July 2003 - Darwin-kernel - lists.apple.com

Re: I/O in my Filesystem causing application crashes.
by Quinn 19 Jul '03

19 Jul '03

At 17:35 -0700 16/7/03, raldrich(a)mac.com wrote: Any ideas? My guess is that this has something to do with how your VOP_PAGEIN entry point is implemented. I did some poking around in Preview with GDB and it seems that it memory maps the image file (using mmap). [NSBMPImageReader getByte] then reads the file by reading the mapped memory. Unfortunately I don't have a recommendation for how you might implement VOP_PAGEIN given your current I/O design. I'd have to study the problem in more detail than I have time for right now. S+E -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Technical Support * Networking, Communications, Hardware _______________________________________________ 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: Increasing the value of kern.ipc.maxsockets?
by Laurent Dumont 18 Jul '03

18 Jul '03

Jeramey, On OS X/ Darwin kern.ipc.maxsockets is the initial value of for the zone size, but it will be grown dynamically as more sockets are created. There is no hard limit for the max number of sockets, it's dependent of your memory configuration. To have an idea of the number of sockets in use, you can watch: "sysctl net.inet.tcp.pcb" which will tell you the number of TCP pcb/sockets that are currently in use (including those in time wait), that's a good reflection of the socket use on a system. Laurent on 07/15/03 15:26, Jeramey Crawford at jeramey(a)onereel.org wrote: > One of our OS X servers (kernel version 6.6) is running a set of > applications that, together, eat up a lot of sockets. The default of > 512 is simply too low of a ceiling for the kind of load we'd like to > throw at the machine. I have tried increasing it by running 'sysctl -w > kern.ipc.maxsockets=xxx', however this is a read-only value in a > running kernel. > > I am wondering if there is a way to increase it at boot time a la > FreeBSD's method of editing /boot/loader.conf with the appropriate > value. I'd prefer to not have to recompile a kernel just to get more > sockets available for the system. > > Thanks! _______________________________________________ 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

Calling physio() with no buf pointer panics
by Sam Vaughan 17 Jul '03

17 Jul '03

In the physio() function in xnu/bsd/kern/physio.c, there's a panic if the supplied buf pointer is NULL. There's also a commented-out call on the previous line to getphysbuf(), and an ifdef'd out call to putphysbuf at the end. According to the algorithm for physio() on p.203, figure 6.3 of The Documentation*, physio should allocate a buffer if one wasn't passed in. Is there any reason for the removal of this code? getphysbuf() and putphysbuf() don't seem to be defined anywhere, which is a pretty good reason I guess. If I allocate my own struct buf for this purpose, is there a recommended call to make to initialise it correctly? Thanks, Sam * The Design and Implementation of the 4.4 BSD Operating System, McKusick et al, Addison Wesley 1996, for those new to Apple's BSD kernel documentation ;-) _______________________________________________ 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

Recommendations for avoiding a bug in devfs?
by Sam Vaughan 17 Jul '03

17 Jul '03

I just reported a bug in devfs (Radar 3330941) that makes it impossible to add a device that contains an intermediate path. Say I wanted to add a device foo/bar, i.e. full path /dev/foo/bar. When I call devfs_make_node() with the "foo/bar" path, devfs_make_node() calls dev_finddir() with "foo" to create the intermediate directory. dev_finddir() then copies the string "foo" into its private 'component' string, but forgets to add a terminating '\0' character. The result is a directory in /dev with a name that starts with "foo", but has some random junk on the end. Sure enough, if you manage to list this directory, the device "bar" will be inside. I need to work around this bug, and I was wondering if anyone else has already done so. Is it possible to add a device "foo" with dummy properties, then add a child device "bar" to it separately? If not, is my only option to include my own fixed version of these functions in my code? The product I'm building is only targetting Jaguar, so I'm not concerned about the fact that this will probably break on future builds of Darwin. Thanks in advance for any suggestions, Sam _______________________________________________ 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: Problems with my foreign file system
by raldrich＠mac.com 17 Jul '03

17 Jul '03

On Tuesday, July 15, 2003, at 08:29 PM, Steven Bytnar wrote: On Tue, Jul 15, 2003 at 06:19:13PM -0700, raldrich(a)mac.com wrote: Somewhere between statfs and PBHGetVolParmsSync, the /dev/ prefix is stripped from the vMDeviceID parameter - however, it is not stripped when my filesystem is called. I wrote a simple program to test PBHGetVolParmsSync - it simply iterates all of the volumes and dumps the vMDevice parameter for each - this is its output: vMDeviceID = disk0s9 vMDeviceID = disk0s10 vMDeviceID = disk1s9 vMDeviceID = disk1s10 vMDeviceID = /dev/disk2s0 vMDeviceID = afp_1Q2ykU1Q2ykU1Q2ykU1Q2ykU-0.2f000002 I'm guessing (based on the icon problem, and on the vMDeviceID for a network volume, that PBHGetVolParmsSync has decided for some reason that mine is a network volume, and that it therefore shouldn't strip the /dev/ prefix. What I don't know, is what flag is messed up which would cause that. Any ideas? How about printing out the other info in the GetVolParmsInfoBuffer struct? It sounds like you might have answered the question yourself... What happens if you make it so your file system has GetVolParmsInfoBuffer.vMServerAdr == 0? I don't know the nature of your file system, so I can only guess that you're acting like a network file system. You might do some IORegistry (ioreg) spelunking. How does your device's IOMedia entry compare to other hard disk entries? It turns out that the MNT_LOCAL flag in mount_ptr->mnt_flag wasn't set, thus everything thought that it was a network disk. Thanks, Ron _______________________________________________ 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

I/O in my Filesystem causing application crashes.
by raldrich＠mac.com 17 Jul '03

17 Jul '03

Folks, I'm running into a problem with I/O which has me puzzled - when I copy a file from a disk mounted by my filesystem, to the boot drive (HFS extended), it works fine, and the resulting file can be opened and read properly. However, when I try to open the same file directly with an application such as Preview, the Preview application crashes as follows: ********** Date/Time: 2003-07-16 16:35:40 -0700 OS Version: 10.2.6 (Build 6L60) Host: schizm.local. Command: Preview PID: 521 Exception: EXC_BAD_ACCESS (0x0001) Code[0]: 0x0000000aCode[1]: 0x01cef000 Thread 0 Crashed: #0 0x9323fc3c in -[NSBMPImageReader getByte] #1 0x9323fdd0 in -[NSBMPImageReader loadBitmapFileHeader] #2 0x932406ec in -[NSBMPImageReader loadImageHeader] #3 0x930ef4a4 in -[NSBitmapImageRep _initWithImageReader:] #4 0x930bd388 in +[NSBitmapImageRep _imagesWithData:hfsFileType:extension:zone:] #5 0x9310e4b8 in +[NSBitmapImageRep _imageRepsWithData:hfsFileType:extension:] #6 0x930e02a4 in +[NSImageRep imageRepsWithContentsOfFile:] #7 0x9312e68c in -[NSImage initWithContentsOfFile:] #8 0x0000c9b4 in 0xc9b4 #9 0x00009c20 in 0x9c20 #10 0x00004e98 in 0x4e98 #11 0x00004ab0 in 0x4ab0 #12 0x931c4a9c in -[NSDocumentController _openDocumentFileAt:display:] #13 0x000112a4 in 0x112a4 #14 0x930f9cac in -[NSApplication sendAction:to:from:] #15 0x9316ec5c in -[NSMenu performActionForItemAtIndex:] #16 0x931af2fc in -[NSCarbonMenuImpl performActionWithHighlightingForItemAtIndex:] #17 0x930ec730 in _NSHandleCarbonMenuEvent #18 0x9308e22c in _DPSNextEvent #19 0x930a0158 in -[NSApplication nextEventMatchingMask:untilDate:inMode:dequeue:] #20 0x930b1d88 in -[NSApplication run] #21 0x9315fc58 in NSApplicationMain #22 0x000046e0 in 0x46e0 #23 0x00004560 in 0x4560 Thread 1: #0 0x90014d28 in syscall_thread_switch #1 0x97e03ef4 in +[NSThread sleepUntilDate:] #2 0x93081cac in -[NSUIHeartBeat _heartBeatThread:] #3 0x97e2cc50 in forkThreadForFunction #4 0x90020d48 in _pthread_body PPC Thread State: srr0: 0x9323fc3c srr1: 0x0200f030 vrsave: 0x00000000 xer: 0x00000000 lr: 0x9323fdd0 ctr: 0x9323fc0c mq: 0x00000000 r0: 0x00000000 r1: 0xbfffe460 r2: 0x00000000 r3: 0x01c4a250 r4: 0x906b679c r5: 0x01c4a250 r6: 0x00000010 r7: 0x00000010 r8: 0x000a1010 r9: 0x00000000 r10: 0x906852a8 r11: 0x01cef000 r12: 0x9323fc0c r13: 0x00000000 r14: 0x00000000 r15: 0x00000000 r16: 0x00000000 r17: 0xbffff770 r18: 0x00000000 r19: 0xffffffff r20: 0x00000000 r21: 0x01c4a250 r22: 0xa30a0060 r23: 0x00000000 r24: 0x001ef1a0 r25: 0x01c4afd0 r26: 0x01cef000 r27: 0x01c4afd0 r28: 0x00028c38 r29: 0x01c4a250 r30: 0x01c4a250 r31: 0x9323fd9c The exception (EXC_BAD_ACCESS) would indicate to me that Preview attempted to read a buffer which had been freed, but I'm not sure how this could have occurred. I should note here that my FS has a somewhat different I/O model from most BSD filesystems - because the FS core was written to be operating system independent, all of the buffer pointers passed through the FS core have to be located in kernel memory. As a result, my vnode_read routine works something like this: 1) MALLOC a buffer in kernel space the size of the read (uio_resid). 2) Call the FS core to read the data into the kernel buffer. 3) Call uiomove to copy the data to user memory. 4) FREE the kernel buffer. I assumed when I wrote that code, that it was going to be pretty inefficient, but according to "Kernel Programming: Boundary Crossings: Memory Mapping and Block Copying", its the best way to handle the problem: > In the Mac OS X kernel, data is most easily copied into kernel space > with the BSD copyin function, and back out to user space with the > copyout function. For large blocks of data, entire pages will be > memory mapped using copy-on-write. For this reason, it is generally > not useful to do memory mapping by hand. Apparently, when I call uiomove, the kernel buffer is co-mapped into both kernel and user space, and then when I FREE the kernel buffer, it is un-mapped from kernel space, while the user space mapping remains. At least that's the theory. The code is so simple, that I wouldn't expect it be able to break, and if it did, I would expect a kernel panic. I certainly wouldn't expect it to work for Finder, and not for other applications. Any ideas? Thanks, Ron Aldrich Software Architects, Inc. _______________________________________________ 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: Problems with my foreign file system
by Steven Bytnar 16 Jul '03

16 Jul '03

On Tue, Jul 15, 2003 at 06:19:13PM -0700, raldrich(a)mac.com wrote: > Somewhere between statfs and PBHGetVolParmsSync, the /dev/ prefix is > stripped from the vMDeviceID parameter - however, it is not stripped > when my filesystem is called. > > I wrote a simple program to test PBHGetVolParmsSync - it simply > iterates all of the volumes and dumps the vMDevice parameter for each - > this is its output: > > vMDeviceID = disk0s9 > vMDeviceID = disk0s10 > vMDeviceID = disk1s9 > vMDeviceID = disk1s10 > vMDeviceID = /dev/disk2s0 > vMDeviceID = afp_1Q2ykU1Q2ykU1Q2ykU1Q2ykU-0.2f000002 > > I'm guessing (based on the icon problem, and on the vMDeviceID for a > network volume, that PBHGetVolParmsSync has decided for some reason > that mine is a network volume, and that it therefore shouldn't strip > the /dev/ prefix. What I don't know, is what flag is messed up which > would cause that. Any ideas? How about printing out the other info in the GetVolParmsInfoBuffer struct? It sounds like you might have answered the question yourself... What happens if you make it so your file system has GetVolParmsInfoBuffer.vMServerAdr == 0? I don't know the nature of your file system, so I can only guess that you're acting like a network file system. You might do some IORegistry (ioreg) spelunking. How does your device's IOMedia entry compare to other hard disk entries? --Steve _______________________________________________ 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: Problems with my foreign file system
by Marek Kozubal 16 Jul '03

16 Jul '03

On Tue, 15 Jul 2003 raldrich(a)mac.com wrote: > I'm guessing (based on the icon problem, and on the vMDeviceID for a > network volume, that PBHGetVolParmsSync has decided for some reason > that mine is a network volume, and that it therefore shouldn't strip > the /dev/ prefix. What I don't know, is what flag is messed up which > would cause that. Any ideas? How are you mounting this volume? If you are doing it manually and then telling DiskArb that the disk has now appeared, DiskArb will assume this is a network volume since you can't tell it otherwise. If DiskArb itself is doing the mount via a *.util program in your .filesystem bundle in /System/Library/Filesystems/, then its probably not figuring out what kind of file system you are in a meaningful enough way to know that you're a local happy disk. Also your file system type might be coming up as a "UNIX" file system to Carbon instead of an HFS volume, if you want it to be HFS you'll have to put into your mnt_stat.f_fstypename field the string "HFS" for the file system ID numbers to work out, plus return the HFS+ code in getattrlist(). But this is irrelevent if you're not supporting volfs with your file system. It'd be a lot easier if you could tell us what your file system is trying to accomplish. --- Marek Kozubal marek(a)portents.com _______________________________________________ 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: Problems with my foreign file system
by raldrich＠mac.com 16 Jul '03

16 Jul '03

On Tuesday, July 15, 2003, at 07:50 AM, Steven Bytnar wrote: > From: Steven Bytnar <sbytnar(a)kagi.com> > Date: Tue Jul 15, 2003 7:50:42 AM US/Pacific > To: raldrich(a)mac.com > Cc: Darwin-Kernel <darwin-kernel(a)lists.apple.com> > Subject: Re: Problems with my foreign file system > > On Mon, Jul 14, 2003 at 06:58:07PM -0700, raldrich(a)mac.com wrote: >> What I don't understand is why Finder hasn't stripped the '/dev/' >> prefix off of the identifier. How does Finder determine what the disk >> identifier for a volume should be? > > The Finder isn't supposed to strip /dev/ away, is it? However you're > returning the BSD name from the file system, you're probably not > supposed to be prepending the "/dev/" in the first place. (Just a > guess based on what PBHGetVolParmsSync is supposed to return.) > I've checked against other (running) file systems to see if the /dev/ prefix is supposed to be returned, and yes, it is. The name of the device node is stored in volume's mount structure, in the form /dev/disk*s* (mount_ptr->mnt_stat.f_mntfromname). There are two FileSystem APIs which access that string - statfs and getattrlist. They both copy the entire string, without modifying it an any way. Somewhere between statfs and PBHGetVolParmsSync, the /dev/ prefix is stripped from the vMDeviceID parameter - however, it is not stripped when my filesystem is called. I wrote a simple program to test PBHGetVolParmsSync - it simply iterates all of the volumes and dumps the vMDevice parameter for each - this is its output: vMDeviceID = disk0s9 vMDeviceID = disk0s10 vMDeviceID = disk1s9 vMDeviceID = disk1s10 vMDeviceID = /dev/disk2s0 vMDeviceID = afp_1Q2ykU1Q2ykU1Q2ykU1Q2ykU-0.2f000002 I'm guessing (based on the icon problem, and on the vMDeviceID for a network volume, that PBHGetVolParmsSync has decided for some reason that mine is a network volume, and that it therefore shouldn't strip the /dev/ prefix. What I don't know, is what flag is messed up which would cause that. Any ideas? Thanks, Ron Aldrich Software Architects, Inc. _______________________________________________ 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

Increasing the value of kern.ipc.maxsockets?
by Jeramey Crawford 15 Jul '03

15 Jul '03

One of our OS X servers (kernel version 6.6) is running a set of applications that, together, eat up a lot of sockets. The default of 512 is simply too low of a ceiling for the kind of load we'd like to throw at the machine. I have tried increasing it by running 'sysctl -w kern.ipc.maxsockets=xxx', however this is a read-only value in a running kernel. I am wondering if there is a way to increase it at boot time a la FreeBSD's method of editing /boot/loader.conf with the appropriate value. I'd prefer to not have to recompile a kernel just to get more sockets available for the system. Thanks! -- Jeramey Crawford One Reel Systems and Network Administrator (206) 281-7788 ext. 214 _______________________________________________ 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