September 2008 - Darwin-dev

Re: AHCI Error?
by Chris Sarcone 10 Sep '08

10 Sep '08

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

AHCI Error?
by Duane Murphy 10 Sep '08

10 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com What is error 0xe0030005? I decoded this with mach/error.h and it is an iokit error in the ahci family, error code 5. So what is error code 5 for the ahci family? (I've been digging in the open source files, but can't find any ahci drivers.) Thanks, ...Duane _______________________________________________ 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

kDADiskOptionEjectUponLogout equivalent for Fast User Switching?
by Tim Schooley 09 Sep '08

09 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi folks, Anyone know of a way to make the OS eject disks when a user "loses console" during Fast User Switching? Currently, our code listens for the kEventSystemUserSessionDeactivated event, but most of the time, our code isn't executed before the user logging in has access to the system. Any ideas? Thanks for any help! Tim _______________________________________________ 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: CFFileDescriptor and stdin
by Terry Lambert 09 Sep '08

09 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com I agree. But I'm not using stdio. I shouldn't have said "stdin"... http://lists.apple.com/archives/Unix-porting/2006/Jan/msg00072.html http://lists.apple.com/archives/Unix-porting/2006/Jan/msg00084.html -- Terry _______________________________________________ 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… On Sep 6, 2008, at 1:37 PM, Nicolas Roussel wrote: In general, you should avoid multiple layers of buffering. I would suggest turning buffering off at the stdio layer using setbuf(3). I'm simply passing 0 as the file descriptor parameter to CFFileDescriptorCreate. And I'm not using any Carbon or Coca stuff for the moment (check the code in my first message). It seems others have had problems try to detect user input on the terminal with kqueue: Could this be related to my problem with the CFFileDescriptor? Try fstating fd 0 and looking at the S_IFMT bits in the st_mode field and see what flags are set in it. If it's a character or block special device, look at the major and minor fields of the st_dev and match them to a device in /dev. This email sent to site_archiver(a)lists.apple.com

1 0

Re: DiskArbitration, unmounting busy due to fseventsd?
by Tim Schooley 08 Sep '08

08 Sep '08

site_archiver(a)lists.apple.com Delivered-To: Darwin-dev(a)lists.apple.com Hi Terry, Many thanks for your response. On Fri, 05 Sep 2008 12:01:58 -0700, Terry Lambert wrote: > Are you maybe using command line mounts and unmounts other than > diskutil, which talks to DA? DA will run your command line tools > under the covers after doing the right things. As long as you go > through DA to do your mounts and unmount rather than explicit command > line utilities or programs not going through DA, you should not need > the nobrowse. As I said before, DA will notify fseventsd to stop > looking. All mounts and unmounts are performed via DADiskMountWithArguments and DADiskUnmount respectively. See below for a code snippet of the unmount function. > If Spolight and Time Machine are undesirable, you can explicitly > disable them programattically by updating settings for "don't look > here". Pretty sure this is well documented. Yes - I'm aware that this is possible, but until I've fixed the busy unmount issue, I didn't see much point looking into this. So if I can get DADiskUnmount to succeed on mounts to specified locations, I'd be a super-happy-bunny. It seems though that others are also having issues: http://zfs.macosforge.org/trac/wiki/faq (search for fseventsd), although it looks like they're discussing command-line umount. Below is our unmount function (with some slight changes to make it more readable without having to add more definitions etc). If you can spot anything wrong with the way we're approving the unmount, please let me know! Note that the function is declared in a static class. Many thanks again for any help. Kind regards, Tim ---- /* Callback context struct. */ typedef struct { bool completed; int result; CFRunLoopRef runloop; bool mount; } MOUNT_RESULT; /* Generic callback function for mounts and unmounts. */ static void mount_disk_da_callback( DADiskRef /*disk*/, DADissenterRef dissenter, void* context ) { DAReturn rc; MOUNT_RESULT* result = static_cast<MOUNT_RESULT*>(context); if (NULL == result) { LOG_ERR("Failed to obtain context during DA callback."); goto out; // Should never happen. } if (NULL == dissenter) goto out; rc = DADissenterGetStatus(dissenter); switch (rc) { /* error handling that sets result.result */ } out: CFRunLoopStop(result->runloop); result->completed = true; } /* Unmount approval callback. We accept by returning NULL. */ static DADissenterRef approve_unmount_da_callback( DADiskRef /*disk*/, void* /*context*/ ) { LOG_DEBUG("approve_unmount_da_callback(): approving unmount."); return NULL; } /* Unmount function. */ int Mounter::unmount( const char* device_path, const bool& force ) { int result = 0; DADiskRef disk_ref = NULL; DASessionRef session_ref = NULL; DAApprovalSessionRef app_session_ref = NULL; MOUNT_RESULT unmount_result; unmount_result.completed = false; unmount_result.result = 0; unmount_result.mount = false; // Create DA session. session_ref = DASessionCreate(kCFAllocatorDefault); if (NULL == session_ref) { LOG_ERR("Failed to create DA session during unmount."); result = ERR_UNMOUNT; goto out; } // Create DA approval session. app_session_ref = DAApprovalSessionCreate(kCFAllocatorDefault); if (NULL == app_session_ref) { LOG_ERR("Failed to create DA approval session during unmount."); result = ERR_UMOUNT; goto out; } // Grab the current runloop and add it to the struct. unmount_result.runloop = CFRunLoopGetCurrent(); CFRetain(unmount_result.runloop); // Register our unmount approval callback. DARegisterDiskUnmountApprovalCallback( app_session_ref, kDADiskDescriptionMatchVolumeMountable, approve_unmount_da_callback, NULL); // Add our DA session to the runloop. DASessionScheduleWithRunLoop(session_ref, //CFRunLoopGetCurrent(), unmount_result.runloop, kCFRunLoopDefaultMode); // Add our DA approval session to the runloop. DAApprovalSessionScheduleWithRunLoop(app_session_ref, unmount_result.runloop, kCFRunLoopDefaultMode); // Get a reference to the disk, using the disk name. disk_ref = DADiskCreateFromBSDName(kCFAllocatorDefault, session_ref, device_path); if (NULL == disk_ref) { LOG_ERR("Failed to create DA disk reference during unmount."); result = ERR_UNMOUNT; goto out; } // Actually call the DA service to unmount the disk DADiskUnmountOptions options = force ? kDADiskUnmountOptionForce : kDADiskUnmountOptionDefault; DADiskUnmount(disk_ref, options, &mount_disk_da_callback, &unmount_result); LOG_DEBUG("Waiting for unmount callback from DA service..."); (void)CFRunLoopRun(); CFRelease(unmount_result.runloop); if (0 != (result = unmount_result.result)) { LOG_ERR("The DA service failed to unmount the device [%s].", error_str(unmount_result.result)); goto out; } LOG_DEBUG("unmount successful"); out: if (NULL != session_ref) { DASessionUnscheduleFromRunLoop(session_ref, CFRunLoopGetCurrent(), kCFRunLoopDefaultMode); CFRelease(session_ref); } if (NULL != app_session_ref) { DAUnregisterApprovalCallback(app_session_ref, (void*)approve_unmount_da_callback, NULL); DAApprovalSessionUnscheduleFromRunLoop(app_session_ref, CFRunLoopGetCurrent(), kCFRunLoopDefaultMode); CFRelease(app_session_ref); } if (NULL != disk_ref) CFRelease(disk_ref); return result; } ---- _______________________________________________ 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: multiple symbols of select(), do they matter?
by Terry Lambert 07 Sep '08

07 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Le 7 sept. 08 à 10:59, Luke Daley a écrit : On 07/09/2008, at 6:42 PM, Jean-Daniel Dupas wrote: for example $1050 => Mac OS X 10.5 and later behavior change $DARWIN_EXTSN => Extended behavior beyond standards $INODE64 => 64-bit ino_t values $LDBL128 => 128-bit long double support (32-bit PowerPC only) $UNIX2003 => UNIX™ conformance $NOCANCEL => (used internally) -- Terry _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… On Sep 7, 2008, at 2:49 AM, Jean-Daniel Dupas <devlists(a)shadowlab.org> wrote: Why you do no simply call the matching function instead of using the same select call for all variants ? int select_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { my_before_select(); int result; int (*real_select)(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) = dlsym(RTLD_NEXT, "select$UNIX2003"); result = real_select(nfds, readfds, writefds, errorfds, timeout); my_after_select(); return result; } That's what I am looking to do, was more curious about what the _actual_ differences are with the different symbols. -- I don't know is there is a doc that explains all differences, but at least, there is this: Generaly, the NOCANCEL variant is a variant that does not have posix thread cancelation point (but should according to the standard). This was done to avoid binary incompatibility with people who used the test check. It was probably overcautios, but it was for you. This email sent to site_archiver(a)lists.apple.com

1 0

Re: multiple symbols of select(), do they matter?
by Jean-Daniel Dupas 07 Sep '08

07 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Le 7 sept. 08 à 10:59, Luke Daley a écrit : On 07/09/2008, at 6:42 PM, Jean-Daniel Dupas wrote: for example $1050 => Mac OS X 10.5 and later behavior change $DARWIN_EXTSN => Extended behavior beyond standards $INODE64 => 64-bit ino_t values $LDBL128 => 128-bit long double support (32-bit PowerPC only) $UNIX2003 => UNIX™ conformance $NOCANCEL => (used internally) _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… Why you do no simply call the matching function instead of using the same select call for all variants ? int select_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { my_before_select(); int result; int (*real_select)(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) = dlsym(RTLD_NEXT, "select$UNIX2003"); result = real_select(nfds, readfds, writefds, errorfds, timeout); my_after_select(); return result; } That's what I am looking to do, was more curious about what the _actual_ differences are with the different symbols. -- I don't know is there is a doc that explains all differences, but at least, there is this: Generaly, the NOCANCEL variant is a variant that does not have posix thread cancelation point (but should according to the standard). UNIX2003 signifie a change in the function behavior. For select, this is only a change in the way it handle the timeout argument. I think one of the variant update it, but not the other. If you realy want details about the changes, you may have a look at the libc sources. This email sent to site_archiver(a)lists.apple.com

1 0

Re: multiple symbols of select(), do they matter?
by Luke Daley 07 Sep '08

07 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 07/09/2008, at 6:42 PM, Jean-Daniel Dupas wrote: for example -- LD. _______________________________________________ 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… Why you do no simply call the matching function instead of using the same select call for all variants ? int select_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { my_before_select(); int result; int (*real_select)(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) = dlsym(RTLD_NEXT, "select$UNIX2003"); result = real_select(nfds, readfds, writefds, errorfds, timeout); my_after_select(); return result; } That's what I am looking to do, was more curious about what the _actual_ differences are with the different symbols. This email sent to site_archiver(a)lists.apple.com

1 0

Re: multiple symbols of select(), do they matter?
by Jean-Daniel Dupas 07 Sep '08

07 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Le 7 sept. 08 à 02:57, Luke Daley a écrit : Hi, I am asking this for read(), write(), open() and close() as well. for example _______________________________________________ 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… For my library that interposes read(), write(), open(), close() and select(), I am wondering if I need to be a bit smarter about what I do depending on which _variant_ of each of these is actually called. My select() stuff looks like this… int select(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select"); int select_darwinextsn(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select$DARWIN_EXTSN"); int select_darwinextsn_nocancel(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select$DARWIN_EXTSN$NOCANCEL"); int select_nocancel_unix2003(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select$NOCANCEL$UNIX2003"); int select_unix2003(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select $UNIX2003"); int select(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { // my code int (*real_select)(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) = dlsym(RTLD_NEXT, "select"); result = real_select(nfds, readfds, writefds, errorfds, timeout); // my code return result; } int select_darwinextsn(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } int select_darwinextsn_nocancel(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } int select_nocancel_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } int select_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } So my question is really two parts; Do they different variants of select() in libSystem actually do different things? And should I be actually fetching the implementation with dlsym() that matches what was actually called by the client of my code? For example, the 'NOCANCEL' variant makes me nervous, seems like I should be doing something there. But, I have no idea what. Thanks, Why you do no simply call the matching function instead of using the same select call for all variants ? int select_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { my_before_select(); int result; int (*real_select)(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) = dlsym(RTLD_NEXT, "select$UNIX2003"); result = real_select(nfds, readfds, writefds, errorfds, timeout); my_after_select(); return result; } This email sent to site_archiver(a)lists.apple.com

1 0

multiple symbols of select(), do they matter?
by Luke Daley 07 Sep '08

07 Sep '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi, I am asking this for read(), write(), open() and close() as well. Thanks, -- LD. _______________________________________________ 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… For my library that interposes read(), write(), open(), close() and select(), I am wondering if I need to be a bit smarter about what I do depending on which _variant_ of each of these is actually called. My select() stuff looks like this… int select(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select"); int select_darwinextsn(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select $DARWIN_EXTSN"); int select_darwinextsn_nocancel(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select$DARWIN_EXTSN$NOCANCEL"); int select_nocancel_unix2003(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select$NOCANCEL$UNIX2003"); int select_unix2003(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict) __asm("_select $UNIX2003"); int select(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { // my code int (*real_select)(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) = dlsym(RTLD_NEXT, "select"); result = real_select(nfds, readfds, writefds, errorfds, timeout); // my code return result; } int select_darwinextsn(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } int select_darwinextsn_nocancel(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } int select_nocancel_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } int select_unix2003(int nfds, fd_set * __restrict readfds, fd_set * __restrict writefds, fd_set * __restrict errorfds, struct timeval * __restrict timeout) { return select(nfds, readfds, writefds, errorfds, timeout); } So my question is really two parts; Do they different variants of select() in libSystem actually do different things? And should I be actually fetching the implementation with dlsym() that matches what was actually called by the client of my code? For example, the 'NOCANCEL' variant makes me nervous, seems like I should be doing something there. But, I have no idea what. This email sent to site_archiver(a)lists.apple.com

1 0