PDMAsyncCompletionFile.cpp@ 26638

Last change on this file since 26638 was 26527, checked in by vboxsync, 15 years ago
dang
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 31.9 KB

Line
1	/* $Id: PDMAsyncCompletionFile.cpp 26527 2010-02-15 03:42:05Z vboxsync $ */
2	/** @file
3	* PDM Async I/O - Transport data asynchronous in R3 using EMT.
4	*/
5
6	/*
7	* Copyright (C) 2006-2009 Sun Microsystems, Inc.
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	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22
23	/*******************************************************************************
24	* Header Files *
25	*******************************************************************************/
26	#define LOG_GROUP LOG_GROUP_PDM_ASYNC_COMPLETION
27	#define RT_STRICT
28	//#define DEBUG
29	#include "PDMInternal.h"
30	#include <VBox/pdm.h>
31	#include <VBox/mm.h>
32	#include <VBox/vm.h>
33	#include <VBox/err.h>
34	#include <VBox/log.h>
35
36	#include <iprt/asm.h>
37	#include <iprt/assert.h>
38	#include <iprt/critsect.h>
39	#include <iprt/env.h>
40	#include <iprt/file.h>
41	#include <iprt/mem.h>
42	#include <iprt/semaphore.h>
43	#include <iprt/string.h>
44	#include <iprt/thread.h>
45	#include <iprt/path.h>
46
47	#include "PDMAsyncCompletionFileInternal.h"
48
49	/**
50	* Frees a task.
51	*
52	* @returns nothing.
53	* @param pEndpoint Pointer to the endpoint the segment was for.
54	* @param pTask The task to free.
55	*/
56	void pdmacFileTaskFree(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint,
57	PPDMACTASKFILE pTask)
58	{
59	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
60
61	LogFlowFunc((": pEndpoint=%p pTask=%p\n", pEndpoint, pTask));
62
63	/* Try the per endpoint cache first. */
64	if (pEndpoint->cTasksCached < pEpClass->cTasksCacheMax)
65	{
66	/* Add it to the list. */
67	pEndpoint->pTasksFreeTail->pNext = pTask;
68	pEndpoint->pTasksFreeTail = pTask;
69	ASMAtomicIncU32(&pEndpoint->cTasksCached);
70	}
71	else if (false)
72	{
73	/* Bigger class cache */
74	}
75	else
76	{
77	Log(("Freeing task %p because all caches are full\n", pTask));
78	MMR3HeapFree(pTask);
79	}
80	}
81
82	/**
83	* Allocates a task segment
84	*
85	* @returns Pointer to the new task segment or NULL
86	* @param pEndpoint Pointer to the endpoint
87	*/
88	PPDMACTASKFILE pdmacFileTaskAlloc(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
89	{
90	PPDMACTASKFILE pTask = NULL;
91
92	/* Try the small per endpoint cache first. */
93	if (pEndpoint->pTasksFreeHead == pEndpoint->pTasksFreeTail)
94	{
95	/* Try the bigger endpoint class cache. */
96	PPDMASYNCCOMPLETIONEPCLASSFILE pEndpointClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
97
98	#if 0
99	/* We start with the assigned slot id to distribute the load when allocating new tasks. */
100	unsigned iSlot = pEndpoint->iSlotStart;
101	do
102	{
103	pTask = (PPDMASYNCCOMPLETIONTASK)ASMAtomicXchgPtr((void * volatile *)&pEndpointClass->apTaskCache[iSlot], NULL);
104	if (pTask)
105	break;
106
107	iSlot = (iSlot + 1) % RT_ELEMENTS(pEndpointClass->apTaskCache);
108	} while (iSlot != pEndpoint->iSlotStart);
109	#endif
110	if (!pTask)
111	{
112	/*
113	* Allocate completely new.
114	* If this fails we return NULL.
115	*/
116	int rc = MMR3HeapAllocZEx(pEndpointClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
117	sizeof(PDMACTASKFILE),
118	(void **)&pTask);
119	if (RT_FAILURE(rc))
120	pTask = NULL;
121
122	LogFlow(("Allocated task %p\n", pTask));
123	}
124	#if 0
125	else
126	{
127	/* Remove the first element and put the rest into the slot again. */
128	PPDMASYNCCOMPLETIONTASK pTaskHeadNew = pTask->pNext;
129
130	pTaskHeadNew->pPrev = NULL;
131
132	/* Put back into the list adding any new tasks. */
133	while (true)
134	{
135	bool fChanged = ASMAtomicCmpXchgPtr((void * volatile *)&pEndpointClass->apTaskCache[iSlot], pTaskHeadNew, NULL);
136
137	if (fChanged)
138	break;
139
140	PPDMASYNCCOMPLETIONTASK pTaskHead = (PPDMASYNCCOMPLETIONTASK)ASMAtomicXchgPtr((void * volatile *)&pEndpointClass->apTaskCache[iSlot], NULL);
141
142	/* The new task could be taken inbetween */
143	if (pTaskHead)
144	{
145	/* Go to the end of the probably much shorter new list. */
146	PPDMASYNCCOMPLETIONTASK pTaskTail = pTaskHead;
147	while (pTaskTail->pNext)
148	pTaskTail = pTaskTail->pNext;
149
150	/* Concatenate */
151	pTaskTail->pNext = pTaskHeadNew;
152
153	pTaskHeadNew = pTaskHead;
154	}
155	/* Another round trying to change the list. */
156	}
157	/* We got a task from the global cache so decrement the counter */
158	ASMAtomicDecU32(&pEndpointClass->cTasksCached);
159	}
160	#endif
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	}
171
172	pTask->pNext = NULL;
173
174	return pTask;
175	}
176
177	PPDMACTASKFILE pdmacFileEpGetNewTasks(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
178	{
179	PPDMACTASKFILE pTasks = NULL;
180
181	/*
182	* Get pending tasks.
183	*/
184	pTasks = (PPDMACTASKFILE)ASMAtomicXchgPtr((void * volatile *)&pEndpoint->pTasksNewHead, NULL);
185
186	/* Reverse the list to process in FIFO order. */
187	if (pTasks)
188	{
189	PPDMACTASKFILE pTask = pTasks;
190
191	pTasks = NULL;
192
193	while (pTask)
194	{
195	PPDMACTASKFILE pCur = pTask;
196	pTask = pTask->pNext;
197	pCur->pNext = pTasks;
198	pTasks = pCur;
199	}
200	}
201
202	return pTasks;
203	}
204
205	static void pdmacFileAioMgrWakeup(PPDMACEPFILEMGR pAioMgr)
206	{
207	bool fWokenUp = ASMAtomicXchgBool(&pAioMgr->fWokenUp, true);
208
209	if (!fWokenUp)
210	{
211	int rc = VINF_SUCCESS;
212	bool fWaitingEventSem = ASMAtomicReadBool(&pAioMgr->fWaitingEventSem);
213
214	if (fWaitingEventSem)
215	rc = RTSemEventSignal(pAioMgr->EventSem);
216
217	AssertRC(rc);
218	}
219	}
220
221	static int pdmacFileAioMgrWaitForBlockingEvent(PPDMACEPFILEMGR pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT enmEvent)
222	{
223	int rc = VINF_SUCCESS;
224
225	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, enmEvent);
226	Assert(!pAioMgr->fBlockingEventPending);
227	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, true);
228
229	/* Wakeup the async I/O manager */
230	pdmacFileAioMgrWakeup(pAioMgr);
231
232	/* Wait for completion. */
233	rc = RTSemEventWait(pAioMgr->EventSemBlock, RT_INDEFINITE_WAIT);
234	AssertRC(rc);
235
236	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, false);
237	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, PDMACEPFILEAIOMGRBLOCKINGEVENT_INVALID);
238
239	return rc;
240	}
241
242	int pdmacFileAioMgrAddEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
243	{
244	int rc;
245
246	rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
247	AssertRCReturn(rc, rc);
248
249	ASMAtomicWritePtr((void * volatile *)&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint, pEndpoint);
250	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_ADD_ENDPOINT);
251
252	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
253
254	if (RT_SUCCESS(rc))
255	ASMAtomicWritePtr((void * volatile *)&pEndpoint->pAioMgr, pAioMgr);
256
257	return rc;
258	}
259
260	static int pdmacFileAioMgrRemoveEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
261	{
262	int rc;
263
264	rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
265	AssertRCReturn(rc, rc);
266
267	ASMAtomicWritePtr((void * volatile *)&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint, pEndpoint);
268	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_REMOVE_ENDPOINT);
269
270	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
271
272	return rc;
273	}
274
275	static int pdmacFileAioMgrCloseEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
276	{
277	int rc;
278
279	rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
280	AssertRCReturn(rc, rc);
281
282	ASMAtomicWritePtr((void * volatile *)&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint, pEndpoint);
283	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_CLOSE_ENDPOINT);
284
285	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
286
287	return rc;
288	}
289
290	static int pdmacFileAioMgrShutdown(PPDMACEPFILEMGR pAioMgr)
291	{
292	int rc;
293
294	rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
295	AssertRCReturn(rc, rc);
296
297	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_SHUTDOWN);
298
299	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
300
301	return rc;
302	}
303
304	int pdmacFileEpAddTask(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
305	{
306	PPDMACTASKFILE pNext;
307	do
308	{
309	pNext = pEndpoint->pTasksNewHead;
310	pTask->pNext = pNext;
311	} while (!ASMAtomicCmpXchgPtr((void * volatile )&pEndpoint->pTasksNewHead, (void )pTask, (void *)pNext));
312
313	pdmacFileAioMgrWakeup((PPDMACEPFILEMGR)ASMAtomicReadPtr((void * volatile *)&pEndpoint->pAioMgr));
314
315	return VINF_SUCCESS;
316	}
317
318	void pdmacFileEpTaskCompleted(PPDMACTASKFILE pTask, void *pvUser)
319	{
320	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pvUser;
321
322	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH)
323	{
324	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, true);
325	}
326	else
327	{
328	Assert((uint32_t)pTask->DataSeg.cbSeg == pTask->DataSeg.cbSeg && (int32_t)pTask->DataSeg.cbSeg >= 0);
329	uint32_t uOld = ASMAtomicSubS32(&pTaskFile->cbTransferLeft, (int32_t)pTask->DataSeg.cbSeg);
330
331	if (!(uOld - pTask->DataSeg.cbSeg)
332	&& !ASMAtomicXchgBool(&pTaskFile->fCompleted, true))
333	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, true);
334	}
335	}
336
337	int pdmacFileEpTaskInitiate(PPDMASYNCCOMPLETIONTASK pTask,
338	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
339	PCPDMDATASEG paSegments, size_t cSegments,
340	size_t cbTransfer, PDMACTASKFILETRANSFER enmTransfer)
341	{
342	int rc = VINF_SUCCESS;
343	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
344	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
345	PPDMACEPFILEMGR pAioMgr = pEpFile->pAioMgr;
346
347	Assert( (enmTransfer == PDMACTASKFILETRANSFER_READ)
348	\|\| (enmTransfer == PDMACTASKFILETRANSFER_WRITE));
349
350	Assert((uint32_t)cbTransfer == cbTransfer && (int32_t)cbTransfer >= 0);
351	ASMAtomicWriteS32(&pTaskFile->cbTransferLeft, (int32_t)cbTransfer);
352	ASMAtomicWriteBool(&pTaskFile->fCompleted, false);
353
354	for (unsigned i = 0; i < cSegments; i++)
355	{
356	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
357	AssertPtr(pIoTask);
358
359	pIoTask->pEndpoint = pEpFile;
360	pIoTask->enmTransferType = enmTransfer;
361	pIoTask->Off = off;
362	pIoTask->DataSeg.cbSeg = paSegments[i].cbSeg;
363	pIoTask->DataSeg.pvSeg = paSegments[i].pvSeg;
364	pIoTask->pvUser = pTaskFile;
365	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
366
367	/* Send it off to the I/O manager. */
368	pdmacFileEpAddTask(pEpFile, pIoTask);
369	off += paSegments[i].cbSeg;
370	cbTransfer -= paSegments[i].cbSeg;
371	}
372
373	AssertMsg(!cbTransfer, ("Incomplete transfer %u bytes left\n", cbTransfer));
374
375	if (ASMAtomicReadS32(&pTaskFile->cbTransferLeft) == 0
376	&& !ASMAtomicXchgBool(&pTaskFile->fCompleted, true))
377	pdmR3AsyncCompletionCompleteTask(pTask, false);
378	else
379	rc = VINF_AIO_TASK_PENDING;
380
381	return rc;
382	}
383
384	/**
385	* Creates a new async I/O manager.
386	*
387	* @returns VBox status code.
388	* @param pEpClass Pointer to the endpoint class data.
389	* @param ppAioMgr Where to store the pointer to the new async I/O manager on success.
390	* @param fFailsafe Flag to force a failsafe manager even if the global flag is not set.
391	*/
392	int pdmacFileAioMgrCreate(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass, PPPDMACEPFILEMGR ppAioMgr, bool fFailsafe)
393	{
394	int rc = VINF_SUCCESS;
395	PPDMACEPFILEMGR pAioMgrNew;
396
397	LogFlowFunc((": Entered\n"));
398
399	rc = MMR3HeapAllocZEx(pEpClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION, sizeof(PDMACEPFILEMGR), (void **)&pAioMgrNew);
400	if (RT_SUCCESS(rc))
401	{
402	pAioMgrNew->fFailsafe = fFailsafe \|\| pEpClass->fFailsafe;
403
404	rc = RTSemEventCreate(&pAioMgrNew->EventSem);
405	if (RT_SUCCESS(rc))
406	{
407	rc = RTSemEventCreate(&pAioMgrNew->EventSemBlock);
408	if (RT_SUCCESS(rc))
409	{
410	rc = RTCritSectInit(&pAioMgrNew->CritSectBlockingEvent);
411	if (RT_SUCCESS(rc))
412	{
413	/* Init the rest of the manager. */
414	if (!pAioMgrNew->fFailsafe)
415	rc = pdmacFileAioMgrNormalInit(pAioMgrNew);
416
417	if (RT_SUCCESS(rc))
418	{
419	pAioMgrNew->enmState = PDMACEPFILEMGRSTATE_RUNNING;
420
421	rc = RTThreadCreateF(&pAioMgrNew->Thread,
422	pAioMgrNew->fFailsafe
423	? pdmacFileAioMgrFailsafe
424	: pdmacFileAioMgrNormal,
425	pAioMgrNew,
426	0,
427	RTTHREADTYPE_IO,
428	0,
429	"AioMgr%d-%s", pEpClass->cAioMgrs,
430	pAioMgrNew->fFailsafe
431	? "F"
432	: "N");
433	if (RT_SUCCESS(rc))
434	{
435	/* Link it into the list. */
436	RTCritSectEnter(&pEpClass->CritSect);
437	pAioMgrNew->pNext = pEpClass->pAioMgrHead;
438	if (pEpClass->pAioMgrHead)
439	pEpClass->pAioMgrHead->pPrev = pAioMgrNew;
440	pEpClass->pAioMgrHead = pAioMgrNew;
441	pEpClass->cAioMgrs++;
442	RTCritSectLeave(&pEpClass->CritSect);
443
444	*ppAioMgr = pAioMgrNew;
445
446	Log(("PDMAC: Successfully created new file AIO Mgr {%s}\n", RTThreadGetName(pAioMgrNew->Thread)));
447	return VINF_SUCCESS;
448	}
449	pdmacFileAioMgrNormalDestroy(pAioMgrNew);
450	}
451	RTCritSectDelete(&pAioMgrNew->CritSectBlockingEvent);
452	}
453	RTSemEventDestroy(pAioMgrNew->EventSem);
454	}
455	RTSemEventDestroy(pAioMgrNew->EventSemBlock);
456	}
457	MMR3HeapFree(pAioMgrNew);
458	}
459
460	LogFlowFunc((": Leave rc=%Rrc\n", rc));
461
462	return rc;
463	}
464
465	/**
466	* Destroys a async I/O manager.
467	*
468	* @returns nothing.
469	* @param pAioMgr The async I/O manager to destroy.
470	*/
471	static void pdmacFileAioMgrDestroy(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile, PPDMACEPFILEMGR pAioMgr)
472	{
473	int rc = pdmacFileAioMgrShutdown(pAioMgr);
474	AssertRC(rc);
475
476	/* Unlink from the list. */
477	rc = RTCritSectEnter(&pEpClassFile->CritSect);
478	AssertRC(rc);
479
480	PPDMACEPFILEMGR pPrev = pAioMgr->pPrev;
481	PPDMACEPFILEMGR pNext = pAioMgr->pNext;
482
483	if (pPrev)
484	pPrev->pNext = pNext;
485	else
486	pEpClassFile->pAioMgrHead = pNext;
487
488	if (pNext)
489	pNext->pPrev = pPrev;
490
491	pEpClassFile->cAioMgrs--;
492
493	rc = RTCritSectLeave(&pEpClassFile->CritSect);
494	AssertRC(rc);
495
496	/* Free the ressources. */
497	RTCritSectDelete(&pAioMgr->CritSectBlockingEvent);
498	RTSemEventDestroy(pAioMgr->EventSem);
499	if (!pAioMgr->fFailsafe)
500	pdmacFileAioMgrNormalDestroy(pAioMgr);
501
502	MMR3HeapFree(pAioMgr);
503	}
504
505	static int pdmacFileInitialize(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals, PCFGMNODE pCfgNode)
506	{
507	int rc = VINF_SUCCESS;
508	RTFILEAIOLIMITS AioLimits; /** < Async I/O limitations. */
509
510	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
511
512	rc = RTFileAioGetLimits(&AioLimits);
513	#ifdef DEBUG
514	if (RT_SUCCESS(rc) && RTEnvExist("VBOX_ASYNC_IO_FAILBACK"))
515	rc = VERR_ENV_VAR_NOT_FOUND;
516	#endif
517	if (RT_FAILURE(rc))
518	{
519	LogRel(("AIO: Async I/O manager not supported (rc=%Rrc). Falling back to failsafe manager\n",
520	rc));
521	pEpClassFile->fFailsafe = true;
522	}
523	else
524	{
525	pEpClassFile->uBitmaskAlignment = AioLimits.cbBufferAlignment ? ~((RTR3UINTPTR)AioLimits.cbBufferAlignment - 1) : RTR3UINTPTR_MAX;
526	pEpClassFile->cReqsOutstandingMax = AioLimits.cReqsOutstandingMax;
527
528	/* The user can force the failsafe manager. */
529	rc = CFGMR3QueryBoolDef(pCfgNode, "UseFailsafeIo", &pEpClassFile->fFailsafe, false);
530	AssertLogRelRCReturn(rc, rc);
531
532	if (pEpClassFile->fFailsafe)
533	LogRel(("AIOMgr: Failsafe I/O was requested by user\n"));
534	}
535
536	/* Init critical section. */
537	rc = RTCritSectInit(&pEpClassFile->CritSect);
538	if (RT_SUCCESS(rc))
539	{
540	/* Check if the host cache should be used too. */
541	#ifndef RT_OS_LINUX
542	rc = CFGMR3QueryBoolDef(pCfgNode, "HostCacheEnabled", &pEpClassFile->fHostCacheEnabled, false);
543	AssertLogRelRCReturn(rc, rc);
544	#else
545	/*
546	* Host cache + async I/O is not supported on Linux. Check if the user enabled the cache,
547	* leave a warning and disable it always.
548	*/
549	bool fDummy;
550	rc = CFGMR3QueryBool(pCfgNode, "HostCacheEnabled", &fDummy);
551	if (RT_SUCCESS(rc))
552	LogRel(("AIOMgr: The host cache is not supported with async I/O on Linux\n"));
553
554	pEpClassFile->fHostCacheEnabled = false;
555	#endif
556
557	/* Check if the cache was disabled by the user. */
558	rc = CFGMR3QueryBoolDef(pCfgNode, "CacheEnabled", &pEpClassFile->fCacheEnabled, true);
559	AssertLogRelRCReturn(rc, rc);
560
561	if (pEpClassFile->fCacheEnabled)
562	{
563	/* Init cache structure */
564	rc = pdmacFileCacheInit(pEpClassFile, pCfgNode);
565	if (RT_FAILURE(rc))
566	{
567	RTCritSectDelete(&pEpClassFile->CritSect);
568	pEpClassFile->fCacheEnabled = false;
569	LogRel(("AIOMgr: Failed to initialise the cache (rc=%Rrc), disabled caching\n"));
570	}
571	}
572	else
573	LogRel(("AIOMgr: Cache was globally disabled\n"));
574	}
575
576	return rc;
577	}
578
579	static void pdmacFileTerminate(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals)
580	{
581	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
582
583	/* All endpoints should be closed at this point. */
584	AssertMsg(!pEpClassFile->Core.pEndpointsHead, ("There are still endpoints left\n"));
585
586	/* Destroy all left async I/O managers. */
587	while (pEpClassFile->pAioMgrHead)
588	pdmacFileAioMgrDestroy(pEpClassFile, pEpClassFile->pAioMgrHead);
589
590	/* Destroy the cache. */
591	if (pEpClassFile->fCacheEnabled)
592	pdmacFileCacheDestroy(pEpClassFile);
593
594	RTCritSectDelete(&pEpClassFile->CritSect);
595	}
596
597	static int pdmacFileEpInitialize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint,
598	const char *pszUri, uint32_t fFlags)
599	{
600	int rc = VINF_SUCCESS;
601	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
602	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
603	bool fUseFailsafeManager = pEpClassFile->fFailsafe;
604
605	AssertMsgReturn((fFlags & ~(PDMACEP_FILE_FLAGS_READ_ONLY \| PDMACEP_FILE_FLAGS_CACHING)) == 0,
606	("PDMAsyncCompletion: Invalid flag specified\n"), VERR_INVALID_PARAMETER);
607
608	unsigned fFileFlags = fFlags & PDMACEP_FILE_FLAGS_READ_ONLY
609	? RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE
610	: RTFILE_O_READWRITE \| RTFILE_O_OPEN \| RTFILE_O_DENY_WRITE;
611
612	if (!pEpClassFile->fFailsafe)
613	{
614	fFileFlags \|= (RTFILE_O_ASYNC_IO \| RTFILE_O_WRITE_THROUGH);
615
616	/*
617	* We only disable the cache if the size of the file is a multiple of 512.
618	* Certain hosts like Windows, Linux and Solaris require that transfer sizes
619	* are aligned to the volume sector size.
620	* If not we just make sure that the data is written to disk with RTFILE_O_WRITE_THROUGH
621	* which will trash the host cache but ensures that the host cache will not
622	* contain dirty buffers.
623	*/
624	RTFILE File = NIL_RTFILE;
625
626	rc = RTFileOpen(&File, pszUri, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
627	if (RT_SUCCESS(rc))
628	{
629	uint64_t cbSize;
630
631	rc = RTFileGetSize(File, &cbSize);
632	if (RT_SUCCESS(rc) && ((cbSize % 512) == 0))
633	{
634	fFileFlags &= ~RTFILE_O_WRITE_THROUGH;
635
636	#if defined(RT_OS_LINUX)
637	AssertMsg(!pEpClassFile->fHostCacheEnabled, ("Host cache + async I/O is not supported on Linux\n"));
638	fFileFlags \|= RTFILE_O_NO_CACHE;
639	#else
640	if (!pEpClassFile->fHostCacheEnabled)
641	fFileFlags \|= RTFILE_O_NO_CACHE;
642	#endif
643	}
644
645	pEpFile->cbFile = cbSize;
646
647	RTFileClose(File);
648	}
649	}
650
651	/* Open with final flags. */
652	rc = RTFileOpen(&pEpFile->File, pszUri, fFileFlags);
653	if ((rc == VERR_INVALID_FUNCTION) \|\| (rc == VERR_INVALID_PARAMETER))
654	{
655	LogRel(("pdmacFileEpInitialize: RTFileOpen %s / %08x failed with %Rrc\n",
656	pszUri, fFileFlags, rc));
657	/*
658	* Solaris doesn't support directio on ZFS so far. :-\
659	* Trying to enable it returns VERR_INVALID_FUNCTION
660	* (ENOTTY). Remove it and hope for the best.
661	* ZFS supports write throttling in case applications
662	* write more data than can be synced to the disk
663	* without blocking the whole application.
664	*
665	* On Linux we have the same problem with cifs.
666	* Have to disable async I/O here too because it requires O_DIRECT.
667	*/
668	fFileFlags &= ~RTFILE_O_NO_CACHE;
669
670	#ifdef RT_OS_LINUX
671	fFileFlags &= ~RTFILE_O_ASYNC_IO;
672	fUseFailsafeManager = true;
673	#endif
674
675	/* Open again. */
676	rc = RTFileOpen(&pEpFile->File, pszUri, fFileFlags);
677
678	if (RT_FAILURE(rc))
679	{
680	LogRel(("pdmacFileEpInitialize: RTFileOpen %s / %08x failed AGAIN(!) with %Rrc\n",
681	pszUri, fFileFlags, rc));
682	}
683	}
684
685	if (RT_SUCCESS(rc))
686	{
687	pEpFile->fFlags = fFileFlags;
688
689	rc = RTFileGetSize(pEpFile->File, (uint64_t *)&pEpFile->cbFile);
690	if (RT_SUCCESS(rc) && (pEpFile->cbFile == 0))
691	{
692	/* Could be a block device */
693	rc = RTFileSeek(pEpFile->File, 0, RTFILE_SEEK_END, (uint64_t *)&pEpFile->cbFile);
694	}
695
696	if (RT_SUCCESS(rc))
697	{
698	/* Initialize the segment cache */
699	rc = MMR3HeapAllocZEx(pEpClassFile->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
700	sizeof(PDMACTASKFILE),
701	(void **)&pEpFile->pTasksFreeHead);
702	if (RT_SUCCESS(rc))
703	{
704	PPDMACEPFILEMGR pAioMgr = NULL;
705
706	pEpFile->pTasksFreeTail = pEpFile->pTasksFreeHead;
707	pEpFile->cTasksCached = 0;
708
709	if (fUseFailsafeManager)
710	{
711	/* Safe mode. Every file has its own async I/O manager. */
712	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, true);
713	AssertRC(rc);
714	}
715	else
716	{
717	if ( (fFlags & PDMACEP_FILE_FLAGS_CACHING)
718	&& (pEpClassFile->fCacheEnabled))
719	{
720	pEpFile->fCaching = true;
721	rc = pdmacFileEpCacheInit(pEpFile, pEpClassFile);
722	if (RT_FAILURE(rc))
723	{
724	LogRel(("AIOMgr: Endpoint for \"%s\" was opened with caching but initializing cache failed. Disabled caching\n", pszUri));
725	pEpFile->fCaching = false;
726	}
727	}
728
729	pAioMgr = pEpClassFile->pAioMgrHead;
730
731	/* Check for an idling not failsafe one or create new if not found */
732	while (pAioMgr && pAioMgr->fFailsafe)
733	pAioMgr = pAioMgr->pNext;
734
735	if (!pAioMgr)
736	{
737	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, false);
738	AssertRC(rc);
739	}
740	}
741
742	pEpFile->AioMgr.pTreeRangesLocked = (PAVLRFOFFTREE)RTMemAllocZ(sizeof(AVLRFOFFTREE));
743	if (!pEpFile->AioMgr.pTreeRangesLocked)
744	rc = VERR_NO_MEMORY;
745	else
746	{
747	pEpFile->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE;
748
749	/* Assign the endpoint to the thread. */
750	rc = pdmacFileAioMgrAddEndpoint(pAioMgr, pEpFile);
751	if (RT_FAILURE(rc))
752	{
753	RTMemFree(pEpFile->AioMgr.pTreeRangesLocked);
754	MMR3HeapFree(pEpFile->pTasksFreeHead);
755	}
756	}
757	}
758	}
759
760	if (RT_FAILURE(rc))
761	RTFileClose(pEpFile->File);
762	}
763
764	#ifdef VBOX_WITH_STATISTICS
765	if (RT_SUCCESS(rc))
766	{
767	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatRead,
768	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
769	STAMUNIT_TICKS_PER_CALL, "Time taken to read from the endpoint",
770	"/PDM/AsyncCompletion/File/%s/Read", RTPathFilename(pEpFile->Core.pszUri));
771
772	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatWrite,
773	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
774	STAMUNIT_TICKS_PER_CALL, "Time taken to write to the endpoint",
775	"/PDM/AsyncCompletion/File/%s/Write", RTPathFilename(pEpFile->Core.pszUri));
776	}
777	#endif
778
779	return rc;
780	}
781
782	static int pdmacFileEpRangesLockedDestroy(PAVLRFOFFNODECORE pNode, void *pvUser)
783	{
784	AssertMsgFailed(("The locked ranges tree should be empty at that point\n"));
785	return VINF_SUCCESS;
786	}
787
788	static int pdmacFileEpClose(PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
789	{
790	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
791	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
792
793	/* Make sure that all tasks finished for this endpoint. */
794	int rc = pdmacFileAioMgrCloseEndpoint(pEpFile->pAioMgr, pEpFile);
795	AssertRC(rc);
796
797	/*
798	* If the async I/O manager is in failsafe mode this is the only endpoint
799	* he processes and thus can be destroyed now.
800	*/
801	if (pEpFile->pAioMgr->fFailsafe)
802	pdmacFileAioMgrDestroy(pEpClassFile, pEpFile->pAioMgr);
803
804	/* Free cached tasks. */
805	PPDMACTASKFILE pTask = pEpFile->pTasksFreeHead;
806
807	while (pTask)
808	{
809	PPDMACTASKFILE pTaskFree = pTask;
810	pTask = pTask->pNext;
811	MMR3HeapFree(pTaskFree);
812	}
813
814	/* Free the cached data. */
815	if (pEpFile->fCaching)
816	pdmacFileEpCacheDestroy(pEpFile);
817
818	/* Destroy the locked ranges tree now. */
819	RTAvlrFileOffsetDestroy(pEpFile->AioMgr.pTreeRangesLocked, pdmacFileEpRangesLockedDestroy, NULL);
820
821	RTFileClose(pEpFile->File);
822
823	#ifdef VBOX_WITH_STATISTICS
824	STAMR3Deregister(pEpClassFile->Core.pVM, &pEpFile->StatRead);
825	STAMR3Deregister(pEpClassFile->Core.pVM, &pEpFile->StatWrite);
826	#endif
827
828	return VINF_SUCCESS;
829	}
830
831	static int pdmacFileEpRead(PPDMASYNCCOMPLETIONTASK pTask,
832	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
833	PCPDMDATASEG paSegments, size_t cSegments,
834	size_t cbRead)
835	{
836	int rc = VINF_SUCCESS;
837	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
838
839	STAM_PROFILE_ADV_START(&pEpFile->StatRead, Read);
840
841	if (pEpFile->fCaching)
842	rc = pdmacFileEpCacheRead(pEpFile, (PPDMASYNCCOMPLETIONTASKFILE)pTask,
843	off, paSegments, cSegments, cbRead);
844	else
845	rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbRead,
846	PDMACTASKFILETRANSFER_READ);
847
848	STAM_PROFILE_ADV_STOP(&pEpFile->StatRead, Read);
849
850	return rc;
851	}
852
853	static int pdmacFileEpWrite(PPDMASYNCCOMPLETIONTASK pTask,
854	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
855	PCPDMDATASEG paSegments, size_t cSegments,
856	size_t cbWrite)
857	{
858	int rc = VINF_SUCCESS;
859	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
860
861	if (RT_UNLIKELY(pEpFile->fReadonly))
862	return VERR_NOT_SUPPORTED;
863
864	STAM_PROFILE_ADV_START(&pEpFile->StatWrite, Write);
865
866	if (pEpFile->fCaching)
867	rc = pdmacFileEpCacheWrite(pEpFile, (PPDMASYNCCOMPLETIONTASKFILE)pTask,
868	off, paSegments, cSegments, cbWrite);
869	else
870	rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbWrite,
871	PDMACTASKFILETRANSFER_WRITE);
872
873	STAM_PROFILE_ADV_STOP(&pEpFile->StatWrite, Write);
874
875	return rc;
876	}
877
878	static int pdmacFileEpFlush(PPDMASYNCCOMPLETIONTASK pTask,
879	PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
880	{
881	int rc = VINF_SUCCESS;
882	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
883	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
884
885	if (RT_UNLIKELY(pEpFile->fReadonly))
886	return VERR_NOT_SUPPORTED;
887
888	pTaskFile->cbTransferLeft = 0;
889
890	if (pEpFile->fCaching)
891	rc = pdmacFileEpCacheFlush(pEpFile, pTaskFile);
892	else
893	{
894	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
895	AssertPtr(pIoTask);
896
897	pIoTask->pEndpoint = pEpFile;
898	pIoTask->enmTransferType = PDMACTASKFILETRANSFER_FLUSH;
899	pIoTask->pvUser = pTaskFile;
900	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
901	pdmacFileEpAddTask(pEpFile, pIoTask);
902	rc = VINF_AIO_TASK_PENDING;
903	}
904
905	return rc;
906	}
907
908	static int pdmacFileEpGetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t *pcbSize)
909	{
910	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
911
912	*pcbSize = ASMAtomicReadU64(&pEpFile->cbFile);
913
914	return VINF_SUCCESS;
915	}
916
917	const PDMASYNCCOMPLETIONEPCLASSOPS g_PDMAsyncCompletionEndpointClassFile =
918	{
919	/* u32Version */
920	PDMAC_EPCLASS_OPS_VERSION,
921	/* pcszName */
922	"File",
923	/* enmClassType */
924	PDMASYNCCOMPLETIONEPCLASSTYPE_FILE,
925	/* cbEndpointClassGlobal */
926	sizeof(PDMASYNCCOMPLETIONEPCLASSFILE),
927	/* cbEndpoint */
928	sizeof(PDMASYNCCOMPLETIONENDPOINTFILE),
929	/* cbTask */
930	sizeof(PDMASYNCCOMPLETIONTASKFILE),
931	/* pfnInitialize */
932	pdmacFileInitialize,
933	/* pfnTerminate */
934	pdmacFileTerminate,
935	/* pfnEpInitialize. */
936	pdmacFileEpInitialize,
937	/* pfnEpClose */
938	pdmacFileEpClose,
939	/* pfnEpRead */
940	pdmacFileEpRead,
941	/* pfnEpWrite */
942	pdmacFileEpWrite,
943	/* pfnEpFlush */
944	pdmacFileEpFlush,
945	/* pfnEpGetSize */
946	pdmacFileEpGetSize,
947	/* u32VersionEnd */
948	PDMAC_EPCLASS_OPS_VERSION
949	};
950

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

source: vbox/trunk/src/VBox/VMM/PDMAsyncCompletionFile.cpp@ 26638

Download in other formats: