February 2004 - Darwin-kernel

Re: opening files in the kernel
by Marek Kozubal 24 Feb '04

24 Feb '04

On Sun, 22 Feb 2004, Chris Bergmann wrote: > Greets, > > I've been trying to open/read/write/etc a file from within a KEXT, This is generally considered to be a *bad* idea. > using the vnode interface. I keep running myself into locking trouble > because I lack a thorough understanding of the correct procedures for > dealing with vnodes. I've tried a few different solutions, however > (apart from the odd case where it "just worked" for a while) they all > end up either deadlocking the vfs or causing a (locking-related) panic. Generally if you need to read/write to files from a KEXT in mass, you use the IOKit messages or sockets or sysctl depending on what you're doing to a user daemon to do file IO. > As a basic background to where i'm coming from- I've converted Apple's > HFS+ code to compile with C++ and made the necessary modifications to > run as a KEXT; now i'm trying to create some additional "system" files. I'm baffled why you converted this to C++, file systems are not IOKit objects, and are generally C only entities. > Apple's existing system files (extents tree, etc) are treated rather > specially by the filesystem and I don't want to go down the same path- > instead I would like to open a file in much the same way as a userland > process would (except using the vfs directly since i'm in a KEXT.) This is generally considered a bad thing to do. The locking issues are complex in a file system and you can easily bypass a lot of the locking rules by accessing these functions out of context. Its incredably easy to leave a vnode locked somewhere and not have the system panic till much later on. > I'm looking for any suggestions on how to go about this task.. or any > good references such as books describing the usage of the vfs would be > great (i purchased McKusick's 4.4 BSD OS book but it's not overly > helpful in this area.) Pointers to existing source from which I can > learn by example would also be helpful. What are you trying to accomplish anyways? If you really are trying to add another file to the "special" area of HFS+ in CNID's 2 though 15 then I'd imagine you'd probably have to make your own HFS+ file system and build that into a new kernel, or try to make it a loadable file system and put it at a higher priority for HFS in the filesystem bundles. Either way this can get quite annoying pretty quickly. --- 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: profiling the kernel
by chuck remes 23 Feb '04

23 Feb '04

On Feb 23, 2004, at 4:54 PM, chuck remes wrote: While doing some performance tuning on a driver, I noticed that darwin-x86 is significantly slower than darwin-ppc on roughly equivalent hardware. When doing some pretty common operations (like running "sum <600 MB file>") most of the system time is charged to the kernel (e.g. kernel_task). If I wanted to profile the kernel and try to narrow down the sections of code that are problematic, how would I go about doing this? Is it safe to assume that 2-machine debugging is a necessary prerequisite? Is there any chance that the tools Apple uses internally for this operation will see the light of day? If this is NOT feasible, can any Apple kernel engineers suggest places in the code that could use some scrutiny? Any and all suggestions are welcome. Five minutes after sending this (ain't that always the way it goes?) I stumbled across "kgmon" which looks like it does what I want. Is this the right place to start? If so, all I need is a kernel and its support frameworks compiled with debugging symbols on x86... (I'll get help from Felix@opendarwin with this part). cr _______________________________________________ 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

profiling the kernel
by chuck remes 23 Feb '04

23 Feb '04

While doing some performance tuning on a driver, I noticed that darwin-x86 is significantly slower than darwin-ppc on roughly equivalent hardware. When doing some pretty common operations (like running "sum <600 MB file>") most of the system time is charged to the kernel (e.g. kernel_task). If I wanted to profile the kernel and try to narrow down the sections of code that are problematic, how would I go about doing this? Is it safe to assume that 2-machine debugging is a necessary prerequisite? Is there any chance that the tools Apple uses internally for this operation will see the light of day? If this is NOT feasible, can any Apple kernel engineers suggest places in the code that could use some scrutiny? Any and all suggestions are welcome. cr _______________________________________________ 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] What happened to IOREG TOC item?
by Lance Drake 23 Feb '04

23 Feb '04

Excuse me - half my foot is blown off on this one... there is no problem except in my cerebral cortext. Apologies! Lance Drake _______________________________________________ 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: IPC with Sockets
by Matt Jaffa 23 Feb '04

23 Feb '04

Yeah, sorry I meant this to the list I don't know if you guys would know this. My app is working pretty good now, except for when the computer goes to sleep, and you have it set to where it must authenticate before coming out of sleep mode. This is where my app doesn't cut it, because when you wake up your machine and before typing in the password, it still gets on the internet and does communication, for instance if you have a messenger client or mail.app opened, it will still login into those even though you didn't type your password in to get out of the locked screen. Well since this occurs, my app tried to pop up a CFUserNotification window, which since they haven't got out of the secure screen yet, will never be able to be seen causing my computer to get the wheel of death, the spinning circle thing. Is there a way to avoid this, like know when you are in the wakeup process and stuff? Matt On Feb 22, 2004, at 9:06 PM, Robert Gordon wrote: Think Matt meant this to go to the list.. If not DOH! Robert.. Begin forwarded message: I don't know if you guys would know this. My app is working pretty good now, except for when the computer goes to sleep, and you have it set to where it must authenticate before coming out of sleep mode. This is where my app doesn't cut it, because when you wake up your machine and before typing in the password, it still gets on the internet and does communication, for instance if you have a messenger client or mail.app opened, it will still login into those even though you didn't type your password in to get out of the locked screen. Well since this occurs, my app tried to pop up a CFUserNotification window, which since they haven't got out of the secure screen yet, will never be able to be seen causing my computer to get the wheel of death, the spinning circle thing. Is there a way to avoid this, like know when you are in the wakeup process and stuff? Matt _______________________________________________ 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. _______________________________________________ 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

What happened to IOREG TOC item?
by Lance Drake 23 Feb '04

23 Feb '04

You used to be able to apply a function that would take an IOREG entry and do something like CFDictionaryApplyFunction(properties, CheckFor_TOC, (void *)&buf); where the dictionary would use your supplied routine to scan thru the various IOREG entries in the 'properties' dict and return an entry for the TOC for an audio CD. Now - (10.3.2) - everything EXCEPT that item is returned. I'm sure I have the correct IOREG dictionary. Can anyone explain what happened here? Thanks, Lance Drake _______________________________________________ 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

Fwd: IPC with Sockets
by Robert Gordon 23 Feb '04

23 Feb '04

Think Matt meant this to go to the list.. If not DOH! Robert.. Begin forwarded message: I don't know if you guys would know this. My app is working pretty good now, except for when the computer goes to sleep, and you have it set to where it must authenticate before coming out of sleep mode. This is where my app doesn't cut it, because when you wake up your machine and before typing in the password, it still gets on the internet and does communication, for instance if you have a messenger client or mail.app opened, it will still login into those even though you didn't type your password in to get out of the locked screen. Well since this occurs, my app tried to pop up a CFUserNotification window, which since they haven't got out of the secure screen yet, will never be able to be seen causing my computer to get the wheel of death, the spinning circle thing. Is there a way to avoid this, like know when you are in the wakeup process and stuff? Matt _______________________________________________ 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

opening files in the kernel
by Chris Bergmann 22 Feb '04

22 Feb '04

Greets, I've been trying to open/read/write/etc a file from within a KEXT, using the vnode interface. I keep running myself into locking trouble because I lack a thorough understanding of the correct procedures for dealing with vnodes. I've tried a few different solutions, however (apart from the odd case where it "just worked" for a while) they all end up either deadlocking the vfs or causing a (locking-related) panic. As a basic background to where i'm coming from- I've converted Apple's HFS+ code to compile with C++ and made the necessary modifications to run as a KEXT; now i'm trying to create some additional "system" files. Apple's existing system files (extents tree, etc) are treated rather specially by the filesystem and I don't want to go down the same path- instead I would like to open a file in much the same way as a userland process would (except using the vfs directly since i'm in a KEXT.) I notice that the paging system also works this way- the dynamic pager passes in a file path which the kernel then opens and uses- however I have tried copying its file access techniques without any luck. I'm also considering the possibility that i'm lacking the "correct" userland context that I would need in order to access the filesystem, however I don't know what would be required here if there is anything. I'm looking for any suggestions on how to go about this task.. or any good references such as books describing the usage of the vfs would be great (i purchased McKusick's 4.4 BSD OS book but it's not overly helpful in this area.) Pointers to existing source from which I can learn by example would also be helpful. It's also worth mentioning that my code generally appears to be not crashing directly- instead, it's leaving the system in a state where another process is panicking/deadlocking (Finder, when accessing the same vnode to gather file stats, or some such.) thx, chris _______________________________________________ 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: IPC with Sockets
by Justin Walker 22 Feb '04

22 Feb '04

On Saturday, February 21, 2004, at 06:55 PM, Mike Smith wrote: If all you are doing is logging the kext's current condition, use a sysctl. The massive overhead involved in the socket, and the fact that you're not maintaining any of the daemon's state in the kext makes it a bad choice for this model. >From what little I've picked up on Matt's intentions, I don't think the sysctl will work well. I think he needs the KEXT to inform the daemon of something, and the only way to do this with sysctls is to poll. You haven't really helped any of us help you by not describing your application and design; it's very hard to second-guess something like this. Matt is learning this stuff the hard way. He's still trying to figure out what information is useful and what is not (and he tends to shoot first and ask himself questions later :-}). Cheers, Justin -- Justin C. Walker, Curmudgeon-At-Large * Institute for General Semantics | "Weaseling out of things is what | separates us from the animals. | Well, except the weasel." | - Homer J Simpson *--------------------------------------*-------------------------------* _______________________________________________ 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: IPC with Sockets
by Mike Smith 22 Feb '04

22 Feb '04

If all you are doing is logging the kext's current condition, use a sysctl. The massive overhead involved in the socket, and the fact that you're not maintaining any of the daemon's state in the kext makes it a bad choice for this model. You haven't really helped any of us help you by not describing your application and design; it's very hard to second-guess something like this. = Mike On Feb 21, 2004, at 2:23 PM, Matt Jaffa wrote: Well I have discovered the cost of having to do that In my system log where it logs the KEXT's current condition, I am noticing that since there are so many requests on the KEXT side of things, that some get to the soconnect before the daemon has a chance to set up again and be ready to accept new connections. Any thoughts on how to over come this? Thanks, Matt On Feb 21, 2004, at 2:40 PM, Justin Walker wrote: On Saturday, February 21, 2004, at 01:30 PM, Matt Jaffa wrote: I think I have it working now, The KEXT disconnected from it before, but after 250 connections it still started that mess, Keep in mind that even though one end of a 'channel' closes it, the other end remains open until that end is closed. but now In the daemon I have it creating a socket then accepting and after it has done the communication, I close the the socket and it is working fine now. FWIW, I think that opening and closing a connection for each exchange may be overkill (or inefficient). You have to evaluate this based on the frequency of exchange, and the cost of the open/close operations. I'm happy you got it fixed. Regards, Justin -- Justin C. Walker, Curmudgeon-At-Large * Institute for General Semantics | "Weaseling out of things is what | separates us from the animals. | Well, except the weasel." | - Homer J Simpson *-------------------------------------- *-------------------------------* _______________________________________________ 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. _______________________________________________ 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. _______________________________________________ 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