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

Re: CFMessagePortSendRequest seems to ignore sendTimeout
by Stéphane Sudre 13 Mar '05

13 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On dimanche, mars 13, 2005, at 11:11 PM, David Niemeijer wrote: Note that I am not interested in getting a reply. Shouldn't you use Maybe, but this is what I found in the online docs: The documentation is correct and you are too. _______________________________________________ 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… At 22:37 +0100 13/3/05, Stéphane Sudre wrote: On dimanche, mars 13, 2005, at 10:31 PM, David Niemeijer wrote: Does anyone here have experience with timeouts and CFMessagePortSendRequest? As far as I understand CFTimeInterval is in seconds, so if I pass 0.02 for sendTimeout when calling CFMessagePortSendRequest I would expect a kCFMessagePortSendTimeout error if more than 0.02 secs are needed. However, what I am seeing is that CFMessagePortSendRequest can take for example 6 ticks (= 0.1 sec) and still return with noErr, so not timing out as I would have hoped. Am I doing something wrong, is the documentation incorrect, or is this a bug? I am running on 10.3.8 and calling it like this: result = CFMessagePortSendRequest( sMessagePort, inMessageID, inData, 0.02, 0, NULL, &replyData ) ; A related question: What kind of things influence the duration of this call. To what degree does what happens at the receiving end influence the duration of the call. I am not working with return values so I am presuming that as soon as the CFMessagePortCallBack at the receiving end is called CFMessagePortSendRequest is able to return and not delayed by any processing inside the callback, is this a correct assumption? result = CFMessagePortSendRequest( sMessagePort, inMessageID, inData, 0.02, 0, kCFRunLoopDefaultMode, NULL ) ; if you don't want to deal with the return value and return as soon as the message is sent? /* NULL replyMode argument means no return value expected, dont wait for it */ replyMode The run loop mode in which the function should wait for a reply. Pass NULL if you do not want or expect a reply. returnData On return, a pointer to a CFData containing the reply data. You are responsible for releasing the data. Which suggests that I may pass NULL to the second last parameter (if I don't want a reply) and does not tell me whether NULL is allowed for the last one. But perhaps I misread this or perhaps the docs are wrong or incomplete. I had a look at the CFMessagePortSendRequest source code and the time-out is used for the mach_msg call. So the time-out does not apply to the whole CFMessagePortSendRequest code and there does not seem to be any atomic restrictions in the code which would prevent other processes to take CPU time during the execution of this method till the call to mach_msg. Not sure it would explain the situation. This email sent to site_archiver(a)lists.apple.com

1 0

Re: CFMessagePortSendRequest seems to ignore sendTimeout
by David Niemeijer 13 Mar '05

13 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Note that I am not interested in getting a reply. Shouldn't you use Maybe, but this is what I found in the online docs: david. _______________________________________________ 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… At 22:37 +0100 13/3/05, Stéphane Sudre wrote: On dimanche, mars 13, 2005, at 10:31 PM, David Niemeijer wrote: Does anyone here have experience with timeouts and CFMessagePortSendRequest? As far as I understand CFTimeInterval is in seconds, so if I pass 0.02 for sendTimeout when calling CFMessagePortSendRequest I would expect a kCFMessagePortSendTimeout error if more than 0.02 secs are needed. However, what I am seeing is that CFMessagePortSendRequest can take for example 6 ticks (= 0.1 sec) and still return with noErr, so not timing out as I would have hoped. Am I doing something wrong, is the documentation incorrect, or is this a bug? I am running on 10.3.8 and calling it like this: result = CFMessagePortSendRequest( sMessagePort, inMessageID, inData, 0.02, 0, NULL, &replyData ) ; A related question: What kind of things influence the duration of this call. To what degree does what happens at the receiving end influence the duration of the call. I am not working with return values so I am presuming that as soon as the CFMessagePortCallBack at the receiving end is called CFMessagePortSendRequest is able to return and not delayed by any processing inside the callback, is this a correct assumption? result = CFMessagePortSendRequest( sMessagePort, inMessageID, inData, 0.02, 0, kCFRunLoopDefaultMode, NULL ) ; if you don't want to deal with the return value and return as soon as the message is sent? /* NULL replyMode argument means no return value expected, dont wait for it */ replyMode The run loop mode in which the function should wait for a reply. Pass NULL if you do not want or expect a reply. returnData On return, a pointer to a CFData containing the reply data. You are responsible for releasing the data. Which suggests that I may pass NULL to the second last parameter (if I don't want a reply) and does not tell me whether NULL is allowed for the last one. But perhaps I misread this or perhaps the docs are wrong or incomplete. This email sent to site_archiver(a)lists.apple.com

1 0

Re: CFMessagePortSendRequest seems to ignore sendTimeout
by Stéphane Sudre 13 Mar '05

13 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On dimanche, mars 13, 2005, at 10:31 PM, David Niemeijer wrote: Note that I am not interested in getting a reply. Shouldn't you use _______________________________________________ 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 got a suggestion to post this question here rather than on Carbon-dev; apologies if you got it there too. Does anyone here have experience with timeouts and CFMessagePortSendRequest? As far as I understand CFTimeInterval is in seconds, so if I pass 0.02 for sendTimeout when calling CFMessagePortSendRequest I would expect a kCFMessagePortSendTimeout error if more than 0.02 secs are needed. However, what I am seeing is that CFMessagePortSendRequest can take for example 6 ticks (= 0.1 sec) and still return with noErr, so not timing out as I would have hoped. Am I doing something wrong, is the documentation incorrect, or is this a bug? I am running on 10.3.8 and calling it like this: result = CFMessagePortSendRequest( sMessagePort, inMessageID, inData, 0.02, 0, NULL, &replyData ) ; A related question: What kind of things influence the duration of this call. To what degree does what happens at the receiving end influence the duration of the call. I am not working with return values so I am presuming that as soon as the CFMessagePortCallBack at the receiving end is called CFMessagePortSendRequest is able to return and not delayed by any processing inside the callback, is this a correct assumption? result = CFMessagePortSendRequest( sMessagePort, inMessageID, inData, 0.02, 0, kCFRunLoopDefaultMode, NULL ) ; if you don't want to deal with the return value and return as soon as the message is sent? /* NULL replyMode argument means no return value expected, dont wait for it */ This email sent to site_archiver(a)lists.apple.com

1 0

CFMessagePortSendRequest seems to ignore sendTimeout
by David Niemeijer 13 Mar '05

13 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Note that I am not interested in getting a reply. Thanks, david. _______________________________________________ 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 got a suggestion to post this question here rather than on Carbon-dev; apologies if you got it there too. Does anyone here have experience with timeouts and CFMessagePortSendRequest? As far as I understand CFTimeInterval is in seconds, so if I pass 0.02 for sendTimeout when calling CFMessagePortSendRequest I would expect a kCFMessagePortSendTimeout error if more than 0.02 secs are needed. However, what I am seeing is that CFMessagePortSendRequest can take for example 6 ticks (= 0.1 sec) and still return with noErr, so not timing out as I would have hoped. Am I doing something wrong, is the documentation incorrect, or is this a bug? I am running on 10.3.8 and calling it like this: result = CFMessagePortSendRequest( sMessagePort, inMessageID, inData, 0.02, 0, NULL, &replyData ) ; A related question: What kind of things influence the duration of this call. To what degree does what happens at the receiving end influence the duration of the call. I am not working with return values so I am presuming that as soon as the CFMessagePortCallBack at the receiving end is called CFMessagePortSendRequest is able to return and not delayed by any processing inside the callback, is this a correct assumption? This email sent to site_archiver(a)lists.apple.com

1 0

Re: Listusers command
by Stéphane Sudre 12 Mar '05

12 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On samedi, mars 12, 2005, at 12:13 AM, Jason Townsend wrote: From: Stephane Sudre <ssudre(a)intego.com> Well, I was using getpwent in a software too and for one user (not a user account, a real user), it would not work (the accounts were on a Windows server IIRC). Sample was showing a hang in getpw: --------------8<----------------8<--------------8<------------------- --------------8<----------------8<--------------8<------------------- _______________________________________________ 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… Le 11 mars 05, à 14:42, Chuck Gagnon a écrit : It makes use of getpwent from libc, I would hope there is a backend bridge to DirectoryServices just like there is to NIS as such, perhaps I'm wrong there. My account which is not in /etc/passwd shows up in the list it. if the libc functions are to be abandon in favor of an API then all the BSD tools that make use of them (users, groups, who...) will need to be retooled. With the DirectoryServices, it worked. In some cases it may not be possible to iterate over all users like this due to scale reasons. For example, many LDAP servers limit query results to 500 records, so if there are thousands of users in the server you won't get all of them. If you instead do a getpwnam() which is an exact match lookup those users should show up. Going through the Directory Services APIs allows you to do searches that will translate into the equivalent LDAP search, while calling getpwent() over and over and filtering the results will not. Was that the problem, or was it something else? Analysis of sampling pid 795 every 10.000000 milliseconds Call graph: 98 Thread_0e0b 98 start 98 _start 98 NSApplicationMain 98 +[NSBundle(NSNibLoading) loadNibNamed:owner:] 98 +[NSBundle(NSNibLoading) loadNibFile:externalNameTable:withZone:] 98 +[NSBundle(NSNibLoading) _loadNibFile:nameTable:withZone:ownerBundle:] 98 old_loadNib 98 -[NSIBObjectData nibInstantiateWithOwner:topLevelObjects:] 98 -[NSSet makeObjectsPerformSelector:] 98 -[xxxxxxxxxxxxxxxxxxxx awakeFromNib] 98 getpw 98 getpw_internal 98 lu_getpwent 98 _lookup_all 98 _lookup_all_secure 98 mach_msg 98 mach_msg_trap 98 mach_msg_trap The Server was a Windows OS running ActiveDirectory. On the Mac, ADmitMac was running. My $0.02 This email sent to site_archiver(a)lists.apple.com

1 0

Re: Listusers command
by Jason Townsend 11 Mar '05

11 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com From: Stephane Sudre <ssudre(a)intego.com> Well, I was using getpwent in a software too and for one user (not a user account, a real user), it would not work (the accounts were on a Windows server IIRC). Was that the problem, or was it something else? -Jason _______________________________________________ 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… Le 11 mars 05, à 14:42, Chuck Gagnon a écrit : It makes use of getpwent from libc, I would hope there is a backend bridge to DirectoryServices just like there is to NIS as such, perhaps I'm wrong there. My account which is not in /etc/passwd shows up in the list it. if the libc functions are to be abandon in favor of an API then all the BSD tools that make use of them (users, groups, who...) will need to be retooled. With the DirectoryServices, it worked. In some cases it may not be possible to iterate over all users like this due to scale reasons. For example, many LDAP servers limit query results to 500 records, so if there are thousands of users in the server you won't get all of them. If you instead do a getpwnam() which is an exact match lookup those users should show up. Going through the Directory Services APIs allows you to do searches that will translate into the equivalent LDAP search, while calling getpwent() over and over and filtering the results will not. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Listusers command
by Jason Townsend 11 Mar '05

11 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com So this is implementation should work fine. -Jason From: Chuck Gagnon <chuck.gagnon(a)mac.com> It makes use of getpwent from libc, I would hope there is a backend bridge to DirectoryServices just like there is to NIS as such, perhaps I'm wrong there. My account which is not in /etc/passwd shows up in the list it. if the libc functions are to be abandon in favor of an API then all the BSD tools that make use of them (users, groups, who...) will need to be retooled. I'm not sure if this is the appropriate way to do this, but for ease and compatibility with the various flavors of unix I administer at work I wrote this code. It is under a BSD style license. Code is at: http://homepage.mac.com/gagnocg/downloads/listusers.zip Hum, I'm not sure it would work correctly on OS X in every case. AFAIK, the only safe way to have the correct list of users is to use the Directory Services API. But I can be wrong. _______________________________________________ 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… There is a DSAgent in lookupd which bridges between POSIX style calls like getpwnam()/getpwent() and others and the Directory Services framework. It would not make much sense to drop support for these POSIX APIs since so much existing software uses them. On Mar 11, 2005, at 4:30 AM, Stephane Sudre wrote: On 9 mars 05, at 05:45, Chuck Gagnon wrote: It is basically a work alike to the SYSV command of the same name (man page included), with no arguments it lists all users in a sorted list showing username and comment (gecos) field. the to options are [ -l users ] to list specific users and or [ -g groups ] to list all users in the given group(s). Users are defined as ID's with a uid greater than 99. We have a lot of scripts that made use of this command on Solaris and HP-UX, and I couldn't find a similar command in the BSD world. The real difficult part with the Directory Services API IMHO is that the APIs are a bit strange from a spelling point of view (Bug filed on this matter and closed as duplicate) Happily, a SDK (not provided with the DevTools) exists at http://developer.apple.com/sdk/ and contains a really good sample code showing how to list users. My $0.02 This email sent to site_archiver(a)lists.apple.com

1 0

Re: Leaking Mach Ports?
by Chris Sarcone 11 Mar '05

11 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dirk -- if (NULL != portRef) { // we have found valid client port if (CFMessagePortIsValid(portRef)) { success = CFMessagePortSendRequest ( portRef, kMyMessageID, sendDataRef, kSendTimeout, kReceiveTimeout, NULL, NULL); CheckError( success); } CFRelease(portRef); } -- Chris -------- 6 Infinite Loop M/S 306-2MS Cupertino CA 95014 phone: (408) 974-4033 fax: (408) 862-7577 email: sarcone(a)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… CFMessagePortRef portRef = CFMessagePortCreateRemote( kCFAllocatorDefault, portNameRef); After stepping through this code the number of send rights remain increased. Is there anything wrong with this code? I'm definitely no expert on CFMessagePort (was just trying to lend a helping hand). However, looking at the source code, it appears it will create a remote port regardless of whether you have specified wanting a reply or not. In your case, it appears you don't want to wait for a reply. It looks like CFMessagePortInvalidate() might need to be called after sending the message in order to get the reply port released. I'm not sure if this works as intended or if it's a bug. Looking at the documentation doesn't provide any extra insight either (at least for me). I'd suggest filing a bug with this example and see what those more familiar with CFMessagePort say... This email sent to site_archiver(a)lists.apple.com

1 0

Re: Leaking Mach Ports?
by Matt Watson 11 Mar '05

11 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com matt. _______________________________________________ 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 seem to recall that you have to call mach_port_deallocate() for any port returned from mach_thread_self()...Are you doing that? No one should be using mach_thread_self(), unless you are creating raw mach threads. Use pthread_mach_thread_np() instead. This does not require a mach_port_deallocate(). I believe the port leak being seen here is normal, as a thread port remains after the thread is detached, to be cleaned up by the next pthread_create() or pthread_exit() that is called. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Leaking Mach Ports?
by Chris Sarcone 11 Mar '05

11 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dirk -- 0x6303 3 0x6403 1 0x6503 1 1 0x6603 1 0x6703 1 0x6303 2 which gives 2 DeadName rights. -- Chris ------------------ 6 Infinite Loop M/S 306-2MS Cupertino CA 95014 phone: (408) 974-4033 fax: (408) 862-7577 email: sarcone(a)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… After calling pthread_create() and pthread_detach() MachPortDump shows these additional ports: This makes up for five ports with a total of 4 send rights, 3 receive rights and a PortSet. 0x6303 is the result for mach_thread_self() in this example. After terminating the thread these rights still exist: How could I care for releasing these references? I seem to recall that you have to call mach_port_deallocate() for any port returned from mach_thread_self()...Are you doing that? This email sent to site_archiver(a)lists.apple.com

1 0