December 2007 - Darwin-kernel

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Terry Lambert 13 Dec '07

13 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com -- 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 Dec 12, 2007, at 5:08 PM, Nick Blievers <nickb(a)sgi.com> wrote: Specifically, creation of subdirectories in /dev is not officially supported, don't do that. The change of function precludes kernel buffer overflow based crashes for some device drivers using name information supplied from user space. Even with the "off by one" introduced in the changeover, the buffet overrun is avoided, which is a good thing for kernel stability. In general, device node length is constrained overall to a value "reasonable" for a single path component, which means the more /'s you used, the shorter the name allowed for the eventual device anyway. I think the overall limit is something like 32 characters; if that's not exactly correct, whatever it is is a heck of a lot smaller than PATH_MAX at 1024 bytes. -- Terry >From what I remember the limit discussed was 32, however looking at the code it appears the limit is DEVMAXPATHSIZE (128) for the full path and DEVMAXNAMESIZE (32) for each element. Also I believe this limit existed pre-Leopard, just wasn't enforced, and hence buffer- overrun's where possible. So currently our solution is to test the kernel version and add a bogus character to the path element in out devfs_make_node() call... this is the only way to do it without using functions that are not exported. Or you can choose to not try creating subdirectores in /dev, and put the information into node names themselves instead. For example, a SCSI disc device on controller 0 with unit 3 and lun 8 with three partitions on it, where you wanted the second one could be named something like "sdc0u3l8s3". UNIX systems have used this type of convention forever. Think of the letters other than the initial "sd" as path component separators, if it helps you sleep. Given that subdirectories in /dev are not supported, and may be removed, if we end up needing to do that for some technical reason, it's probably better to just go that route on both 10.4 and 10.5 instead. This email sent to site_archiver(a)lists.apple.com

1 0

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Nick Blievers 13 Dec '07

13 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Thunderbird 2.0.0.0 (Macintosh/20070326) -- Terry >From what I remember the limit discussed was 32, however looking at the code it appears the limit is DEVMAXPATHSIZE (128) for the full path and DEVMAXNAMESIZE (32) for each element. Also I believe this limit existed pre-Leopard, just wasn't enforced, and hence buffer-overrun's where possible. So currently our solution is to test the kernel version and add a bogus character to the path element in out devfs_make_node() call... this is the only way to do it without using functions that are not exported. Hope this helps, Nick _______________________________________________ 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.… Specifically, creation of subdirectories in /dev is not officially supported, don't do that. The change of function precludes kernel buffer overflow based crashes for some device drivers using name information supplied from user space. Even with the "off by one" introduced in the changeover, the buffet overrun is avoided, which is a good thing for kernel stability. In general, device node length is constrained overall to a value "reasonable" for a single path component, which means the more /'s you used, the shorter the name allowed for the eventual device anyway. I think the overall limit is something like 32 characters; if that's not exactly correct, whatever it is is a heck of a lot smaller than PATH_MAX at 1024 bytes. This email sent to site_archiver(a)lists.apple.com

1 0

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Terry Lambert 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com under OS X 10.5, when I install a file system device with -- 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 Dec 12, 2007, at 8:27 AM, Steve Checkoway <s(a)pahtak.org> wrote: On Dec 12, 2007, at 8:17 AM, Alexander v. Below wrote: procDevice = devfs_make_node(myDevice, DEVFS_CHAR, UID_ROOT, GID_WHEEL, 0666, path) where "path" is an actual path (like foo/bar/myDevice), then the device gets created in /dev/fo/ba/myDevice. This does not happen in 10.4 Is this just a plain bug, or am I doing something wrong? This has come up a number of times on the list recently. As I recall, this is both not supported and an off-by-one error introduced when switching from some standard library function to some supposedly "safe" version which was supposed to prevent these sorts of errors. See the list archive for more details. Specifically, creation of subdirectories in /dev is not officially supported, don't do that. The change of function precludes kernel buffer overflow based crashes for some device drivers using name information supplied from user space. Even with the "off by one" introduced in the changeover, the buffet overrun is avoided, which is a good thing for kernel stability. In general, device node length is constrained overall to a value "reasonable" for a single path component, which means the more /'s you used, the shorter the name allowed for the eventual device anyway. I think the overall limit is something like 32 characters; if that's not exactly correct, whatever it is is a heck of a lot smaller than PATH_MAX at 1024 bytes. This email sent to site_archiver(a)lists.apple.com

1 0

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Steve Checkoway 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Thunderbird 2.0.0.4 (X11/20070711) -- Steve Checkoway _______________________________________________ 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.… Alexander v. Below wrote: Thanks. Would you have a Radar ID so I can just hop on that bandwagon? I'm not sure which one of these two it is, probably the first: <rdar://5589808> <rdar://5584196>. This email sent to site_archiver(a)lists.apple.com

1 0

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Alexander v. Below 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Thanks. Would you have a Radar ID so I can just hop on that bandwagon? Alex On 12.12.2007, at 17:41, Steve Checkoway wrote: On Dec 12, 2007, at 8:32 AM, Alexander v. Below wrote: http://search.lists.apple.com/?q=%22devfs_make_node%22 At any rate, the bug was given as Looks like there is already a radar filed about this. Hope this helps. -- Steve Checkoway "Anyone who says that the solution is to educate the users hasn't ever met an actual user." -- Bruce Schneier _______________________________________________ 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/below%40mac.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.… Sorry, I had tried to search the list before, but my keywords did not yield anything releated to the issue. For example, searching for "devfs_make_node" on all lists only yields six results, none of which appear to be related: Could you give me a hint on what to search the archives for? Turns out I can't find it when searching the archives either. But the first thread I saw about it was written by Nick Blievers on November 8 with the subject, "devfs_make_node()". No, I take that back, he posted it first in May, but was told not to talk about Leopard on a public list. --- /Developer/Darwin-8.9.0/xnu-792.17.14/bsd/miscfs/devfs/ devfs_tree.c 2007-04-18 09:19:56.000000000 +0800 +++ /Developer/Darwin-9.0.0/xnu-1228//bsd/miscfs/devfs/ devfs_tree.c 2007-10-30 11:58:56.000000000 +0800 @@ -266,8 +292,7 @@ while (*scan != '/' && *scan) scan++; - strncpy(component, start, scan - start); - component[ scan - start ] = '\0'; + strlcpy(component, start, scan - start); if (*scan == '/') scan++; The second time it came up was by David Rose on December 5 with the subject, "Checking if a file exists from the kernel." This email sent to below(a)mac.com This email sent to site_archiver(a)lists.apple.com

1 0

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Steve Checkoway 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Dec 12, 2007, at 8:32 AM, Alexander v. Below wrote: http://search.lists.apple.com/?q=%22devfs_make_node%22 At any rate, the bug was given as Looks like there is already a radar filed about this. Hope this helps. -- Steve Checkoway "Anyone who says that the solution is to educate the users hasn't ever met an actual user." -- Bruce Schneier _______________________________________________ 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 Sorry, I had tried to search the list before, but my keywords did not yield anything releated to the issue. For example, searching for "devfs_make_node" on all lists only yields six results, none of which appear to be related: Could you give me a hint on what to search the archives for? Turns out I can't find it when searching the archives either. But the first thread I saw about it was written by Nick Blievers on November 8 with the subject, "devfs_make_node()". No, I take that back, he posted it first in May, but was told not to talk about Leopard on a public list. --- /Developer/Darwin-8.9.0/xnu-792.17.14/bsd/miscfs/devfs/ devfs_tree.c 2007-04-18 09:19:56.000000000 +0800 +++ /Developer/Darwin-9.0.0/xnu-1228//bsd/miscfs/devfs/ devfs_tree.c 2007-10-30 11:58:56.000000000 +0800 @@ -266,8 +292,7 @@ while (*scan != '/' && *scan) scan++; - strncpy(component, start, scan - start); - component[ scan - start ] = '\0'; + strlcpy(component, start, scan - start); if (*scan == '/') scan++; The second time it came up was by David Rose on December 5 with the subject, "Checking if a file exists from the kernel." smime.p7s

1 0

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Alexander v. Below 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com http://search.lists.apple.com/?q=%22devfs_make_node%22 Could you give me a hint on what to search the archives for? Thanks Alex On 12.12.2007, at 17:27, Steve Checkoway wrote: On Dec 12, 2007, at 8:17 AM, Alexander v. Below wrote: under OS X 10.5, when I install a file system device with -- Steve Checkoway "Anyone who says that the solution is to educate the users hasn't ever met an actual user." -- Bruce Schneier _______________________________________________ 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/below%40mac.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.… Sorry, I had tried to search the list before, but my keywords did not yield anything releated to the issue. For example, searching for "devfs_make_node" on all lists only yields six results, none of which appear to be related: procDevice = devfs_make_node(myDevice, DEVFS_CHAR, UID_ROOT, GID_WHEEL, 0666, path) where "path" is an actual path (like foo/bar/myDevice), then the device gets created in /dev/fo/ba/myDevice. This does not happen in 10.4 Is this just a plain bug, or am I doing something wrong? This has come up a number of times on the list recently. As I recall, this is both not supported and an off-by-one error introduced when switching from some standard library function to some supposedly "safe" version which was supposed to prevent these sorts of errors. See the list archive for more details. This email sent to below(a)mac.com This email sent to site_archiver(a)lists.apple.com

1 0

Re: [darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Steve Checkoway 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Dec 12, 2007, at 8:17 AM, Alexander v. Below wrote: under OS X 10.5, when I install a file system device with -- Steve Checkoway "Anyone who says that the solution is to educate the users hasn't ever met an actual user." -- Bruce Schneier _______________________________________________ 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 procDevice = devfs_make_node(myDevice, DEVFS_CHAR, UID_ROOT, GID_WHEEL, 0666, path) where "path" is an actual path (like foo/bar/myDevice), then the device gets created in /dev/fo/ba/myDevice. This does not happen in 10.4 Is this just a plain bug, or am I doing something wrong? This has come up a number of times on the list recently. As I recall, this is both not supported and an off-by-one error introduced when switching from some standard library function to some supposedly "safe" version which was supposed to prevent these sorts of errors. See the list archive for more details. smime.p7s

1 0

[darwin-kernel] devfs_make_node mutilates pathname on Leopard?
by Alexander v. Below 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.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: clear_wait() using thread address
by Vishal Shetye 12 Dec '07

12 Dec '07

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:received:received:from:to:cc:references:subject:date:message-id:mime-version:content-type:content-transfer-encoding:x-mailer:thread-index:x-mimeole:in-reply-to; bh=9zfOYp/Yy1xW/HkJ41dV1RzYgqKwgUmOgivr8J549bg=; b=Jyuvu4k8pHQDrHf6nkko1IZ2yTnko/RtBBK8WiINXmR8xRzOnNcyzlBWMyFhsrfxU1dp/lCeG1M2iy4sTS1OQGyLOqBlpuMG+EWTpl2HPGUUFO7QveFIH9rT/nq5qi1RGXSUHewpyWtvwv5z928LH1i9ryxsCSePUI+iOSPFGME= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=from:to:cc:references:subject:date:message-id:mime-version:content-type:content-transfer-encoding:x-mailer:thread-index:x-mimeole:in-reply-to; b=t38E4ibPau3NJLeZaxFrd+z1Hn+YH95j2L8Mqk8K9xbCPSAVmYaOPmQ3sNPw4k3J8E8HlOWKJ66EoKyP74Cu8/qElh1Wfn8pEMChUHvQAHUu0H+lMWYB6EAvmYhWKftFI3UJc3uUjofoCm0rf2Yz6zzYMyuGQzpY0x/tVLmbUyA= Thread-index: Acg8hoVQX3A+hFzfQKe26TA0KFS0uwALx25g Hi, >> I didn't find any function for getting thread state (like sleeping). >> For now I just do a proc_find() and assume true return value = thread >> sleeping (bad assumption). Please suggest alternative??? >You must maintain this state yourself. >That being said, condvars were invented for a reason. Attempting to >determine the state of one thread from within another is intrinsically >race-prone, and if you ever find yourself thinking this would be a >good idea, it's time to toss out your design and start again. May be the design is not good but at least for now I'm only interested in Checking if thread is sleeping on a particular event on not. >These are not symmetrical. msleep() pairs with wakeup(), which you >should probably be using instead. Using wakeup caused some error while compiling (symbol DEBUG not found) I used thread_wakeup_prim() as wakeup() has been defined in XNU as void wakeup(chan) register void *chan; { thread_wakeup_prim((caddr_t)chan, FALSE, THREAD_AWAKENED); } _______________________________________________ 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