October 2002 - Darwin-kernel

Socket problems in NKE
by Michael Cashwell 18 Oct '02

18 Oct '02

Greetings all, I can almost talk with my NKE from userland but I'm having socket problems. I can't get any data from the NKE. I'm using the AF_SYSTEM domain from <sys/kern_control.h>. It essentially is registering a protocol handler as a SOCK_DGRAM type. Then from userland I do this (I've deleted the error checking code below for brevity): struct sockaddr_ctl sf; memset(&sf, 0, sizeof(sf)); m_rpcSock = socket(AF_SYSTEM, SOCK_DGRAM, SYSPROTO_CONTROL); sf.sc_len = sizeof(sf); sf.sc_family = AF_SYSTEM; sf.ss_sysaddr = AF_SYS_CONTROL; sf.sc_id = A1_DARWIN_NKE_SIG; sf.sc_unit = A1_NKE_UNIT_RPC; connect(m_rpcSock, (struct sockaddr *)&sf, sizeof(sf); This all works. I also successfully send one byte of data to the NKE via: char dummy = 0; error = send(m_rpcSock, &dummy, 1, 0); But then when the userland code call recv() with zero flags it blocks forever. I then added this diagnostic loop just to verify and recv() always returns -1 and EAGAIN: do { sleep(2); error = recv(m_rpcSock, ver, sizeof(ver), MSG_DONTWAIT | MSG_PEEK); printf("version MSG_PEEK returned %08X, error %d\n", error, errno); } while (error != sizeof(ver)); But the log shows that the NKE does receive the one zero byte and in response does this: unsigned short ver[2]; ver[0] = IPC_VERSION; ver[1] = IPC_MIN_SUPPORTED_VERSION; error = ctl_enqueuedata(ctlref, ver, sizeof(ver), 0); printf("RPC version send result = %d\n", error); The log also shows that the ctl_enqueuedata() call returns zero. (I've also verified that ctlref is the same one the NKE registered.) So it should have written data to the socket. I can't figure out why the userland code cannot receive anything. I've done this sort of thing in the past using SOCK_RAW but the infrastructure is forcing me to use SOCK_DGRAM. I expect this to raise some issues in terms of how I read data from the socket but I can handle that with MSG_PEEK. But only if I can get data to arrive. If anyone's done this sort of socket code and could assist I'd appreciate it. Thanks, Mike _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: socket owner ( pid )
by Justin C. Walker 16 Oct '02

16 Oct '02

On Wednesday, October 16, 2002, at 06:29 AM, Peter Lovell wrote: On Wednesday, October 16, 2002, at 07:23 AM, ollie(a)mac.com wrote: Hi fokes, I have a socket filter and I'm trying to get the pid that created or currently "ownes" the socket. I'd like to be able to get this info from the kernel without having to go out to user space... Any suggestions appreciated !!! -Ollie. It's ugly and not part of the sanctioned kext interface, but sometimes you just have to ... struct proc* p; p = current_proc(); now look at p->p_cred->p_ruid and p->p_ucred->cr_uid To get the creator you have to get the proc pointer at socreate time. The current owner might be different. Whether or not that matters to you depends on what you're doing, of course. Keep in mind that socket descriptors, like file descriptors, can be passed around from process to process, either with fork or explicitly, so as Peter says, you can't be sure whether you have the creator. Also, for the same reason, there is no "owner", since multiple processes can have valid access to that "socket". To make your life even more complicated, there's no guarantee, in general, that when your piece of the kernel is active, any of the processes that have access to that socket are actually active, or that you are executing in one of their threads. If the activity is, for example, transmission of TCP datagrams (segments), this could be triggered by incoming TCP ACK's, so that your extension might be running in a "kernel thread". Exercise caution :-}. Regards, Justin -- Justin C. Walker, Curmudgeon-At-Large * Institute for General Semantics | It's not whether you win or lose... | It's whether *I* win or lose. *--------------------------------------*-------------------------------* _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: socket owner ( pid )
by Peter Lovell 16 Oct '02

16 Oct '02

On Wednesday, October 16, 2002, at 07:23 AM, ollie(a)mac.com wrote: Hi fokes, I have a socket filter and I'm trying to get the pid that created or currently "ownes" the socket. I'd like to be able to get this info from the kernel without having to go out to user space... Any suggestions appreciated !!! -Ollie. It's ugly and not part of the sanctioned kext interface, but sometimes you just have to ... struct proc* p; p = current_proc(); now look at p->p_cred->p_ruid and p->p_ucred->cr_uid To get the creator you have to get the proc pointer at socreate time. The current owner might be different. Whether or not that matters to you depends on what you're doing, of course. Cheers.....Peter _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

socket owner ( pid )
by ollie＠mac.com 16 Oct '02

16 Oct '02

Hi fokes, I have a socket filter and I'm trying to get the pid that created or currently "ownes" the socket. I'd like to be able to get this info from the kernel without having to go out to user space... Any suggestions appreciated !!! -Ollie. _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: sharing custom filesystem
by Marek Kozubal 16 Oct '02

16 Oct '02

On Tue, 15 Oct 2002, Umesh Vaishampayan wrote: > On Monday, October 14, 2002, at 09:28 PM, Marek Kozubal wrote: > > > We are creating a custom file system to view and access the contents > > of a locally-connected (Firewire) storage device from the Mac OS X > > desktop. We would like to share this device and its contents over AFP > > to other macs, but have been unsuccessful so far. > > What is the mode of failure? AppleFileServer just never shows it. We have an icon on the Finder desktop showing our file system and we can look at it from the Finder and from the command line. Right now we're just working with a file system that puts up "." and ".." with 555 permissions owned by root:wheel and a local read-only file system. We do support getattrlist but this doesnt seem to be a requirement for AFP since UFS doesnt support that VNOP call. AppleFileServer has log files in /Library/Log/AppleFileService but they seem to be rather a low level of info. Is there a way to up the debugging level to see it scanning disks to be shown for sharing? There's no man page for AppleFileServer that i can find unfortunately. Also on a side note, how does one notify the system that the disk is now available? For normal disks with a /dev/disk# entry DiskArbitration just does the same thing that disktool -A disk1s1s2 does. However this requires a /dev entry which our file system doesnt have at the moment. Or is this not a nessessity? Does this notification have anything to do with AppleFileServer picking up our filesystem. > > Or is AppleFileServer "hard coded" for sharing only HFS+, UFS, and > > cd9660 file systems? > > I doubt it. I know at least 1 instance of AFP serving NFS files.... There are hardcoded strings in AppleFileServer that list /nfs_reshares so I'm wondering if this implies that it is in fact hard coded. > AFP server is a user land daemon. Volfs support should not a be > requirement because UFS and NFS do not support it and those are > sharable using AFP.... I would agree with this after the things I've tried today. However it doesnt seem to want to share a FAT12 filesystem off a CF card but will share a normal audio CD. Thanks --- Marek Kozubal marek(a)portents.com _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: sharing custom filesystem
by Umesh Vaishampayan 16 Oct '02

16 Oct '02

On Monday, October 14, 2002, at 09:28 PM, Marek Kozubal wrote: > We are creating a custom file system to view and access the contents > of a > locally-connected (Firewire) storage device from the Mac OS X desktop. > We would like to share this device and its contents over AFP to other > macs, but have been unsuccessful so far. What is the mode of failure? > > Is it possible to share a proprietary file system over AFP from Mac OS > X? As long as your proprietary file system supports POSIX semantics, I believe it is possible to share it over AFP. > Or is AppleFileServer "hard coded" for sharing only HFS+, UFS, and > cd9660 > file systems? I doubt it. I know at least 1 instance of AFP serving NFS files.... > If it is possible, is there a particular function or attribute/flag we > need to add to our file system or vnodes? I'm guessing volfs support > (as > described in Technical Q&A QA1113) is required by AppleFileServer, but > is > there anything else? > AFP server is a user land daemon. Volfs support should not a be requirement because UFS and NFS do not support it and those are sharable using AFP.... Must be something else. --Umesh -- Umesh Vaishampayan Apple Computer, Inc. Mac OS X Kernel Ph: (408) 974 0229 _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: NKE Manager
by Christophe Allie 15 Oct '02

15 Oct '02

On 10/15/02 3:48 PM, "Michael Cashwell" <mboards(a)prograde.net> wrote: > On Friday, October 11, 2002, at 06:32 PM, Joshua Graessley wrote: > >> In Mac OS X 10.2, the header sys/kern_control.h defines the >> AF_SYS_CONTROL protocol of PF_SYSTEM. This can be used to communicate >> with and control kernel extensions. It has some limitations. >> sys/kern_control.h has some documentation that should help you get >> started. Unlike the NKE manager (PF_NKE), there is no sample code. > > Thanks Josh (and Justin), this has been helpful. > > Because my NKE needs the ability to distinguish multiple channels (for > the lack of a better term) I've decided to go with PF_SYSTEM as Josh > described. It supports the concept of a unit number (per NKE signature) > and that's just what I need. > > The header sys/kern_control.h is pretty complete as to the use of the > API but I have one question. When userland code writes data to the NKE > and its ctl_write() function is called is the NKE responsible for > freeing the mbuf? My hunch is yes since it's consuming the buffer (ie: > not passing it on to anything else) but I wanted to confirm. This is your responsibility to free the mbuf when you are done with it. Christophe. _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: NKE Manager
by Michael Cashwell 15 Oct '02

15 Oct '02

On Friday, October 11, 2002, at 06:32 PM, Joshua Graessley wrote: > In Mac OS X 10.2, the header sys/kern_control.h defines the > AF_SYS_CONTROL protocol of PF_SYSTEM. This can be used to communicate > with and control kernel extensions. It has some limitations. > sys/kern_control.h has some documentation that should help you get > started. Unlike the NKE manager (PF_NKE), there is no sample code. Thanks Josh (and Justin), this has been helpful. Because my NKE needs the ability to distinguish multiple channels (for the lack of a better term) I've decided to go with PF_SYSTEM as Josh described. It supports the concept of a unit number (per NKE signature) and that's just what I need. The header sys/kern_control.h is pretty complete as to the use of the API but I have one question. When userland code writes data to the NKE and its ctl_write() function is called is the NKE responsible for freeing the mbuf? My hunch is yes since it's consuming the buffer (ie: not passing it on to anything else) but I wanted to confirm. Thanks, Mike _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

sharing custom filesystem
by Marek Kozubal 15 Oct '02

15 Oct '02

We are creating a custom file system to view and access the contents of a locally-connected (Firewire) storage device from the Mac OS X desktop. We would like to share this device and its contents over AFP to other macs, but have been unsuccessful so far. Is it possible to share a proprietary file system over AFP from Mac OS X? Or is AppleFileServer "hard coded" for sharing only HFS+, UFS, and cd9660 file systems? If it is possible, is there a particular function or attribute/flag we need to add to our file system or vnodes? I'm guessing volfs support (as described in Technical Q&A QA1113) is required by AppleFileServer, but is there anything else? Thanks! --- Marek Kozubal marek(a)portents.com _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

[SOLVED]KEXTs & Altivec stack corruption
by Mike Vannorsdel 12 Oct '02

12 Oct '02

I have somewhat solved my problem with stack corruption. It seems that some AltiVec-less sources compiled with the -faltivec flag can have bad side effects. What happens with some of these sources is that the compiler creates four functions (saveVEC, restVEC, saveVEC_vr11, restVEC_vr10). Some of my functions in the sources will have calls to these (usually saveVEC_vr11 and restVEC_vr10) inserted into them. These new functions are what seem to be corrupting the stack. So my solution was to compile all non-AltiVec sources without the -faltivec flag and the AltiVec sources with it. This prevented the extra functions from being created and inserted into my functions. And lots of testing has shown the AltiVec code running very well without problems in my KEXT with this method. I'd be interested to know why these functions are created and inserted into other functions (so far functions with no vector types or instructions). It's definitely a GCC problem. _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0