/**
* Executes one (or perhaps a few more) IO instruction(s).
*
* @returns VBox status code suitable for EM.
* @param pVM The cross context VM structure.
* @param pVCpu The cross context virtual CPU structure.
*/
static int emR3HmExecuteIOInstruction(PVM pVM, PVMCPU pVCpu)
{
PCPUMCTX pCtx = pVCpu->em.s.pCtx;
STAM_PROFILE_START(&pVCpu->em.s.StatIOEmu, a);
/*
* Try to restart the io instruction that was refused in ring-0.
*/
VBOXSTRICTRC rcStrict = HMR3RestartPendingIOInstr(pVM, pVCpu, pCtx);
if (IOM_SUCCESS(rcStrict))
{
STAM_COUNTER_INC(&pVCpu->em.s.CTX_SUFF(pStats)->StatIoRestarted);
STAM_PROFILE_STOP(&pVCpu->em.s.StatIOEmu, a);
return VBOXSTRICTRC_TODO(rcStrict); /* rip already updated. */
}
AssertMsgReturn(rcStrict == VERR_NOT_FOUND, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)),
RT_SUCCESS_NP(rcStrict) ? VERR_IPE_UNEXPECTED_INFO_STATUS : VBOXSTRICTRC_TODO(rcStrict));
/*
* Hand it over to the interpreter.
*/
rcStrict = IEMExecOne(pVCpu);
LogFlow(("emR3HmExecuteIOInstruction: %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
STAM_COUNTER_INC(&pVCpu->em.s.CTX_SUFF(pStats)->StatIoIem);
STAM_PROFILE_STOP(&pVCpu->em.s.StatIOEmu, a);
return VBOXSTRICTRC_TODO(rcStrict);
}
/**
* Initializes the tracing.
*
* @returns VBox status code
* @param pVM The cross context VM structure.
* @param cbEntry The trace entry size.
* @param cEntries The number of entries.
*/
static int dbgfR3TraceEnable(PVM pVM, uint32_t cbEntry, uint32_t cEntries)
{
/*
* Don't enable it twice.
*/
if (pVM->hTraceBufR3 != NIL_RTTRACEBUF)
return VERR_ALREADY_EXISTS;
/*
* Resolve default parameter values.
*/
int rc;
if (!cbEntry)
{
rc = CFGMR3QueryU32Def(CFGMR3GetChild(CFGMR3GetRoot(pVM), "DBGF"), "TraceBufEntrySize", &cbEntry, 128);
AssertRCReturn(rc, rc);
}
if (!cEntries)
{
rc = CFGMR3QueryU32Def(CFGMR3GetChild(CFGMR3GetRoot(pVM), "DBGF"), "TraceBufEntries", &cEntries, 4096);
AssertRCReturn(rc, rc);
}
/*
* Figure the required size.
*/
RTTRACEBUF hTraceBuf;
size_t cbBlock = 0;
rc = RTTraceBufCarve(&hTraceBuf, cEntries, cbEntry, 0 /*fFlags*/, NULL, &cbBlock);
if (rc != VERR_BUFFER_OVERFLOW)
{
AssertReturn(!RT_SUCCESS_NP(rc), VERR_IPE_UNEXPECTED_INFO_STATUS);
return rc;
}
/*
* Allocate a hyper heap block and carve a trace buffer out of it.
*
* Note! We ASSUME that the returned trace buffer handle has the same value
* as the heap block.
*/
cbBlock = RT_ALIGN_Z(cbBlock, PAGE_SIZE);
void *pvBlock;
rc = MMR3HyperAllocOnceNoRel(pVM, cbBlock, PAGE_SIZE, MM_TAG_DBGF, &pvBlock);
if (RT_FAILURE(rc))
return rc;
rc = RTTraceBufCarve(&hTraceBuf, cEntries, cbEntry, 0 /*fFlags*/, pvBlock, &cbBlock);
AssertRCReturn(rc, rc);
AssertRelease(hTraceBuf == (RTTRACEBUF)pvBlock);
AssertRelease((void *)hTraceBuf == pvBlock);
pVM->hTraceBufR3 = hTraceBuf;
pVM->hTraceBufR0 = MMHyperCCToR0(pVM, hTraceBuf);
pVM->hTraceBufRC = MMHyperCCToRC(pVM, hTraceBuf);
return VINF_SUCCESS;
}
RTDECL(bool) RTSymlinkIsDangling(const char *pszSymlink)
{
bool fRc = false;
RTFSOBJINFO ObjInfo;
int rc = RTPathQueryInfoEx(pszSymlink, &ObjInfo, RTFSOBJATTRADD_NOTHING, RTPATH_F_ON_LINK);
if (RT_SUCCESS(rc))
{
fRc = RTFS_IS_SYMLINK(ObjInfo.Attr.fMode);
if (fRc)
{
rc = RTPathQueryInfoEx(pszSymlink, &ObjInfo, RTFSOBJATTRADD_NOTHING, RTPATH_F_FOLLOW_LINK);
fRc = !RT_SUCCESS_NP(rc);
}
}
LogFlow(("RTSymlinkIsDangling(%p={%s}): returns %RTbool\n", pszSymlink, pszSymlink, fRc));
return fRc;
}
RTDECL(bool) RTManifestEntryExists(RTMANIFEST hManifest, const char *pszEntry)
{
RTMANIFESTINT *pThis = hManifest;
AssertPtrReturn(pThis, false);
AssertReturn(pThis->u32Magic == RTMANIFEST_MAGIC, false);
AssertPtr(pszEntry);
bool fNeedNormalization;
size_t cchEntry;
int rc = rtManifestValidateNameEntry(pszEntry, &fNeedNormalization, &cchEntry);
AssertRCReturn(rc, false);
/*
* Check if it exists.
*/
PRTMANIFESTENTRY pEntry;
rc = rtManifestGetEntry(pThis, pszEntry, fNeedNormalization, cchEntry, &pEntry);
return RT_SUCCESS_NP(rc);
}
/**
* Executes one (or perhaps a few more) IO instruction(s).
*
* @returns VBox status code suitable for EM.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
static int emR3ExecuteIOInstruction(PVM pVM, PVMCPU pVCpu)
{
PCPUMCTX pCtx = pVCpu->em.s.pCtx;
STAM_PROFILE_START(&pVCpu->em.s.StatIOEmu, a);
/* Try to restart the io instruction that was refused in ring-0. */
VBOXSTRICTRC rcStrict = HWACCMR3RestartPendingIOInstr(pVM, pVCpu, pCtx);
if (IOM_SUCCESS(rcStrict))
{
STAM_COUNTER_INC(&pVCpu->em.s.CTX_SUFF(pStats)->StatIoRestarted);
STAM_PROFILE_STOP(&pVCpu->em.s.StatIOEmu, a);
return VBOXSTRICTRC_TODO(rcStrict); /* rip already updated. */
}
AssertMsgReturn(rcStrict == VERR_NOT_FOUND, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)),
RT_SUCCESS_NP(rcStrict) ? VERR_IPE_UNEXPECTED_INFO_STATUS : VBOXSTRICTRC_TODO(rcStrict));
/** @todo probably we should fall back to the recompiler; otherwise we'll go back and forth between HC & GC
* as io instructions tend to come in packages of more than one
*/
DISCPUSTATE Cpu;
int rc2 = CPUMR3DisasmInstrCPU(pVM, pVCpu, pCtx, pCtx->rip, &Cpu, "IO EMU");
if (RT_SUCCESS(rc2))
{
rcStrict = VINF_EM_RAW_EMULATE_INSTR;
if (!(Cpu.fPrefix & (DISPREFIX_REP | DISPREFIX_REPNE)))
{
switch (Cpu.pCurInstr->uOpcode)
{
case OP_IN:
{
STAM_COUNTER_INC(&pVCpu->em.s.CTX_SUFF(pStats)->StatIn);
rcStrict = IOMInterpretIN(pVM, CPUMCTX2CORE(pCtx), &Cpu);
break;
}
case OP_OUT:
{
STAM_COUNTER_INC(&pVCpu->em.s.CTX_SUFF(pStats)->StatOut);
rcStrict = IOMInterpretOUT(pVM, CPUMCTX2CORE(pCtx), &Cpu);
break;
}
}
}
else if (Cpu.fPrefix & DISPREFIX_REP)
{
switch (Cpu.pCurInstr->uOpcode)
{
case OP_INSB:
case OP_INSWD:
{
STAM_COUNTER_INC(&pVCpu->em.s.CTX_SUFF(pStats)->StatIn);
rcStrict = IOMInterpretINS(pVM, CPUMCTX2CORE(pCtx), &Cpu);
break;
}
case OP_OUTSB:
case OP_OUTSWD:
{
STAM_COUNTER_INC(&pVCpu->em.s.CTX_SUFF(pStats)->StatOut);
rcStrict = IOMInterpretOUTS(pVM, CPUMCTX2CORE(pCtx), &Cpu);
break;
}
}
}
/*
* Handled the I/O return codes.
* (The unhandled cases end up with rcStrict == VINF_EM_RAW_EMULATE_INSTR.)
*/
if (IOM_SUCCESS(rcStrict))
{
pCtx->rip += Cpu.cbInstr;
STAM_PROFILE_STOP(&pVCpu->em.s.StatIOEmu, a);
return VBOXSTRICTRC_TODO(rcStrict);
}
if (rcStrict == VINF_EM_RAW_GUEST_TRAP)
{
/* The active trap will be dispatched. */
Assert(TRPMHasTrap(pVCpu));
STAM_PROFILE_STOP(&pVCpu->em.s.StatIOEmu, a);
return VINF_SUCCESS;
}
AssertMsg(rcStrict != VINF_TRPM_XCPT_DISPATCHED, ("Handle VINF_TRPM_XCPT_DISPATCHED\n"));
if (RT_FAILURE(rcStrict))
{
STAM_PROFILE_STOP(&pVCpu->em.s.StatIOEmu, a);
return VBOXSTRICTRC_TODO(rcStrict);
}
AssertMsg(rcStrict == VINF_EM_RAW_EMULATE_INSTR || rcStrict == VINF_EM_RESCHEDULE_REM, ("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
}
STAM_PROFILE_STOP(&pVCpu->em.s.StatIOEmu, a);
return emR3ExecuteInstruction(pVM, pVCpu, "IO: ");
}
RTDECL(int) RTFileAioCtxWait(RTFILEAIOCTX hAioCtx, size_t cMinReqs, RTMSINTERVAL cMillies,
PRTFILEAIOREQ pahReqs, size_t cReqs, uint32_t *pcReqs)
{
int rc = VINF_SUCCESS;
int cRequestsCompleted = 0;
PRTFILEAIOCTXINTERNAL pCtxInt = (PRTFILEAIOCTXINTERNAL)hAioCtx;
struct timespec Timeout;
struct timespec *pTimeout = NULL;
uint64_t StartNanoTS = 0;
LogFlowFunc(("hAioCtx=%#p cMinReqs=%zu cMillies=%u pahReqs=%#p cReqs=%zu pcbReqs=%#p\n",
hAioCtx, cMinReqs, cMillies, pahReqs, cReqs, pcReqs));
/* Check parameters. */
AssertPtrReturn(pCtxInt, VERR_INVALID_HANDLE);
AssertPtrReturn(pcReqs, VERR_INVALID_POINTER);
AssertPtrReturn(pahReqs, VERR_INVALID_POINTER);
AssertReturn(cReqs != 0, VERR_INVALID_PARAMETER);
AssertReturn(cReqs >= cMinReqs, VERR_OUT_OF_RANGE);
rtFileAioCtxDump(pCtxInt);
int32_t cRequestsWaiting = ASMAtomicReadS32(&pCtxInt->cRequests);
if ( RT_UNLIKELY(cRequestsWaiting <= 0)
&& !(pCtxInt->fFlags & RTFILEAIOCTX_FLAGS_WAIT_WITHOUT_PENDING_REQUESTS))
return VERR_FILE_AIO_NO_REQUEST;
if (RT_UNLIKELY(cMinReqs > (uint32_t)cRequestsWaiting))
return VERR_INVALID_PARAMETER;
if (cMillies != RT_INDEFINITE_WAIT)
{
Timeout.tv_sec = cMillies / 1000;
Timeout.tv_nsec = (cMillies % 1000) * 1000000;
pTimeout = &Timeout;
StartNanoTS = RTTimeNanoTS();
}
/* Wait for at least one. */
if (!cMinReqs)
cMinReqs = 1;
/* For the wakeup call. */
Assert(pCtxInt->hThreadWait == NIL_RTTHREAD);
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, RTThreadSelf());
/* Update the waiting list once before we enter the loop. */
rc = rtFileAioCtxProcessEvents(pCtxInt);
while ( cMinReqs
&& RT_SUCCESS_NP(rc))
{
#ifdef RT_STRICT
if (RT_UNLIKELY(!pCtxInt->iFirstFree))
{
for (unsigned i = 0; i < pCtxInt->cReqsWaitMax; i++)
RTAssertMsg2Weak("wait[%d] = %#p\n", i, pCtxInt->apReqs[i]);
AssertMsgFailed(("No request to wait for. pReqsWaitHead=%#p pReqsWaitTail=%#p\n",
pCtxInt->pReqsWaitHead, pCtxInt->pReqsWaitTail));
}
#endif
LogFlow(("Waiting for %d requests to complete\n", pCtxInt->iFirstFree));
rtFileAioCtxDump(pCtxInt);
ASMAtomicXchgBool(&pCtxInt->fWaiting, true);
int rcPosix = aio_suspend((const struct aiocb * const *)pCtxInt->apReqs,
pCtxInt->iFirstFree, pTimeout);
ASMAtomicXchgBool(&pCtxInt->fWaiting, false);
if (rcPosix < 0)
{
LogFlow(("aio_suspend failed %d nent=%u\n", errno, pCtxInt->iFirstFree));
/* Check that this is an external wakeup event. */
if (errno == EINTR)
rc = rtFileAioCtxProcessEvents(pCtxInt);
else
rc = RTErrConvertFromErrno(errno);
}
else
{
/* Requests finished. */
unsigned iReqCurr = 0;
unsigned cDone = 0;
/* Remove completed requests from the waiting list. */
while ( (iReqCurr < pCtxInt->iFirstFree)
&& (cDone < cReqs))
{
PRTFILEAIOREQINTERNAL pReq = pCtxInt->apReqs[iReqCurr];
int rcReq = aio_error(&pReq->AioCB);
if (rcReq != EINPROGRESS)
{
/* Completed store the return code. */
if (rcReq == 0)
{
pReq->Rc = VINF_SUCCESS;
/* Call aio_return() to free resources. */
pReq->cbTransfered = aio_return(&pReq->AioCB);
}
else
{
#if defined(RT_OS_DARWIN) || defined(RT_OS_FREEBSD)
pReq->Rc = RTErrConvertFromErrno(errno);
#else
pReq->Rc = RTErrConvertFromErrno(rcReq);
#endif
}
/* Mark the request as finished. */
RTFILEAIOREQ_SET_STATE(pReq, COMPLETED);
cDone++;
/* If there are other entries waiting put the head into the now free entry. */
if (pCtxInt->pReqsWaitHead)
{
PRTFILEAIOREQINTERNAL pReqInsert = pCtxInt->pReqsWaitHead;
pCtxInt->pReqsWaitHead = pReqInsert->pNext;
if (!pCtxInt->pReqsWaitHead)
{
/* List is empty now. Clear tail too. */
pCtxInt->pReqsWaitTail = NULL;
}
pReqInsert->iWaitingList = pReq->iWaitingList;
pCtxInt->apReqs[pReqInsert->iWaitingList] = pReqInsert;
iReqCurr++;
}
else
{
/*
* Move the last entry into the current position to avoid holes
* but only if it is not the last element already.
*/
if (pReq->iWaitingList < pCtxInt->iFirstFree - 1)
{
pCtxInt->apReqs[pReq->iWaitingList] = pCtxInt->apReqs[--pCtxInt->iFirstFree];
pCtxInt->apReqs[pReq->iWaitingList]->iWaitingList = pReq->iWaitingList;
}
else
pCtxInt->iFirstFree--;
pCtxInt->apReqs[pCtxInt->iFirstFree] = NULL;
}
/* Put the request into the completed list. */
pahReqs[cRequestsCompleted++] = pReq;
pReq->iWaitingList = RTFILEAIOCTX_WAIT_ENTRY_INVALID;
}
else
iReqCurr++;
}
AssertMsg((cDone <= cReqs), ("Overflow cReqs=%u cMinReqs=%u cDone=%u\n",
cReqs, cDone));
cReqs -= cDone;
cMinReqs = RT_MAX(cMinReqs, cDone) - cDone;
ASMAtomicSubS32(&pCtxInt->cRequests, cDone);
AssertMsg(pCtxInt->cRequests >= 0, ("Finished more requests than currently active\n"));
if (!cMinReqs)
break;
if (cMillies != RT_INDEFINITE_WAIT)
{
uint64_t TimeDiff;
/* Recalculate the timeout. */
TimeDiff = RTTimeSystemNanoTS() - StartNanoTS;
Timeout.tv_sec = Timeout.tv_sec - (TimeDiff / 1000000);
Timeout.tv_nsec = Timeout.tv_nsec - (TimeDiff % 1000000);
}
/* Check for new elements. */
rc = rtFileAioCtxProcessEvents(pCtxInt);
}
}
*pcReqs = cRequestsCompleted;
Assert(pCtxInt->hThreadWait == RTThreadSelf());
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, NIL_RTTHREAD);
rtFileAioCtxDump(pCtxInt);
return rc;
}
RTDECL(int) RTFileAioCtxSubmit(RTFILEAIOCTX hAioCtx, PRTFILEAIOREQ pahReqs, size_t cReqs)
{
int rc = VINF_SUCCESS;
PRTFILEAIOCTXINTERNAL pCtxInt = hAioCtx;
/* Parameter checks */
AssertPtrReturn(pCtxInt, VERR_INVALID_HANDLE);
AssertReturn(cReqs != 0, VERR_INVALID_POINTER);
AssertPtrReturn(pahReqs, VERR_INVALID_PARAMETER);
rtFileAioCtxDump(pCtxInt);
/* Check that we don't exceed the limit */
if (ASMAtomicUoReadS32(&pCtxInt->cRequests) + cReqs > pCtxInt->cMaxRequests)
return VERR_FILE_AIO_LIMIT_EXCEEDED;
PRTFILEAIOREQINTERNAL pHead = NULL;
do
{
int rcPosix = 0;
size_t cReqsSubmit = 0;
size_t i = 0;
PRTFILEAIOREQINTERNAL pReqInt;
while ( (i < cReqs)
&& (i < AIO_LISTIO_MAX))
{
pReqInt = pahReqs[i];
if (RTFILEAIOREQ_IS_NOT_VALID(pReqInt))
{
/* Undo everything and stop submitting. */
for (size_t iUndo = 0; iUndo < i; iUndo++)
{
pReqInt = pahReqs[iUndo];
RTFILEAIOREQ_SET_STATE(pReqInt, PREPARED);
pReqInt->pCtxInt = NULL;
/* Unlink from the list again. */
PRTFILEAIOREQINTERNAL pNext, pPrev;
pNext = pReqInt->pNext;
pPrev = pReqInt->pPrev;
if (pNext)
pNext->pPrev = pPrev;
if (pPrev)
pPrev->pNext = pNext;
else
pHead = pNext;
}
rc = VERR_INVALID_HANDLE;
break;
}
pReqInt->pCtxInt = pCtxInt;
if (pReqInt->fFlush)
break;
/* Link them together. */
pReqInt->pNext = pHead;
if (pHead)
pHead->pPrev = pReqInt;
pReqInt->pPrev = NULL;
pHead = pReqInt;
RTFILEAIOREQ_SET_STATE(pReqInt, SUBMITTED);
cReqsSubmit++;
i++;
}
if (cReqsSubmit)
{
rcPosix = lio_listio(LIO_NOWAIT, (struct aiocb **)pahReqs, cReqsSubmit, NULL);
if (RT_UNLIKELY(rcPosix < 0))
{
size_t cReqsSubmitted = cReqsSubmit;
if (errno == EAGAIN)
rc = VERR_FILE_AIO_INSUFFICIENT_RESSOURCES;
else
rc = RTErrConvertFromErrno(errno);
/* Check which ones were not submitted. */
for (i = 0; i < cReqsSubmit; i++)
{
pReqInt = pahReqs[i];
rcPosix = aio_error(&pReqInt->AioCB);
if ((rcPosix != EINPROGRESS) && (rcPosix != 0))
{
cReqsSubmitted--;
#if defined(RT_OS_DARWIN) || defined(RT_OS_FREEBSD)
if (errno == EINVAL)
#else
if (rcPosix == EINVAL)
#endif
{
/* Was not submitted. */
RTFILEAIOREQ_SET_STATE(pReqInt, PREPARED);
}
else
{
/* An error occurred. */
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
/*
* Looks like Apple and glibc interpret the standard in different ways.
* glibc returns the error code which would be in errno but Apple returns
* -1 and sets errno to the appropriate value
*/
#if defined(RT_OS_DARWIN) || defined(RT_OS_FREEBSD)
Assert(rcPosix == -1);
pReqInt->Rc = RTErrConvertFromErrno(errno);
#elif defined(RT_OS_LINUX)
pReqInt->Rc = RTErrConvertFromErrno(rcPosix);
#endif
pReqInt->cbTransfered = 0;
}
/* Unlink from the list. */
PRTFILEAIOREQINTERNAL pNext, pPrev;
pNext = pReqInt->pNext;
pPrev = pReqInt->pPrev;
if (pNext)
pNext->pPrev = pPrev;
if (pPrev)
pPrev->pNext = pNext;
else
pHead = pNext;
pReqInt->pNext = NULL;
pReqInt->pPrev = NULL;
}
}
ASMAtomicAddS32(&pCtxInt->cRequests, cReqsSubmitted);
AssertMsg(pCtxInt->cRequests >= 0, ("Adding requests resulted in overflow\n"));
break;
}
ASMAtomicAddS32(&pCtxInt->cRequests, cReqsSubmit);
AssertMsg(pCtxInt->cRequests >= 0, ("Adding requests resulted in overflow\n"));
cReqs -= cReqsSubmit;
pahReqs += cReqsSubmit;
}
/*
* Check if we have a flush request now.
* If not we hit the AIO_LISTIO_MAX limit
* and will continue submitting requests
* above.
*/
if (cReqs && RT_SUCCESS_NP(rc))
{
pReqInt = pahReqs[0];
if (pReqInt->fFlush)
{
/*
* lio_listio does not work with flush requests so
* we have to use aio_fsync directly.
*/
rcPosix = aio_fsync(O_SYNC, &pReqInt->AioCB);
if (RT_UNLIKELY(rcPosix < 0))
{
if (errno == EAGAIN)
{
rc = VERR_FILE_AIO_INSUFFICIENT_RESSOURCES;
RTFILEAIOREQ_SET_STATE(pReqInt, PREPARED);
}
else
{
rc = RTErrConvertFromErrno(errno);
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
pReqInt->Rc = rc;
}
pReqInt->cbTransfered = 0;
break;
}
/* Link them together. */
pReqInt->pNext = pHead;
if (pHead)
pHead->pPrev = pReqInt;
pReqInt->pPrev = NULL;
pHead = pReqInt;
RTFILEAIOREQ_SET_STATE(pReqInt, SUBMITTED);
ASMAtomicIncS32(&pCtxInt->cRequests);
AssertMsg(pCtxInt->cRequests >= 0, ("Adding requests resulted in overflow\n"));
cReqs--;
pahReqs++;
}
}
} while ( cReqs
&& RT_SUCCESS_NP(rc));
if (pHead)
{
/*
* Forward successfully submitted requests to the thread waiting for requests.
* We search for a free slot first and if we don't find one
* we will grab the first one and append our list to the existing entries.
*/
unsigned iSlot = 0;
while ( (iSlot < RT_ELEMENTS(pCtxInt->apReqsNewHead))
&& !ASMAtomicCmpXchgPtr(&pCtxInt->apReqsNewHead[iSlot], pHead, NULL))
iSlot++;
if (iSlot == RT_ELEMENTS(pCtxInt->apReqsNewHead))
{
/* Nothing found. */
PRTFILEAIOREQINTERNAL pOldHead = ASMAtomicXchgPtrT(&pCtxInt->apReqsNewHead[0], NULL, PRTFILEAIOREQINTERNAL);
/* Find the end of the current head and link the old list to the current. */
PRTFILEAIOREQINTERNAL pTail = pHead;
while (pTail->pNext)
pTail = pTail->pNext;
pTail->pNext = pOldHead;
ASMAtomicWritePtr(&pCtxInt->apReqsNewHead[0], pHead);
}
/* Set the internal wakeup flag and wakeup the thread if possible. */
bool fWokenUp = ASMAtomicXchgBool(&pCtxInt->fWokenUpInternal, true);
if (!fWokenUp)
rtFileAioCtxWakeup(pCtxInt);
}
rtFileAioCtxDump(pCtxInt);
return rc;
}
RTDECL(int) RTFileAioCtxWait(RTFILEAIOCTX hAioCtx, size_t cMinReqs, RTMSINTERVAL cMillies,
PRTFILEAIOREQ pahReqs, size_t cReqs, uint32_t *pcReqs)
{
int rc = VINF_SUCCESS;
int cRequestsCompleted = 0;
/*
* Validate the parameters, making sure to always set pcReqs.
*/
AssertPtrReturn(pcReqs, VERR_INVALID_POINTER);
*pcReqs = 0; /* always set */
PRTFILEAIOCTXINTERNAL pCtxInt = hAioCtx;
RTFILEAIOCTX_VALID_RETURN(pCtxInt);
AssertPtrReturn(pahReqs, VERR_INVALID_POINTER);
AssertReturn(cReqs != 0, VERR_INVALID_PARAMETER);
AssertReturn(cReqs >= cMinReqs, VERR_OUT_OF_RANGE);
if (RT_UNLIKELY(ASMAtomicReadS32(&pCtxInt->cRequests) == 0))
return VERR_FILE_AIO_NO_REQUEST;
/*
* Convert the timeout if specified.
*/
struct timespec *pTimeout = NULL;
struct timespec Timeout = {0,0};
uint64_t StartNanoTS = 0;
if (cMillies != RT_INDEFINITE_WAIT)
{
Timeout.tv_sec = cMillies / 1000;
Timeout.tv_nsec = cMillies % 1000 * 1000000;
pTimeout = &Timeout;
StartNanoTS = RTTimeNanoTS();
}
/* Wait for at least one. */
if (!cMinReqs)
cMinReqs = 1;
/* For the wakeup call. */
Assert(pCtxInt->hThreadWait == NIL_RTTHREAD);
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, RTThreadSelf());
while ( cMinReqs
&& RT_SUCCESS_NP(rc))
{
struct kevent aKEvents[AIO_MAXIMUM_REQUESTS_PER_CONTEXT];
int cRequestsToWait = cMinReqs < AIO_MAXIMUM_REQUESTS_PER_CONTEXT ? cReqs : AIO_MAXIMUM_REQUESTS_PER_CONTEXT;
int rcBSD;
uint64_t StartTime;
ASMAtomicXchgBool(&pCtxInt->fWaiting, true);
rcBSD = kevent(pCtxInt->iKQueue, NULL, 0, aKEvents, cRequestsToWait, pTimeout);
ASMAtomicXchgBool(&pCtxInt->fWaiting, false);
if (RT_UNLIKELY(rcBSD < 0))
{
rc = RTErrConvertFromErrno(errno);
break;
}
uint32_t const cDone = rcBSD;
/* Process received events. */
for (uint32_t i = 0; i < cDone; i++)
{
PRTFILEAIOREQINTERNAL pReqInt = (PRTFILEAIOREQINTERNAL)aKEvents[i].udata;
AssertPtr(pReqInt);
Assert(pReqInt->u32Magic == RTFILEAIOREQ_MAGIC);
/*
* Retrieve the status code here already because the
* user may omit the RTFileAioReqGetRC() call and
* we will leak kernel resources then.
* This will result in errors during submission
* of other requests as soon as the max_aio_queue_per_proc
* limit is reached.
*/
int cbTransfered = aio_return(&pReqInt->AioCB);
if (cbTransfered < 0)
{
pReqInt->Rc = RTErrConvertFromErrno(cbTransfered);
pReqInt->cbTransfered = 0;
}
else
{
pReqInt->Rc = VINF_SUCCESS;
pReqInt->cbTransfered = cbTransfered;
}
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
pahReqs[cRequestsCompleted++] = (RTFILEAIOREQ)pReqInt;
}
/*
* Done Yet? If not advance and try again.
*/
if (cDone >= cMinReqs)
break;
cMinReqs -= cDone;
cReqs -= cDone;
if (cMillies != RT_INDEFINITE_WAIT)
{
/* The API doesn't return ETIMEDOUT, so we have to fix that ourselves. */
uint64_t NanoTS = RTTimeNanoTS();
uint64_t cMilliesElapsed = (NanoTS - StartNanoTS) / 1000000;
if (cMilliesElapsed >= cMillies)
{
rc = VERR_TIMEOUT;
break;
}
/* The syscall supposedly updates it, but we're paranoid. :-) */
Timeout.tv_sec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) / 1000;
Timeout.tv_nsec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) % 1000 * 1000000;
}
}
/*
* Update the context state and set the return value.
*/
*pcReqs = cRequestsCompleted;
ASMAtomicSubS32(&pCtxInt->cRequests, cRequestsCompleted);
Assert(pCtxInt->hThreadWait == RTThreadSelf());
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, NIL_RTTHREAD);
/*
* Clear the wakeup flag and set rc.
*/
if ( pCtxInt->fWokenUp
&& RT_SUCCESS(rc))
{
ASMAtomicXchgBool(&pCtxInt->fWokenUp, false);
rc = VERR_INTERRUPTED;
}
return rc;
}
RTDECL(int) RTFileAioCtxSubmit(RTFILEAIOCTX hAioCtx, PRTFILEAIOREQ pahReqs, size_t cReqs)
{
/*
* Parameter validation.
*/
int rc = VINF_SUCCESS;
PRTFILEAIOCTXINTERNAL pCtxInt = hAioCtx;
RTFILEAIOCTX_VALID_RETURN(pCtxInt);
AssertReturn(cReqs > 0, VERR_INVALID_PARAMETER);
AssertPtrReturn(pahReqs, VERR_INVALID_POINTER);
do
{
int rcBSD = 0;
size_t cReqsSubmit = 0;
size_t i = 0;
PRTFILEAIOREQINTERNAL pReqInt;
while ( (i < cReqs)
&& (i < AIO_LISTIO_MAX))
{
pReqInt = pahReqs[i];
if (RTFILEAIOREQ_IS_NOT_VALID(pReqInt))
{
/* Undo everything and stop submitting. */
for (size_t iUndo = 0; iUndo < i; iUndo++)
{
pReqInt = pahReqs[iUndo];
RTFILEAIOREQ_SET_STATE(pReqInt, PREPARED);
pReqInt->pCtxInt = NULL;
pReqInt->AioCB.aio_sigevent.sigev_notify_kqueue = 0;
}
rc = VERR_INVALID_HANDLE;
break;
}
pReqInt->AioCB.aio_sigevent.sigev_notify_kqueue = pCtxInt->iKQueue;
pReqInt->pCtxInt = pCtxInt;
RTFILEAIOREQ_SET_STATE(pReqInt, SUBMITTED);
if (pReqInt->fFlush)
break;
cReqsSubmit++;
i++;
}
if (cReqsSubmit)
{
rcBSD = lio_listio(LIO_NOWAIT, (struct aiocb **)pahReqs, cReqsSubmit, NULL);
if (RT_UNLIKELY(rcBSD < 0))
{
if (errno == EAGAIN)
rc = VERR_FILE_AIO_INSUFFICIENT_RESSOURCES;
else
rc = RTErrConvertFromErrno(errno);
/* Check which requests got actually submitted and which not. */
for (i = 0; i < cReqs; i++)
{
pReqInt = pahReqs[i];
rcBSD = aio_error(&pReqInt->AioCB);
if ( rcBSD == -1
&& errno == EINVAL)
{
/* Was not submitted. */
RTFILEAIOREQ_SET_STATE(pReqInt, PREPARED);
pReqInt->pCtxInt = NULL;
}
else if (rcBSD != EINPROGRESS)
{
/* The request encountered an error. */
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
pReqInt->Rc = RTErrConvertFromErrno(rcBSD);
pReqInt->pCtxInt = NULL;
pReqInt->cbTransfered = 0;
}
}
break;
}
ASMAtomicAddS32(&pCtxInt->cRequests, cReqsSubmit);
cReqs -= cReqsSubmit;
pahReqs += cReqsSubmit;
}
/* Check if we have a flush request now. */
if (cReqs && RT_SUCCESS_NP(rc))
{
pReqInt = pahReqs[0];
RTFILEAIOREQ_VALID_RETURN(pReqInt);
if (pReqInt->fFlush)
{
/*
* lio_listio does not work with flush requests so
* we have to use aio_fsync directly.
*/
rcBSD = aio_fsync(O_SYNC, &pReqInt->AioCB);
if (RT_UNLIKELY(rcBSD < 0))
{
if (rcBSD == EAGAIN)
{
/* Was not submitted. */
RTFILEAIOREQ_SET_STATE(pReqInt, PREPARED);
pReqInt->pCtxInt = NULL;
return VERR_FILE_AIO_INSUFFICIENT_RESSOURCES;
}
else
{
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
pReqInt->Rc = RTErrConvertFromErrno(errno);
pReqInt->cbTransfered = 0;
return pReqInt->Rc;
}
}
ASMAtomicIncS32(&pCtxInt->cRequests);
cReqs--;
pahReqs++;
}
}
} while (cReqs);
return rc;
}
RTDECL(int) RTFileAioCtxWait(RTFILEAIOCTX hAioCtx, size_t cMinReqs, RTMSINTERVAL cMillies,
PRTFILEAIOREQ pahReqs, size_t cReqs, uint32_t *pcReqs)
{
int rc = VINF_SUCCESS;
int cRequestsCompleted = 0;
/*
* Validate the parameters, making sure to always set pcReqs.
*/
AssertPtrReturn(pcReqs, VERR_INVALID_POINTER);
*pcReqs = 0; /* always set */
PRTFILEAIOCTXINTERNAL pCtxInt = hAioCtx;
RTFILEAIOCTX_VALID_RETURN(pCtxInt);
AssertPtrReturn(pahReqs, VERR_INVALID_POINTER);
AssertReturn(cReqs != 0, VERR_INVALID_PARAMETER);
AssertReturn(cReqs >= cMinReqs, VERR_OUT_OF_RANGE);
if ( RT_UNLIKELY(ASMAtomicReadS32(&pCtxInt->cRequests) == 0)
&& !(pCtxInt->fFlags & RTFILEAIOCTX_FLAGS_WAIT_WITHOUT_PENDING_REQUESTS))
return VERR_FILE_AIO_NO_REQUEST;
/*
* Convert the timeout if specified.
*/
struct timespec *pTimeout = NULL;
struct timespec Timeout = {0,0};
uint64_t StartNanoTS = 0;
if (cMillies != RT_INDEFINITE_WAIT)
{
Timeout.tv_sec = cMillies / 1000;
Timeout.tv_nsec = cMillies % 1000 * 1000000;
pTimeout = &Timeout;
StartNanoTS = RTTimeNanoTS();
}
/* Wait for at least one. */
if (!cMinReqs)
cMinReqs = 1;
while ( cMinReqs
&& RT_SUCCESS_NP(rc))
{
port_event_t aPortEvents[AIO_MAXIMUM_REQUESTS_PER_CONTEXT];
uint_t cRequests = cMinReqs;
int cRequestsToWait = RT_MIN(cReqs, AIO_MAXIMUM_REQUESTS_PER_CONTEXT);
int rcSol;
uint64_t StartTime;
rcSol = port_getn(pCtxInt->iPort, &aPortEvents[0], cRequestsToWait, &cRequests, pTimeout);
if (RT_UNLIKELY(rcSol < 0))
rc = RTErrConvertFromErrno(errno);
/* Process received events. */
for (uint_t i = 0; i < cRequests; i++)
{
if (aPortEvents[i].portev_source == PORT_SOURCE_ALERT)
{
Assert(aPortEvents[i].portev_events == AIO_CONTEXT_WAKEUP_EVENT);
rc = VERR_INTERRUPTED; /* We've got interrupted. */
/* Reset the port. */
port_alert(pCtxInt->iPort, PORT_ALERT_SET, 0, NULL);
}
else
{
PRTFILEAIOREQINTERNAL pReqInt = (PRTFILEAIOREQINTERNAL)aPortEvents[i].portev_user;
AssertPtr(pReqInt);
Assert(pReqInt->u32Magic == RTFILEAIOREQ_MAGIC);
/* A request has finished. */
pahReqs[cRequestsCompleted++] = pReqInt;
/* Mark the request as finished. */
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
}
}
/*
* Done Yet? If not advance and try again.
*/
if (cRequests >= cMinReqs)
break;
cMinReqs -= cRequests;
cReqs -= cRequests;
if (cMillies != RT_INDEFINITE_WAIT)
{
uint64_t NanoTS = RTTimeNanoTS();
uint64_t cMilliesElapsed = (NanoTS - StartNanoTS) / 1000000;
/* The syscall supposedly updates it, but we're paranoid. :-) */
if (cMilliesElapsed < cMillies)
{
Timeout.tv_sec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) / 1000;
Timeout.tv_nsec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) % 1000 * 1000000;
}
else
{
Timeout.tv_sec = 0;
Timeout.tv_nsec = 0;
}
}
}
/*
* Update the context state and set the return value.
*/
*pcReqs = cRequestsCompleted;
ASMAtomicSubS32(&pCtxInt->cRequests, cRequestsCompleted);
return rc;
}
/**
* Attempts to start the service, creating it if necessary.
*
* @returns VBox status code.
*/
static int suplibOsStartService(void)
{
/*
* Check if the driver service is there.
*/
SC_HANDLE hSMgr = OpenSCManager(NULL, NULL, SERVICE_QUERY_STATUS | SERVICE_START);
if (hSMgr == NULL)
{
DWORD dwErr = GetLastError();
AssertMsgFailed(("couldn't open service manager in SERVICE_QUERY_CONFIG | SERVICE_QUERY_STATUS mode! (dwErr=%d)\n", dwErr));
return RTErrConvertFromWin32(dwErr);
}
/*
* Try open our service to check it's status.
*/
SC_HANDLE hService = OpenService(hSMgr, SERVICE_NAME, SERVICE_QUERY_STATUS | SERVICE_START);
if (!hService)
{
/*
* Create the service.
*/
int rc = suplibOsCreateService();
if (RT_FAILURE(rc))
return rc;
/*
* Try open the service.
*/
hService = OpenService(hSMgr, SERVICE_NAME, SERVICE_QUERY_STATUS | SERVICE_START);
}
/*
* Check if open and on demand create succeeded.
*/
int rc;
if (hService)
{
/*
* Query service status to see if we need to start it or not.
*/
SERVICE_STATUS Status;
BOOL fRc = QueryServiceStatus(hService, &Status);
Assert(fRc);
if (Status.dwCurrentState == SERVICE_RUNNING)
rc = VINF_ALREADY_INITIALIZED;
else
{
if (Status.dwCurrentState == SERVICE_START_PENDING)
rc = VINF_SUCCESS;
else
{
/*
* Start it.
*/
if (StartService(hService, 0, NULL))
rc = VINF_SUCCESS;
else
{
DWORD dwErr = GetLastError();
AssertMsg(fRc, ("StartService failed with dwErr=%Rwa\n", dwErr));
rc = RTErrConvertFromWin32(dwErr);
}
}
/*
* Wait for the service to finish starting.
* We'll wait for 10 seconds then we'll give up.
*/
QueryServiceStatus(hService, &Status);
if (Status.dwCurrentState == SERVICE_START_PENDING)
{
int iWait;
for (iWait = 100; iWait > 0 && Status.dwCurrentState == SERVICE_START_PENDING; iWait--)
{
Sleep(100);
QueryServiceStatus(hService, &Status);
}
DWORD dwErr = GetLastError(); NOREF(dwErr);
AssertMsg(Status.dwCurrentState != SERVICE_RUNNING,
("Failed to start. dwErr=%Rwa iWait=%d status=%d\n", dwErr, iWait, Status.dwCurrentState));
}
if (Status.dwCurrentState == SERVICE_RUNNING)
rc = VINF_SUCCESS;
else if (RT_SUCCESS_NP(rc))
rc = VERR_GENERAL_FAILURE;
}
/*
* Close open handles.
*/
CloseServiceHandle(hService);
}
else
{
DWORD dwErr = GetLastError();
AssertMsgFailed(("OpenService failed! LastError=%Rwa\n", dwErr));
rc = RTErrConvertFromWin32(dwErr);
}
if (!CloseServiceHandle(hSMgr))
AssertFailed();
return rc;
}
