May 2005 - Darwin-dev - lists.apple.com

Re: Patch to mount_nfs to make NFSMNT_RESVPORT default
by Craig Rodrigues 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Mutt/1.5.9i On Thu, May 12, 2005 at 04:04:46PM -0400, Andrew Gallatin wrote: > AFAIK, the problem is really Apple / Linux. FreeBSD and Solaris don't Nope, I encountered this between MacOS X 10.3.9 (NFS client) and FreeBSD 4.9 (NFS server) and SuSE Enterprise Linux 9 (NFS server). i.e. on MacOS X, I had to do a mount_nfs -P in order to mount exports on both of those servers. -- Craig Rodrigues rodrigc(a)crodrigues.org _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Problem with dlsym
by Jordan K. Hubbard 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Problem with dlsym
by J. Davison de St. Germain 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com We have been using the 'dlcompat' library on OSX for a while and it worked great. However, with Tiger 10.4, it appears that the dl functions (eg: dlsym) are now built in. Unfortunately, when we try to find a non-existent symbol in one of our large shared libraries, it now takes forever (instead of returning null fairly quickly/immediately). After about 20 minutes of 'working' on the dlsym call, our program will return and say that it couldn't find the symbol. When we used dlcompat before, we didn't have this problem. We have read the man page and tried all the different options that are presented, but nothing seems to help. (Note, if we try the same lookup of a non-existent symbol on a system library, it returns very quickly. The library that we are searching is part of a large framework of other libraries and (apparently) for some reason loads all of the other libraries in its search for the symbol. This does seem strange as we specify the exact dylib that we want to searh through.) In a last ditch effort, we tried compiling and using the original 'dlcompat' libraries. Unfortunately, the compilation (using gcc 4.0) died on the #pragma CALL_ON_LOAD dlcompat_init_func. We commented this line out and the compilation succeeded. However, when we go to link against the libdl.dylib that is produced, we get this warning: symbol _dlsym used from dynamic library /usr/lib/gcc/powerpc-apple-darwin8/4.0.0/../../../libSystem.dylib(dyldAPIsInLibSystem.o) not from earlier dynamic \ library /usr/local/SCIRun/Thirdparty/1.24.1/Darwin/gcc-4.0-32bit/lib/libdl.1.dylib(dlfcn.o) I haven't been able to find a way to force the compiler/linker to use the dlsym symbol from our library. Do you have any thoughts on what is going wrong or how we could go about fixing this? Any help on this will be very much appreciated. Sincerely, J. Davison de St. Germain ----------------------------------------------------------------------- - J. Davison de St. Germain dav(a)cs.utah.edu (801) 581-4078 - - Chief Software Engineer http://www.cs.utah.edu/~dav - - SCI Institute, SE C-SAFE University of Utah - ----------------------------------------------------------------------- _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: mutex PTHREAD_MUTEX_ERRORCHECK and pthread_cancel issues
by Matt Watson 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: Darwin-dev(a)lists.apple.com On May 12, 2005, at 11:25 AM, Stéphane Letz wrote: Are you sure your code has reached a cancellation point? The following works fine for me on Tiger: % cat test.c #include <assert.h> #include <pthread.h> #include <stdio.h> static pthread_mutex_t m = PTHREAD_MUTEX_INITIALIZER; static pthread_cond_t c = PTHREAD_COND_INITIALIZER; static pthread_key_t k; void * func(void *arg) { assert(!pthread_setspecific(k, arg)); assert(!pthread_mutex_lock(&m)); while (1) { assert(!pthread_cond_wait(&c, &m)); } /* NOT REACHED */ assert(!pthread_mutex_unlock(&m)); return NULL; } void cleanup_func(void *arg) { printf("cleanup called with %p\n", arg); } int main(void) { pthread_t t; int i = 42; void *ret; assert(!pthread_key_create(&k, cleanup_func)); printf("creating a thread with arg %p\n", &i); assert(!pthread_create(&t, NULL, func, &i)); assert(!pthread_cancel(t)); assert(!pthread_join(t, &ret)); assert(PTHREAD_CANCELED == ret); return 0; } % make test CFLAGS='-Os -Wall -W -D_APPLE_C_SOURCE' cc -Os -Wall -W -D_APPLE_C_SOURCE test.c -o test % ./test creating a thread with arg 0xbffffc7c cleanup called with 0xbffffc7c _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… But after a pthread_cancel() has been called, the __pthread_destructor code is never called. Is there any known issue in using the pthread_key_create system? This email sent to site_archiver(a)lists.apple.com

1 0

Re: Patch to mount_nfs to make NFSMNT_RESVPORT default
by Andrew Gallatin 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Mike Mackovitch writes: > On Thu, May 12, 2005 at 10:49:21AM -0400, Craig Rodrigues wrote: > > > > Can someone comment on the attached patch? > > This patch makes NFSMNT_RESVPORT a default > > option for mount_nfs, which greatly > > improves interoperability with Linux > > and FreeBSD NFS servers. > > Improving interoperability is a good thing. > Unfortunately, this patch alone isn't a good answer. > > 1) All non-root NFS mount attempts would now fail. > > 2) The maximum number of NFS mounts would now be limited > by the limited number of "reserved ports". > > Also, I would like to point out that the requirement that > NFS connections come from reserved ports doesn't actually > improve security... it just limits interoperability. AFAIK, the problem is really Apple / Linux. FreeBSD and Solaris don't require reserved ports by default. However, there are a *lot* of Linux NFS servers out there (no matter how much some of us would like to replace them.. ;) Craig's first patch to make automount use reserved ports shouldn't cause problems, and would save tons of man hours from poor transplanted unix admins who are astonished that any unix could fail to use a reserved port for automount. That was the last thing I looked for. It took a tcpdump on the server for me to realize what was happening. IMHO, if there are really people who have more than a few hundred automounts *at once*, and who are dealing with non-linux servers which permit mount requests from non-reserved ports, then you can make an argument to automount to revert to the current, non-interoperable behaviour. Drew _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Problems building automount
by Jonas Maebe 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 12 mei 2005, at 18:00, Craig Rodrigues wrote: Hmmm....thanks for the tip! Do you know how to pass options like -P using the "-host" option in automount? I invoke automount like this: Jonas _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… FWIW, if you're on Mac OS X, I would strongly recommend to use NFS Manager. It can configure NFS mounts so that the requests always originate from "secure" ports. See http://www.bresink.de/osx/NFSManager.html This email sent to site_archiver(a)lists.apple.com

1 0

Re: Problems building automount
by Mike Mackovitch 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Mutt/1.4i On Thu, May 12, 2005 at 12:00:17PM -0400, Craig Rodrigues wrote: > On Thu, May 12, 2005 at 08:17:37AM -0400, Andrew Gallatin wrote: > > In the meantime, you can work around this by adding the -P flag to your > > automount maps. Eg: > > > > # cat /etc/auto.home > > home -P,-nosuid sw:/home > > common -P,-nosuid sw:/common > > <...> > > Hmmm....thanks for the tip! Do you know how to pass options like -P > using the "-host" option in automount? I invoke automount like this: > > automount -m /net -host > > What this allows me to do is: > cd /net/[hostname]/[mntpoint] > > I don't have any map file. If there's currently no way to specify mount options for that map, automount should be changed to make it so that there is a way. Thanks --macko _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Patch to mount_nfs to make NFSMNT_RESVPORT default
by Mike Mackovitch 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Mutt/1.4i On Thu, May 12, 2005 at 10:49:21AM -0400, Craig Rodrigues wrote: > > Can someone comment on the attached patch? > This patch makes NFSMNT_RESVPORT a default > option for mount_nfs, which greatly > improves interoperability with Linux > and FreeBSD NFS servers. Improving interoperability is a good thing. Unfortunately, this patch alone isn't a good answer. 1) All non-root NFS mount attempts would now fail. 2) The maximum number of NFS mounts would now be limited by the limited number of "reserved ports". Also, I would like to point out that the requirement that NFS connections come from reserved ports doesn't actually improve security... it just limits interoperability. Thanks --macko _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Patch to mount_nfs to make NFSMNT_RESVPORT default
by Kevin Van Vechten 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com - Kevin On May 12, 2005, at 10:44 AM, Markus Hitter wrote: Am 12.05.2005 um 16:49 schrieb Craig Rodrigues: I am a bit new to Darwin, so how would I submit this patch to Apple? You have at least three choices: 3) Both. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… (3) Both, is the best way. Apple is interested in the feedback. The silence Markus mentions in 1) usually arises from the fact that Apple employees cannot comment on future releases. 1) Report a bug/enhancement request at <http://developer.apple.com/ bugreporter/> and wait for Apple to incorporate it into one of their next releases. Apple takes bug reports seriously, but is usually next to mute to their reporters[1], even canned/automated messages were dropped about a year ago. 2) Head over to <http://www.opendarwin.org/>, join the hackers lists and ask there. One of the goals of OpenDarwin is to get enhancements/bugfixes/experimental stuff out to the public more quickly. Making your patch compatible to darwinbuild would help a lot, probably. This email sent to site_archiver(a)lists.apple.com

1 0

Re: mutex PTHREAD_MUTEX_ERRORCHECK and pthread_cancel issues
by Stéphane Letz 12 May '05

12 May '05

site_archiver(a)lists.apple.com Delivered-To: Darwin-dev(a)lists.apple.com Le 12 mai 05, à 19:07, Matt Watson a écrit : On May 12, 2005, at 8:40 AM, Stéphane Letz wrote: Is there a workaround for that? This works fine for me on Tiger: #include <assert.h> #include <pthread.h> Thanks a lot. Actually is works also on Panther... mach_port_t machThread = pthread_mach_thread_np(fThread); thread_terminate(machThread); matt. Thanks. pthread_key_create(&__thread_destructor_key, __pthread_destructor); pthread_setspecific(__thread_destructor_key, this); pthread_key_create and pthread_setspecific return correctly 0. Then void __pthread_destructor(void* thread) { Thread* t; t = (Thread*) thread; t->Destructor(); } Thanks Stephane Letz _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I am porting code that needs the PTHREAD_MUTEX_ERRORCHECK type of pthread_mutex. But it seems that PTHREAD_MUTEX_ERRORCHECK is not supported on Darwin. I get an EINVAL error when trying to use pthread_mutexattr_settype. int main (void) { pthread_mutexattr_t a; assert(!pthread_mutexattr_init(&a)); assert(!pthread_mutexattr_settype(&a, PTHREAD_MUTEX_ERRORCHECK)); return 0; } A second problem on the behaviour of pthread_cancel that is not correct. Is the use of explicit pthread_testcancel the way to go or can the following termination code be used safely? No, it can't. In previous versions of Mac OS X, pthread_testcancel() was the only cancellation point. Some code depended on this, so we had to preserve this behavior by default. If you define _APPLE_C_SOURCE when building your pthreads code on Tiger, you should get the new (proper) cancellation behavior. Now I have another problem (tested on Tiger), the following code is used to setup a destruction function in the thread before actually starting it: But after a pthread_cancel() has been called, the __pthread_destructor code is never called. Is there any known issue in using the pthread_key_create system? This email sent to site_archiver(a)lists.apple.com

1 0