PDMAsyncCompletionFile.cpp@ 53349

Last change on this file since 53349 was 46493, checked in by vboxsync, 12 years ago
STAMR3Deregister* optimizations. Relevant for both startup and shutdown times.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 45.8 KB

Line
1	/* $Id: PDMAsyncCompletionFile.cpp 46493 2013-06-11 13:34:40Z vboxsync $ */
2	/** @file
3	* PDM Async I/O - Transport data asynchronous in R3 using EMT.
4	*/
5
6	/*
7	* Copyright (C) 2006-2013 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*******************************************************************************
20	* Header Files *
21	*******************************************************************************/
22	#define LOG_GROUP LOG_GROUP_PDM_ASYNC_COMPLETION
23	#include "PDMInternal.h"
24	#include <VBox/vmm/pdm.h>
25	#include <VBox/vmm/mm.h>
26	#include <VBox/vmm/vm.h>
27	#include <VBox/err.h>
28	#include <VBox/log.h>
29	#include <VBox/dbg.h>
30	#include <VBox/vmm/uvm.h>
31	#include <VBox/vmm/tm.h>
32
33	#include <iprt/asm.h>
34	#include <iprt/assert.h>
35	#include <iprt/critsect.h>
36	#include <iprt/env.h>
37	#include <iprt/file.h>
38	#include <iprt/mem.h>
39	#include <iprt/semaphore.h>
40	#include <iprt/string.h>
41	#include <iprt/thread.h>
42	#include <iprt/path.h>
43	#include <iprt/rand.h>
44
45	#include "PDMAsyncCompletionFileInternal.h"
46
47	/*******************************************************************************
48	* Internal Functions *
49	*******************************************************************************/
50	#ifdef VBOX_WITH_DEBUGGER
51	static FNDBGCCMD pdmacEpFileErrorInject;
52	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
53	static FNDBGCCMD pdmacEpFileDelayInject;
54	# endif
55	#endif
56
57	/*******************************************************************************
58	* Global Variables *
59	*******************************************************************************/
60	#ifdef VBOX_WITH_DEBUGGER
61	static const DBGCVARDESC g_aInjectErrorArgs[] =
62	{
63	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
64	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write/read." },
65	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
66	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "errcode", "VBox status code." },
67	};
68
69	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
70	static const DBGCVARDESC g_aInjectDelayArgs[] =
71	{
72	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
73	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write\|read\|flush\|any." },
74	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
75	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "delay", "Delay in milliseconds." },
76	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "jitter", "Jitter of the delay." },
77	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "reqs", "Number of requests to delay." }
78
79	};
80	# endif
81
82	/** Command descriptors. */
83	static const DBGCCMD g_aCmds[] =
84	{
85	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax,.pszDescription */
86	{ "injecterror", 3, 3, &g_aInjectErrorArgs[0], 3, 0, pdmacEpFileErrorInject, "", "Inject error into I/O subsystem." }
87	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
88	,{ "injectdelay", 3, 5, &g_aInjectDelayArgs[0], RT_ELEMENTS(g_aInjectDelayArgs), 0, pdmacEpFileDelayInject, "", "Inject a delay of a request." }
89	# endif
90	};
91	#endif
92
93
94	/**
95	* Frees a task.
96	*
97	* @returns nothing.
98	* @param pEndpoint Pointer to the endpoint the segment was for.
99	* @param pTask The task to free.
100	*/
101	void pdmacFileTaskFree(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
102	{
103	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
104
105	LogFlowFunc((": pEndpoint=%p pTask=%p\n", pEndpoint, pTask));
106
107	/* Try the per endpoint cache first. */
108	if (pEndpoint->cTasksCached < pEpClass->cTasksCacheMax)
109	{
110	/* Add it to the list. */
111	pEndpoint->pTasksFreeTail->pNext = pTask;
112	pEndpoint->pTasksFreeTail = pTask;
113	ASMAtomicIncU32(&pEndpoint->cTasksCached);
114	}
115	else
116	{
117	Log(("Freeing task %p because all caches are full\n", pTask));
118	MMR3HeapFree(pTask);
119	}
120	}
121
122	/**
123	* Allocates a task segment
124	*
125	* @returns Pointer to the new task segment or NULL
126	* @param pEndpoint Pointer to the endpoint
127	*/
128	PPDMACTASKFILE pdmacFileTaskAlloc(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
129	{
130	PPDMACTASKFILE pTask = NULL;
131
132	/* Try the small per endpoint cache first. */
133	if (pEndpoint->pTasksFreeHead == pEndpoint->pTasksFreeTail)
134	{
135	/* Try the bigger endpoint class cache. */
136	PPDMASYNCCOMPLETIONEPCLASSFILE pEndpointClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
137
138	/*
139	* Allocate completely new.
140	* If this fails we return NULL.
141	*/
142	int rc = MMR3HeapAllocZEx(pEndpointClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
143	sizeof(PDMACTASKFILE),
144	(void **)&pTask);
145	if (RT_FAILURE(rc))
146	pTask = NULL;
147
148	LogFlow(("Allocated task %p\n", pTask));
149	}
150	else
151	{
152	/* Grab a free task from the head. */
153	AssertMsg(pEndpoint->cTasksCached > 0, ("No tasks cached but list contains more than one element\n"));
154
155	pTask = pEndpoint->pTasksFreeHead;
156	pEndpoint->pTasksFreeHead = pTask->pNext;
157	ASMAtomicDecU32(&pEndpoint->cTasksCached);
158	}
159
160	pTask->pNext = NULL;
161
162	return pTask;
163	}
164
165	PPDMACTASKFILE pdmacFileEpGetNewTasks(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
166	{
167	/*
168	* Get pending tasks.
169	*/
170	PPDMACTASKFILE pTasks = ASMAtomicXchgPtrT(&pEndpoint->pTasksNewHead, NULL, PPDMACTASKFILE);
171
172	/* Reverse the list to process in FIFO order. */
173	if (pTasks)
174	{
175	PPDMACTASKFILE pTask = pTasks;
176
177	pTasks = NULL;
178
179	while (pTask)
180	{
181	PPDMACTASKFILE pCur = pTask;
182	pTask = pTask->pNext;
183	pCur->pNext = pTasks;
184	pTasks = pCur;
185	}
186	}
187
188	return pTasks;
189	}
190
191	static void pdmacFileAioMgrWakeup(PPDMACEPFILEMGR pAioMgr)
192	{
193	bool fWokenUp = ASMAtomicXchgBool(&pAioMgr->fWokenUp, true);
194	if (!fWokenUp)
195	{
196	bool fWaitingEventSem = ASMAtomicReadBool(&pAioMgr->fWaitingEventSem);
197	if (fWaitingEventSem)
198	{
199	int rc = RTSemEventSignal(pAioMgr->EventSem);
200	AssertRC(rc);
201	}
202	}
203	}
204
205	static int pdmacFileAioMgrWaitForBlockingEvent(PPDMACEPFILEMGR pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT enmEvent)
206	{
207	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, enmEvent);
208	Assert(!pAioMgr->fBlockingEventPending);
209	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, true);
210
211	/* Wakeup the async I/O manager */
212	pdmacFileAioMgrWakeup(pAioMgr);
213
214	/* Wait for completion. */
215	int rc = RTSemEventWait(pAioMgr->EventSemBlock, RT_INDEFINITE_WAIT);
216	AssertRC(rc);
217
218	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, false);
219	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, PDMACEPFILEAIOMGRBLOCKINGEVENT_INVALID);
220
221	return rc;
222	}
223
224	int pdmacFileAioMgrAddEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
225	{
226	LogFlowFunc(("pAioMgr=%#p pEndpoint=%#p{%s}\n", pAioMgr, pEndpoint, pEndpoint->Core.pszUri));
227
228	/* Update the assigned I/O manager. */
229	ASMAtomicWritePtr(&pEndpoint->pAioMgr, pAioMgr);
230
231	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
232	AssertRCReturn(rc, rc);
233
234	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint, pEndpoint);
235	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_ADD_ENDPOINT);
236	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint);
237
238	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
239
240	return rc;
241	}
242
243	#ifdef SOME_UNUSED_FUNCTION
244	static int pdmacFileAioMgrRemoveEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
245	{
246	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
247	AssertRCReturn(rc, rc);
248
249	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint, pEndpoint);
250	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_REMOVE_ENDPOINT);
251	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint);
252
253	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
254
255	return rc;
256	}
257	#endif
258
259	static int pdmacFileAioMgrCloseEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
260	{
261	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
262	AssertRCReturn(rc, rc);
263
264	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint, pEndpoint);
265	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_CLOSE_ENDPOINT);
266	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint);
267
268	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
269
270	return rc;
271	}
272
273	static int pdmacFileAioMgrShutdown(PPDMACEPFILEMGR pAioMgr)
274	{
275	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
276	AssertRCReturn(rc, rc);
277
278	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_SHUTDOWN);
279
280	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
281
282	return rc;
283	}
284
285	int pdmacFileEpAddTask(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
286	{
287	PPDMACTASKFILE pNext;
288	do
289	{
290	pNext = pEndpoint->pTasksNewHead;
291	pTask->pNext = pNext;
292	} while (!ASMAtomicCmpXchgPtr(&pEndpoint->pTasksNewHead, pTask, pNext));
293
294	pdmacFileAioMgrWakeup(ASMAtomicReadPtrT(&pEndpoint->pAioMgr, PPDMACEPFILEMGR));
295
296	return VINF_SUCCESS;
297	}
298
299	void pdmacFileEpTaskCompleted(PPDMACTASKFILE pTask, void *pvUser, int rc)
300	{
301	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pvUser;
302
303	LogFlowFunc(("pTask=%#p pvUser=%#p rc=%Rrc\n", pTask, pvUser, rc));
304
305	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH)
306	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, rc, true);
307	else
308	{
309	Assert((uint32_t)pTask->DataSeg.cbSeg == pTask->DataSeg.cbSeg && (int32_t)pTask->DataSeg.cbSeg >= 0);
310	uint32_t uOld = ASMAtomicSubS32(&pTaskFile->cbTransferLeft, (int32_t)pTask->DataSeg.cbSeg);
311
312	/* The first error will be returned. */
313	if (RT_FAILURE(rc))
314	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
315	#ifdef VBOX_WITH_DEBUGGER
316	else
317	{
318	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
319
320	/* Overwrite with injected error code. */
321	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_READ)
322	rc = ASMAtomicXchgS32(&pEpFile->rcReqRead, VINF_SUCCESS);
323	else
324	rc = ASMAtomicXchgS32(&pEpFile->rcReqWrite, VINF_SUCCESS);
325
326	if (RT_FAILURE(rc))
327	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
328	}
329	#endif
330
331	if (!(uOld - pTask->DataSeg.cbSeg)
332	&& !ASMAtomicXchgBool(&pTaskFile->fCompleted, true))
333	{
334	#ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
335	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
336	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEpFile->Core.pEpClass;
337
338	/* Check if we should delay completion of the request. */
339	if ( ASMAtomicReadU32(&pEpFile->msDelay) > 0
340	&& ASMAtomicReadU32(&pEpFile->cReqsDelay) > 0)
341	{
342	uint64_t tsDelay = pEpFile->msDelay;
343
344	if (pEpFile->msJitter)
345	tsDelay = (RTRandU32() % 100) > 50 ? pEpFile->msDelay + (RTRandU32() % pEpFile->msJitter)
346	: pEpFile->msDelay - (RTRandU32() % pEpFile->msJitter);
347	ASMAtomicDecU32(&pEpFile->cReqsDelay);
348
349	/* Arm the delay. */
350	pTaskFile->tsDelayEnd = RTTimeProgramMilliTS() + tsDelay;
351
352	/* Append to the list. */
353	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
354	do
355	{
356	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
357	pTaskFile->pDelayedNext = pHead;
358	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTaskFile, pHead));
359
360	if (tsDelay < pEpClassFile->cMilliesNext)
361	{
362	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, tsDelay);
363	TMTimerSetMillies(pEpClassFile->pTimer, tsDelay);
364	}
365
366	LogRel(("AIOMgr: Delaying request %#p for %u ms\n", pTaskFile, tsDelay));
367	}
368	else
369	#endif
370	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, pTaskFile->rc, true);
371	}
372	}
373	}
374
375	DECLINLINE(void) pdmacFileEpTaskInit(PPDMASYNCCOMPLETIONTASK pTask, size_t cbTransfer)
376	{
377	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
378
379	Assert((uint32_t)cbTransfer == cbTransfer && (int32_t)cbTransfer >= 0);
380	ASMAtomicWriteS32(&pTaskFile->cbTransferLeft, (int32_t)cbTransfer);
381	ASMAtomicWriteBool(&pTaskFile->fCompleted, false);
382	ASMAtomicWriteS32(&pTaskFile->rc, VINF_SUCCESS);
383	}
384
385	int pdmacFileEpTaskInitiate(PPDMASYNCCOMPLETIONTASK pTask,
386	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
387	PCRTSGSEG paSegments, size_t cSegments,
388	size_t cbTransfer, PDMACTASKFILETRANSFER enmTransfer)
389	{
390	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
391	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
392
393	Assert( (enmTransfer == PDMACTASKFILETRANSFER_READ)
394	\|\| (enmTransfer == PDMACTASKFILETRANSFER_WRITE));
395
396	for (size_t i = 0; i < cSegments; i++)
397	{
398	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
399	AssertPtr(pIoTask);
400
401	pIoTask->pEndpoint = pEpFile;
402	pIoTask->enmTransferType = enmTransfer;
403	pIoTask->Off = off;
404	pIoTask->DataSeg.cbSeg = paSegments[i].cbSeg;
405	pIoTask->DataSeg.pvSeg = paSegments[i].pvSeg;
406	pIoTask->pvUser = pTaskFile;
407	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
408
409	/* Send it off to the I/O manager. */
410	pdmacFileEpAddTask(pEpFile, pIoTask);
411	off += paSegments[i].cbSeg;
412	cbTransfer -= paSegments[i].cbSeg;
413	}
414
415	AssertMsg(!cbTransfer, ("Incomplete transfer %u bytes left\n", cbTransfer));
416
417	return VINF_AIO_TASK_PENDING;
418	}
419
420	/**
421	* Creates a new async I/O manager.
422	*
423	* @returns VBox status code.
424	* @param pEpClass Pointer to the endpoint class data.
425	* @param ppAioMgr Where to store the pointer to the new async I/O manager on success.
426	* @param enmMgrType Wanted manager type - can be overwritten by the global override.
427	*/
428	int pdmacFileAioMgrCreate(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass, PPPDMACEPFILEMGR ppAioMgr,
429	PDMACEPFILEMGRTYPE enmMgrType)
430	{
431	LogFlowFunc((": Entered\n"));
432
433	PPDMACEPFILEMGR pAioMgrNew;
434	int rc = MMR3HeapAllocZEx(pEpClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION, sizeof(PDMACEPFILEMGR), (void **)&pAioMgrNew);
435	if (RT_SUCCESS(rc))
436	{
437	if (enmMgrType < pEpClass->enmMgrTypeOverride)
438	pAioMgrNew->enmMgrType = enmMgrType;
439	else
440	pAioMgrNew->enmMgrType = pEpClass->enmMgrTypeOverride;
441
442	pAioMgrNew->msBwLimitExpired = RT_INDEFINITE_WAIT;
443
444	rc = RTSemEventCreate(&pAioMgrNew->EventSem);
445	if (RT_SUCCESS(rc))
446	{
447	rc = RTSemEventCreate(&pAioMgrNew->EventSemBlock);
448	if (RT_SUCCESS(rc))
449	{
450	rc = RTCritSectInit(&pAioMgrNew->CritSectBlockingEvent);
451	if (RT_SUCCESS(rc))
452	{
453	/* Init the rest of the manager. */
454	if (pAioMgrNew->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
455	rc = pdmacFileAioMgrNormalInit(pAioMgrNew);
456
457	if (RT_SUCCESS(rc))
458	{
459	pAioMgrNew->enmState = PDMACEPFILEMGRSTATE_RUNNING;
460
461	rc = RTThreadCreateF(&pAioMgrNew->Thread,
462	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
463	? pdmacFileAioMgrFailsafe
464	: pdmacFileAioMgrNormal,
465	pAioMgrNew,
466	0,
467	RTTHREADTYPE_IO,
468	0,
469	"AioMgr%d-%s", pEpClass->cAioMgrs,
470	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
471	? "F"
472	: "N");
473	if (RT_SUCCESS(rc))
474	{
475	/* Link it into the list. */
476	RTCritSectEnter(&pEpClass->CritSect);
477	pAioMgrNew->pNext = pEpClass->pAioMgrHead;
478	if (pEpClass->pAioMgrHead)
479	pEpClass->pAioMgrHead->pPrev = pAioMgrNew;
480	pEpClass->pAioMgrHead = pAioMgrNew;
481	pEpClass->cAioMgrs++;
482	RTCritSectLeave(&pEpClass->CritSect);
483
484	*ppAioMgr = pAioMgrNew;
485
486	Log(("PDMAC: Successfully created new file AIO Mgr {%s}\n", RTThreadGetName(pAioMgrNew->Thread)));
487	return VINF_SUCCESS;
488	}
489	pdmacFileAioMgrNormalDestroy(pAioMgrNew);
490	}
491	RTCritSectDelete(&pAioMgrNew->CritSectBlockingEvent);
492	}
493	RTSemEventDestroy(pAioMgrNew->EventSem);
494	}
495	RTSemEventDestroy(pAioMgrNew->EventSemBlock);
496	}
497	MMR3HeapFree(pAioMgrNew);
498	}
499
500	LogFlowFunc((": Leave rc=%Rrc\n", rc));
501
502	return rc;
503	}
504
505	/**
506	* Destroys a async I/O manager.
507	*
508	* @returns nothing.
509	* @param pAioMgr The async I/O manager to destroy.
510	*/
511	static void pdmacFileAioMgrDestroy(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile, PPDMACEPFILEMGR pAioMgr)
512	{
513	int rc = pdmacFileAioMgrShutdown(pAioMgr);
514	AssertRC(rc);
515
516	/* Unlink from the list. */
517	rc = RTCritSectEnter(&pEpClassFile->CritSect);
518	AssertRC(rc);
519
520	PPDMACEPFILEMGR pPrev = pAioMgr->pPrev;
521	PPDMACEPFILEMGR pNext = pAioMgr->pNext;
522
523	if (pPrev)
524	pPrev->pNext = pNext;
525	else
526	pEpClassFile->pAioMgrHead = pNext;
527
528	if (pNext)
529	pNext->pPrev = pPrev;
530
531	pEpClassFile->cAioMgrs--;
532	rc = RTCritSectLeave(&pEpClassFile->CritSect);
533	AssertRC(rc);
534
535	/* Free the resources. */
536	RTCritSectDelete(&pAioMgr->CritSectBlockingEvent);
537	RTSemEventDestroy(pAioMgr->EventSem);
538	if (pAioMgr->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
539	pdmacFileAioMgrNormalDestroy(pAioMgr);
540
541	MMR3HeapFree(pAioMgr);
542	}
543
544	static int pdmacFileMgrTypeFromName(const char *pszVal, PPDMACEPFILEMGRTYPE penmMgrType)
545	{
546	int rc = VINF_SUCCESS;
547
548	if (!RTStrCmp(pszVal, "Simple"))
549	*penmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
550	else if (!RTStrCmp(pszVal, "Async"))
551	*penmMgrType = PDMACEPFILEMGRTYPE_ASYNC;
552	else
553	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
554
555	return rc;
556	}
557
558	static const char *pdmacFileMgrTypeToName(PDMACEPFILEMGRTYPE enmMgrType)
559	{
560	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
561	return "Simple";
562	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
563	return "Async";
564
565	return NULL;
566	}
567
568	static int pdmacFileBackendTypeFromName(const char *pszVal, PPDMACFILEEPBACKEND penmBackendType)
569	{
570	int rc = VINF_SUCCESS;
571
572	if (!RTStrCmp(pszVal, "Buffered"))
573	*penmBackendType = PDMACFILEEPBACKEND_BUFFERED;
574	else if (!RTStrCmp(pszVal, "NonBuffered"))
575	*penmBackendType = PDMACFILEEPBACKEND_NON_BUFFERED;
576	else
577	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
578
579	return rc;
580	}
581
582	static const char *pdmacFileBackendTypeToName(PDMACFILEEPBACKEND enmBackendType)
583	{
584	if (enmBackendType == PDMACFILEEPBACKEND_BUFFERED)
585	return "Buffered";
586	if (enmBackendType == PDMACFILEEPBACKEND_NON_BUFFERED)
587	return "NonBuffered";
588
589	return NULL;
590	}
591
592	/**
593	* Get the size of the given file.
594	* Works for block devices too.
595	*
596	* @returns VBox status code.
597	* @param hFile The file handle.
598	* @param pcbSize Where to store the size of the file on success.
599	*/
600	static int pdmacFileEpNativeGetSize(RTFILE hFile, uint64_t *pcbSize)
601	{
602	uint64_t cbFile;
603	int rc = RTFileGetSize(hFile, &cbFile);
604	if (RT_SUCCESS(rc))
605	*pcbSize = cbFile;
606
607	return rc;
608	}
609
610	#ifdef VBOX_WITH_DEBUGGER
611
612	/**
613	* @callback_method_impl{FNDBGCCMD, The '.injecterror' command.}
614	*/
615	static DECLCALLBACK(int) pdmacEpFileErrorInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
616	{
617	/*
618	* Validate input.
619	*/
620	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
621	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 3);
622	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
623	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
624	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
625
626	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
627	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
628
629	/* Syntax is "read\|write <filename> <status code>" */
630	bool fWrite;
631	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
632	fWrite = false;
633	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
634	fWrite = true;
635	else
636	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
637
638	int32_t rcToInject = (int32_t)pArgs[2].u.u64Number;
639	if ((uint64_t)rcToInject != pArgs[2].u.u64Number)
640	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The status code '%lld' is out of range", pArgs[0].u.u64Number);
641
642	/*
643	* Search for the matching endpoint.
644	*/
645	RTCritSectEnter(&pEpClassFile->Core.CritSect);
646
647	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
648	while (pEpFile)
649	{
650	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
651	break;
652	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
653	}
654
655	if (pEpFile)
656	{
657	/*
658	* Do the job.
659	*/
660	if (fWrite)
661	ASMAtomicXchgS32(&pEpFile->rcReqWrite, rcToInject);
662	else
663	ASMAtomicXchgS32(&pEpFile->rcReqRead, rcToInject);
664
665	DBGCCmdHlpPrintf(pCmdHlp, "Injected %Rrc into '%s' for %s\n",
666	(int)rcToInject, pArgs[1].u.pszString, pArgs[0].u.pszString);
667	}
668
669	RTCritSectLeave(&pEpClassFile->Core.CritSect);
670
671	if (!pEpFile)
672	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
673	return VINF_SUCCESS;
674	}
675
676	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
677	/**
678	* @callback_method_impl{FNDBGCCMD, The '.injectdelay' command.}
679	*/
680	static DECLCALLBACK(int) pdmacEpFileDelayInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
681	{
682	/*
683	* Validate input.
684	*/
685	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
686	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs >= 3);
687	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
688	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
689	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
690
691	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
692	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
693
694	/* Syntax is "read\|write\|flush\|any <filename> <delay> [reqs]" */
695	PDMACFILEREQTYPEDELAY enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
696	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
697	enmDelayType = PDMACFILEREQTYPEDELAY_READ;
698	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
699	enmDelayType = PDMACFILEREQTYPEDELAY_WRITE;
700	else if (!RTStrCmp(pArgs[0].u.pszString, "flush"))
701	enmDelayType = PDMACFILEREQTYPEDELAY_FLUSH;
702	else if (!RTStrCmp(pArgs[0].u.pszString, "any"))
703	enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
704	else
705	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
706
707	uint32_t msDelay = (uint32_t)pArgs[2].u.u64Number;
708	if ((uint64_t)msDelay != pArgs[2].u.u64Number)
709	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The delay '%lld' is out of range", pArgs[0].u.u64Number);
710
711	uint32_t cReqsDelay = 1;
712	uint32_t msJitter = 0;
713	if (cArgs >= 4)
714	msJitter = (uint32_t)pArgs[3].u.u64Number;
715	if (cArgs == 5)
716	cReqsDelay = (uint32_t)pArgs[4].u.u64Number;
717
718	/*
719	* Search for the matching endpoint.
720	*/
721	RTCritSectEnter(&pEpClassFile->Core.CritSect);
722
723	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
724	while (pEpFile)
725	{
726	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
727	break;
728	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
729	}
730
731	if (pEpFile)
732	{
733	ASMAtomicWriteSize(&pEpFile->enmTypeDelay, enmDelayType);
734	ASMAtomicWriteU32(&pEpFile->msDelay, msDelay);
735	ASMAtomicWriteU32(&pEpFile->msJitter, msJitter);
736	ASMAtomicWriteU32(&pEpFile->cReqsDelay, cReqsDelay);
737
738	DBGCCmdHlpPrintf(pCmdHlp, "Injected delay for the next %u requests of %u ms into '%s' for %s\n",
739	cReqsDelay, msDelay, pArgs[1].u.pszString, pArgs[0].u.pszString);
740	}
741
742	RTCritSectLeave(&pEpClassFile->Core.CritSect);
743
744	if (!pEpFile)
745	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
746	return VINF_SUCCESS;
747	}
748
749	static DECLCALLBACK(void) pdmacR3TimerCallback(PVM pVM, PTMTIMER pTimer, void *pvUser)
750	{
751	uint64_t tsCur = RTTimeProgramMilliTS();
752	uint64_t cMilliesNext = UINT64_MAX;
753	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pvUser;
754
755	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, UINT64_MAX);
756
757	/* Go through all endpoints and check for expired requests. */
758	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
759
760	while (pEpFile)
761	{
762	/* Check for an expired delay. */
763	if (pEpFile->pDelayedHead != NULL)
764	{
765	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = ASMAtomicXchgPtrT(&pEpFile->pDelayedHead, NULL, PPDMASYNCCOMPLETIONTASKFILE);
766
767	while (pTaskFile)
768	{
769	PPDMASYNCCOMPLETIONTASKFILE pTmp = pTaskFile;
770	pTaskFile = pTaskFile->pDelayedNext;
771
772	if (tsCur >= pTmp->tsDelayEnd)
773	{
774	LogRel(("AIOMgr: Delayed request %#p completed\n", pTmp));
775	pdmR3AsyncCompletionCompleteTask(&pTmp->Core, pTmp->rc, true);
776	}
777	else
778	{
779	/* Prepend to the delayed list again. */
780	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
781
782	if (pTmp->tsDelayEnd - tsCur < cMilliesNext)
783	cMilliesNext = pTmp->tsDelayEnd - tsCur;
784
785	do
786	{
787	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
788	pTmp->pDelayedNext = pHead;
789	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTmp, pHead));
790	}
791	}
792	}
793
794	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
795	}
796
797	if (cMilliesNext < pEpClassFile->cMilliesNext)
798	{
799	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, cMilliesNext);
800	TMTimerSetMillies(pEpClassFile->pTimer, cMilliesNext);
801	}
802	}
803
804	# endif /* PDM_ASYNC_COMPLETION_FILE_WITH_DELAY */
805
806	#endif /* VBOX_WITH_DEBUGGER */
807
808	static int pdmacFileInitialize(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals, PCFGMNODE pCfgNode)
809	{
810	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
811	RTFILEAIOLIMITS AioLimits; /** < Async I/O limitations. */
812
813	int rc = RTFileAioGetLimits(&AioLimits);
814	#ifdef DEBUG
815	if (RT_SUCCESS(rc) && RTEnvExist("VBOX_ASYNC_IO_FAILBACK"))
816	rc = VERR_ENV_VAR_NOT_FOUND;
817	#endif
818	if (RT_FAILURE(rc))
819	{
820	LogRel(("AIO: Async I/O manager not supported (rc=%Rrc). Falling back to simple manager\n",
821	rc));
822	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_SIMPLE;
823	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_BUFFERED;
824	}
825	else
826	{
827	pEpClassFile->uBitmaskAlignment = AioLimits.cbBufferAlignment ? ~((RTR3UINTPTR)AioLimits.cbBufferAlignment - 1) : RTR3UINTPTR_MAX;
828	pEpClassFile->cReqsOutstandingMax = AioLimits.cReqsOutstandingMax;
829
830	if (pCfgNode)
831	{
832	/* Query the default manager type */
833	char *pszVal = NULL;
834	rc = CFGMR3QueryStringAllocDef(pCfgNode, "IoMgr", &pszVal, "Async");
835	AssertLogRelRCReturn(rc, rc);
836
837	rc = pdmacFileMgrTypeFromName(pszVal, &pEpClassFile->enmMgrTypeOverride);
838	MMR3HeapFree(pszVal);
839	if (RT_FAILURE(rc))
840	return rc;
841
842	LogRel(("AIOMgr: Default manager type is \"%s\"\n", pdmacFileMgrTypeToName(pEpClassFile->enmMgrTypeOverride)));
843
844	/* Query default backend type */
845	rc = CFGMR3QueryStringAllocDef(pCfgNode, "FileBackend", &pszVal, "NonBuffered");
846	AssertLogRelRCReturn(rc, rc);
847
848	rc = pdmacFileBackendTypeFromName(pszVal, &pEpClassFile->enmEpBackendDefault);
849	MMR3HeapFree(pszVal);
850	if (RT_FAILURE(rc))
851	return rc;
852
853	LogRel(("AIOMgr: Default file backend is \"%s\"\n", pdmacFileBackendTypeToName(pEpClassFile->enmEpBackendDefault)));
854
855	#ifdef RT_OS_LINUX
856	if ( pEpClassFile->enmMgrTypeOverride == PDMACEPFILEMGRTYPE_ASYNC
857	&& pEpClassFile->enmEpBackendDefault == PDMACFILEEPBACKEND_BUFFERED)
858	{
859	LogRel(("AIOMgr: Linux does not support buffered async I/O, changing to non buffered\n"));
860	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
861	}
862	#endif
863	}
864	else
865	{
866	/* No configuration supplied, set defaults */
867	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
868	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_ASYNC;
869	}
870	}
871
872	/* Init critical section. */
873	rc = RTCritSectInit(&pEpClassFile->CritSect);
874
875	#ifdef VBOX_WITH_DEBUGGER
876	/* Install the error injection handler. */
877	if (RT_SUCCESS(rc))
878	{
879	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
880	AssertRC(rc);
881	}
882
883	#ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
884	rc = TMR3TimerCreateInternal(pEpClassFile->Core.pVM, TMCLOCK_REAL, pdmacR3TimerCallback, pEpClassFile, "AC Delay", &pEpClassFile->pTimer);
885	AssertRC(rc);
886	pEpClassFile->cMilliesNext = UINT64_MAX;
887	#endif
888	#endif
889
890	return rc;
891	}
892
893	static void pdmacFileTerminate(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals)
894	{
895	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
896
897	/* All endpoints should be closed at this point. */
898	AssertMsg(!pEpClassFile->Core.pEndpointsHead, ("There are still endpoints left\n"));
899
900	/* Destroy all left async I/O managers. */
901	while (pEpClassFile->pAioMgrHead)
902	pdmacFileAioMgrDestroy(pEpClassFile, pEpClassFile->pAioMgrHead);
903
904	RTCritSectDelete(&pEpClassFile->CritSect);
905	}
906
907	static int pdmacFileEpInitialize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint,
908	const char *pszUri, uint32_t fFlags)
909	{
910	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
911	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
912	PDMACEPFILEMGRTYPE enmMgrType = pEpClassFile->enmMgrTypeOverride;
913	PDMACFILEEPBACKEND enmEpBackend = pEpClassFile->enmEpBackendDefault;
914
915	AssertMsgReturn((fFlags & ~(PDMACEP_FILE_FLAGS_READ_ONLY \| PDMACEP_FILE_FLAGS_DONT_LOCK \| PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)) == 0,
916	("PDMAsyncCompletion: Invalid flag specified\n"), VERR_INVALID_PARAMETER);
917
918	unsigned fFileFlags = RTFILE_O_OPEN;
919
920	/*
921	* Revert to the simple manager and the buffered backend if
922	* the host cache should be enabled.
923	*/
924	if (fFlags & PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)
925	{
926	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
927	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
928	}
929
930	if (fFlags & PDMACEP_FILE_FLAGS_READ_ONLY)
931	fFileFlags \|= RTFILE_O_READ \| RTFILE_O_DENY_NONE;
932	else
933	{
934	fFileFlags \|= RTFILE_O_READWRITE;
935
936	/*
937	* Opened in read/write mode. Check whether the caller wants to
938	* avoid the lock. Return an error in case caching is enabled
939	* because this can lead to data corruption.
940	*/
941	if (fFlags & PDMACEP_FILE_FLAGS_DONT_LOCK)
942	fFileFlags \|= RTFILE_O_DENY_NONE;
943	else
944	fFileFlags \|= RTFILE_O_DENY_WRITE;
945	}
946
947	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
948	fFileFlags \|= RTFILE_O_ASYNC_IO;
949
950	int rc;
951	if (enmEpBackend == PDMACFILEEPBACKEND_NON_BUFFERED)
952	{
953	/*
954	* We only disable the cache if the size of the file is a multiple of 512.
955	* Certain hosts like Windows, Linux and Solaris require that transfer sizes
956	* are aligned to the volume sector size.
957	* If not we just make sure that the data is written to disk with RTFILE_O_WRITE_THROUGH
958	* which will trash the host cache but ensures that the host cache will not
959	* contain dirty buffers.
960	*/
961	RTFILE hFile;
962	rc = RTFileOpen(&hFile, pszUri, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
963	if (RT_SUCCESS(rc))
964	{
965	uint64_t cbSize;
966
967	rc = pdmacFileEpNativeGetSize(hFile, &cbSize);
968
969	if (RT_SUCCESS(rc) && ((cbSize % 512) == 0))
970	fFileFlags \|= RTFILE_O_NO_CACHE;
971	else
972	{
973	/* Downgrade to the buffered backend */
974	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
975
976	#ifdef RT_OS_LINUX
977	fFileFlags &= ~RTFILE_O_ASYNC_IO;
978	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
979	#endif
980	}
981	RTFileClose(hFile);
982	}
983	}
984
985	/* Open with final flags. */
986	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
987	if ( rc == VERR_INVALID_FUNCTION
988	\|\| rc == VERR_INVALID_PARAMETER)
989	{
990	LogRel(("pdmacFileEpInitialize: RTFileOpen %s / %08x failed with %Rrc\n",
991	pszUri, fFileFlags, rc));
992	/*
993	* Solaris doesn't support directio on ZFS so far. :-\
994	* Trying to enable it returns VERR_INVALID_FUNCTION
995	* (ENOTTY). Remove it and hope for the best.
996	* ZFS supports write throttling in case applications
997	* write more data than can be synced to the disk
998	* without blocking the whole application.
999	*
1000	* On Linux we have the same problem with cifs.
1001	* Have to disable async I/O here too because it requires O_DIRECT.
1002	*/
1003	fFileFlags &= ~RTFILE_O_NO_CACHE;
1004	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
1005
1006	#ifdef RT_OS_LINUX
1007	fFileFlags &= ~RTFILE_O_ASYNC_IO;
1008	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
1009	#endif
1010
1011	/* Open again. */
1012	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
1013
1014	if (RT_FAILURE(rc))
1015	{
1016	LogRel(("pdmacFileEpInitialize: RTFileOpen %s / %08x failed AGAIN(!) with %Rrc\n",
1017	pszUri, fFileFlags, rc));
1018	}
1019	}
1020
1021	if (RT_SUCCESS(rc))
1022	{
1023	pEpFile->fFlags = fFileFlags;
1024
1025	rc = pdmacFileEpNativeGetSize(pEpFile->hFile, (uint64_t *)&pEpFile->cbFile);
1026	if (RT_SUCCESS(rc))
1027	{
1028	/* Initialize the segment cache */
1029	rc = MMR3HeapAllocZEx(pEpClassFile->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
1030	sizeof(PDMACTASKFILE),
1031	(void **)&pEpFile->pTasksFreeHead);
1032	if (RT_SUCCESS(rc))
1033	{
1034	PPDMACEPFILEMGR pAioMgr = NULL;
1035
1036	pEpFile->pTasksFreeTail = pEpFile->pTasksFreeHead;
1037	pEpFile->cTasksCached = 0;
1038	pEpFile->enmBackendType = enmEpBackend;
1039	/*
1040	* Disable async flushes on Solaris for now.
1041	* They cause weird hangs which needs more investigations.
1042	*/
1043	#ifndef RT_OS_SOLARIS
1044	pEpFile->fAsyncFlushSupported = true;
1045	#else
1046	pEpFile->fAsyncFlushSupported = false;
1047	#endif
1048
1049	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1050	{
1051	/* Simple mode. Every file has its own async I/O manager. */
1052	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, PDMACEPFILEMGRTYPE_SIMPLE);
1053	AssertRC(rc);
1054	}
1055	else
1056	{
1057	pAioMgr = pEpClassFile->pAioMgrHead;
1058
1059	/* Check for an idling manager of the same type */
1060	while (pAioMgr)
1061	{
1062	if (pAioMgr->enmMgrType == enmMgrType)
1063	break;
1064	pAioMgr = pAioMgr->pNext;
1065	}
1066
1067	if (!pAioMgr)
1068	{
1069	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, enmMgrType);
1070	AssertRC(rc);
1071	}
1072	}
1073
1074	pEpFile->AioMgr.pTreeRangesLocked = (PAVLRFOFFTREE)RTMemAllocZ(sizeof(AVLRFOFFTREE));
1075	if (!pEpFile->AioMgr.pTreeRangesLocked)
1076	rc = VERR_NO_MEMORY;
1077	else
1078	{
1079	pEpFile->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE;
1080
1081	/* Assign the endpoint to the thread. */
1082	rc = pdmacFileAioMgrAddEndpoint(pAioMgr, pEpFile);
1083	if (RT_FAILURE(rc))
1084	{
1085	RTMemFree(pEpFile->AioMgr.pTreeRangesLocked);
1086	MMR3HeapFree(pEpFile->pTasksFreeHead);
1087	}
1088	}
1089	}
1090	}
1091
1092	if (RT_FAILURE(rc))
1093	RTFileClose(pEpFile->hFile);
1094	}
1095
1096	#ifdef VBOX_WITH_STATISTICS
1097	if (RT_SUCCESS(rc))
1098	{
1099	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatRead,
1100	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1101	STAMUNIT_TICKS_PER_CALL, "Time taken to read from the endpoint",
1102	"/PDM/AsyncCompletion/File/%s/Read", RTPathFilename(pEpFile->Core.pszUri));
1103
1104	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatWrite,
1105	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1106	STAMUNIT_TICKS_PER_CALL, "Time taken to write to the endpoint",
1107	"/PDM/AsyncCompletion/File/%s/Write", RTPathFilename(pEpFile->Core.pszUri));
1108	}
1109	#endif
1110
1111	if (RT_SUCCESS(rc))
1112	LogRel(("AIOMgr: Endpoint for file '%s' (flags %08x) created successfully\n", pszUri, pEpFile->fFlags));
1113
1114	return rc;
1115	}
1116
1117	static int pdmacFileEpRangesLockedDestroy(PAVLRFOFFNODECORE pNode, void *pvUser)
1118	{
1119	NOREF(pNode); NOREF(pvUser);
1120	AssertMsgFailed(("The locked ranges tree should be empty at that point\n"));
1121	return VINF_SUCCESS;
1122	}
1123
1124	static int pdmacFileEpClose(PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1125	{
1126	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1127	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
1128
1129	/* Make sure that all tasks finished for this endpoint. */
1130	int rc = pdmacFileAioMgrCloseEndpoint(pEpFile->pAioMgr, pEpFile);
1131	AssertRC(rc);
1132
1133	/*
1134	* If the async I/O manager is in failsafe mode this is the only endpoint
1135	* he processes and thus can be destroyed now.
1136	*/
1137	if (pEpFile->pAioMgr->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1138	pdmacFileAioMgrDestroy(pEpClassFile, pEpFile->pAioMgr);
1139
1140	/* Free cached tasks. */
1141	PPDMACTASKFILE pTask = pEpFile->pTasksFreeHead;
1142
1143	while (pTask)
1144	{
1145	PPDMACTASKFILE pTaskFree = pTask;
1146	pTask = pTask->pNext;
1147	MMR3HeapFree(pTaskFree);
1148	}
1149
1150	/* Destroy the locked ranges tree now. */
1151	RTAvlrFileOffsetDestroy(pEpFile->AioMgr.pTreeRangesLocked, pdmacFileEpRangesLockedDestroy, NULL);
1152
1153	RTFileClose(pEpFile->hFile);
1154
1155	#ifdef VBOX_WITH_STATISTICS
1156	/* Not sure if this might be unnecessary because of similar statement in pdmR3AsyncCompletionStatisticsDeregister? */
1157	STAMR3DeregisterF(pEpClassFile->Core.pVM->pUVM, "/PDM/AsyncCompletion/File/%s/*", RTPathFilename(pEpFile->Core.pszUri));
1158	#endif
1159
1160	return VINF_SUCCESS;
1161	}
1162
1163	static int pdmacFileEpRead(PPDMASYNCCOMPLETIONTASK pTask,
1164	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1165	PCRTSGSEG paSegments, size_t cSegments,
1166	size_t cbRead)
1167	{
1168	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1169
1170	LogFlowFunc(("pTask=%#p pEndpoint=%#p off=%RTfoff paSegments=%#p cSegments=%zu cbRead=%zu\n",
1171	pTask, pEndpoint, off, paSegments, cSegments, cbRead));
1172
1173	if (RT_UNLIKELY((uint64_t)off + cbRead > pEpFile->cbFile))
1174	return VERR_EOF;
1175
1176	STAM_PROFILE_ADV_START(&pEpFile->StatRead, Read);
1177	pdmacFileEpTaskInit(pTask, cbRead);
1178	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbRead,
1179	PDMACTASKFILETRANSFER_READ);
1180	STAM_PROFILE_ADV_STOP(&pEpFile->StatRead, Read);
1181
1182	return rc;
1183	}
1184
1185	static int pdmacFileEpWrite(PPDMASYNCCOMPLETIONTASK pTask,
1186	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1187	PCRTSGSEG paSegments, size_t cSegments,
1188	size_t cbWrite)
1189	{
1190	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1191
1192	if (RT_UNLIKELY(pEpFile->fReadonly))
1193	return VERR_NOT_SUPPORTED;
1194
1195	STAM_PROFILE_ADV_START(&pEpFile->StatWrite, Write);
1196
1197	pdmacFileEpTaskInit(pTask, cbWrite);
1198
1199	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbWrite,
1200	PDMACTASKFILETRANSFER_WRITE);
1201
1202	STAM_PROFILE_ADV_STOP(&pEpFile->StatWrite, Write);
1203
1204	return rc;
1205	}
1206
1207	static int pdmacFileEpFlush(PPDMASYNCCOMPLETIONTASK pTask,
1208	PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1209	{
1210	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1211	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
1212
1213	if (RT_UNLIKELY(pEpFile->fReadonly))
1214	return VERR_NOT_SUPPORTED;
1215
1216	pdmacFileEpTaskInit(pTask, 0);
1217
1218	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
1219	if (RT_UNLIKELY(!pIoTask))
1220	return VERR_NO_MEMORY;
1221
1222	pIoTask->pEndpoint = pEpFile;
1223	pIoTask->enmTransferType = PDMACTASKFILETRANSFER_FLUSH;
1224	pIoTask->pvUser = pTaskFile;
1225	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
1226	pdmacFileEpAddTask(pEpFile, pIoTask);
1227
1228	return VINF_AIO_TASK_PENDING;
1229	}
1230
1231	static int pdmacFileEpGetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t *pcbSize)
1232	{
1233	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1234
1235	*pcbSize = ASMAtomicReadU64(&pEpFile->cbFile);
1236
1237	return VINF_SUCCESS;
1238	}
1239
1240	static int pdmacFileEpSetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t cbSize)
1241	{
1242	int rc;
1243	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1244
1245	rc = RTFileSetSize(pEpFile->hFile, cbSize);
1246	if (RT_SUCCESS(rc))
1247	ASMAtomicWriteU64(&pEpFile->cbFile, cbSize);
1248
1249	return rc;
1250	}
1251
1252	const PDMASYNCCOMPLETIONEPCLASSOPS g_PDMAsyncCompletionEndpointClassFile =
1253	{
1254	/* u32Version */
1255	PDMAC_EPCLASS_OPS_VERSION,
1256	/* pcszName */
1257	"File",
1258	/* enmClassType */
1259	PDMASYNCCOMPLETIONEPCLASSTYPE_FILE,
1260	/* cbEndpointClassGlobal */
1261	sizeof(PDMASYNCCOMPLETIONEPCLASSFILE),
1262	/* cbEndpoint */
1263	sizeof(PDMASYNCCOMPLETIONENDPOINTFILE),
1264	/* cbTask */
1265	sizeof(PDMASYNCCOMPLETIONTASKFILE),
1266	/* pfnInitialize */
1267	pdmacFileInitialize,
1268	/* pfnTerminate */
1269	pdmacFileTerminate,
1270	/* pfnEpInitialize. */
1271	pdmacFileEpInitialize,
1272	/* pfnEpClose */
1273	pdmacFileEpClose,
1274	/* pfnEpRead */
1275	pdmacFileEpRead,
1276	/* pfnEpWrite */
1277	pdmacFileEpWrite,
1278	/* pfnEpFlush */
1279	pdmacFileEpFlush,
1280	/* pfnEpGetSize */
1281	pdmacFileEpGetSize,
1282	/* pfnEpSetSize */
1283	pdmacFileEpSetSize,
1284	/* u32VersionEnd */
1285	PDMAC_EPCLASS_OPS_VERSION
1286	};
1287

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/VMM/VMMR3/PDMAsyncCompletionFile.cpp@ 53349

Download in other formats: