March 2008 - Darwin-kernel - lists.apple.com

Re: current_proc() issue on leopard
by Terry Lambert 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com The following code gives the current process name. struct proc *p = current_proc(); p->p_comm // contains current process name On Tiger, it is working fine, but on leopard, p_comm is empty. -- Terry _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… On Mar 20, 2008, at 2:34 AM, Quinn wrote: At 14:38 +0530 20/3/08, Praveen Kumar wrote: Please let me know if there are alternate ways to get the current process name. (struct proc) is private, and has been since 10.4. You should /not/ be trying to access fields within it. In fact, I'm confused as to how this even compiles! Fortunately, there is a much better way to do this. Check out proc_name defined in <sys/proc.h> in the Kernel framework. Also, proc_self (from the same header) is a more modern way to get the current process. If you switch to that, please note that it gives you a reference to the current process that you then have to release using proc_rele. There is also the KPI function proc_selfname(), which returns the name of the current process in a string buffer supplied by the caller. Note that Mike Smith's point about no using the name to make any decisions, particularly security decisions, is a good one. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Can not copy files >=4GB. Leopard (for NFS)
by Mike Mackovitch 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Mutt/1.5.17 (2007-11-01) This is a known problem with a fix in the pipeline. It is quite unfortunate that the capability defined to indicate "large file support" (i.e. supports files >4GB) was misnamed/misdocumented as the capability of supporting files >=2TB. Combine that with an NFSv3 file server that reports a maximum file size less than 2TB and confusion ensues... Thanks --macko On Thu, Mar 20, 2008 at 11:30:31AM -0700, Doug Rudoff wrote: > >From an OSX 10.5.2 system I have not been able to copy a file larger > than 4Gb using drag-and-drop to an NFS server. The error message is > "Sorry, the operation could not be completed because an unexpected > error occurred (Error Code 0)" > > I've confirmed the same problem with these NFS servers: > - Linux CentOS 4.6 > - FreeBSD 5.4 > - FreeBSD 6.1 > - Solaris 9 > > There has been no problem copying using the command line. > > It seems likely that this is related to Quinn's message from November: > http://lists.apple.com/archives/darwin-kernel/2007/Nov/msg00092.html > > Quinn wrote: > > At 15:01 +0400 20/9/07, Alexander Obuschenko wrote: > >> There is a problem when copying a file greater than 4 GB to > partitions > >> mounted by our fs-plugin (NTFS for Mac OS X driver) using the Finder > >> application. > ... > > After a bit of digging I've discovered that this error indicates > > that the file is too large for the destination volume. Specifically, > > the file is 'large' (4 GB or more) but the destination volume does not > > indicate that it supports large files. > ... > > The kernel's getattrlist compatibility code sets this attribute > > based on the value you put in the f_capabilities attribute returned by > > your VFS plug-in's getattr VFS entry point. In your case you're > > stating that your volume understands the VOL_CAP_FMT_2TB_FILESIZE > > capability (the value is set in the "valid" field) but it does not > > support that capability (the value is clear in the "capabilities" > > field). > > > > If your VFS plug-in does actually support files of 4 GB or larger on > > disk, I recommend that you claim to support this by returning the > > VOL_CAP_FMT_2TB_FILESIZE capability set. > > >From this I can infer that it appears that Apple's VFS plug-in for NFS > doesn't have doesn't have the VOL_CAP_FMT_2TB_FILESIZE capability set. > > My questions are: > 1) Is there any way around this NFS 4Gb limitation now? > 2) Is Apple aware of this problem? > 3) Will this be fixed in an upcoming release? > > Many thanks for any help. > > Doug Rudoff | Software Engineer > Isilon Systems P +1-206-315-7500 F +1-206-315-7485 > www.isilon.com D +1-206-315-7674 > How breakthroughs begin.(tm) > > > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/darwin-kernel/macko%40apple.com > > This email sent to macko(a)apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Can not copy files >=4GB. Leopard (for NFS)
by Rick Macklem 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Thu, 20 Mar 2008, Doug Rudoff wrote: than 4Gb using drag-and-drop to an NFS server. The error message is "Sorry, the operation could not be completed because an unexpected error occurred (Error Code 0)" I've confirmed the same problem with these NFS servers: - Linux CentOS 4.6 - FreeBSD 5.4 - FreeBSD 6.1 - Solaris 9 There has been no problem copying using the command line. It seems likely that this is related to Quinn's message from November: http://lists.apple.com/archives/darwin-kernel/2007/Nov/msg00092.html mounted by our fs-plugin (NTFS for Mac OS X driver) using the Finder application. After a bit of digging I've discovered that this error indicates that the file is too large for the destination volume. Specifically, the file is 'large' (4 GB or more) but the destination volume does not indicate that it supports large files. The kernel's getattrlist compatibility code sets this attribute based on the value you put in the f_capabilities attribute returned by your VFS plug-in's getattr VFS entry point. In your case you're stating that your volume understands the VOL_CAP_FMT_2TB_FILESIZE capability (the value is set in the "valid" field) but it does not support that capability (the value is clear in the "capabilities" field). If your VFS plug-in does actually support files of 4 GB or larger on disk, I recommend that you claim to support this by returning the VOL_CAP_FMT_2TB_FILESIZE capability set. doesn't have doesn't have the VOL_CAP_FMT_2TB_FILESIZE capability set. My questions are: 1) Is there any way around this NFS 4Gb limitation now? 2) Is Apple aware of this problem? 3) Will this be fixed in an upcoming release? Obviously I can't answer for Apple, but a quick glance at the xnu-1228 nfs sources shows that VOL_CAP_FMT_2TB_FILESIZE is set when an NFSv3 server replies 0x20000000000ULL (2TByte) or greater in the maxfilesize reply field of the FSINFO RPC. My guess is that servers you list above don't report that large a value. (I supose you could argue that they should actually test for > 4GB, but??) If you want it to work against your server, you could "cheat" and report a maxfilesize >= 2GB for NFSv3, assuming you support > 4GB. I just took a quick glance at the sources, so figure the above is worth the $0.00 you paid for it:-) rick _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… >From an OSX 10.5.2 system I have not been able to copy a file larger Quinn wrote: At 15:01 +0400 20/9/07, Alexander Obuschenko wrote: There is a problem when copying a file greater than 4 GB to partitions ... ... >From this I can infer that it appears that Apple's VFS plug-in for NFS This email sent to site_archiver(a)lists.apple.com

1 0

Re: Can not copy files >=4GB. Leopard (for NFS)
by Doug Rudoff 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Thread-index: AciKuHqVJzZ45gfrSFS3aMnGFr1eXg== Thread-topic: Re: Can not copy files >=4GB. Leopard (for NFS) >From an OSX 10.5.2 system I have not been able to copy a file larger than 4Gb using drag-and-drop to an NFS server. The error message is "Sorry, the operation could not be completed because an unexpected error occurred (Error Code 0)" I've confirmed the same problem with these NFS servers: - Linux CentOS 4.6 - FreeBSD 5.4 - FreeBSD 6.1 - Solaris 9 There has been no problem copying using the command line. It seems likely that this is related to Quinn's message from November: http://lists.apple.com/archives/darwin-kernel/2007/Nov/msg00092.html Quinn wrote: > At 15:01 +0400 20/9/07, Alexander Obuschenko wrote: >> There is a problem when copying a file greater than 4 GB to partitions >> mounted by our fs-plugin (NTFS for Mac OS X driver) using the Finder >> application. ... > After a bit of digging I've discovered that this error indicates > that the file is too large for the destination volume. Specifically, > the file is 'large' (4 GB or more) but the destination volume does not > indicate that it supports large files. ... > The kernel's getattrlist compatibility code sets this attribute > based on the value you put in the f_capabilities attribute returned by > your VFS plug-in's getattr VFS entry point. In your case you're > stating that your volume understands the VOL_CAP_FMT_2TB_FILESIZE > capability (the value is set in the "valid" field) but it does not > support that capability (the value is clear in the "capabilities" > field). > > If your VFS plug-in does actually support files of 4 GB or larger on > disk, I recommend that you claim to support this by returning the > VOL_CAP_FMT_2TB_FILESIZE capability set. >From this I can infer that it appears that Apple's VFS plug-in for NFS doesn't have doesn't have the VOL_CAP_FMT_2TB_FILESIZE capability set. My questions are: 1) Is there any way around this NFS 4Gb limitation now? 2) Is Apple aware of this problem? 3) Will this be fixed in an upcoming release? Many thanks for any help. Doug Rudoff | Software Engineer Isilon Systems P +1-206-315-7500 F +1-206-315-7485 www.isilon.com D +1-206-315-7674 How breakthroughs begin.(tm) _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Re: per process file attributes
by Todd Heberlein 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Todd _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… If you do, you need to be willing to update your application every time we do a software update of any kind, just like we have to be prepared to update ps, lsof, and anything else that uses SPI any time we do a Software Update. Way back in the early 1990s during the SunOS 4.x days we experienced the same problem in our code that looked at internal kernel structures. Having your code die every time the customer updated their software kind of sucked. This email sent to site_archiver(a)lists.apple.com

1 0

Re: current_proc() issue on leopard
by Michael Smith 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Mar 20, 2008, at 2:08 AM, Praveen Kumar wrote: The following code gives the current process name. struct proc *p = current_proc(); p->p_comm // contains current process name = Mike _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… On Tiger, it is working fine, but on leopard, p_comm is empty. Please let me know if there are alternate ways to get the current process name. Some unsolicited advice: do not use the process name for anything other than debugging purposes. In particular, do not make any decisions whatsoever based on the name that a given process might have. The name is not guaranteed to be unique, and it can be changed trivially. This email sent to site_archiver(a)lists.apple.com

1 0

Re: current_proc() issue on leopard
by Quinn 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com The following code gives the current process name. struct proc *p = current_proc(); p->p_comm // contains current process name On Tiger, it is working fine, but on leopard, p_comm is empty. S+E -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Relations, Developer Technical Support, Core OS/Hardware _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… At 14:38 +0530 20/3/08, Praveen Kumar wrote: Please let me know if there are alternate ways to get the current process name. (struct proc) is private, and has been since 10.4. You should /not/ be trying to access fields within it. In fact, I'm confused as to how this even compiles! Fortunately, there is a much better way to do this. Check out proc_name defined in <sys/proc.h> in the Kernel framework. Also, proc_self (from the same header) is a more modern way to get the current process. If you switch to that, please note that it gives you a reference to the current process that you then have to release using proc_rele. This email sent to site_archiver(a)lists.apple.com

1 0

Re: per process file attributes
by Quinn 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com S+E -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Relations, Developer Technical Support, Core OS/Hardware _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… At 23:31 -0700 19/3/08, Terry Lambert wrote: My advice is to parse the output. Also, lsof has an option to create output that's easily parseable. Check out the "OUTPUT FOR OTHER PROGRAMS" section of <x-man-page://8/lsof>. If you can require Mac OS X 10.5 or later, you can also use the <libproc.h> API to do this. This email sent to site_archiver(a)lists.apple.com

1 0

current_proc() issue on leopard
by Praveen Kumar 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Hi, The following code gives the current process name. struct proc *p = current_proc(); p->p_comm // contains current process name - M Praveen Kumar. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… On Tiger, it is working fine, but on leopard, p_comm is empty. Please let me know if there are alternate ways to get the current process name. This email sent to site_archiver(a)lists.apple.com

1 0

Re: per process file attributes
by Terry Lambert 20 Mar '08

20 Mar '08

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com It will probably not be helpful. My advice is to parse the output. -- Terry Manish -----Original Message----- From: Terry Lambert [mailto:tlambert@apple.com] Sent: Thursday, March 20, 2008 11:24 AM To: Manish Chaturvedi Cc: darwin-kernel(a)lists.apple.com Subject: Re: per process file attributes The man page is cribbed from BSD, and not updated. We consider sysctl as SPI due to not having gone through API review, and therefore both unstable and not for use by third parties. If you need this information, consider using "lsof". -- Terry Hi All, I am writing application to list all process and opened files by the process, where I am able to get the process information like Pid,ppid, path, process commadline arguments etc using sysctl() ( followed this link http://developer.apple.com/qa/qa2001/qa1123.html) . I am using the following code to get the file information about every process. int mib[3]; struct extern_file *file = NULL ; // Is it the correct structure ? mib[0] = CTL_KERN; mib[1] = KERN_MAXFILES; mib[2] = 1693 ; size = sizeof(argmax) ; if (sysctl((int *)mib, 2, &argmax, &size, NULL, 0) < 0) { printf("No args found\n"); printf("Error is %s\n", strerror(errno)) ; } else { printf("size is Given as %d\n",argmax) ; // argmax is giving me the correct values (i.e. count of all opened files ) size = (size_t)argmax ; mib[0] = CTL_KERN; // Here I am resetting the mib arry for getting per process file information . mib[1] = KERN_FILE; mib[2] = 1693 ;//1693 ;//procList[i].kp_proc.p_pid; file =(extern_file*) malloc(argmax); if(file == NULL) { printf("file pointe not allocated yes\n") ; } if (sysctl((int *)mib, 3, fileNew, &size, NULL, 0) < 0) { printf("Error in getting file struct %s\n", strerror(errno)) ; } else { // Print file attributes } I want to list all file descriptors file paths and othet file attributes , which I am not getting with this program, Getting an error "Not a directory" during second sysctl call, where as getting the correct file count after first sysctl() call. In the man page of sysctl() KERN_FILE will return a file structure which is not at all documented ( not confirm about the structure to use ), there is no file structure exposed on Mac OS X (apart from extern_file and filedesc), I am not sure which structure to use ? is it possible to get per process file information using KERN_FILE? OS version is 10.4.7 Thanks in advance! Regards, Manish. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/tlambert%40apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… If you do, you need to be willing to update your application every time we do a software update of any kind, just like we have to be prepared to update ps, lsof, and anything else that uses SPI any time we do a Software Update. Most vendors like to keep their application from breaking every month or so, though, since SPI can (and does) change all the time. The bright spot is that we update those things that depend on SPI in lock-step with software updates, so things like lsof, ps, netstat, and so on keep working. Generally, people are happier talking to the programs that use SPI (and stay working between updates) than they are talking to SPI (even if that means having to parse program output). It means they don't have support calls and suffer being broken a couple of weeks during which they scramble to fix their application to use the new SPI for their irate customers. This gets even more complicated if you specify a release target, since the SDK headers are not updated as a result of an OS SU, rather only as a result of an XCode SU. The XCode and OS schedules usually don't sync up, so you'll be using the wrong header files until they do, if you end up using an SDK. So unless you are working from the xnu and/ or utility sources for something that already knows how to output what you are trying to generate, you won't necessarily have the internal headers you need until the sources post, which is yet more time you will be broken. On Mar 19, 2008, at 11:08 PM, Manish Chaturvedi wrote: Thanks Terry for quick reply, but using lsof in our program will make me to parse the output, which seems to be a complicated job, is there is any alternate solution to this. Will be helpful if I follow the source code of lsof to do the same? On Mar 19, 2008, at 10:36 PM, Manish Chaturvedi wrote: This email sent to tlambert(a)apple.com This email sent to site_archiver(a)lists.apple.com

1 0