PDMAsyncCompletionFile.cpp@ 97188

Last change on this file since 97188 was 96407, checked in by vboxsync, 2 years ago
scm copyright and license note update
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 47.9 KB

Line
1	/* $Id: PDMAsyncCompletionFile.cpp 96407 2022-08-22 17:43:14Z vboxsync $ */
2	/** @file
3	* PDM Async I/O - Transport data asynchronous in R3 using EMT.
4	*/
5
6	/*
7	* Copyright (C) 2006-2022 Oracle and/or its affiliates.
8	*
9	* This file is part of VirtualBox base platform packages, as
10	* available from https://www.virtualbox.org.
11	*
12	* This program is free software; you can redistribute it and/or
13	* modify it under the terms of the GNU General Public License
14	* as published by the Free Software Foundation, in version 3 of the
15	* License.
16	*
17	* This program is distributed in the hope that it will be useful, but
18	* WITHOUT ANY WARRANTY; without even the implied warranty of
19	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20	* General Public License for more details.
21	*
22	* You should have received a copy of the GNU General Public License
23	* along with this program; if not, see <https://www.gnu.org/licenses>.
24	*
25	* SPDX-License-Identifier: GPL-3.0-only
26	*/
27
28
29	/*********************************************************************************************************************************
30	* Header Files *
31	*********************************************************************************************************************************/
32	#define LOG_GROUP LOG_GROUP_PDM_ASYNC_COMPLETION
33	#include "PDMInternal.h"
34	#include <VBox/vmm/pdm.h>
35	#include <VBox/vmm/mm.h>
36	#include <VBox/vmm/vm.h>
37	#include <VBox/err.h>
38	#include <VBox/log.h>
39	#include <VBox/dbg.h>
40	#include <VBox/vmm/uvm.h>
41	#include <VBox/vmm/tm.h>
42
43	#include <iprt/asm.h>
44	#include <iprt/assert.h>
45	#include <iprt/critsect.h>
46	#include <iprt/env.h>
47	#include <iprt/file.h>
48	#include <iprt/mem.h>
49	#include <iprt/semaphore.h>
50	#include <iprt/string.h>
51	#include <iprt/thread.h>
52	#include <iprt/path.h>
53	#include <iprt/rand.h>
54
55	#include "PDMAsyncCompletionFileInternal.h"
56
57
58	/*********************************************************************************************************************************
59	* Internal Functions *
60	*********************************************************************************************************************************/
61	#ifdef VBOX_WITH_DEBUGGER
62	static FNDBGCCMD pdmacEpFileErrorInject;
63	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
64	static FNDBGCCMD pdmacEpFileDelayInject;
65	# endif
66	#endif
67
68
69	/*********************************************************************************************************************************
70	* Global Variables *
71	*********************************************************************************************************************************/
72	#ifdef VBOX_WITH_DEBUGGER
73	static const DBGCVARDESC g_aInjectErrorArgs[] =
74	{
75	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
76	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write/read." },
77	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
78	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "errcode", "VBox status code." },
79	};
80
81	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
82	static const DBGCVARDESC g_aInjectDelayArgs[] =
83	{
84	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
85	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write\|read\|flush\|any." },
86	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
87	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "delay", "Delay in milliseconds." },
88	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "jitter", "Jitter of the delay." },
89	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "reqs", "Number of requests to delay." }
90
91	};
92	# endif
93
94	/** Command descriptors. */
95	static const DBGCCMD g_aCmds[] =
96	{
97	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax,.pszDescription */
98	{ "injecterror", 3, 3, &g_aInjectErrorArgs[0], 3, 0, pdmacEpFileErrorInject, "", "Inject error into I/O subsystem." }
99	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
100	,{ "injectdelay", 3, 5, &g_aInjectDelayArgs[0], RT_ELEMENTS(g_aInjectDelayArgs), 0, pdmacEpFileDelayInject, "", "Inject a delay of a request." }
101	# endif
102	};
103	#endif
104
105
106	/**
107	* Frees a task.
108	*
109	* @returns nothing.
110	* @param pEndpoint Pointer to the endpoint the segment was for.
111	* @param pTask The task to free.
112	*/
113	void pdmacFileTaskFree(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
114	{
115	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
116
117	LogFlowFunc((": pEndpoint=%p pTask=%p\n", pEndpoint, pTask));
118
119	/* Try the per endpoint cache first. */
120	if (pEndpoint->cTasksCached < pEpClass->cTasksCacheMax)
121	{
122	/* Add it to the list. */
123	pEndpoint->pTasksFreeTail->pNext = pTask;
124	pEndpoint->pTasksFreeTail = pTask;
125	ASMAtomicIncU32(&pEndpoint->cTasksCached);
126	}
127	else
128	{
129	Log(("Freeing task %p because all caches are full\n", pTask));
130	MMR3HeapFree(pTask);
131	}
132	}
133
134	/**
135	* Allocates a task segment
136	*
137	* @returns Pointer to the new task segment or NULL
138	* @param pEndpoint Pointer to the endpoint
139	*/
140	PPDMACTASKFILE pdmacFileTaskAlloc(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
141	{
142	PPDMACTASKFILE pTask = NULL;
143
144	/* Try the small per endpoint cache first. */
145	if (pEndpoint->pTasksFreeHead == pEndpoint->pTasksFreeTail)
146	{
147	/* Try the bigger endpoint class cache. */
148	PPDMASYNCCOMPLETIONEPCLASSFILE pEndpointClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
149
150	/*
151	* Allocate completely new.
152	* If this fails we return NULL.
153	*/
154	int rc = MMR3HeapAllocZEx(pEndpointClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
155	sizeof(PDMACTASKFILE),
156	(void **)&pTask);
157	if (RT_FAILURE(rc))
158	pTask = NULL;
159
160	LogFlow(("Allocated task %p -> %Rrc\n", pTask, rc));
161	}
162	else
163	{
164	/* Grab a free task from the head. */
165	AssertMsg(pEndpoint->cTasksCached > 0, ("No tasks cached but list contains more than one element\n"));
166
167	pTask = pEndpoint->pTasksFreeHead;
168	pEndpoint->pTasksFreeHead = pTask->pNext;
169	ASMAtomicDecU32(&pEndpoint->cTasksCached);
170	pTask->pNext = NULL;
171	}
172
173	return pTask;
174	}
175
176	PPDMACTASKFILE pdmacFileEpGetNewTasks(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
177	{
178	/*
179	* Get pending tasks.
180	*/
181	PPDMACTASKFILE pTasks = ASMAtomicXchgPtrT(&pEndpoint->pTasksNewHead, NULL, PPDMACTASKFILE);
182
183	/* Reverse the list to process in FIFO order. */
184	if (pTasks)
185	{
186	PPDMACTASKFILE pTask = pTasks;
187
188	pTasks = NULL;
189
190	while (pTask)
191	{
192	PPDMACTASKFILE pCur = pTask;
193	pTask = pTask->pNext;
194	pCur->pNext = pTasks;
195	pTasks = pCur;
196	}
197	}
198
199	return pTasks;
200	}
201
202	static void pdmacFileAioMgrWakeup(PPDMACEPFILEMGR pAioMgr)
203	{
204	bool fWokenUp = ASMAtomicXchgBool(&pAioMgr->fWokenUp, true);
205	if (!fWokenUp)
206	{
207	bool fWaitingEventSem = ASMAtomicReadBool(&pAioMgr->fWaitingEventSem);
208	if (fWaitingEventSem)
209	{
210	int rc = RTSemEventSignal(pAioMgr->EventSem);
211	AssertRC(rc);
212	}
213	}
214	}
215
216	static int pdmacFileAioMgrWaitForBlockingEvent(PPDMACEPFILEMGR pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT enmEvent)
217	{
218	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, enmEvent);
219	Assert(!pAioMgr->fBlockingEventPending);
220	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, true);
221
222	/* Wakeup the async I/O manager */
223	pdmacFileAioMgrWakeup(pAioMgr);
224
225	/* Wait for completion. */
226	int rc = RTSemEventWait(pAioMgr->EventSemBlock, RT_INDEFINITE_WAIT);
227	AssertRC(rc);
228
229	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, false);
230	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, PDMACEPFILEAIOMGRBLOCKINGEVENT_INVALID);
231
232	return rc;
233	}
234
235	int pdmacFileAioMgrAddEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
236	{
237	LogFlowFunc(("pAioMgr=%#p pEndpoint=%#p{%s}\n", pAioMgr, pEndpoint, pEndpoint->Core.pszUri));
238
239	/* Update the assigned I/O manager. */
240	ASMAtomicWritePtr(&pEndpoint->pAioMgr, pAioMgr);
241
242	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
243	AssertRCReturn(rc, rc);
244
245	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint, pEndpoint);
246	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_ADD_ENDPOINT);
247	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint);
248
249	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
250
251	return rc;
252	}
253
254	#ifdef SOME_UNUSED_FUNCTION
255	static int pdmacFileAioMgrRemoveEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
256	{
257	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
258	AssertRCReturn(rc, rc);
259
260	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint, pEndpoint);
261	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_REMOVE_ENDPOINT);
262	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint);
263
264	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
265
266	return rc;
267	}
268	#endif
269
270	static int pdmacFileAioMgrCloseEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
271	{
272	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
273	AssertRCReturn(rc, rc);
274
275	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint, pEndpoint);
276	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_CLOSE_ENDPOINT);
277	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint);
278
279	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
280
281	return rc;
282	}
283
284	static int pdmacFileAioMgrShutdown(PPDMACEPFILEMGR pAioMgr)
285	{
286	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
287	AssertRCReturn(rc, rc);
288
289	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_SHUTDOWN);
290
291	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
292
293	return rc;
294	}
295
296	int pdmacFileEpAddTask(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
297	{
298	PPDMACTASKFILE pNext;
299	do
300	{
301	pNext = pEndpoint->pTasksNewHead;
302	pTask->pNext = pNext;
303	} while (!ASMAtomicCmpXchgPtr(&pEndpoint->pTasksNewHead, pTask, pNext));
304
305	pdmacFileAioMgrWakeup(ASMAtomicReadPtrT(&pEndpoint->pAioMgr, PPDMACEPFILEMGR));
306
307	return VINF_SUCCESS;
308	}
309
310	static DECLCALLBACK(void) pdmacFileEpTaskCompleted(PPDMACTASKFILE pTask, void *pvUser, int rc)
311	{
312	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pvUser;
313
314	LogFlowFunc(("pTask=%#p pvUser=%#p rc=%Rrc\n", pTask, pvUser, rc));
315
316	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH)
317	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, rc, true);
318	else
319	{
320	Assert((uint32_t)pTask->DataSeg.cbSeg == pTask->DataSeg.cbSeg && (int32_t)pTask->DataSeg.cbSeg >= 0);
321	uint32_t uOld = ASMAtomicSubS32(&pTaskFile->cbTransferLeft, (int32_t)pTask->DataSeg.cbSeg);
322
323	/* The first error will be returned. */
324	if (RT_FAILURE(rc))
325	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
326	#ifdef VBOX_WITH_DEBUGGER
327	else
328	{
329	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
330
331	/* Overwrite with injected error code. */
332	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_READ)
333	rc = ASMAtomicXchgS32(&pEpFile->rcReqRead, VINF_SUCCESS);
334	else
335	rc = ASMAtomicXchgS32(&pEpFile->rcReqWrite, VINF_SUCCESS);
336
337	if (RT_FAILURE(rc))
338	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
339	}
340	#endif
341
342	if (!(uOld - pTask->DataSeg.cbSeg)
343	&& !ASMAtomicXchgBool(&pTaskFile->fCompleted, true))
344	{
345	#ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
346	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
347	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEpFile->Core.pEpClass;
348
349	/* Check if we should delay completion of the request. */
350	if ( ASMAtomicReadU32(&pEpFile->msDelay) > 0
351	&& ASMAtomicReadU32(&pEpFile->cReqsDelay) > 0)
352	{
353	uint64_t tsDelay = pEpFile->msDelay;
354
355	if (pEpFile->msJitter)
356	tsDelay = (RTRandU32() % 100) > 50 ? pEpFile->msDelay + (RTRandU32() % pEpFile->msJitter)
357	: pEpFile->msDelay - (RTRandU32() % pEpFile->msJitter);
358	ASMAtomicDecU32(&pEpFile->cReqsDelay);
359
360	/* Arm the delay. */
361	pTaskFile->tsDelayEnd = RTTimeProgramMilliTS() + tsDelay;
362
363	/* Append to the list. */
364	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
365	do
366	{
367	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
368	pTaskFile->pDelayedNext = pHead;
369	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTaskFile, pHead));
370
371	if (tsDelay < pEpClassFile->cMilliesNext)
372	{
373	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, tsDelay);
374	TMTimerSetMillies(pVM, pEpClassFile->hTimer, tsDelay);
375	}
376
377	LogRel(("AIOMgr: Delaying request %#p for %u ms\n", pTaskFile, tsDelay));
378	}
379	else
380	#endif
381	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, pTaskFile->rc, true);
382	}
383	}
384	}
385
386	DECLINLINE(void) pdmacFileEpTaskInit(PPDMASYNCCOMPLETIONTASK pTask, size_t cbTransfer)
387	{
388	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
389
390	Assert((uint32_t)cbTransfer == cbTransfer && (int32_t)cbTransfer >= 0);
391	ASMAtomicWriteS32(&pTaskFile->cbTransferLeft, (int32_t)cbTransfer);
392	ASMAtomicWriteBool(&pTaskFile->fCompleted, false);
393	ASMAtomicWriteS32(&pTaskFile->rc, VINF_SUCCESS);
394	}
395
396	int pdmacFileEpTaskInitiate(PPDMASYNCCOMPLETIONTASK pTask,
397	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
398	PCRTSGSEG paSegments, size_t cSegments,
399	size_t cbTransfer, PDMACTASKFILETRANSFER enmTransfer)
400	{
401	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
402	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
403
404	Assert( (enmTransfer == PDMACTASKFILETRANSFER_READ)
405	\|\| (enmTransfer == PDMACTASKFILETRANSFER_WRITE));
406
407	for (size_t i = 0; i < cSegments; i++)
408	{
409	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
410	AssertPtr(pIoTask);
411
412	pIoTask->pEndpoint = pEpFile;
413	pIoTask->enmTransferType = enmTransfer;
414	pIoTask->Off = off;
415	pIoTask->DataSeg.cbSeg = paSegments[i].cbSeg;
416	pIoTask->DataSeg.pvSeg = paSegments[i].pvSeg;
417	pIoTask->pvUser = pTaskFile;
418	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
419
420	/* Send it off to the I/O manager. */
421	pdmacFileEpAddTask(pEpFile, pIoTask);
422	off += paSegments[i].cbSeg;
423	cbTransfer -= paSegments[i].cbSeg;
424	}
425
426	AssertMsg(!cbTransfer, ("Incomplete transfer %u bytes left\n", cbTransfer));
427
428	return VINF_AIO_TASK_PENDING;
429	}
430
431	/**
432	* Creates a new async I/O manager.
433	*
434	* @returns VBox status code.
435	* @param pEpClass Pointer to the endpoint class data.
436	* @param ppAioMgr Where to store the pointer to the new async I/O manager on success.
437	* @param enmMgrType Wanted manager type - can be overwritten by the global override.
438	*/
439	int pdmacFileAioMgrCreate(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass, PPPDMACEPFILEMGR ppAioMgr,
440	PDMACEPFILEMGRTYPE enmMgrType)
441	{
442	LogFlowFunc((": Entered\n"));
443
444	PPDMACEPFILEMGR pAioMgrNew;
445	int rc = MMR3HeapAllocZEx(pEpClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION, sizeof(PDMACEPFILEMGR), (void **)&pAioMgrNew);
446	if (RT_SUCCESS(rc))
447	{
448	if (enmMgrType < pEpClass->enmMgrTypeOverride)
449	pAioMgrNew->enmMgrType = enmMgrType;
450	else
451	pAioMgrNew->enmMgrType = pEpClass->enmMgrTypeOverride;
452
453	pAioMgrNew->msBwLimitExpired = RT_INDEFINITE_WAIT;
454
455	rc = RTSemEventCreate(&pAioMgrNew->EventSem);
456	if (RT_SUCCESS(rc))
457	{
458	rc = RTSemEventCreate(&pAioMgrNew->EventSemBlock);
459	if (RT_SUCCESS(rc))
460	{
461	rc = RTCritSectInit(&pAioMgrNew->CritSectBlockingEvent);
462	if (RT_SUCCESS(rc))
463	{
464	/* Init the rest of the manager. */
465	if (pAioMgrNew->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
466	rc = pdmacFileAioMgrNormalInit(pAioMgrNew);
467
468	if (RT_SUCCESS(rc))
469	{
470	pAioMgrNew->enmState = PDMACEPFILEMGRSTATE_RUNNING;
471
472	rc = RTThreadCreateF(&pAioMgrNew->Thread,
473	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
474	? pdmacFileAioMgrFailsafe
475	: pdmacFileAioMgrNormal,
476	pAioMgrNew,
477	0,
478	RTTHREADTYPE_IO,
479	0,
480	"AioMgr%d-%s", pEpClass->cAioMgrs,
481	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
482	? "F"
483	: "N");
484	if (RT_SUCCESS(rc))
485	{
486	/* Link it into the list. */
487	RTCritSectEnter(&pEpClass->CritSect);
488	pAioMgrNew->pNext = pEpClass->pAioMgrHead;
489	if (pEpClass->pAioMgrHead)
490	pEpClass->pAioMgrHead->pPrev = pAioMgrNew;
491	pEpClass->pAioMgrHead = pAioMgrNew;
492	pEpClass->cAioMgrs++;
493	RTCritSectLeave(&pEpClass->CritSect);
494
495	*ppAioMgr = pAioMgrNew;
496
497	Log(("PDMAC: Successfully created new file AIO Mgr {%s}\n", RTThreadGetName(pAioMgrNew->Thread)));
498	return VINF_SUCCESS;
499	}
500	pdmacFileAioMgrNormalDestroy(pAioMgrNew);
501	}
502	RTCritSectDelete(&pAioMgrNew->CritSectBlockingEvent);
503	}
504	RTSemEventDestroy(pAioMgrNew->EventSem);
505	}
506	RTSemEventDestroy(pAioMgrNew->EventSemBlock);
507	}
508	MMR3HeapFree(pAioMgrNew);
509	}
510
511	LogFlowFunc((": Leave rc=%Rrc\n", rc));
512
513	return rc;
514	}
515
516	/**
517	* Destroys a async I/O manager.
518	*
519	* @returns nothing.
520	* @param pEpClassFile Pointer to globals for the file endpoint class.
521	* @param pAioMgr The async I/O manager to destroy.
522	*/
523	static void pdmacFileAioMgrDestroy(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile, PPDMACEPFILEMGR pAioMgr)
524	{
525	int rc = pdmacFileAioMgrShutdown(pAioMgr);
526	AssertRC(rc);
527
528	/* Unlink from the list. */
529	rc = RTCritSectEnter(&pEpClassFile->CritSect);
530	AssertRC(rc);
531
532	PPDMACEPFILEMGR pPrev = pAioMgr->pPrev;
533	PPDMACEPFILEMGR pNext = pAioMgr->pNext;
534
535	if (pPrev)
536	pPrev->pNext = pNext;
537	else
538	pEpClassFile->pAioMgrHead = pNext;
539
540	if (pNext)
541	pNext->pPrev = pPrev;
542
543	pEpClassFile->cAioMgrs--;
544	rc = RTCritSectLeave(&pEpClassFile->CritSect);
545	AssertRC(rc);
546
547	/* Free the resources. */
548	RTCritSectDelete(&pAioMgr->CritSectBlockingEvent);
549	RTSemEventDestroy(pAioMgr->EventSem);
550	RTSemEventDestroy(pAioMgr->EventSemBlock);
551	if (pAioMgr->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
552	pdmacFileAioMgrNormalDestroy(pAioMgr);
553
554	MMR3HeapFree(pAioMgr);
555	}
556
557	static int pdmacFileMgrTypeFromName(const char *pszVal, PPDMACEPFILEMGRTYPE penmMgrType)
558	{
559	int rc = VINF_SUCCESS;
560
561	if (!RTStrCmp(pszVal, "Simple"))
562	*penmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
563	else if (!RTStrCmp(pszVal, "Async"))
564	*penmMgrType = PDMACEPFILEMGRTYPE_ASYNC;
565	else
566	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
567
568	return rc;
569	}
570
571	static const char *pdmacFileMgrTypeToName(PDMACEPFILEMGRTYPE enmMgrType)
572	{
573	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
574	return "Simple";
575	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
576	return "Async";
577
578	return NULL;
579	}
580
581	static int pdmacFileBackendTypeFromName(const char *pszVal, PPDMACFILEEPBACKEND penmBackendType)
582	{
583	int rc = VINF_SUCCESS;
584
585	if (!RTStrCmp(pszVal, "Buffered"))
586	*penmBackendType = PDMACFILEEPBACKEND_BUFFERED;
587	else if (!RTStrCmp(pszVal, "NonBuffered"))
588	*penmBackendType = PDMACFILEEPBACKEND_NON_BUFFERED;
589	else
590	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
591
592	return rc;
593	}
594
595	static const char *pdmacFileBackendTypeToName(PDMACFILEEPBACKEND enmBackendType)
596	{
597	if (enmBackendType == PDMACFILEEPBACKEND_BUFFERED)
598	return "Buffered";
599	if (enmBackendType == PDMACFILEEPBACKEND_NON_BUFFERED)
600	return "NonBuffered";
601
602	return NULL;
603	}
604
605	#ifdef VBOX_WITH_DEBUGGER
606
607	/**
608	* @callback_method_impl{FNDBGCCMD, The '.injecterror' command.}
609	*/
610	static DECLCALLBACK(int) pdmacEpFileErrorInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
611	{
612	/*
613	* Validate input.
614	*/
615	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
616	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 3);
617	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
618	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
619	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
620
621	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
622	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
623
624	/* Syntax is "read\|write <filename> <status code>" */
625	bool fWrite;
626	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
627	fWrite = false;
628	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
629	fWrite = true;
630	else
631	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
632
633	int32_t rcToInject = (int32_t)pArgs[2].u.u64Number;
634	if ((uint64_t)rcToInject != pArgs[2].u.u64Number)
635	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The status code '%lld' is out of range", pArgs[0].u.u64Number);
636
637	/*
638	* Search for the matching endpoint.
639	*/
640	RTCritSectEnter(&pEpClassFile->Core.CritSect);
641
642	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
643	while (pEpFile)
644	{
645	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
646	break;
647	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
648	}
649
650	if (pEpFile)
651	{
652	/*
653	* Do the job.
654	*/
655	if (fWrite)
656	ASMAtomicXchgS32(&pEpFile->rcReqWrite, rcToInject);
657	else
658	ASMAtomicXchgS32(&pEpFile->rcReqRead, rcToInject);
659
660	DBGCCmdHlpPrintf(pCmdHlp, "Injected %Rrc into '%s' for %s\n",
661	(int)rcToInject, pArgs[1].u.pszString, pArgs[0].u.pszString);
662	}
663
664	RTCritSectLeave(&pEpClassFile->Core.CritSect);
665
666	if (!pEpFile)
667	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
668	return VINF_SUCCESS;
669	}
670
671	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
672	/**
673	* @callback_method_impl{FNDBGCCMD, The '.injectdelay' command.}
674	*/
675	static DECLCALLBACK(int) pdmacEpFileDelayInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
676	{
677	/*
678	* Validate input.
679	*/
680	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
681	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs >= 3);
682	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
683	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
684	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
685
686	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
687	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
688
689	/* Syntax is "read\|write\|flush\|any <filename> <delay> [reqs]" */
690	PDMACFILEREQTYPEDELAY enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
691	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
692	enmDelayType = PDMACFILEREQTYPEDELAY_READ;
693	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
694	enmDelayType = PDMACFILEREQTYPEDELAY_WRITE;
695	else if (!RTStrCmp(pArgs[0].u.pszString, "flush"))
696	enmDelayType = PDMACFILEREQTYPEDELAY_FLUSH;
697	else if (!RTStrCmp(pArgs[0].u.pszString, "any"))
698	enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
699	else
700	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
701
702	uint32_t msDelay = (uint32_t)pArgs[2].u.u64Number;
703	if ((uint64_t)msDelay != pArgs[2].u.u64Number)
704	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The delay '%lld' is out of range", pArgs[0].u.u64Number);
705
706	uint32_t cReqsDelay = 1;
707	uint32_t msJitter = 0;
708	if (cArgs >= 4)
709	msJitter = (uint32_t)pArgs[3].u.u64Number;
710	if (cArgs == 5)
711	cReqsDelay = (uint32_t)pArgs[4].u.u64Number;
712
713	/*
714	* Search for the matching endpoint.
715	*/
716	RTCritSectEnter(&pEpClassFile->Core.CritSect);
717
718	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
719	while (pEpFile)
720	{
721	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
722	break;
723	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
724	}
725
726	if (pEpFile)
727	{
728	ASMAtomicWriteSize(&pEpFile->enmTypeDelay, enmDelayType);
729	ASMAtomicWriteU32(&pEpFile->msDelay, msDelay);
730	ASMAtomicWriteU32(&pEpFile->msJitter, msJitter);
731	ASMAtomicWriteU32(&pEpFile->cReqsDelay, cReqsDelay);
732
733	DBGCCmdHlpPrintf(pCmdHlp, "Injected delay for the next %u requests of %u ms into '%s' for %s\n",
734	cReqsDelay, msDelay, pArgs[1].u.pszString, pArgs[0].u.pszString);
735	}
736
737	RTCritSectLeave(&pEpClassFile->Core.CritSect);
738
739	if (!pEpFile)
740	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
741	return VINF_SUCCESS;
742	}
743
744	/**
745	* @callback_method_impl{FNTMTIMERINT, }
746	*/
747	static DECLCALLBACK(void) pdmacR3TimerCallback(PVM pVM, TMTIMERHANDLE hTimer, void *pvUser)
748	{
749	Assert(hTimer == pEpClassFile->hTimer);
750	uint64_t tsCur = RTTimeProgramMilliTS();
751	uint64_t cMilliesNext = UINT64_MAX;
752	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pvUser;
753
754	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, UINT64_MAX);
755
756	/* Go through all endpoints and check for expired requests. */
757	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
758
759	while (pEpFile)
760	{
761	/* Check for an expired delay. */
762	if (pEpFile->pDelayedHead != NULL)
763	{
764	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = ASMAtomicXchgPtrT(&pEpFile->pDelayedHead, NULL, PPDMASYNCCOMPLETIONTASKFILE);
765
766	while (pTaskFile)
767	{
768	PPDMASYNCCOMPLETIONTASKFILE pTmp = pTaskFile;
769	pTaskFile = pTaskFile->pDelayedNext;
770
771	if (tsCur >= pTmp->tsDelayEnd)
772	{
773	LogRel(("AIOMgr: Delayed request %#p completed\n", pTmp));
774	pdmR3AsyncCompletionCompleteTask(&pTmp->Core, pTmp->rc, true);
775	}
776	else
777	{
778	/* Prepend to the delayed list again. */
779	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
780
781	if (pTmp->tsDelayEnd - tsCur < cMilliesNext)
782	cMilliesNext = pTmp->tsDelayEnd - tsCur;
783
784	do
785	{
786	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
787	pTmp->pDelayedNext = pHead;
788	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTmp, pHead));
789	}
790	}
791	}
792
793	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
794	}
795
796	if (cMilliesNext < pEpClassFile->cMilliesNext)
797	{
798	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, cMilliesNext);
799	TMTimerSetMillies(pVM, hTimer, cMilliesNext);
800	}
801	}
802
803	# endif /* PDM_ASYNC_COMPLETION_FILE_WITH_DELAY */
804
805	#endif /* VBOX_WITH_DEBUGGER */
806
807	static DECLCALLBACK(int) pdmacFileInitialize(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals, PCFGMNODE pCfgNode)
808	{
809	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
810	RTFILEAIOLIMITS AioLimits; /** < Async I/O limitations. */
811
812	int rc = RTFileAioGetLimits(&AioLimits);
813	#ifdef DEBUG
814	if (RT_SUCCESS(rc) && RTEnvExist("VBOX_ASYNC_IO_FAILBACK"))
815	rc = VERR_ENV_VAR_NOT_FOUND;
816	#endif
817	if (RT_FAILURE(rc))
818	{
819	LogRel(("AIO: Async I/O manager not supported (rc=%Rrc). Falling back to simple manager\n", rc));
820	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_SIMPLE;
821	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_BUFFERED;
822	}
823	else
824	{
825	pEpClassFile->uBitmaskAlignment = AioLimits.cbBufferAlignment ? ~((RTR3UINTPTR)AioLimits.cbBufferAlignment - 1) : RTR3UINTPTR_MAX;
826	pEpClassFile->cReqsOutstandingMax = AioLimits.cReqsOutstandingMax;
827
828	if (pCfgNode)
829	{
830	/* Query the default manager type */
831	char *pszVal = NULL;
832	rc = CFGMR3QueryStringAllocDef(pCfgNode, "IoMgr", &pszVal, "Async");
833	AssertLogRelRCReturn(rc, rc);
834
835	rc = pdmacFileMgrTypeFromName(pszVal, &pEpClassFile->enmMgrTypeOverride);
836	MMR3HeapFree(pszVal);
837	if (RT_FAILURE(rc))
838	return rc;
839
840	LogRel(("AIOMgr: Default manager type is '%s'\n", pdmacFileMgrTypeToName(pEpClassFile->enmMgrTypeOverride)));
841
842	/* Query default backend type */
843	rc = CFGMR3QueryStringAllocDef(pCfgNode, "FileBackend", &pszVal, "NonBuffered");
844	AssertLogRelRCReturn(rc, rc);
845
846	rc = pdmacFileBackendTypeFromName(pszVal, &pEpClassFile->enmEpBackendDefault);
847	MMR3HeapFree(pszVal);
848	if (RT_FAILURE(rc))
849	return rc;
850
851	LogRel(("AIOMgr: Default file backend is '%s'\n", pdmacFileBackendTypeToName(pEpClassFile->enmEpBackendDefault)));
852
853	#ifdef RT_OS_LINUX
854	if ( pEpClassFile->enmMgrTypeOverride == PDMACEPFILEMGRTYPE_ASYNC
855	&& pEpClassFile->enmEpBackendDefault == PDMACFILEEPBACKEND_BUFFERED)
856	{
857	LogRel(("AIOMgr: Linux does not support buffered async I/O, changing to non buffered\n"));
858	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
859	}
860	#endif
861	}
862	else
863	{
864	/* No configuration supplied, set defaults */
865	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
866	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_ASYNC;
867	}
868	}
869
870	/* Init critical section. */
871	rc = RTCritSectInit(&pEpClassFile->CritSect);
872
873	#ifdef VBOX_WITH_DEBUGGER
874	/* Install the error injection handler. */
875	if (RT_SUCCESS(rc))
876	{
877	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
878	AssertRC(rc);
879	}
880
881	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
882	rc = TMR3TimerCreate(pEpClassFile->Core.pVM, TMCLOCK_REAL, pdmacR3TimerCallback, pEpClassFile,
883	TMTIMER_FLAGS_NO_RING0, "AC Delay", &pEpClassFile->hTimer);
884	AssertRC(rc);
885	pEpClassFile->cMilliesNext = UINT64_MAX;
886	# endif
887	#endif
888
889	return rc;
890	}
891
892	static DECLCALLBACK(void) pdmacFileTerminate(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals)
893	{
894	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
895
896	/* All endpoints should be closed at this point. */
897	AssertMsg(!pEpClassFile->Core.pEndpointsHead, ("There are still endpoints left\n"));
898
899	/* Destroy all left async I/O managers. */
900	while (pEpClassFile->pAioMgrHead)
901	pdmacFileAioMgrDestroy(pEpClassFile, pEpClassFile->pAioMgrHead);
902
903	RTCritSectDelete(&pEpClassFile->CritSect);
904	}
905
906	static DECLCALLBACK(int) pdmacFileEpInitialize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint,
907	const char *pszUri, uint32_t fFlags)
908	{
909	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
910	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
911	PDMACEPFILEMGRTYPE enmMgrType = pEpClassFile->enmMgrTypeOverride;
912	PDMACFILEEPBACKEND enmEpBackend = pEpClassFile->enmEpBackendDefault;
913
914	AssertMsgReturn((fFlags & ~(PDMACEP_FILE_FLAGS_READ_ONLY \| PDMACEP_FILE_FLAGS_DONT_LOCK \| PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)) == 0,
915	("PDMAsyncCompletion: Invalid flag specified\n"), VERR_INVALID_PARAMETER);
916
917	unsigned fFileFlags = RTFILE_O_OPEN;
918
919	/*
920	* Revert to the simple manager and the buffered backend if
921	* the host cache should be enabled.
922	*/
923	if (fFlags & PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)
924	{
925	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
926	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
927	}
928
929	if (fFlags & PDMACEP_FILE_FLAGS_READ_ONLY)
930	fFileFlags \|= RTFILE_O_READ \| RTFILE_O_DENY_NONE;
931	else
932	{
933	fFileFlags \|= RTFILE_O_READWRITE;
934
935	/*
936	* Opened in read/write mode. Check whether the caller wants to
937	* avoid the lock. Return an error in case caching is enabled
938	* because this can lead to data corruption.
939	*/
940	if (fFlags & PDMACEP_FILE_FLAGS_DONT_LOCK)
941	fFileFlags \|= RTFILE_O_DENY_NONE;
942	else
943	fFileFlags \|= RTFILE_O_DENY_WRITE;
944	}
945
946	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
947	fFileFlags \|= RTFILE_O_ASYNC_IO;
948
949	int rc;
950	if (enmEpBackend == PDMACFILEEPBACKEND_NON_BUFFERED)
951	{
952	/*
953	* We only disable the cache if the size of the file is a multiple of 512.
954	* Certain hosts like Windows, Linux and Solaris require that transfer sizes
955	* are aligned to the volume sector size.
956	* If not we just make sure that the data is written to disk with RTFILE_O_WRITE_THROUGH
957	* which will trash the host cache but ensures that the host cache will not
958	* contain dirty buffers.
959	*/
960	RTFILE hFile;
961	rc = RTFileOpen(&hFile, pszUri, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
962	if (RT_SUCCESS(rc))
963	{
964	uint64_t cbSize;
965
966	rc = RTFileQuerySize(hFile, &cbSize);
967
968	if (RT_SUCCESS(rc) && ((cbSize % 512) == 0))
969	fFileFlags \|= RTFILE_O_NO_CACHE;
970	else
971	{
972	/* Downgrade to the buffered backend */
973	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
974
975	#ifdef RT_OS_LINUX
976	fFileFlags &= ~RTFILE_O_ASYNC_IO;
977	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
978	#endif
979	}
980	RTFileClose(hFile);
981	}
982	}
983
984	/* Open with final flags. */
985	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
986	if ( rc == VERR_INVALID_FUNCTION
987	\|\| rc == VERR_INVALID_PARAMETER)
988	{
989	LogRel(("AIOMgr: pdmacFileEpInitialize: RTFileOpen %s / %08x failed with %Rrc\n",
990	pszUri, fFileFlags, rc));
991	/*
992	* Solaris doesn't support directio on ZFS so far. :-\
993	* Trying to enable it returns VERR_INVALID_FUNCTION
994	* (ENOTTY). Remove it and hope for the best.
995	* ZFS supports write throttling in case applications
996	* write more data than can be synced to the disk
997	* without blocking the whole application.
998	*
999	* On Linux we have the same problem with cifs.
1000	* Have to disable async I/O here too because it requires O_DIRECT.
1001	*/
1002	fFileFlags &= ~RTFILE_O_NO_CACHE;
1003	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
1004
1005	#ifdef RT_OS_LINUX
1006	fFileFlags &= ~RTFILE_O_ASYNC_IO;
1007	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
1008	#endif
1009
1010	/* Open again. */
1011	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
1012
1013	if (RT_FAILURE(rc))
1014	{
1015	LogRel(("AIOMgr: pdmacFileEpInitialize: RTFileOpen %s / %08x failed AGAIN(!) with %Rrc\n",
1016	pszUri, fFileFlags, rc));
1017	}
1018	}
1019
1020	if (RT_SUCCESS(rc))
1021	{
1022	pEpFile->fFlags = fFileFlags;
1023
1024	rc = RTFileQuerySize(pEpFile->hFile, (uint64_t *)&pEpFile->cbFile);
1025	if (RT_SUCCESS(rc))
1026	{
1027	/* Initialize the segment cache */
1028	rc = MMR3HeapAllocZEx(pEpClassFile->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
1029	sizeof(PDMACTASKFILE),
1030	(void **)&pEpFile->pTasksFreeHead);
1031	if (RT_SUCCESS(rc))
1032	{
1033	PPDMACEPFILEMGR pAioMgr = NULL;
1034
1035	pEpFile->pTasksFreeTail = pEpFile->pTasksFreeHead;
1036	pEpFile->cTasksCached = 0;
1037	pEpFile->enmBackendType = enmEpBackend;
1038	/*
1039	* Disable async flushes on Solaris for now.
1040	* They cause weird hangs which needs more investigations.
1041	*/
1042	#ifndef RT_OS_SOLARIS
1043	pEpFile->fAsyncFlushSupported = true;
1044	#else
1045	pEpFile->fAsyncFlushSupported = false;
1046	#endif
1047
1048	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1049	{
1050	/* Simple mode. Every file has its own async I/O manager. */
1051	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, PDMACEPFILEMGRTYPE_SIMPLE);
1052	}
1053	else
1054	{
1055	pAioMgr = pEpClassFile->pAioMgrHead;
1056
1057	/* Check for an idling manager of the same type */
1058	while (pAioMgr)
1059	{
1060	if (pAioMgr->enmMgrType == enmMgrType)
1061	break;
1062	pAioMgr = pAioMgr->pNext;
1063	}
1064
1065	if (!pAioMgr)
1066	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, enmMgrType);
1067	}
1068
1069	if (RT_SUCCESS(rc))
1070	{
1071	pEpFile->AioMgr.pTreeRangesLocked = (PAVLRFOFFTREE)RTMemAllocZ(sizeof(AVLRFOFFTREE));
1072	if (!pEpFile->AioMgr.pTreeRangesLocked)
1073	rc = VERR_NO_MEMORY;
1074	else
1075	{
1076	pEpFile->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE;
1077
1078	/* Assign the endpoint to the thread. */
1079	rc = pdmacFileAioMgrAddEndpoint(pAioMgr, pEpFile);
1080	if (RT_FAILURE(rc))
1081	{
1082	RTMemFree(pEpFile->AioMgr.pTreeRangesLocked);
1083	MMR3HeapFree(pEpFile->pTasksFreeHead);
1084	}
1085	}
1086	}
1087	else if (rc == VERR_FILE_AIO_INSUFFICIENT_EVENTS)
1088	{
1089	PUVM pUVM = VMR3GetUVM(pEpClassFile->Core.pVM);
1090	#if defined(RT_OS_LINUX)
1091	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1092	N_("Failed to create I/O manager for VM due to insufficient resources on the host. "
1093	"Either increase the amount of allowed events in /proc/sys/fs/aio-max-nr or enable "
1094	"the host I/O cache"));
1095	#else
1096	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1097	N_("Failed to create I/O manager for VM due to insufficient resources on the host. "
1098	"Enable the host I/O cache"));
1099	#endif
1100	}
1101	else
1102	{
1103	PUVM pUVM = VMR3GetUVM(pEpClassFile->Core.pVM);
1104	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1105	N_("Failed to create I/O manager for VM due to an unknown error"));
1106	}
1107	}
1108	}
1109
1110	if (RT_FAILURE(rc))
1111	RTFileClose(pEpFile->hFile);
1112	}
1113
1114	#ifdef VBOX_WITH_STATISTICS
1115	if (RT_SUCCESS(rc))
1116	{
1117	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatRead,
1118	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1119	STAMUNIT_TICKS_PER_CALL, "Time taken to read from the endpoint",
1120	"/PDM/AsyncCompletion/File/%s/%d/Read", RTPathFilename(pEpFile->Core.pszUri), pEpFile->Core.iStatId);
1121
1122	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatWrite,
1123	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1124	STAMUNIT_TICKS_PER_CALL, "Time taken to write to the endpoint",
1125	"/PDM/AsyncCompletion/File/%s/%d/Write", RTPathFilename(pEpFile->Core.pszUri), pEpFile->Core.iStatId);
1126	}
1127	#endif
1128
1129	if (RT_SUCCESS(rc))
1130	LogRel(("AIOMgr: Endpoint for file '%s' (flags %08x) created successfully\n", pszUri, pEpFile->fFlags));
1131
1132	return rc;
1133	}
1134
1135	static DECLCALLBACK(int) pdmacFileEpRangesLockedDestroy(PAVLRFOFFNODECORE pNode, void *pvUser)
1136	{
1137	NOREF(pNode); NOREF(pvUser);
1138	AssertMsgFailed(("The locked ranges tree should be empty at that point\n"));
1139	return VINF_SUCCESS;
1140	}
1141
1142	static DECLCALLBACK(int) pdmacFileEpClose(PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1143	{
1144	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1145	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
1146
1147	/* Make sure that all tasks finished for this endpoint. */
1148	int rc = pdmacFileAioMgrCloseEndpoint(pEpFile->pAioMgr, pEpFile);
1149	AssertRC(rc);
1150
1151	/*
1152	* If the async I/O manager is in failsafe mode this is the only endpoint
1153	* he processes and thus can be destroyed now.
1154	*/
1155	if (pEpFile->pAioMgr->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1156	pdmacFileAioMgrDestroy(pEpClassFile, pEpFile->pAioMgr);
1157
1158	/* Free cached tasks. */
1159	PPDMACTASKFILE pTask = pEpFile->pTasksFreeHead;
1160
1161	while (pTask)
1162	{
1163	PPDMACTASKFILE pTaskFree = pTask;
1164	pTask = pTask->pNext;
1165	MMR3HeapFree(pTaskFree);
1166	}
1167
1168	/* Destroy the locked ranges tree now. */
1169	RTAvlrFileOffsetDestroy(pEpFile->AioMgr.pTreeRangesLocked, pdmacFileEpRangesLockedDestroy, NULL);
1170	RTMemFree(pEpFile->AioMgr.pTreeRangesLocked);
1171	pEpFile->AioMgr.pTreeRangesLocked = NULL;
1172
1173	RTFileClose(pEpFile->hFile);
1174
1175	#ifdef VBOX_WITH_STATISTICS
1176	/* Not sure if this might be unnecessary because of similar statement in pdmR3AsyncCompletionStatisticsDeregister? */
1177	STAMR3DeregisterF(pEpClassFile->Core.pVM->pUVM, "/PDM/AsyncCompletion/File/%s/*", RTPathFilename(pEpFile->Core.pszUri));
1178	#endif
1179
1180	return VINF_SUCCESS;
1181	}
1182
1183	static DECLCALLBACK(int) pdmacFileEpRead(PPDMASYNCCOMPLETIONTASK pTask,
1184	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1185	PCRTSGSEG paSegments, size_t cSegments,
1186	size_t cbRead)
1187	{
1188	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1189
1190	LogFlowFunc(("pTask=%#p pEndpoint=%#p off=%RTfoff paSegments=%#p cSegments=%zu cbRead=%zu\n",
1191	pTask, pEndpoint, off, paSegments, cSegments, cbRead));
1192
1193	if (RT_UNLIKELY((uint64_t)off + cbRead > pEpFile->cbFile))
1194	return VERR_EOF;
1195
1196	STAM_PROFILE_ADV_START(&pEpFile->StatRead, Read);
1197	pdmacFileEpTaskInit(pTask, cbRead);
1198	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbRead,
1199	PDMACTASKFILETRANSFER_READ);
1200	STAM_PROFILE_ADV_STOP(&pEpFile->StatRead, Read);
1201
1202	return rc;
1203	}
1204
1205	static DECLCALLBACK(int) pdmacFileEpWrite(PPDMASYNCCOMPLETIONTASK pTask,
1206	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1207	PCRTSGSEG paSegments, size_t cSegments,
1208	size_t cbWrite)
1209	{
1210	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1211
1212	if (RT_UNLIKELY(pEpFile->fReadonly))
1213	return VERR_NOT_SUPPORTED;
1214
1215	STAM_PROFILE_ADV_START(&pEpFile->StatWrite, Write);
1216
1217	pdmacFileEpTaskInit(pTask, cbWrite);
1218
1219	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbWrite,
1220	PDMACTASKFILETRANSFER_WRITE);
1221
1222	STAM_PROFILE_ADV_STOP(&pEpFile->StatWrite, Write);
1223
1224	return rc;
1225	}
1226
1227	static DECLCALLBACK(int) pdmacFileEpFlush(PPDMASYNCCOMPLETIONTASK pTask,
1228	PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1229	{
1230	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1231	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
1232
1233	if (RT_UNLIKELY(pEpFile->fReadonly))
1234	return VERR_NOT_SUPPORTED;
1235
1236	pdmacFileEpTaskInit(pTask, 0);
1237
1238	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
1239	if (RT_UNLIKELY(!pIoTask))
1240	return VERR_NO_MEMORY;
1241
1242	pIoTask->pEndpoint = pEpFile;
1243	pIoTask->enmTransferType = PDMACTASKFILETRANSFER_FLUSH;
1244	pIoTask->pvUser = pTaskFile;
1245	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
1246	pdmacFileEpAddTask(pEpFile, pIoTask);
1247
1248	return VINF_AIO_TASK_PENDING;
1249	}
1250
1251	static DECLCALLBACK(int) pdmacFileEpGetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t *pcbSize)
1252	{
1253	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1254
1255	*pcbSize = ASMAtomicReadU64(&pEpFile->cbFile);
1256
1257	return VINF_SUCCESS;
1258	}
1259
1260	static DECLCALLBACK(int) pdmacFileEpSetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t cbSize)
1261	{
1262	int rc;
1263	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1264
1265	rc = RTFileSetSize(pEpFile->hFile, cbSize);
1266	if (RT_SUCCESS(rc))
1267	ASMAtomicWriteU64(&pEpFile->cbFile, cbSize);
1268
1269	return rc;
1270	}
1271
1272	const PDMASYNCCOMPLETIONEPCLASSOPS g_PDMAsyncCompletionEndpointClassFile =
1273	{
1274	/* u32Version */
1275	PDMAC_EPCLASS_OPS_VERSION,
1276	/* pcszName */
1277	"File",
1278	/* enmClassType */
1279	PDMASYNCCOMPLETIONEPCLASSTYPE_FILE,
1280	/* cbEndpointClassGlobal */
1281	sizeof(PDMASYNCCOMPLETIONEPCLASSFILE),
1282	/* cbEndpoint */
1283	sizeof(PDMASYNCCOMPLETIONENDPOINTFILE),
1284	/* cbTask */
1285	sizeof(PDMASYNCCOMPLETIONTASKFILE),
1286	/* pfnInitialize */
1287	pdmacFileInitialize,
1288	/* pfnTerminate */
1289	pdmacFileTerminate,
1290	/* pfnEpInitialize. */
1291	pdmacFileEpInitialize,
1292	/* pfnEpClose */
1293	pdmacFileEpClose,
1294	/* pfnEpRead */
1295	pdmacFileEpRead,
1296	/* pfnEpWrite */
1297	pdmacFileEpWrite,
1298	/* pfnEpFlush */
1299	pdmacFileEpFlush,
1300	/* pfnEpGetSize */
1301	pdmacFileEpGetSize,
1302	/* pfnEpSetSize */
1303	pdmacFileEpSetSize,
1304	/* u32VersionEnd */
1305	PDMAC_EPCLASS_OPS_VERSION
1306	};
1307

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

source: vbox/trunk/src/VBox/VMM/VMMR3/PDMAsyncCompletionFile.cpp@ 97188

Download in other formats: