PDMAsyncCompletionFile.cpp@ 87758

Last change on this file since 87758 was 82968, checked in by vboxsync, 5 years ago
Copyright year updates by scm.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 47.4 KB

Line
1	/* $Id: PDMAsyncCompletionFile.cpp 82968 2020-02-04 10:35:17Z vboxsync $ */
2	/** @file
3	* PDM Async I/O - Transport data asynchronous in R3 using EMT.
4	*/
5
6	/*
7	* Copyright (C) 2006-2020 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	/*********************************************************************************************************************************
49	* Internal Functions *
50	*********************************************************************************************************************************/
51	#ifdef VBOX_WITH_DEBUGGER
52	static FNDBGCCMD pdmacEpFileErrorInject;
53	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
54	static FNDBGCCMD pdmacEpFileDelayInject;
55	# endif
56	#endif
57
58
59	/*********************************************************************************************************************************
60	* Global Variables *
61	*********************************************************************************************************************************/
62	#ifdef VBOX_WITH_DEBUGGER
63	static const DBGCVARDESC g_aInjectErrorArgs[] =
64	{
65	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
66	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write/read." },
67	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
68	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "errcode", "VBox status code." },
69	};
70
71	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
72	static const DBGCVARDESC g_aInjectDelayArgs[] =
73	{
74	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
75	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write\|read\|flush\|any." },
76	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
77	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "delay", "Delay in milliseconds." },
78	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "jitter", "Jitter of the delay." },
79	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "reqs", "Number of requests to delay." }
80
81	};
82	# endif
83
84	/** Command descriptors. */
85	static const DBGCCMD g_aCmds[] =
86	{
87	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax,.pszDescription */
88	{ "injecterror", 3, 3, &g_aInjectErrorArgs[0], 3, 0, pdmacEpFileErrorInject, "", "Inject error into I/O subsystem." }
89	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
90	,{ "injectdelay", 3, 5, &g_aInjectDelayArgs[0], RT_ELEMENTS(g_aInjectDelayArgs), 0, pdmacEpFileDelayInject, "", "Inject a delay of a request." }
91	# endif
92	};
93	#endif
94
95
96	/**
97	* Frees a task.
98	*
99	* @returns nothing.
100	* @param pEndpoint Pointer to the endpoint the segment was for.
101	* @param pTask The task to free.
102	*/
103	void pdmacFileTaskFree(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
104	{
105	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
106
107	LogFlowFunc((": pEndpoint=%p pTask=%p\n", pEndpoint, pTask));
108
109	/* Try the per endpoint cache first. */
110	if (pEndpoint->cTasksCached < pEpClass->cTasksCacheMax)
111	{
112	/* Add it to the list. */
113	pEndpoint->pTasksFreeTail->pNext = pTask;
114	pEndpoint->pTasksFreeTail = pTask;
115	ASMAtomicIncU32(&pEndpoint->cTasksCached);
116	}
117	else
118	{
119	Log(("Freeing task %p because all caches are full\n", pTask));
120	MMR3HeapFree(pTask);
121	}
122	}
123
124	/**
125	* Allocates a task segment
126	*
127	* @returns Pointer to the new task segment or NULL
128	* @param pEndpoint Pointer to the endpoint
129	*/
130	PPDMACTASKFILE pdmacFileTaskAlloc(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
131	{
132	PPDMACTASKFILE pTask = NULL;
133
134	/* Try the small per endpoint cache first. */
135	if (pEndpoint->pTasksFreeHead == pEndpoint->pTasksFreeTail)
136	{
137	/* Try the bigger endpoint class cache. */
138	PPDMASYNCCOMPLETIONEPCLASSFILE pEndpointClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
139
140	/*
141	* Allocate completely new.
142	* If this fails we return NULL.
143	*/
144	int rc = MMR3HeapAllocZEx(pEndpointClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
145	sizeof(PDMACTASKFILE),
146	(void **)&pTask);
147	if (RT_FAILURE(rc))
148	pTask = NULL;
149
150	LogFlow(("Allocated task %p\n", pTask));
151	}
152	else
153	{
154	/* Grab a free task from the head. */
155	AssertMsg(pEndpoint->cTasksCached > 0, ("No tasks cached but list contains more than one element\n"));
156
157	pTask = pEndpoint->pTasksFreeHead;
158	pEndpoint->pTasksFreeHead = pTask->pNext;
159	ASMAtomicDecU32(&pEndpoint->cTasksCached);
160	}
161
162	pTask->pNext = NULL;
163
164	return pTask;
165	}
166
167	PPDMACTASKFILE pdmacFileEpGetNewTasks(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
168	{
169	/*
170	* Get pending tasks.
171	*/
172	PPDMACTASKFILE pTasks = ASMAtomicXchgPtrT(&pEndpoint->pTasksNewHead, NULL, PPDMACTASKFILE);
173
174	/* Reverse the list to process in FIFO order. */
175	if (pTasks)
176	{
177	PPDMACTASKFILE pTask = pTasks;
178
179	pTasks = NULL;
180
181	while (pTask)
182	{
183	PPDMACTASKFILE pCur = pTask;
184	pTask = pTask->pNext;
185	pCur->pNext = pTasks;
186	pTasks = pCur;
187	}
188	}
189
190	return pTasks;
191	}
192
193	static void pdmacFileAioMgrWakeup(PPDMACEPFILEMGR pAioMgr)
194	{
195	bool fWokenUp = ASMAtomicXchgBool(&pAioMgr->fWokenUp, true);
196	if (!fWokenUp)
197	{
198	bool fWaitingEventSem = ASMAtomicReadBool(&pAioMgr->fWaitingEventSem);
199	if (fWaitingEventSem)
200	{
201	int rc = RTSemEventSignal(pAioMgr->EventSem);
202	AssertRC(rc);
203	}
204	}
205	}
206
207	static int pdmacFileAioMgrWaitForBlockingEvent(PPDMACEPFILEMGR pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT enmEvent)
208	{
209	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, enmEvent);
210	Assert(!pAioMgr->fBlockingEventPending);
211	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, true);
212
213	/* Wakeup the async I/O manager */
214	pdmacFileAioMgrWakeup(pAioMgr);
215
216	/* Wait for completion. */
217	int rc = RTSemEventWait(pAioMgr->EventSemBlock, RT_INDEFINITE_WAIT);
218	AssertRC(rc);
219
220	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, false);
221	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, PDMACEPFILEAIOMGRBLOCKINGEVENT_INVALID);
222
223	return rc;
224	}
225
226	int pdmacFileAioMgrAddEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
227	{
228	LogFlowFunc(("pAioMgr=%#p pEndpoint=%#p{%s}\n", pAioMgr, pEndpoint, pEndpoint->Core.pszUri));
229
230	/* Update the assigned I/O manager. */
231	ASMAtomicWritePtr(&pEndpoint->pAioMgr, pAioMgr);
232
233	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
234	AssertRCReturn(rc, rc);
235
236	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint, pEndpoint);
237	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_ADD_ENDPOINT);
238	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint);
239
240	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
241
242	return rc;
243	}
244
245	#ifdef SOME_UNUSED_FUNCTION
246	static int pdmacFileAioMgrRemoveEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
247	{
248	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
249	AssertRCReturn(rc, rc);
250
251	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint, pEndpoint);
252	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_REMOVE_ENDPOINT);
253	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint);
254
255	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
256
257	return rc;
258	}
259	#endif
260
261	static int pdmacFileAioMgrCloseEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
262	{
263	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
264	AssertRCReturn(rc, rc);
265
266	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint, pEndpoint);
267	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_CLOSE_ENDPOINT);
268	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint);
269
270	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
271
272	return rc;
273	}
274
275	static int pdmacFileAioMgrShutdown(PPDMACEPFILEMGR pAioMgr)
276	{
277	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
278	AssertRCReturn(rc, rc);
279
280	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_SHUTDOWN);
281
282	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
283
284	return rc;
285	}
286
287	int pdmacFileEpAddTask(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
288	{
289	PPDMACTASKFILE pNext;
290	do
291	{
292	pNext = pEndpoint->pTasksNewHead;
293	pTask->pNext = pNext;
294	} while (!ASMAtomicCmpXchgPtr(&pEndpoint->pTasksNewHead, pTask, pNext));
295
296	pdmacFileAioMgrWakeup(ASMAtomicReadPtrT(&pEndpoint->pAioMgr, PPDMACEPFILEMGR));
297
298	return VINF_SUCCESS;
299	}
300
301	static DECLCALLBACK(void) pdmacFileEpTaskCompleted(PPDMACTASKFILE pTask, void *pvUser, int rc)
302	{
303	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pvUser;
304
305	LogFlowFunc(("pTask=%#p pvUser=%#p rc=%Rrc\n", pTask, pvUser, rc));
306
307	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH)
308	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, rc, true);
309	else
310	{
311	Assert((uint32_t)pTask->DataSeg.cbSeg == pTask->DataSeg.cbSeg && (int32_t)pTask->DataSeg.cbSeg >= 0);
312	uint32_t uOld = ASMAtomicSubS32(&pTaskFile->cbTransferLeft, (int32_t)pTask->DataSeg.cbSeg);
313
314	/* The first error will be returned. */
315	if (RT_FAILURE(rc))
316	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
317	#ifdef VBOX_WITH_DEBUGGER
318	else
319	{
320	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
321
322	/* Overwrite with injected error code. */
323	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_READ)
324	rc = ASMAtomicXchgS32(&pEpFile->rcReqRead, VINF_SUCCESS);
325	else
326	rc = ASMAtomicXchgS32(&pEpFile->rcReqWrite, VINF_SUCCESS);
327
328	if (RT_FAILURE(rc))
329	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
330	}
331	#endif
332
333	if (!(uOld - pTask->DataSeg.cbSeg)
334	&& !ASMAtomicXchgBool(&pTaskFile->fCompleted, true))
335	{
336	#ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
337	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
338	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEpFile->Core.pEpClass;
339
340	/* Check if we should delay completion of the request. */
341	if ( ASMAtomicReadU32(&pEpFile->msDelay) > 0
342	&& ASMAtomicReadU32(&pEpFile->cReqsDelay) > 0)
343	{
344	uint64_t tsDelay = pEpFile->msDelay;
345
346	if (pEpFile->msJitter)
347	tsDelay = (RTRandU32() % 100) > 50 ? pEpFile->msDelay + (RTRandU32() % pEpFile->msJitter)
348	: pEpFile->msDelay - (RTRandU32() % pEpFile->msJitter);
349	ASMAtomicDecU32(&pEpFile->cReqsDelay);
350
351	/* Arm the delay. */
352	pTaskFile->tsDelayEnd = RTTimeProgramMilliTS() + tsDelay;
353
354	/* Append to the list. */
355	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
356	do
357	{
358	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
359	pTaskFile->pDelayedNext = pHead;
360	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTaskFile, pHead));
361
362	if (tsDelay < pEpClassFile->cMilliesNext)
363	{
364	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, tsDelay);
365	TMTimerSetMillies(pEpClassFile->pTimer, tsDelay);
366	}
367
368	LogRel(("AIOMgr: Delaying request %#p for %u ms\n", pTaskFile, tsDelay));
369	}
370	else
371	#endif
372	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, pTaskFile->rc, true);
373	}
374	}
375	}
376
377	DECLINLINE(void) pdmacFileEpTaskInit(PPDMASYNCCOMPLETIONTASK pTask, size_t cbTransfer)
378	{
379	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
380
381	Assert((uint32_t)cbTransfer == cbTransfer && (int32_t)cbTransfer >= 0);
382	ASMAtomicWriteS32(&pTaskFile->cbTransferLeft, (int32_t)cbTransfer);
383	ASMAtomicWriteBool(&pTaskFile->fCompleted, false);
384	ASMAtomicWriteS32(&pTaskFile->rc, VINF_SUCCESS);
385	}
386
387	int pdmacFileEpTaskInitiate(PPDMASYNCCOMPLETIONTASK pTask,
388	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
389	PCRTSGSEG paSegments, size_t cSegments,
390	size_t cbTransfer, PDMACTASKFILETRANSFER enmTransfer)
391	{
392	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
393	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
394
395	Assert( (enmTransfer == PDMACTASKFILETRANSFER_READ)
396	\|\| (enmTransfer == PDMACTASKFILETRANSFER_WRITE));
397
398	for (size_t i = 0; i < cSegments; i++)
399	{
400	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
401	AssertPtr(pIoTask);
402
403	pIoTask->pEndpoint = pEpFile;
404	pIoTask->enmTransferType = enmTransfer;
405	pIoTask->Off = off;
406	pIoTask->DataSeg.cbSeg = paSegments[i].cbSeg;
407	pIoTask->DataSeg.pvSeg = paSegments[i].pvSeg;
408	pIoTask->pvUser = pTaskFile;
409	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
410
411	/* Send it off to the I/O manager. */
412	pdmacFileEpAddTask(pEpFile, pIoTask);
413	off += paSegments[i].cbSeg;
414	cbTransfer -= paSegments[i].cbSeg;
415	}
416
417	AssertMsg(!cbTransfer, ("Incomplete transfer %u bytes left\n", cbTransfer));
418
419	return VINF_AIO_TASK_PENDING;
420	}
421
422	/**
423	* Creates a new async I/O manager.
424	*
425	* @returns VBox status code.
426	* @param pEpClass Pointer to the endpoint class data.
427	* @param ppAioMgr Where to store the pointer to the new async I/O manager on success.
428	* @param enmMgrType Wanted manager type - can be overwritten by the global override.
429	*/
430	int pdmacFileAioMgrCreate(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass, PPPDMACEPFILEMGR ppAioMgr,
431	PDMACEPFILEMGRTYPE enmMgrType)
432	{
433	LogFlowFunc((": Entered\n"));
434
435	PPDMACEPFILEMGR pAioMgrNew;
436	int rc = MMR3HeapAllocZEx(pEpClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION, sizeof(PDMACEPFILEMGR), (void **)&pAioMgrNew);
437	if (RT_SUCCESS(rc))
438	{
439	if (enmMgrType < pEpClass->enmMgrTypeOverride)
440	pAioMgrNew->enmMgrType = enmMgrType;
441	else
442	pAioMgrNew->enmMgrType = pEpClass->enmMgrTypeOverride;
443
444	pAioMgrNew->msBwLimitExpired = RT_INDEFINITE_WAIT;
445
446	rc = RTSemEventCreate(&pAioMgrNew->EventSem);
447	if (RT_SUCCESS(rc))
448	{
449	rc = RTSemEventCreate(&pAioMgrNew->EventSemBlock);
450	if (RT_SUCCESS(rc))
451	{
452	rc = RTCritSectInit(&pAioMgrNew->CritSectBlockingEvent);
453	if (RT_SUCCESS(rc))
454	{
455	/* Init the rest of the manager. */
456	if (pAioMgrNew->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
457	rc = pdmacFileAioMgrNormalInit(pAioMgrNew);
458
459	if (RT_SUCCESS(rc))
460	{
461	pAioMgrNew->enmState = PDMACEPFILEMGRSTATE_RUNNING;
462
463	rc = RTThreadCreateF(&pAioMgrNew->Thread,
464	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
465	? pdmacFileAioMgrFailsafe
466	: pdmacFileAioMgrNormal,
467	pAioMgrNew,
468	0,
469	RTTHREADTYPE_IO,
470	0,
471	"AioMgr%d-%s", pEpClass->cAioMgrs,
472	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
473	? "F"
474	: "N");
475	if (RT_SUCCESS(rc))
476	{
477	/* Link it into the list. */
478	RTCritSectEnter(&pEpClass->CritSect);
479	pAioMgrNew->pNext = pEpClass->pAioMgrHead;
480	if (pEpClass->pAioMgrHead)
481	pEpClass->pAioMgrHead->pPrev = pAioMgrNew;
482	pEpClass->pAioMgrHead = pAioMgrNew;
483	pEpClass->cAioMgrs++;
484	RTCritSectLeave(&pEpClass->CritSect);
485
486	*ppAioMgr = pAioMgrNew;
487
488	Log(("PDMAC: Successfully created new file AIO Mgr {%s}\n", RTThreadGetName(pAioMgrNew->Thread)));
489	return VINF_SUCCESS;
490	}
491	pdmacFileAioMgrNormalDestroy(pAioMgrNew);
492	}
493	RTCritSectDelete(&pAioMgrNew->CritSectBlockingEvent);
494	}
495	RTSemEventDestroy(pAioMgrNew->EventSem);
496	}
497	RTSemEventDestroy(pAioMgrNew->EventSemBlock);
498	}
499	MMR3HeapFree(pAioMgrNew);
500	}
501
502	LogFlowFunc((": Leave rc=%Rrc\n", rc));
503
504	return rc;
505	}
506
507	/**
508	* Destroys a async I/O manager.
509	*
510	* @returns nothing.
511	* @param pEpClassFile Pointer to globals for the file endpoint class.
512	* @param pAioMgr The async I/O manager to destroy.
513	*/
514	static void pdmacFileAioMgrDestroy(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile, PPDMACEPFILEMGR pAioMgr)
515	{
516	int rc = pdmacFileAioMgrShutdown(pAioMgr);
517	AssertRC(rc);
518
519	/* Unlink from the list. */
520	rc = RTCritSectEnter(&pEpClassFile->CritSect);
521	AssertRC(rc);
522
523	PPDMACEPFILEMGR pPrev = pAioMgr->pPrev;
524	PPDMACEPFILEMGR pNext = pAioMgr->pNext;
525
526	if (pPrev)
527	pPrev->pNext = pNext;
528	else
529	pEpClassFile->pAioMgrHead = pNext;
530
531	if (pNext)
532	pNext->pPrev = pPrev;
533
534	pEpClassFile->cAioMgrs--;
535	rc = RTCritSectLeave(&pEpClassFile->CritSect);
536	AssertRC(rc);
537
538	/* Free the resources. */
539	RTCritSectDelete(&pAioMgr->CritSectBlockingEvent);
540	RTSemEventDestroy(pAioMgr->EventSem);
541	RTSemEventDestroy(pAioMgr->EventSemBlock);
542	if (pAioMgr->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
543	pdmacFileAioMgrNormalDestroy(pAioMgr);
544
545	MMR3HeapFree(pAioMgr);
546	}
547
548	static int pdmacFileMgrTypeFromName(const char *pszVal, PPDMACEPFILEMGRTYPE penmMgrType)
549	{
550	int rc = VINF_SUCCESS;
551
552	if (!RTStrCmp(pszVal, "Simple"))
553	*penmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
554	else if (!RTStrCmp(pszVal, "Async"))
555	*penmMgrType = PDMACEPFILEMGRTYPE_ASYNC;
556	else
557	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
558
559	return rc;
560	}
561
562	static const char *pdmacFileMgrTypeToName(PDMACEPFILEMGRTYPE enmMgrType)
563	{
564	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
565	return "Simple";
566	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
567	return "Async";
568
569	return NULL;
570	}
571
572	static int pdmacFileBackendTypeFromName(const char *pszVal, PPDMACFILEEPBACKEND penmBackendType)
573	{
574	int rc = VINF_SUCCESS;
575
576	if (!RTStrCmp(pszVal, "Buffered"))
577	*penmBackendType = PDMACFILEEPBACKEND_BUFFERED;
578	else if (!RTStrCmp(pszVal, "NonBuffered"))
579	*penmBackendType = PDMACFILEEPBACKEND_NON_BUFFERED;
580	else
581	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
582
583	return rc;
584	}
585
586	static const char *pdmacFileBackendTypeToName(PDMACFILEEPBACKEND enmBackendType)
587	{
588	if (enmBackendType == PDMACFILEEPBACKEND_BUFFERED)
589	return "Buffered";
590	if (enmBackendType == PDMACFILEEPBACKEND_NON_BUFFERED)
591	return "NonBuffered";
592
593	return NULL;
594	}
595
596	#ifdef VBOX_WITH_DEBUGGER
597
598	/**
599	* @callback_method_impl{FNDBGCCMD, The '.injecterror' command.}
600	*/
601	static DECLCALLBACK(int) pdmacEpFileErrorInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
602	{
603	/*
604	* Validate input.
605	*/
606	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
607	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 3);
608	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
609	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
610	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
611
612	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
613	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
614
615	/* Syntax is "read\|write <filename> <status code>" */
616	bool fWrite;
617	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
618	fWrite = false;
619	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
620	fWrite = true;
621	else
622	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
623
624	int32_t rcToInject = (int32_t)pArgs[2].u.u64Number;
625	if ((uint64_t)rcToInject != pArgs[2].u.u64Number)
626	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The status code '%lld' is out of range", pArgs[0].u.u64Number);
627
628	/*
629	* Search for the matching endpoint.
630	*/
631	RTCritSectEnter(&pEpClassFile->Core.CritSect);
632
633	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
634	while (pEpFile)
635	{
636	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
637	break;
638	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
639	}
640
641	if (pEpFile)
642	{
643	/*
644	* Do the job.
645	*/
646	if (fWrite)
647	ASMAtomicXchgS32(&pEpFile->rcReqWrite, rcToInject);
648	else
649	ASMAtomicXchgS32(&pEpFile->rcReqRead, rcToInject);
650
651	DBGCCmdHlpPrintf(pCmdHlp, "Injected %Rrc into '%s' for %s\n",
652	(int)rcToInject, pArgs[1].u.pszString, pArgs[0].u.pszString);
653	}
654
655	RTCritSectLeave(&pEpClassFile->Core.CritSect);
656
657	if (!pEpFile)
658	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
659	return VINF_SUCCESS;
660	}
661
662	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
663	/**
664	* @callback_method_impl{FNDBGCCMD, The '.injectdelay' command.}
665	*/
666	static DECLCALLBACK(int) pdmacEpFileDelayInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
667	{
668	/*
669	* Validate input.
670	*/
671	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
672	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs >= 3);
673	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
674	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
675	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
676
677	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
678	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
679
680	/* Syntax is "read\|write\|flush\|any <filename> <delay> [reqs]" */
681	PDMACFILEREQTYPEDELAY enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
682	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
683	enmDelayType = PDMACFILEREQTYPEDELAY_READ;
684	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
685	enmDelayType = PDMACFILEREQTYPEDELAY_WRITE;
686	else if (!RTStrCmp(pArgs[0].u.pszString, "flush"))
687	enmDelayType = PDMACFILEREQTYPEDELAY_FLUSH;
688	else if (!RTStrCmp(pArgs[0].u.pszString, "any"))
689	enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
690	else
691	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
692
693	uint32_t msDelay = (uint32_t)pArgs[2].u.u64Number;
694	if ((uint64_t)msDelay != pArgs[2].u.u64Number)
695	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The delay '%lld' is out of range", pArgs[0].u.u64Number);
696
697	uint32_t cReqsDelay = 1;
698	uint32_t msJitter = 0;
699	if (cArgs >= 4)
700	msJitter = (uint32_t)pArgs[3].u.u64Number;
701	if (cArgs == 5)
702	cReqsDelay = (uint32_t)pArgs[4].u.u64Number;
703
704	/*
705	* Search for the matching endpoint.
706	*/
707	RTCritSectEnter(&pEpClassFile->Core.CritSect);
708
709	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
710	while (pEpFile)
711	{
712	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
713	break;
714	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
715	}
716
717	if (pEpFile)
718	{
719	ASMAtomicWriteSize(&pEpFile->enmTypeDelay, enmDelayType);
720	ASMAtomicWriteU32(&pEpFile->msDelay, msDelay);
721	ASMAtomicWriteU32(&pEpFile->msJitter, msJitter);
722	ASMAtomicWriteU32(&pEpFile->cReqsDelay, cReqsDelay);
723
724	DBGCCmdHlpPrintf(pCmdHlp, "Injected delay for the next %u requests of %u ms into '%s' for %s\n",
725	cReqsDelay, msDelay, pArgs[1].u.pszString, pArgs[0].u.pszString);
726	}
727
728	RTCritSectLeave(&pEpClassFile->Core.CritSect);
729
730	if (!pEpFile)
731	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
732	return VINF_SUCCESS;
733	}
734
735	static DECLCALLBACK(void) pdmacR3TimerCallback(PVM pVM, PTMTIMER pTimer, void *pvUser)
736	{
737	uint64_t tsCur = RTTimeProgramMilliTS();
738	uint64_t cMilliesNext = UINT64_MAX;
739	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pvUser;
740
741	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, UINT64_MAX);
742
743	/* Go through all endpoints and check for expired requests. */
744	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
745
746	while (pEpFile)
747	{
748	/* Check for an expired delay. */
749	if (pEpFile->pDelayedHead != NULL)
750	{
751	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = ASMAtomicXchgPtrT(&pEpFile->pDelayedHead, NULL, PPDMASYNCCOMPLETIONTASKFILE);
752
753	while (pTaskFile)
754	{
755	PPDMASYNCCOMPLETIONTASKFILE pTmp = pTaskFile;
756	pTaskFile = pTaskFile->pDelayedNext;
757
758	if (tsCur >= pTmp->tsDelayEnd)
759	{
760	LogRel(("AIOMgr: Delayed request %#p completed\n", pTmp));
761	pdmR3AsyncCompletionCompleteTask(&pTmp->Core, pTmp->rc, true);
762	}
763	else
764	{
765	/* Prepend to the delayed list again. */
766	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
767
768	if (pTmp->tsDelayEnd - tsCur < cMilliesNext)
769	cMilliesNext = pTmp->tsDelayEnd - tsCur;
770
771	do
772	{
773	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
774	pTmp->pDelayedNext = pHead;
775	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTmp, pHead));
776	}
777	}
778	}
779
780	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
781	}
782
783	if (cMilliesNext < pEpClassFile->cMilliesNext)
784	{
785	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, cMilliesNext);
786	TMTimerSetMillies(pEpClassFile->pTimer, cMilliesNext);
787	}
788	}
789
790	# endif /* PDM_ASYNC_COMPLETION_FILE_WITH_DELAY */
791
792	#endif /* VBOX_WITH_DEBUGGER */
793
794	static DECLCALLBACK(int) pdmacFileInitialize(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals, PCFGMNODE pCfgNode)
795	{
796	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
797	RTFILEAIOLIMITS AioLimits; /** < Async I/O limitations. */
798
799	int rc = RTFileAioGetLimits(&AioLimits);
800	#ifdef DEBUG
801	if (RT_SUCCESS(rc) && RTEnvExist("VBOX_ASYNC_IO_FAILBACK"))
802	rc = VERR_ENV_VAR_NOT_FOUND;
803	#endif
804	if (RT_FAILURE(rc))
805	{
806	LogRel(("AIO: Async I/O manager not supported (rc=%Rrc). Falling back to simple manager\n", rc));
807	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_SIMPLE;
808	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_BUFFERED;
809	}
810	else
811	{
812	pEpClassFile->uBitmaskAlignment = AioLimits.cbBufferAlignment ? ~((RTR3UINTPTR)AioLimits.cbBufferAlignment - 1) : RTR3UINTPTR_MAX;
813	pEpClassFile->cReqsOutstandingMax = AioLimits.cReqsOutstandingMax;
814
815	if (pCfgNode)
816	{
817	/* Query the default manager type */
818	char *pszVal = NULL;
819	rc = CFGMR3QueryStringAllocDef(pCfgNode, "IoMgr", &pszVal, "Async");
820	AssertLogRelRCReturn(rc, rc);
821
822	rc = pdmacFileMgrTypeFromName(pszVal, &pEpClassFile->enmMgrTypeOverride);
823	MMR3HeapFree(pszVal);
824	if (RT_FAILURE(rc))
825	return rc;
826
827	LogRel(("AIOMgr: Default manager type is '%s'\n", pdmacFileMgrTypeToName(pEpClassFile->enmMgrTypeOverride)));
828
829	/* Query default backend type */
830	rc = CFGMR3QueryStringAllocDef(pCfgNode, "FileBackend", &pszVal, "NonBuffered");
831	AssertLogRelRCReturn(rc, rc);
832
833	rc = pdmacFileBackendTypeFromName(pszVal, &pEpClassFile->enmEpBackendDefault);
834	MMR3HeapFree(pszVal);
835	if (RT_FAILURE(rc))
836	return rc;
837
838	LogRel(("AIOMgr: Default file backend is '%s'\n", pdmacFileBackendTypeToName(pEpClassFile->enmEpBackendDefault)));
839
840	#ifdef RT_OS_LINUX
841	if ( pEpClassFile->enmMgrTypeOverride == PDMACEPFILEMGRTYPE_ASYNC
842	&& pEpClassFile->enmEpBackendDefault == PDMACFILEEPBACKEND_BUFFERED)
843	{
844	LogRel(("AIOMgr: Linux does not support buffered async I/O, changing to non buffered\n"));
845	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
846	}
847	#endif
848	}
849	else
850	{
851	/* No configuration supplied, set defaults */
852	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
853	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_ASYNC;
854	}
855	}
856
857	/* Init critical section. */
858	rc = RTCritSectInit(&pEpClassFile->CritSect);
859
860	#ifdef VBOX_WITH_DEBUGGER
861	/* Install the error injection handler. */
862	if (RT_SUCCESS(rc))
863	{
864	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
865	AssertRC(rc);
866	}
867
868	#ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
869	rc = TMR3TimerCreateInternal(pEpClassFile->Core.pVM, TMCLOCK_REAL, pdmacR3TimerCallback, pEpClassFile, "AC Delay", &pEpClassFile->pTimer);
870	AssertRC(rc);
871	pEpClassFile->cMilliesNext = UINT64_MAX;
872	#endif
873	#endif
874
875	return rc;
876	}
877
878	static DECLCALLBACK(void) pdmacFileTerminate(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals)
879	{
880	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
881
882	/* All endpoints should be closed at this point. */
883	AssertMsg(!pEpClassFile->Core.pEndpointsHead, ("There are still endpoints left\n"));
884
885	/* Destroy all left async I/O managers. */
886	while (pEpClassFile->pAioMgrHead)
887	pdmacFileAioMgrDestroy(pEpClassFile, pEpClassFile->pAioMgrHead);
888
889	RTCritSectDelete(&pEpClassFile->CritSect);
890	}
891
892	static DECLCALLBACK(int) pdmacFileEpInitialize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint,
893	const char *pszUri, uint32_t fFlags)
894	{
895	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
896	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
897	PDMACEPFILEMGRTYPE enmMgrType = pEpClassFile->enmMgrTypeOverride;
898	PDMACFILEEPBACKEND enmEpBackend = pEpClassFile->enmEpBackendDefault;
899
900	AssertMsgReturn((fFlags & ~(PDMACEP_FILE_FLAGS_READ_ONLY \| PDMACEP_FILE_FLAGS_DONT_LOCK \| PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)) == 0,
901	("PDMAsyncCompletion: Invalid flag specified\n"), VERR_INVALID_PARAMETER);
902
903	unsigned fFileFlags = RTFILE_O_OPEN;
904
905	/*
906	* Revert to the simple manager and the buffered backend if
907	* the host cache should be enabled.
908	*/
909	if (fFlags & PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)
910	{
911	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
912	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
913	}
914
915	if (fFlags & PDMACEP_FILE_FLAGS_READ_ONLY)
916	fFileFlags \|= RTFILE_O_READ \| RTFILE_O_DENY_NONE;
917	else
918	{
919	fFileFlags \|= RTFILE_O_READWRITE;
920
921	/*
922	* Opened in read/write mode. Check whether the caller wants to
923	* avoid the lock. Return an error in case caching is enabled
924	* because this can lead to data corruption.
925	*/
926	if (fFlags & PDMACEP_FILE_FLAGS_DONT_LOCK)
927	fFileFlags \|= RTFILE_O_DENY_NONE;
928	else
929	fFileFlags \|= RTFILE_O_DENY_WRITE;
930	}
931
932	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
933	fFileFlags \|= RTFILE_O_ASYNC_IO;
934
935	int rc;
936	if (enmEpBackend == PDMACFILEEPBACKEND_NON_BUFFERED)
937	{
938	/*
939	* We only disable the cache if the size of the file is a multiple of 512.
940	* Certain hosts like Windows, Linux and Solaris require that transfer sizes
941	* are aligned to the volume sector size.
942	* If not we just make sure that the data is written to disk with RTFILE_O_WRITE_THROUGH
943	* which will trash the host cache but ensures that the host cache will not
944	* contain dirty buffers.
945	*/
946	RTFILE hFile;
947	rc = RTFileOpen(&hFile, pszUri, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
948	if (RT_SUCCESS(rc))
949	{
950	uint64_t cbSize;
951
952	rc = RTFileQuerySize(hFile, &cbSize);
953
954	if (RT_SUCCESS(rc) && ((cbSize % 512) == 0))
955	fFileFlags \|= RTFILE_O_NO_CACHE;
956	else
957	{
958	/* Downgrade to the buffered backend */
959	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
960
961	#ifdef RT_OS_LINUX
962	fFileFlags &= ~RTFILE_O_ASYNC_IO;
963	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
964	#endif
965	}
966	RTFileClose(hFile);
967	}
968	}
969
970	/* Open with final flags. */
971	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
972	if ( rc == VERR_INVALID_FUNCTION
973	\|\| rc == VERR_INVALID_PARAMETER)
974	{
975	LogRel(("AIOMgr: pdmacFileEpInitialize: RTFileOpen %s / %08x failed with %Rrc\n",
976	pszUri, fFileFlags, rc));
977	/*
978	* Solaris doesn't support directio on ZFS so far. :-\
979	* Trying to enable it returns VERR_INVALID_FUNCTION
980	* (ENOTTY). Remove it and hope for the best.
981	* ZFS supports write throttling in case applications
982	* write more data than can be synced to the disk
983	* without blocking the whole application.
984	*
985	* On Linux we have the same problem with cifs.
986	* Have to disable async I/O here too because it requires O_DIRECT.
987	*/
988	fFileFlags &= ~RTFILE_O_NO_CACHE;
989	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
990
991	#ifdef RT_OS_LINUX
992	fFileFlags &= ~RTFILE_O_ASYNC_IO;
993	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
994	#endif
995
996	/* Open again. */
997	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
998
999	if (RT_FAILURE(rc))
1000	{
1001	LogRel(("AIOMgr: pdmacFileEpInitialize: RTFileOpen %s / %08x failed AGAIN(!) with %Rrc\n",
1002	pszUri, fFileFlags, rc));
1003	}
1004	}
1005
1006	if (RT_SUCCESS(rc))
1007	{
1008	pEpFile->fFlags = fFileFlags;
1009
1010	rc = RTFileQuerySize(pEpFile->hFile, (uint64_t *)&pEpFile->cbFile);
1011	if (RT_SUCCESS(rc))
1012	{
1013	/* Initialize the segment cache */
1014	rc = MMR3HeapAllocZEx(pEpClassFile->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
1015	sizeof(PDMACTASKFILE),
1016	(void **)&pEpFile->pTasksFreeHead);
1017	if (RT_SUCCESS(rc))
1018	{
1019	PPDMACEPFILEMGR pAioMgr = NULL;
1020
1021	pEpFile->pTasksFreeTail = pEpFile->pTasksFreeHead;
1022	pEpFile->cTasksCached = 0;
1023	pEpFile->enmBackendType = enmEpBackend;
1024	/*
1025	* Disable async flushes on Solaris for now.
1026	* They cause weird hangs which needs more investigations.
1027	*/
1028	#ifndef RT_OS_SOLARIS
1029	pEpFile->fAsyncFlushSupported = true;
1030	#else
1031	pEpFile->fAsyncFlushSupported = false;
1032	#endif
1033
1034	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1035	{
1036	/* Simple mode. Every file has its own async I/O manager. */
1037	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, PDMACEPFILEMGRTYPE_SIMPLE);
1038	}
1039	else
1040	{
1041	pAioMgr = pEpClassFile->pAioMgrHead;
1042
1043	/* Check for an idling manager of the same type */
1044	while (pAioMgr)
1045	{
1046	if (pAioMgr->enmMgrType == enmMgrType)
1047	break;
1048	pAioMgr = pAioMgr->pNext;
1049	}
1050
1051	if (!pAioMgr)
1052	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, enmMgrType);
1053	}
1054
1055	if (RT_SUCCESS(rc))
1056	{
1057	pEpFile->AioMgr.pTreeRangesLocked = (PAVLRFOFFTREE)RTMemAllocZ(sizeof(AVLRFOFFTREE));
1058	if (!pEpFile->AioMgr.pTreeRangesLocked)
1059	rc = VERR_NO_MEMORY;
1060	else
1061	{
1062	pEpFile->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE;
1063
1064	/* Assign the endpoint to the thread. */
1065	rc = pdmacFileAioMgrAddEndpoint(pAioMgr, pEpFile);
1066	if (RT_FAILURE(rc))
1067	{
1068	RTMemFree(pEpFile->AioMgr.pTreeRangesLocked);
1069	MMR3HeapFree(pEpFile->pTasksFreeHead);
1070	}
1071	}
1072	}
1073	else if (rc == VERR_FILE_AIO_INSUFFICIENT_EVENTS)
1074	{
1075	PUVM pUVM = VMR3GetUVM(pEpClassFile->Core.pVM);
1076	#if defined(RT_OS_LINUX)
1077	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1078	N_("Failed to create I/O manager for VM due to insufficient resources on the host. "
1079	"Either increase the amount of allowed events in /proc/sys/fs/aio-max-nr or enable "
1080	"the host I/O cache"));
1081	#else
1082	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1083	N_("Failed to create I/O manager for VM due to insufficient resources on the host. "
1084	"Enable the host I/O cache"));
1085	#endif
1086	}
1087	else
1088	{
1089	PUVM pUVM = VMR3GetUVM(pEpClassFile->Core.pVM);
1090	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1091	N_("Failed to create I/O manager for VM due to an unknown error"));
1092	}
1093	}
1094	}
1095
1096	if (RT_FAILURE(rc))
1097	RTFileClose(pEpFile->hFile);
1098	}
1099
1100	#ifdef VBOX_WITH_STATISTICS
1101	if (RT_SUCCESS(rc))
1102	{
1103	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatRead,
1104	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1105	STAMUNIT_TICKS_PER_CALL, "Time taken to read from the endpoint",
1106	"/PDM/AsyncCompletion/File/%s/%d/Read", RTPathFilename(pEpFile->Core.pszUri), pEpFile->Core.iStatId);
1107
1108	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatWrite,
1109	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1110	STAMUNIT_TICKS_PER_CALL, "Time taken to write to the endpoint",
1111	"/PDM/AsyncCompletion/File/%s/%d/Write", RTPathFilename(pEpFile->Core.pszUri), pEpFile->Core.iStatId);
1112	}
1113	#endif
1114
1115	if (RT_SUCCESS(rc))
1116	LogRel(("AIOMgr: Endpoint for file '%s' (flags %08x) created successfully\n", pszUri, pEpFile->fFlags));
1117
1118	return rc;
1119	}
1120
1121	static DECLCALLBACK(int) pdmacFileEpRangesLockedDestroy(PAVLRFOFFNODECORE pNode, void *pvUser)
1122	{
1123	NOREF(pNode); NOREF(pvUser);
1124	AssertMsgFailed(("The locked ranges tree should be empty at that point\n"));
1125	return VINF_SUCCESS;
1126	}
1127
1128	static DECLCALLBACK(int) pdmacFileEpClose(PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1129	{
1130	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1131	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
1132
1133	/* Make sure that all tasks finished for this endpoint. */
1134	int rc = pdmacFileAioMgrCloseEndpoint(pEpFile->pAioMgr, pEpFile);
1135	AssertRC(rc);
1136
1137	/*
1138	* If the async I/O manager is in failsafe mode this is the only endpoint
1139	* he processes and thus can be destroyed now.
1140	*/
1141	if (pEpFile->pAioMgr->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1142	pdmacFileAioMgrDestroy(pEpClassFile, pEpFile->pAioMgr);
1143
1144	/* Free cached tasks. */
1145	PPDMACTASKFILE pTask = pEpFile->pTasksFreeHead;
1146
1147	while (pTask)
1148	{
1149	PPDMACTASKFILE pTaskFree = pTask;
1150	pTask = pTask->pNext;
1151	MMR3HeapFree(pTaskFree);
1152	}
1153
1154	/* Destroy the locked ranges tree now. */
1155	RTAvlrFileOffsetDestroy(pEpFile->AioMgr.pTreeRangesLocked, pdmacFileEpRangesLockedDestroy, NULL);
1156	RTMemFree(pEpFile->AioMgr.pTreeRangesLocked);
1157	pEpFile->AioMgr.pTreeRangesLocked = NULL;
1158
1159	RTFileClose(pEpFile->hFile);
1160
1161	#ifdef VBOX_WITH_STATISTICS
1162	/* Not sure if this might be unnecessary because of similar statement in pdmR3AsyncCompletionStatisticsDeregister? */
1163	STAMR3DeregisterF(pEpClassFile->Core.pVM->pUVM, "/PDM/AsyncCompletion/File/%s/*", RTPathFilename(pEpFile->Core.pszUri));
1164	#endif
1165
1166	return VINF_SUCCESS;
1167	}
1168
1169	static DECLCALLBACK(int) pdmacFileEpRead(PPDMASYNCCOMPLETIONTASK pTask,
1170	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1171	PCRTSGSEG paSegments, size_t cSegments,
1172	size_t cbRead)
1173	{
1174	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1175
1176	LogFlowFunc(("pTask=%#p pEndpoint=%#p off=%RTfoff paSegments=%#p cSegments=%zu cbRead=%zu\n",
1177	pTask, pEndpoint, off, paSegments, cSegments, cbRead));
1178
1179	if (RT_UNLIKELY((uint64_t)off + cbRead > pEpFile->cbFile))
1180	return VERR_EOF;
1181
1182	STAM_PROFILE_ADV_START(&pEpFile->StatRead, Read);
1183	pdmacFileEpTaskInit(pTask, cbRead);
1184	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbRead,
1185	PDMACTASKFILETRANSFER_READ);
1186	STAM_PROFILE_ADV_STOP(&pEpFile->StatRead, Read);
1187
1188	return rc;
1189	}
1190
1191	static DECLCALLBACK(int) pdmacFileEpWrite(PPDMASYNCCOMPLETIONTASK pTask,
1192	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1193	PCRTSGSEG paSegments, size_t cSegments,
1194	size_t cbWrite)
1195	{
1196	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1197
1198	if (RT_UNLIKELY(pEpFile->fReadonly))
1199	return VERR_NOT_SUPPORTED;
1200
1201	STAM_PROFILE_ADV_START(&pEpFile->StatWrite, Write);
1202
1203	pdmacFileEpTaskInit(pTask, cbWrite);
1204
1205	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbWrite,
1206	PDMACTASKFILETRANSFER_WRITE);
1207
1208	STAM_PROFILE_ADV_STOP(&pEpFile->StatWrite, Write);
1209
1210	return rc;
1211	}
1212
1213	static DECLCALLBACK(int) pdmacFileEpFlush(PPDMASYNCCOMPLETIONTASK pTask,
1214	PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1215	{
1216	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1217	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
1218
1219	if (RT_UNLIKELY(pEpFile->fReadonly))
1220	return VERR_NOT_SUPPORTED;
1221
1222	pdmacFileEpTaskInit(pTask, 0);
1223
1224	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
1225	if (RT_UNLIKELY(!pIoTask))
1226	return VERR_NO_MEMORY;
1227
1228	pIoTask->pEndpoint = pEpFile;
1229	pIoTask->enmTransferType = PDMACTASKFILETRANSFER_FLUSH;
1230	pIoTask->pvUser = pTaskFile;
1231	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
1232	pdmacFileEpAddTask(pEpFile, pIoTask);
1233
1234	return VINF_AIO_TASK_PENDING;
1235	}
1236
1237	static DECLCALLBACK(int) pdmacFileEpGetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t *pcbSize)
1238	{
1239	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1240
1241	*pcbSize = ASMAtomicReadU64(&pEpFile->cbFile);
1242
1243	return VINF_SUCCESS;
1244	}
1245
1246	static DECLCALLBACK(int) pdmacFileEpSetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t cbSize)
1247	{
1248	int rc;
1249	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1250
1251	rc = RTFileSetSize(pEpFile->hFile, cbSize);
1252	if (RT_SUCCESS(rc))
1253	ASMAtomicWriteU64(&pEpFile->cbFile, cbSize);
1254
1255	return rc;
1256	}
1257
1258	const PDMASYNCCOMPLETIONEPCLASSOPS g_PDMAsyncCompletionEndpointClassFile =
1259	{
1260	/* u32Version */
1261	PDMAC_EPCLASS_OPS_VERSION,
1262	/* pcszName */
1263	"File",
1264	/* enmClassType */
1265	PDMASYNCCOMPLETIONEPCLASSTYPE_FILE,
1266	/* cbEndpointClassGlobal */
1267	sizeof(PDMASYNCCOMPLETIONEPCLASSFILE),
1268	/* cbEndpoint */
1269	sizeof(PDMASYNCCOMPLETIONENDPOINTFILE),
1270	/* cbTask */
1271	sizeof(PDMASYNCCOMPLETIONTASKFILE),
1272	/* pfnInitialize */
1273	pdmacFileInitialize,
1274	/* pfnTerminate */
1275	pdmacFileTerminate,
1276	/* pfnEpInitialize. */
1277	pdmacFileEpInitialize,
1278	/* pfnEpClose */
1279	pdmacFileEpClose,
1280	/* pfnEpRead */
1281	pdmacFileEpRead,
1282	/* pfnEpWrite */
1283	pdmacFileEpWrite,
1284	/* pfnEpFlush */
1285	pdmacFileEpFlush,
1286	/* pfnEpGetSize */
1287	pdmacFileEpGetSize,
1288	/* pfnEpSetSize */
1289	pdmacFileEpSetSize,
1290	/* u32VersionEnd */
1291	PDMAC_EPCLASS_OPS_VERSION
1292	};
1293

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

source: vbox/trunk/src/VBox/VMM/VMMR3/PDMAsyncCompletionFile.cpp@ 87758

Download in other formats: