December 2002 - Darwin-kernel

Trapping/inserting keystrokes
by Jim Witte 28 Dec '02

28 Dec '02

I want to write a piece of software (for my own use, maybe shareware distribution if it works well) to do a system-wide auto-type feature that would tell when you started to type something in a pre-set list, or a dynamically generated list of often-typed things. It would ideally work like MS Explorer's autocomplete feature - the inserted text that would be selected, so you could just hit delete if you didn't want it, or continue to type something different. I don't need this to work in Classic apps. I figure that all of this except for the 'select auto-inserted text' feature could be implemented at the Darwin layer. Does this make sense, or is there a better place I could patch (posing for NSEditTextor something?) Thanks, Jim Witte _______________________________________________ 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

Cross-version development
by Peter Lovell 20 Dec '02

20 Dec '02

Happy Holidays, all :) There was a lot of discussion a while back about various versions of headers and libraries, and what things were included automatically during builds. I'm interested in following this but in a slightly different way. Some background: one of the things I develop are network kernel extensions (NKEs). These do not involve IOKit, which I think might be important for what follows. Until now, the NKE is a single extension which works on both 10.1.x and 10.2. There are a couple of places where it has to be sensitive to which kernel is running but I've been able to keep that fairly small. But I can see that having an appropriate extension for each kernel version (major version, at least) would be a better and safer thing and I'm heading in that direction. Obviously it would be nice to be able to build the whole product in a single build cycle. But so far it seems that you have to boot into each OS version to build the extension for that version. This blows away any notion of an automated build system, and wastes a lot of time too during development build-debug-fix cycles. So here's my question: - is there any way I can set up a development environment (I'm using ProjectBuilder for the extension) so that I can build an extension for 10.1.x using 10.1 headers etc and build the 10.2 extension using 10.2 headers, and do both while running 10.2 ?? I have a separate partition for each version already (so I can boot into whichever I need for testing etc) so all the headers etc are already available. Thanks.....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

Re: MACH ports and fork calls
by Jim Magee 19 Dec '02

19 Dec '02

Mach ports are owned at the task (process) level, not the thread level. So all threads in a process see the same Mach port namespace. But there is NO way to force Mach port inheritance across fork(). You can perform a little dance that gets the parent and child to explicitly share some ports after the fork() call (set a special port in the parent, fork(), and then have the child use that special port to make requests of the parent to share certain ports). But there is nothing that will share them implicitly. And most frameworks have no mechanism to re-hook the new port names in the child to the frameworks internal storage of the portname in the parent. --Jim On Wednesday, December 18, 2002, at 12:36 PM, Nathan T. Hjelm wrote: Is there any way to force their inheritance when using fork? How about threads, under darwin, do they share ports? On Wed, 18 Dec 2002, Ed Wynne wrote: On Wednesday, December 18, 2002, at 11:58 AM, Nathan T. Hjelm wrote: I know that standard file descriptors get copied during a fork call, but what about MACH ports? Does the child process inherit them as well? Only the special ports... bootstrap, exception, etc. The normal complement of mach ports that get created over the life of a process are not inherited. -Ed _______________________________________________ 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

Re: MACH ports and fork calls
by Nathan T. Hjelm 18 Dec '02

18 Dec '02

Is there any way to force their inheritance when using fork? How about threads, under darwin, do they share ports? On Wed, 18 Dec 2002, Ed Wynne wrote: > On Wednesday, December 18, 2002, at 11:58 AM, Nathan T. Hjelm wrote: > > > I know that standard file descriptors get copied during a fork call, > > but > > what about MACH ports? Does the child process inherit them as well? > > Only the special ports... bootstrap, exception, etc. The normal > complement > of mach ports that get created over the life of a process are not > inherited. > > -Ed > _______________________________________________ > 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

Re: MACH ports and fork calls
by Ed Wynne 18 Dec '02

18 Dec '02

On Wednesday, December 18, 2002, at 11:58 AM, Nathan T. Hjelm wrote: I know that standard file descriptors get copied during a fork call, but what about MACH ports? Does the child process inherit them as well? Only the special ports... bootstrap, exception, etc. The normal complement of mach ports that get created over the life of a process are not inherited. -Ed _______________________________________________ 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

MACH ports and fork calls
by Nathan T. Hjelm 18 Dec '02

18 Dec '02

I know that standard file descriptors get copied during a fork call, but what about MACH ports? Does the child process inherit them as well? _______________________________________________ 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: mount/umount bug?
by Peter Montagner 18 Dec '02

18 Dec '02

Sorry, thanks for your help. Peter On Wednesday, December 18, 2002, at 07:32 AM, Umesh Vaishampayan wrote: This is not really kernel issue. please use "mount -t testfs ..." to mount your test file system. [mount -t ftp ... for ftpfs] File systems mounted using mount(8) can be unmounted by umount(8). If you use mount_xxx(8) commands directly, umount(8) will fail unless you previously supplied absolute pathname [result from realpath(3)] to mount_xxx(8) See Darwin-Development list archives for discussion about this.... --Umesh On Tuesday, December 17, 2002, at 07:39 AM, Joshua LeVasseur wrote: The umount command is the culprit. You needn't use the umount command though. You can write your own program to directly call the unmount() system call on the appropriate path. I use this tactic to clean nfs mounts if I accidentally use a relative path for the mount_nfs > command. Although I've never investigated whether directly unmounting the file system leaves user level apps in inconsistent states. For example, for systems which maintain mtab files, the mtab file becomes inconsistent. From my point of view, I don't care. Josh On Tuesday, December 17, 2002, at 03:02 PM, Peter Montagner wrote: Hi, I've been mucking around with a dummy filesystem while trying to learn Darwin's VFS. My filesystem has a bug: unless you mount it passing a full path as the mount point, you can never unmount it. I just get an error like: umount: /path/to/mp: not currently mounted I've tried practically every possible combination of path styles and it seems that: 1) The path passed to mount_testfs (my mount program) must be the full path missing the trailing slash eg. /path/to/mp 2) It doesn't matter what I pass to umount, as long as it is a valid path to the mount point. Otherwise it doesn't work. Also, df reports the mount point as "mp/" rather than a full path. I've been wracking my brain trying to find what is wrong with my code. It seems identical to the webdavfs code in all the crucial areas. I tried mounting another filesystem (mount_ftp) to the mount point instead of mine and ftpfs seems to have the same problem. Steps to reproduce: % mkdir mp % sudo mount_ftp ftp://server/ mp % sudo umount mp umount: /path/to/mp: not currently mounted % mkdir mp2 % sudo mount_ftp ftp://server/ /path/to/mp2/ % sudo umount /path/to/mp2/ umount: /path/to/mp2: not currently mounted % mkdir mp3 % sudo mount_ftp ftp://server/ /path/to/mp3 % sudo umount /path/to/mp3 [Succeeds] Notice that only the full path without a trailing slash works. So, is my filesystem buggy (and ftpfs) or is umount or something else? Or are you supposed to use full paths when mounting filesystems (I doubt it)? Can anyone reproduce this? I know this is the kernel list but I have a feeling that the problem is kernel related? Thanks, 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

Re: mount/umount bug?
by Umesh Vaishampayan 17 Dec '02

17 Dec '02

This is not really kernel issue. please use "mount -t testfs ..." to mount your test file system. [mount -t ftp ... for ftpfs] File systems mounted using mount(8) can be unmounted by umount(8). If you use mount_xxx(8) commands directly, umount(8) will fail unless you previously supplied absolute pathname [result from realpath(3)] to mount_xxx(8) See Darwin-Development list archives for discussion about this.... --Umesh On Tuesday, December 17, 2002, at 07:39 AM, Joshua LeVasseur wrote: The umount command is the culprit. You needn't use the umount command though. You can write your own program to directly call the unmount() system call on the appropriate path. I use this tactic to clean nfs mounts if I accidentally use a relative path for the mount_nfs > command. Although I've never investigated whether directly unmounting the file system leaves user level apps in inconsistent states. For example, for systems which maintain mtab files, the mtab file becomes inconsistent. From my point of view, I don't care. Josh On Tuesday, December 17, 2002, at 03:02 PM, Peter Montagner wrote: Hi, I've been mucking around with a dummy filesystem while trying to learn Darwin's VFS. My filesystem has a bug: unless you mount it passing a full path as the mount point, you can never unmount it. I just get an error like: umount: /path/to/mp: not currently mounted I've tried practically every possible combination of path styles and it seems that: 1) The path passed to mount_testfs (my mount program) must be the full path missing the trailing slash eg. /path/to/mp 2) It doesn't matter what I pass to umount, as long as it is a valid path to the mount point. Otherwise it doesn't work. Also, df reports the mount point as "mp/" rather than a full path. I've been wracking my brain trying to find what is wrong with my code. It seems identical to the webdavfs code in all the crucial areas. I tried mounting another filesystem (mount_ftp) to the mount point instead of mine and ftpfs seems to have the same problem. Steps to reproduce: % mkdir mp % sudo mount_ftp ftp://server/ mp % sudo umount mp umount: /path/to/mp: not currently mounted % mkdir mp2 % sudo mount_ftp ftp://server/ /path/to/mp2/ % sudo umount /path/to/mp2/ umount: /path/to/mp2: not currently mounted % mkdir mp3 % sudo mount_ftp ftp://server/ /path/to/mp3 % sudo umount /path/to/mp3 [Succeeds] Notice that only the full path without a trailing slash works. So, is my filesystem buggy (and ftpfs) or is umount or something else? Or are you supposed to use full paths when mounting filesystems (I doubt it)? Can anyone reproduce this? I know this is the kernel list but I have a feeling that the problem is kernel related? Thanks, Peter -- Umesh Vaishampayan Mac OS X - Kernel _______________________________________________ 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: mount/umount bug?
by Joshua LeVasseur 17 Dec '02

17 Dec '02

The umount command is the culprit. You needn't use the umount command though. You can write your own program to directly call the unmount() system call on the appropriate path. I use this tactic to clean nfs mounts if I accidentally use a relative path for the mount_nfs command. Although I've never investigated whether directly unmounting the file system leaves user level apps in inconsistent states. For example, for systems which maintain mtab files, the mtab file becomes inconsistent. >From my point of view, I don't care. Josh On Tuesday, December 17, 2002, at 03:02 PM, Peter Montagner wrote: Hi, I've been mucking around with a dummy filesystem while trying to learn Darwin's VFS. My filesystem has a bug: unless you mount it passing a full path as the mount point, you can never unmount it. I just get an error like: umount: /path/to/mp: not currently mounted I've tried practically every possible combination of path styles and it seems that: 1) The path passed to mount_testfs (my mount program) must be the full path missing the trailing slash eg. /path/to/mp 2) It doesn't matter what I pass to umount, as long as it is a valid path to the mount point. Otherwise it doesn't work. Also, df reports the mount point as "mp/" rather than a full path. I've been wracking my brain trying to find what is wrong with my code. It seems identical to the webdavfs code in all the crucial areas. I tried mounting another filesystem (mount_ftp) to the mount point instead of mine and ftpfs seems to have the same problem. Steps to reproduce: % mkdir mp % sudo mount_ftp ftp://server/ mp % sudo umount mp umount: /path/to/mp: not currently mounted % mkdir mp2 % sudo mount_ftp ftp://server/ /path/to/mp2/ % sudo umount /path/to/mp2/ umount: /path/to/mp2: not currently mounted % mkdir mp3 % sudo mount_ftp ftp://server/ /path/to/mp3 % sudo umount /path/to/mp3 [Succeeds] Notice that only the full path without a trailing slash works. So, is my filesystem buggy (and ftpfs) or is umount or something else? Or are you supposed to use full paths when mounting filesystems (I doubt it)? Can anyone reproduce this? I know this is the kernel list but I have a feeling that the problem is kernel related? Thanks, 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. _______________________________________________ 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

mount/umount bug?
by Peter Montagner 17 Dec '02

17 Dec '02

Hi, I've been mucking around with a dummy filesystem while trying to learn Darwin's VFS. My filesystem has a bug: unless you mount it passing a full path as the mount point, you can never unmount it. I just get an error like: umount: /path/to/mp: not currently mounted I've tried practically every possible combination of path styles and it seems that: 1) The path passed to mount_testfs (my mount program) must be the full path missing the trailing slash eg. /path/to/mp 2) It doesn't matter what I pass to umount, as long as it is a valid path to the mount point. Otherwise it doesn't work. Also, df reports the mount point as "mp/" rather than a full path. I've been wracking my brain trying to find what is wrong with my code. It seems identical to the webdavfs code in all the crucial areas. I tried mounting another filesystem (mount_ftp) to the mount point instead of mine and ftpfs seems to have the same problem. Steps to reproduce: % mkdir mp % sudo mount_ftp ftp://server/ mp % sudo umount mp umount: /path/to/mp: not currently mounted % mkdir mp2 % sudo mount_ftp ftp://server/ /path/to/mp2/ % sudo umount /path/to/mp2/ umount: /path/to/mp2: not currently mounted % mkdir mp3 % sudo mount_ftp ftp://server/ /path/to/mp3 % sudo umount /path/to/mp3 [Succeeds] Notice that only the full path without a trailing slash works. So, is my filesystem buggy (and ftpfs) or is umount or something else? Or are you supposed to use full paths when mounting filesystems (I doubt it)? Can anyone reproduce this? I know this is the kernel list but I have a feeling that the problem is kernel related? Thanks, 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