NULL; rc = pdmacFileAioMgrNormalProcessTaskList(pTasksHead, pAioMgr, pEndpoint); AssertRC(rc); } if (!pEndpoint->pFlushReq && !pEndpoint->AioMgr.pReqsPendingHead) { /* Now the request queue. */ pTasksHead = pdmacFileEpGetNewTasks(pEndpoint); if (pTasksHead) { rc = pdmacFileAioMgrNormalProcessTaskList(pTasksHead, pAioMgr, pEndpoint); AssertRC(rc); } } return rc; } static int pdmacFileAioMgrNormalProcessBlockingEvent(PPDMACEPFILEMGR pAioMgr) { int rc = VINF_SUCCESS; bool fNotifyWaiter = false; LogFlowFunc((": Enter\n")); Assert(pAioMgr->fBlockingEventPending); switch (pAioMgr->enmBlockingEvent) { case PDMACEPFILEAIOMGRBLOCKINGEVENT_ADD_ENDPOINT: { PPDMASYNCCOMPLETIONENDPOINTFILE pEndpointNew = (PPDMASYNCCOMPLETIONENDPOINTFILE)ASMAtomicReadPtr((void * volatile *)&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint); AssertMsg(VALID_PTR(pEndpointNew), ("Adding endpoint event without a endpoint to add\n")); pEndpointNew->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE; pEndpointNew->AioMgr.pEndpointNext = pAioMgr->pEndpointsHead; pEndpointNew->AioMgr.pEndpointPrev = NULL; if (pAioMgr->pEndpointsHead) pAioMgr->pEndpointsHead->AioMgr.pEndpointPrev = pEndpointNew; pAioMgr->pEndpointsHead = pEndpointNew; /* Assign the completion point to this file. */ rc = RTFileAioCtxAssociateWithFile(pAioMgr->hAioCtx, pEndpointNew->File); fNotifyWaiter = true; pAioMgr->cEndpoints++; break; } case PDMACEPFILEAIOMGRBLOCKINGEVENT_REMOVE_ENDPOINT: { PPDMASYNCCOMPLETIONENDPOINTFILE pEndpointRemove = (PPDMASYNCCOMPLETIONENDPOINTFILE)ASMAtomicReadPtr((void * volatile *)&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint); AssertMsg(VALID_PTR(pEndpointRemove), ("Removing endpoint event without a endpoint to remove\n")); pEndpointRemove->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_REMOVING; fNotifyWaiter = !pdmacFileAioMgrNormalRemoveEndpoint(pEndpointRemove); break; } case PDMACEPFILEAIOMGRBLOCKINGEVENT_CLOSE_ENDPOINT: { PPDMASYNCCOMPLETIONENDPOINTFILE pEndpointClose = (PPDMASYNCCOMPLETIONENDPOINTFILE)ASMAtomicReadPtr((void * volatile *)&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint); AssertMsg(VALID_PTR(pEndpointClose), ("Close endpoint event without a endpoint to close\n")); if (pEndpointClose->enmState == PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE) { LogFlowFunc((": Closing endpoint %#p{%s}\n", pEndpointClose, pEndpointClose->Core.pszUri)); /* Make sure all tasks finished. Process the queues a last time first. */ rc = pdmacFileAioMgrNormalQueueReqs(pAioMgr, pEndpointClose); AssertRC(rc); pEndpointClose->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_CLOSING; fNotifyWaiter = !pdmacFileAioMgrNormalRemoveEndpoint(pEndpointClose); } else if ( (pEndpointClose->enmState == PDMASYNCCOMPLETIONENDPOINTFILESTATE_CLOSING) && (!pEndpointClose->AioMgr.cRequestsActive)) fNotifyWaiter = true; break; } case PDMACEPFILEAIOMGRBLOCKINGEVENT_SHUTDOWN: { pAioMgr->enmState = PDMACEPFILEMGRSTATE_SHUTDOWN; if (!pAioMgr->cRequestsActive) fNotifyWaiter = true; break; } case PDMACEPFILEAIOMGRBLOCKINGEVENT_SUSPEND: { pAioMgr->enmState = PDMACEPFILEMGRSTATE_SUSPENDING; break; } case PDMACEPFILEAIOMGRBLOCKINGEVENT_RESUME: { pAioMgr->enmState = PDMACEPFILEMGRSTATE_RUNNING; fNotifyWaiter = true; break; } default: AssertReleaseMsgFailed(("Invalid event type %d\n", pAioMgr->enmBlockingEvent)); } if (fNotifyWaiter) { ASMAtomicWriteBool(&pAioMgr->fBlockingEventPending, false); pAioMgr->enmBlockingEvent = PDMACEPFILEAIOMGRBLOCKINGEVENT_INVALID; /* Release the waiting thread. */ LogFlow(("Signalling waiter\n")); rc = RTSemEventSignal(pAioMgr->EventSemBlock); AssertRC(rc); } LogFlowFunc((": Leave\n")); return rc; } /** * Checks all endpoints for pending events or new requests. * * @returns VBox status code. * @param pAioMgr The I/O manager handle. */ static int pdmacFileAioMgrNormalCheckEndpoints(PPDMACEPFILEMGR pAioMgr) { /* Check the assigned endpoints for new tasks if there isn't a flush request active at the moment. */ int rc = VINF_SUCCESS; PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint = pAioMgr->pEndpointsHead; pAioMgr->fBwLimitReached = false; while (pEndpoint) { if (!pEndpoint->pFlushReq && (pEndpoint->enmState == PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE) && !pEndpoint->AioMgr.fMoving) { rc = pdmacFileAioMgrNormalQueueReqs(pAioMgr, pEndpoint); if (RT_FAILURE(rc)) return rc; } else if ( !pEndpoint->AioMgr.cRequestsActive && pEndpoint->enmState != PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE) { /* Reopen the file so that the new endpoint can reassociate with the file */ RTFileClose(pEndpoint->File); rc = RTFileOpen(&pEndpoint->File, pEndpoint->Core.pszUri, pEndpoint->fFlags); AssertRC(rc); if (pEndpoint->AioMgr.fMoving) { pEndpoint->AioMgr.fMoving = false; pdmacFileAioMgrAddEndpoint(pEndpoint->AioMgr.pAioMgrDst, pEndpoint); } else { Assert(pAioMgr->fBlockingEventPending); ASMAtomicWriteBool(&pAioMgr->fBlockingEventPending, false); /* Release the waiting thread. */ LogFlow(("Signalling waiter\n")); rc = RTSemEventSignal(pAioMgr->EventSemBlock); AssertRC(rc); } } pEndpoint = pEndpoint->AioMgr.pEndpointNext; } return rc; } static void pdmacFileAioMgrNormalReqComplete(PPDMACEPFILEMGR pAioMgr, RTFILEAIOREQ hReq) { int rc = VINF_SUCCESS; PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint; size_t cbTransfered = 0; int rcReq = RTFileAioReqGetRC(hReq, &cbTransfered); PPDMACTASKFILE pTask = (PPDMACTASKFILE)RTFileAioReqGetUser(hReq); PPDMACTASKFILE pTasksWaiting; pEndpoint = pTask->pEndpoint; /* * It is possible that the request failed on Linux with kernels < 2.6.23 * if the passed buffer was allocated with remap_pfn_range or if the file * is on an NFS endpoint which does not support async and direct I/O at the same time. * The endpoint will be migrated to a failsafe manager in case a request fails. */ if (RT_FAILURE(rcReq)) { /* Free bounce buffers and the IPRT request. */ pdmacFileAioMgrNormalRequestFree(pAioMgr, hReq); pAioMgr->cRequestsActive--; pEndpoint->AioMgr.cRequestsActive--; pEndpoint->AioMgr.cReqsProcessed++; if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH) { LogFlow(("Async flushes are not supported for this endpoint, disabling\n")); pEndpoint->fAsyncFlushSupported = false; AssertMsg(pEndpoint->pFlushReq == pTask, ("Failed flush request doesn't match active one\n")); /* The other method will take over now. */ } else { /* Free the lock and process pending tasks if neccessary */ pTasksWaiting = pdmacFileAioMgrNormalRangeLockFree(pAioMgr, pEndpoint, pTask->pRangeLock); rc = pdmacFileAioMgrNormalProcessTaskList(pTasksWaiting, pAioMgr, pEndpoint); AssertRC(rc); if (pTask->cbBounceBuffer) RTMemPageFree(pTask->pvBounceBuffer, pTask->cbBounceBuffer); /* * Fatal errors are reported to the guest and non-fatal errors * will cause a migration to the failsafe manager in the hope * that the error disappears. */ if (!pdmacFileAioMgrNormalRcIsFatal(rcReq)) { /* Queue the request on the pending list. */ pTask->pNext = pEndpoint->AioMgr.pReqsPendingHead; pEndpoint->AioMgr.pReqsPendingHead = pTask; /* Create a new failsafe manager if neccessary. */ if (!pEndpoint->AioMgr.fMoving) { PPDMACEPFILEMGR pAioMgrFailsafe; LogRel(("%s: Request %#p failed with rc=%Rrc, migrating endpoint %s to failsafe manager.\n", RTThreadGetName(pAioMgr->Thread), pTask, rcReq, pEndpoint->Core.pszUri)); pEndpoint->AioMgr.fMoving = true; rc = pdmacFileAioMgrCreate((PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass, &pAioMgrFailsafe, PDMACEPFILEMGRTYPE_SIMPLE); AssertRC(rc); pEndpoint->AioMgr.pAioMgrDst = pAioMgrFailsafe; /* Update the flags to open the file with. Disable async I/O and enable the host cache. */ pEndpoint->fFlags &= ~(RTFILE_O_ASYNC_IO | RTFILE_O_NO_CACHE); } /* If this was the last request for the endpoint migrate it to the new manager. */ if (!pEndpoint->AioMgr.cRequestsActive) { bool fReqsPending = pdmacFileAioMgrNormalRemoveEndpoint(pEndpoint); Assert(!fReqsPending); rc = pdmacFileAioMgrAddEndpoint(pEndpoint->AioMgr.pAioMgrDst, pEndpoint); AssertRC(rc); } } else { pTask->pfnCompleted(pTask, pTask->pvUser, rcReq); pdmacFileTaskFree(pEndpoint, pTask); } } } else { if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH) { /* Clear pending flush */ AssertMsg(pEndpoint->pFlushReq == pTask, ("Completed flush request doesn't match active one\n")); pEndpoint->pFlushReq = NULL; pdmacFileAioMgrNormalRequestFree(pAioMgr, hReq); pAioMgr->cRequestsActive--; pEndpoint->AioMgr.cRequestsActive--; pEndpoint->AioMgr.cReqsProcessed++; /* Call completion callback */ LogFlow(("Flush task=%#p completed with %Rrc\n", pTask, rcReq)); pTask->pfnCompleted(pTask, pTask->pvUser, rcReq); pdmacFileTaskFree(pEndpoint, pTask); } else { /* * Restart an incomplete transfer. * This usually means that the request will return an error now * but to get the cause of the error (disk full, file too big, I/O error, ...) * the transfer needs to be continued. */ if (RT_UNLIKELY( cbTransfered < pTask->DataSeg.cbSeg || ( pTask->cbBounceBuffer && cbTransfered < pTask->cbBounceBuffer))) { RTFOFF offStart; size_t cbToTransfer; uint8_t *pbBuf = NULL; LogFlow(("Restarting incomplete transfer %#p (%zu bytes transfered)\n", pTask, cbTransfered)); Assert(cbTransfered % 512 == 0); if (pTask->cbBounceBuffer) { AssertPtr(pTask->pvBounceBuffer); offStart = (pTask->Off & ~((RTFOFF)512-1)) + cbTransfered; cbToTransfer = pTask->cbBounceBuffer - cbTransfered; pbBuf = (uint8_t *)pTask->pvBounceBuffer + cbTransfered; } else { Assert(!pTask->pvBounceBuffer); offStart = pTask->Off + cbTransfered; cbToTransfer = pTask->DataSeg.cbSeg - cbTransfered; pbBuf = (uint8_t *)pTask->DataSeg.pvSeg + cbTransfered; } if (pTask->fPrefetch || pTask->enmTransferType == PDMACTASKFILETRANSFER_READ) { rc = RTFileAioReqPrepareRead(hReq, pEndpoint->File, offStart, pbBuf, cbToTransfer, pTask); } else { AssertMsg(pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE, ("Invalid transfer type\n")); rc = RTFileAioReqPrepareWrite(hReq, pEndpoint->File, offStart, pbBuf, cbToTransfer, pTask); } AssertRC(rc); rc = RTFileAioCtxSubmit(pAioMgr->hAioCtx, &hReq, 1); AssertRC(rc); } else if (pTask->fPrefetch) { Assert(pTask->enmTransferType == PDMACTASKFILETRANSFER_WRITE); Assert(pTask->cbBounceBuffer); memcpy(((uint8_t *)pTask->pvBounceBuffer) + pTask->offBounceBuffer, pTask->DataSeg.pvSeg, pTask->DataSeg.cbSeg); /* Write it now. */ pTask->fPrefetch = false; size_t cbToTransfer = RT_ALIGN_Z(pTask->DataSeg.cbSeg, 512); RTFOFF offStart = pTask->Off & ~(RTFOFF)(512-1); /* Grow the file if needed. */ if (RT_UNLIKELY((uint64_t)(pTask->Off + pTask->DataSeg.cbSeg) > pEndpoint->cbFile)) { ASMAtomicWriteU64(&pEndpoint->cbFile, pTask->Off + pTask->DataSeg.cbSeg); RTFileSetSize(pEndpoint->File, pTask->Off + pTask->DataSeg.cbSeg); } rc = RTFileAioReqPrepareWrite(hReq, pEndpoint->File, offStart, pTask->pvBounceBuffer, cbToTransfer, pTask); AssertRC(rc); rc = RTFileAioCtxSubmit(pAioMgr->hAioCtx, &hReq, 1); AssertRC(rc); } else { if (RT_SUCCESS(rc) && pTask->cbBounceBuffer) { if (pTask->enmTransferType == PDMACTASKFILETRANSFER_READ) memcpy(pTask->DataSeg.pvSeg, ((uint8_t *)pTask->pvBounceBuffer) + pTask->offBounceBuffer, pTask->DataSeg.cbSeg); RTMemPageFree(pTask->pvBounceBuffer, pTask->cbBounceBuffer); } pdmacFileAioMgrNormalRequestFree(pAioMgr, hReq); pAioMgr->cRequestsActive--; pEndpoint->AioMgr.cRequestsActive--; pEndpoint->AioMgr.cReqsProcessed++; /* Free the lock and process pending tasks if neccessary */ pTasksWaiting = pdmacFileAioMgrNormalRangeLockFree(pAioMgr, pEndpoint, pTask->pRangeLock); rc = pdmacFileAioMgrNormalProcessTaskList(pTasksWaiting, pAioMgr, pEndpoint); AssertRC(rc); /* Call completion callback */ LogFlow(("Task=%#p completed with %Rrc\n", pTask, rcReq)); pTask->pfnCompleted(pTask, pTask->pvUser, rcReq); pdmacFileTaskFree(pEndpoint, pTask); /* * If there is no request left on the endpoint but a flush request is set * it completed now and we notify the owner. * Furthermore we look for new requests and continue. */ if (!pEndpoint->AioMgr.cRequestsActive && pEndpoint->pFlushReq) { /* Call completion callback */ pTask = pEndpoint->pFlushReq; pEndpoint->pFlushReq = NULL; AssertMsg(pTask->pEndpoint == pEndpoint, ("Endpoint of the flush request does not match assigned one\n")); pTask->pfnCompleted(pTask, pTask->pvUser, VINF_SUCCESS); pdmacFileTaskFree(pEndpoint, pTask); } else if (RT_UNLIKELY(!pEndpoint->AioMgr.cRequestsActive && pEndpoint->AioMgr.fMoving)) { /* If the endpoint is about to be migrated do it now. */ bool fReqsPending = pdmacFileAioMgrNormalRemoveEndpoint(pEndpoint); Assert(!fReqsPending); rc = pdmacFileAioMgrAddEndpoint(pEndpoint->AioMgr.pAioMgrDst, pEndpoint); AssertRC(rc); } } } /* Not a flush request */ } /* request completed successfully */ } /** Helper macro for checking for error codes. */ #define CHECK_RC(pAioMgr, rc) \ if (RT_FAILURE(rc)) \ {\ int rc2 = pdmacFileAioMgrNormalErrorHandler(pAioMgr, rc, RT_SRC_POS);\ return rc2;\ } /** * The normal I/O manager using the RTFileAio* API * * @returns VBox status code. * @param ThreadSelf Handle of the thread. * @param pvUser Opaque user data. */ int pdmacFileAioMgrNormal(RTTHREAD ThreadSelf, void *pvUser) { int rc = VINF_SUCCESS; PPDMACEPFILEMGR pAioMgr = (PPDMACEPFILEMGR)pvUser; uint64_t uMillisEnd = RTTimeMilliTS() + PDMACEPFILEMGR_LOAD_UPDATE_PERIOD; while ( (pAioMgr->enmState == PDMACEPFILEMGRSTATE_RUNNING) || (pAioMgr->enmState == PDMACEPFILEMGRSTATE_SUSPENDING) || (pAioMgr->enmState == PDMACEPFILEMGRSTATE_GROWING)) { if (!pAioMgr->cRequestsActive) { ASMAtomicWriteBool(&pAioMgr->fWaitingEventSem, true); if (!ASMAtomicReadBool(&pAioMgr->fWokenUp)) rc = RTSemEventWait(pAioMgr->EventSem, RT_INDEFINITE_WAIT); ASMAtomicWriteBool(&pAioMgr->fWaitingEventSem, false); AssertRC(rc); LogFlow(("Got woken up\n")); ASMAtomicWriteBool(&pAioMgr->fWokenUp, false); } /* Check for an external blocking event first. */ if (pAioMgr->fBlockingEventPending) { rc = pdmacFileAioMgrNormalProcessBlockingEvent(pAioMgr); CHECK_RC(pAioMgr, rc); } if (RT_LIKELY( pAioMgr->enmState == PDMACEPFILEMGRSTATE_RUNNING || pAioMgr->enmState == PDMACEPFILEMGRSTATE_GROWING)) { /* We got woken up because an endpoint issued new requests. Queue them. */ rc = pdmacFileAioMgrNormalCheckEndpoints(pAioMgr); CHECK_RC(pAioMgr, rc); while ( pAioMgr->cRequestsActive || pAioMgr->fBwLimitReached) { if (pAioMgr->cRequestsActive) { RTFILEAIOREQ apReqs[20]; uint32_t cReqsCompleted = 0; size_t cReqsWait; if (pAioMgr->cRequestsActive > RT_ELEMENTS(apReqs)) cReqsWait = RT_ELEMENTS(apReqs); else cReqsWait = pAioMgr->cRequestsActive; LogFlow(("Waiting for %d of %d tasks to complete\n", 1, cReqsWait)); rc = RTFileAioCtxWait(pAioMgr->hAioCtx, 1, RT_INDEFINITE_WAIT, apReqs, cReqsWait, &cReqsCompleted); if (RT_FAILURE(rc) && (rc != VERR_INTERRUPTED)) CHECK_RC(pAioMgr, rc); LogFlow(("%d tasks completed\n", cReqsCompleted)); for (uint32_t i = 0; i < cReqsCompleted; i++) pdmacFileAioMgrNormalReqComplete(pAioMgr, apReqs[i]); /* Check for an external blocking event before we go to sleep again. */ if (pAioMgr->fBlockingEventPending) { rc = pdmacFileAioMgrNormalProcessBlockingEvent(pAioMgr); CHECK_RC(pAioMgr, rc); } /* Update load statistics. */ uint64_t uMillisCurr = RTTimeMilliTS(); if (uMillisCurr > uMillisEnd) { PPDMASYNCCOMPLETIONENDPOINTFILE pEndpointCurr = pAioMgr->pEndpointsHead; /* Calculate timespan. */ uMillisCurr -= uMillisEnd; while (pEndpointCurr) { pEndpointCurr->AioMgr.cReqsPerSec = pEndpointCurr->AioMgr.cReqsProcessed / (uMillisCurr + PDMACEPFILEMGR_LOAD_UPDATE_PERIOD); pEndpointCurr->AioMgr.cReqsProcessed = 0; pEndpointCurr = pEndpointCurr->AioMgr.pEndpointNext; } /* Set new update interval */ uMillisEnd = RTTimeMilliTS() + PDMACEPFILEMGR_LOAD_UPDATE_PERIOD; } } else { /* * Bandwidth limit reached for all endpoints. * Yield and wait until we have enough resources again. */ RTThreadYield(); } /* Check endpoints for new requests. */ if (pAioMgr->enmState != PDMACEPFILEMGRSTATE_GROWING) { rc = pdmacFileAioMgrNormalCheckEndpoints(pAioMgr); CHECK_RC(pAioMgr, rc); } } /* while requests are active. */ if (pAioMgr->enmState == PDMACEPFILEMGRSTATE_GROWING) { rc = pdmacFileAioMgrNormalGrow(pAioMgr); AssertRC(rc); Assert(pAioMgr->enmState == PDMACEPFILEMGRSTATE_RUNNING); rc = pdmacFileAioMgrNormalCheckEndpoints(pAioMgr); CHECK_RC(pAioMgr, rc); } } /* if still running */ } /* while running */ LogFlowFunc(("rc=%Rrc\n", rc)); return rc; } #undef CHECK_RC