Filter Hook
in [Device Drivers]
|
Prev: PacketStackSize
Next: Loading winusb.sys in Window XP
From: hanzhu on 7 Aug 2006 04:10 OK, Seems I got your problem. For XP and the later, the callback routine is been invoked at DISPATCH_LEVEL. And you have to return the policy from your callback routine when it returns. However, just as what have been mentioned by the folks, you couldn't wait at DISPATCH_LEVEL. If you really have to keep such a queue for the packets, then the IP filter driver is not suitable for your needs since you have to process all the packets in the callback routine without any wait action(KeWaitFor...)and return it to the tcpip driver immediately when the callback routine returns. Why not consider the NDISIM driver? It's supported by MS and I'm convinced it will fit for your requirement. x856256(a)yahoo.com ?: > I tried at the beginning to process the packets on the same thread, but > it caused the system to crash every time. > Since I found it very hard to program in kernel-mode, and even harder > to debug it, I wanted to let a user-mode program process the packet and > wait till it finishes in order to return. Was that a bad idea? If it is > a bad idea, it means I should do all the processing in the driver, and > on the same thread. In such a scenario I will have to synchronize > somewhere because my packet process depends on other packets that were > previously processed (so I must have a synchronized queue of packets) > and again I have the same problem as the first one (or do I ?) > > Thanks, > Guy > > hanzhu wrote: >> IIRC, IP filter driver should process these packets through the callback >> routine instead of submitting the IRP to the lower tcpip protocol >> driver. Why do you need to pass the IRPs and wait for its completion? >> >> >> x856256(a)yahoo.com ?: >>> Hello, >>> >>> Either queueing the job or doing it asynchronously I will have to do >>> the job in another thread, and wait for the process of the packet to >>> finish (in order to return the drop, forward or pass reply), and this >>> is exactly where my problem began, when I tried to use the >>> KeWaitForSingleObject function. >>> So let me ask this again: How can I use this wait function, or any >>> other substitution that will allow me process the packet on another >>> thread (queued or not) and wait for it to finish processing in order to >>> return the value? >>> >>> Thanks, >>> Guy >>> >>> Doron Holan [MS] wrote: >>>> you can't reduce IRQL inline, you can only raise it and then lower it. if >>>> you truly want to synchronously process soemthing, then queue it to a work >>>> item. if you want your filter to not negatively affect performance, learn >>>> how to process the results in the completion routine asynchronously >>>> >>>> d >>>> >>>> -- >>>> Please do not send e-mail directly to this alias. this alias is for >>>> newsgroup purposes only. >>>> This posting is provided "AS IS" with no warranties, and confers no rights. >>>> >>>> >>>> <x856256(a)yahoo.com> wrote in message >>>> news:1154873993.596484.241370(a)i42g2000cwa.googlegroups.com... >>>>> No, >>>>> >>>>> The code is not taking any spinlocks. >>>>> Anyone has an idea how can I reduce the IRQL level inside the filter >>>>> hook function? >>>>> >>>>> Thanks, >>>>> Guy >>>>> >>>>> Don Burn wrote: >>>>>> You cannot call a lot of functions at DISPATCH_LEVEL. I am not well >>>>>> versed >>>>>> in filter hooks, so you need to determine if the raising to DISPATCH is >>>>>> avoidable. For instance, if the code takes a spinlock, could it take an >>>>>> ERESOURCE instead. >>>>>> >>>>>> >>>>>> -- >>>>>> Don Burn (MVP, Windows DDK) >>>>>> Windows 2k/XP/2k3 Filesystem and Driver Consulting >>>>>> http://www.windrvr.com >>>>>> Remove StopSpam from the email to reply >>>>>> >>>>>> >>>>>> >>>>>> <x856256(a)yahoo.com> wrote in message >>>>>> news:1154861486.219608.122660(a)m73g2000cwd.googlegroups.com... >>>>>>> Hello and thanks Don for the answer. >>>>>>> >>>>>>> It is running in IRQL_DISPATCH, I will try to reduce it and see what >>>>>>> happens. >>>>>>> >>>>>>> Thanks, >>>>>>> Guy >>>>>>> >>>>>>> Don Burn wrote: >>>>>>>> A filter hook driver should not have an interrupt handler it is only >>>>>>>> concerned with packets as they are presented to it. You should not >>>>>>>> try >>>>>>>> to >>>>>>>> use IoConnectInterrupt with a software interrupt. >>>>>>>> >>>>>>>> What IRQL are you running at when you crash, a common problem here >>>>>>>> would >>>>>>>> be >>>>>>>> calling the functions you mentioned at IRQL_DISPATCH or higher. >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> Don Burn (MVP, Windows DDK) >>>>>>>> Windows 2k/XP/2k3 Filesystem and Driver Consulting >>>>>>>> http://www.windrvr.com >>>>>>>> Remove StopSpam from the email to reply >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> <x856256(a)yahoo.com> wrote in message >>>>>>>> news:1154814887.364085.142850(a)n13g2000cwa.googlegroups.com... >>>>>>>>> Hi, >>>>>>>>> >>>>>>>>> Drivers are very new to me, can someone help me with the next issue >>>>>>>>> please: >>>>>>>>> >>>>>>>>> I have tried to set up a Filter Hook driver. >>>>>>>>> I took an example code from codeproject (DrvFltIp) and it works ok >>>>>>>>> on >>>>>>>>> my computer (Windows XP). >>>>>>>>> >>>>>>>>> When I try to process the packets more thoroughly than in this >>>>>>>>> example >>>>>>>>> code, my computer crashed (freezes or restarts immediately). >>>>>>>>> The crash happens when I do something that involves a >>>>>>>>> KeWaitForSingleObject or a KeDelayExecutionThread call (but I am not >>>>>>>>> sure these are the only scenarios that cause a crash). >>>>>>>>> >>>>>>>>> I looked in the dump file with WinDbg after the crash and saw that >>>>>>>>> one >>>>>>>>> of the last actions of the kernel were sending an interrupt and >>>>>>>>> afterwards an exception. So my guess was that the kernel is sending >>>>>>>>> a >>>>>>>>> trap that was sent to it by the hardware to my driver to handle, but >>>>>>>>> since I didn't set up an interrupt call back function, an exception >>>>>>>>> is >>>>>>>>> raised. >>>>>>>>> The next thing I did was to try and set up an interrupt call back >>>>>>>>> function, but this also caused a similar crash to the system (when I >>>>>>>>> tried to pull out the parameters that I have to pass to the function >>>>>>>>> "IoConnectInterrupt" from the isrStack). I tried to call the >>>>>>>>> function >>>>>>>>> "IoConnectInterrupt" from the function >>>>>>>>> "DriverObject->MajorFunction[IRP_MJ_CREATE]
From: hanzhu on 7 Aug 2006 04:10 OK, Seems I got your problem. For XP and the later, the callback routine is been invoked at DISPATCH_LEVEL. And you have to return the policy from your callback routine when it returns. However, just as what have been mentioned by the folks, you couldn't wait at DISPATCH_LEVEL. If you really have to keep such a queue for the packets, then the IP filter driver is not suitable for your needs since you have to process all the packets in the callback routine without any wait action(KeWaitFor...)and return it to the tcpip driver immediately when the callback routine returns. Why not consider the NDISIM driver? It's supported by MS and I'm convinced it will fit for your requirement. x856256(a)yahoo.com ?: > I tried at the beginning to process the packets on the same thread, but > it caused the system to crash every time. > Since I found it very hard to program in kernel-mode, and even harder > to debug it, I wanted to let a user-mode program process the packet and > wait till it finishes in order to return. Was that a bad idea? If it is > a bad idea, it means I should do all the processing in the driver, and > on the same thread. In such a scenario I will have to synchronize > somewhere because my packet process depends on other packets that were > previously processed (so I must have a synchronized queue of packets) > and again I have the same problem as the first one (or do I ?) > > Thanks, > Guy > > hanzhu wrote: >> IIRC, IP filter driver should process these packets through the callback >> routine instead of submitting the IRP to the lower tcpip protocol >> driver. Why do you need to pass the IRPs and wait for its completion? >> >> >> x856256(a)yahoo.com ?: >>> Hello, >>> >>> Either queueing the job or doing it asynchronously I will have to do >>> the job in another thread, and wait for the process of the packet to >>> finish (in order to return the drop, forward or pass reply), and this >>> is exactly where my problem began, when I tried to use the >>> KeWaitForSingleObject function. >>> So let me ask this again: How can I use this wait function, or any >>> other substitution that will allow me process the packet on another >>> thread (queued or not) and wait for it to finish processing in order to >>> return the value? >>> >>> Thanks, >>> Guy >>> >>> Doron Holan [MS] wrote: >>>> you can't reduce IRQL inline, you can only raise it and then lower it. if >>>> you truly want to synchronously process soemthing, then queue it to a work >>>> item. if you want your filter to not negatively affect performance, learn >>>> how to process the results in the completion routine asynchronously >>>> >>>> d >>>> >>>> -- >>>> Please do not send e-mail directly to this alias. this alias is for >>>> newsgroup purposes only. >>>> This posting is provided "AS IS" with no warranties, and confers no rights. >>>> >>>> >>>> <x856256(a)yahoo.com> wrote in message >>>> news:1154873993.596484.241370(a)i42g2000cwa.googlegroups.com... >>>>> No, >>>>> >>>>> The code is not taking any spinlocks. >>>>> Anyone has an idea how can I reduce the IRQL level inside the filter >>>>> hook function? >>>>> >>>>> Thanks, >>>>> Guy >>>>> >>>>> Don Burn wrote: >>>>>> You cannot call a lot of functions at DISPATCH_LEVEL. I am not well >>>>>> versed >>>>>> in filter hooks, so you need to determine if the raising to DISPATCH is >>>>>> avoidable. For instance, if the code takes a spinlock, could it take an >>>>>> ERESOURCE instead. >>>>>> >>>>>> >>>>>> -- >>>>>> Don Burn (MVP, Windows DDK) >>>>>> Windows 2k/XP/2k3 Filesystem and Driver Consulting >>>>>> http://www.windrvr.com >>>>>> Remove StopSpam from the email to reply >>>>>> >>>>>> >>>>>> >>>>>> <x856256(a)yahoo.com> wrote in message >>>>>> news:1154861486.219608.122660(a)m73g2000cwd.googlegroups.com... >>>>>>> Hello and thanks Don for the answer. >>>>>>> >>>>>>> It is running in IRQL_DISPATCH, I will try to reduce it and see what >>>>>>> happens. >>>>>>> >>>>>>> Thanks, >>>>>>> Guy >>>>>>> >>>>>>> Don Burn wrote: >>>>>>>> A filter hook driver should not have an interrupt handler it is only >>>>>>>> concerned with packets as they are presented to it. You should not >>>>>>>> try >>>>>>>> to >>>>>>>> use IoConnectInterrupt with a software interrupt. >>>>>>>> >>>>>>>> What IRQL are you running at when you crash, a common problem here >>>>>>>> would >>>>>>>> be >>>>>>>> calling the functions you mentioned at IRQL_DISPATCH or higher. >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> Don Burn (MVP, Windows DDK) >>>>>>>> Windows 2k/XP/2k3 Filesystem and Driver Consulting >>>>>>>> http://www.windrvr.com >>>>>>>> Remove StopSpam from the email to reply >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> <x856256(a)yahoo.com> wrote in message >>>>>>>> news:1154814887.364085.142850(a)n13g2000cwa.googlegroups.com... >>>>>>>>> Hi, >>>>>>>>> >>>>>>>>> Drivers are very new to me, can someone help me with the next issue >>>>>>>>> please: >>>>>>>>> >>>>>>>>> I have tried to set up a Filter Hook driver. >>>>>>>>> I took an example code from codeproject (DrvFltIp) and it works ok >>>>>>>>> on >>>>>>>>> my computer (Windows XP). >>>>>>>>> >>>>>>>>> When I try to process the packets more thoroughly than in this >>>>>>>>> example >>>>>>>>> code, my computer crashed (freezes or restarts immediately). >>>>>>>>> The crash happens when I do something that involves a >>>>>>>>> KeWaitForSingleObject or a KeDelayExecutionThread call (but I am not >>>>>>>>> sure these are the only scenarios that cause a crash). >>>>>>>>> >>>>>>>>> I looked in the dump file with WinDbg after the crash and saw that >>>>>>>>> one >>>>>>>>> of the last actions of the kernel were sending an interrupt and >>>>>>>>> afterwards an exception. So my guess was that the kernel is sending >>>>>>>>> a >>>>>>>>> trap that was sent to it by the hardware to my driver to handle, but >>>>>>>>> since I didn't set up an interrupt call back function, an exception >>>>>>>>> is >>>>>>>>> raised. >>>>>>>>> The next thing I did was to try and set up an interrupt call back >>>>>>>>> function, but this also caused a similar crash to the system (when I >>>>>>>>> tried to pull out the parameters that I have to pass to the function >>>>>>>>> "IoConnectInterrupt" from the isrStack). I tried to call the >>>>>>>>> function >>>>>>>>> "IoConnectInterrupt" from the function >>>>>>>>> "DriverObject->MajorFunction[IRP_MJ_CREATE]
From: Maxim S. Shatskih on 8 Aug 2006 00:22 >Since I found it very hard to program in kernel-mode, and even harder >to debug it, I wanted to let a user-mode program process the packet and >wait till it finishes in order to return. Was that a bad idea? Yes, this will drop the network performance a lot. >on the same thread. In such a scenario I will have to synchronize >somewhere because my packet process depends on other packets that were >previously processed (so I must have a synchronized queue of packets) Correct. -- Maxim Shatskih, Windows DDK MVP StorageCraft Corporation maxim(a)storagecraft.com http://www.storagecraft.com
From: Maxim S. Shatskih on 8 Aug 2006 00:22
>Since I found it very hard to program in kernel-mode, and even harder >to debug it, I wanted to let a user-mode program process the packet and >wait till it finishes in order to return. Was that a bad idea? Yes, this will drop the network performance a lot. >on the same thread. In such a scenario I will have to synchronize >somewhere because my packet process depends on other packets that were >previously processed (so I must have a synchronized queue of packets) Correct. -- Maxim Shatskih, Windows DDK MVP StorageCraft Corporation maxim(a)storagecraft.com http://www.storagecraft.com |