November 2009 - Darwin-dev - lists.apple.com

Re: kevents, waitpid() with WNOHANG race
by Godfrey van der Linden 22 Nov '09

22 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Yes, please. A kernel race between kevent engine and the process state is really important to understand, especially if you have a test case. The hardest part of fixing a race is a solid test case. Godfrey PS. I've been waxing philosophical on race conditions for a while. The nature of a race condition is that it hides until you uncover it by changing the timing between a pair of threads, a classic case of "you milage may vary". (I've been exploring evolutionary metaphors for a while in which a programmer is the mutagens; sorry I can be a boring on this subject) On 2009-11-22, at 15:27 , Damien Sorresso wrote: > On Nov 21, 2009, at 8:23 PM, Dave Keck wrote: >> For the archives: >> >> After some stress testing, I've found that there is in fact a race >> between when kevent() reports that a process exited, and when >> waitpid() can successfully reap the process. Therefore, it is not safe >> to use the WNOHANG option with waitpid() and friends to reap a child >> process, after observing that the child process exited using kevent(). > > > Could you file a bug on this and attach a test case? > -- > Damien Sorresso > BSD Engineering > Apple Inc. _______________________________________________ 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: kevents, waitpid() with WNOHANG race
by Damien Sorresso 22 Nov '09

22 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 21, 2009, at 8:23 PM, Dave Keck wrote: > For the archives: > > After some stress testing, I've found that there is in fact a race > between when kevent() reports that a process exited, and when > waitpid() can successfully reap the process. Therefore, it is not safe > to use the WNOHANG option with waitpid() and friends to reap a child > process, after observing that the child process exited using kevent(). Could you file a bug on this and attach a test case? -- Damien Sorresso BSD Engineering Apple Inc. _______________________________________________ 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: kevents, waitpid() with WNOHANG race
by Dave Keck 22 Nov '09

22 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:sender:received:in-reply-to :references:from:date:x-google-sender-auth:message-id:subject:to :content-type; bh=hOCk86tauaNFfA/pc1HZ6301xhYr3GOOyxFWgzWY4vQ=; b=Knm9yEPEJoSS1yFRtH5wk4gBf9/D5J1iILEey70yLEnTQSxI6WlfZX8VCFQzFy8lHb gRBTLNcFVbiXwv8RjOoZx4gqaAfaNvCefeeqKPIOVfzX9ZucRviuuhLE7dL2PzVeA/Gw V9eA+q+YRZuD4FoC5nJUxNCtSj9/Tv7KHsu28= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:sender:in-reply-to:references:from:date :x-google-sender-auth:message-id:subject:to:content-type; b=uwjgkEqj0k9jZR3fr4eUeKghFiA8jt+YFCYmXE+C2z0ctF9ee5YcSwkoM6dX6164iz CoTqzBy8Mr+GmTHkLJ+Rd23n3yNmaqEHqJ0ksA9onD6Ag9vunDs4SQnkZcseicLCNvQy 2ZFuTnoNQAPmaM3/LE/PeY0WDz2xXQnK6ub54= For the archives: After some stress testing, I've found that there is in fact a race between when kevent() reports that a process exited, and when waitpid() can successfully reap the process. Therefore, it is not safe to use the WNOHANG option with waitpid() and friends to reap a child process, after observing that the child process exited using kevent(). _______________________________________________ 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

Difference between cu.* and tty.*
by Godfrey van der Linden 21 Nov '09

21 Nov '09

site_archiver(a)lists.apple.com Delivered-To: Darwin-dev(a)lists.apple.com For those that are interested. Serial ports are commonly misunderstood. Having been involved with them for a while I thought I clear up some of the confusion (to that end I'm cross posting this to darwin dev from the usb list thread). The two ports are actually named after their traditional usage. A tty.* port was intended to run getty on a directly connected terminal, a terminal that, if wired correctly, asserts DTR (one of the RS232 signals). A getty process using a tty.* serial connection would block in open until the TTY asserted DTR (that is the terminal was switched on). The cu (call unix) port was used mostly by the uucp family of unix communication tools, that is the unix email system used before the advent of the internet. To cause a typical modem to dial, it was necessary to open the port without waiting for DTR, which is typically only asserted one a connection has been established after dialing. WIth an open port uucp could then issue an AT command and wait for a connection with the other side. With an established connection the modem asserts DTR, from that point on the cu and tty ports are exactly the same. Note with many modems the DTR signal often tracks DCD (carrier detect). With either port type once a raised DTR signal is dropped the port is disconnected and will start returning EIO errors. This is basically a hint to the client process that things have come to an end and need to terminate all processes in the process group connected to the terminal port. One problem with this basic design is that sometimes tty or other serial devices are connected using only 3-wires, that is ground, txdata, rxdata. In this case the standard tty port will not open. You have two options in this circumstance. The more difficult option is to loopback the DSR to the DTR on the computer side of the rs232 plug, or alternatively you can open the port with O_NONBLOCK then use tcsetattr to set the CLOCAL flag indicating that DTR should be ignore. Finally although this is the canonical interface the only serial drivers that I know implement the full RS232 signal protocol are Apple's 16x50 drivers. I'm not sure how much of this signalling is implemented by any other driver, so your milage will vary. In retrospect it perhaps would have been a good idea to document this somewhere ;-). Cheers Godfrey On 2009-11-21, at 7:37 , Jonathon Kuo wrote: > Hi all, > > I managed to procure an FTDI-based dongle, and downloaded the 21-Oct-09 kext for it from FTDI's web site. It works! Compared to the Prolific (PL2303) devices, FTDI is really well supported on OS X. BTW, I get the same results using either the /dev/tty.* or /dev/cu.* ports. I know the .cu is intended as a modem port, but is there a functional difference between it and the .tty*? > > Many thanks to everyone who offered suggestions, both on and off list! > Jon _______________________________________________ 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: Alternative to bootstrap_register?
by Damien Sorresso 20 Nov '09

20 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 20, 2009, at 7:42 AM, Uli Kusterer wrote: > On 18.11.2009, at 16:35, Alastair Houghton wrote: >> Another alternative is just to use bootstrap_checkin() from the child process, and then you can use the port that that API returns. The parent would then use bootstrap_look_up(). >> >> (That's what my sample program does.) > > > bootstrap_checkin() requires root privileges for the calling application (at least on 10.5 and earlier), which seems like overkill in our case. Apparently, your code uncovered that it works on 10.6 without root permissions, but sadly that won't help in our current situation :-( bootstrap_check_in() doesn't require root privileges. It only requires that you be advertising the named service in your launchd.plist. -- Damien Sorresso BSD Engineering Apple Inc. _______________________________________________ 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: Alternative to bootstrap_register?
by Damien Sorresso 20 Nov '09

20 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 20, 2009, at 6:43 AM, Uli Kusterer wrote: > On 19.11.2009, at 18:31, Damien Sorresso wrote: >> launchd doesn't provide support for multiple job instances currently. We are very aware of this need, however, so don't take this to mean that we see that use case as outside of our model. >> >> For now though, you can sort of achieve this by using the ServiceManagement API (in Snow Leopard) to submit unique-ified jobs to launchd on-the-fly. > > And on 10.4 and 10.5? You could use the undocumented (but public) launchd API in launch.h, but we don't really like people using it for anything but checking in to get socket file descriptors. That's why we introduced ServiceManagement. -- Damien Sorresso BSD Engineering Apple Inc. _______________________________________________ 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: Alternative to bootstrap_register?
by Uli Kusterer 20 Nov '09

20 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 18.11.2009, at 22:59, Godfrey van der Linden wrote: > Since you are launching a server process directly, have you considered the typical unix solution of using pipes to communicate with the child. See popen(3), which is a single API that that allows a parent to create a child and connect to it. Considered, implemented and discarded. They just don't provide the performance. Cheers, -- Uli Kusterer "The witnesses of TeachText are everywhere..." _______________________________________________ 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: Alternative to bootstrap_register?
by Alastair Houghton 20 Nov '09

20 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha1; c=relaxed; d=alastairs-place.net; h= subject:mime-version:content-type:from:in-reply-to:date:cc: content-transfer-encoding:message-id:references:to; q=dns/txt; s= aug07; bh=Z0Ap0Enkgf8OAC4eNkbPpr3CLdE=; b=YIvEY5q+RhWV/K0c7ps1TB rseIiQj6WiwxaWnhRVWTOg/tCYi8j4FQix9yWPA/ot4dLuycM6qZp/PuwNpuPP4Y xtcrN0R9fG1HvzKb4pxaQ2tcu2QNcQVeL4Lp58wHd5aFN/1pzKFk83oiWcez4TSR CyHdTe0BTb30Y7iE7dEEA= On 20 Nov 2009, at 15:42, Uli Kusterer wrote: > On 18.11.2009, at 16:35, Alastair Houghton wrote: >> Another alternative is just to use bootstrap_checkin() from the child process, and then you can use the port that that API returns. The parent would then use bootstrap_look_up(). >> >> (That's what my sample program does.) > > bootstrap_checkin() requires root privileges for the calling application (at least on 10.5 and earlier), which seems like overkill in our case. Apparently, your code uncovered that it works on 10.6 without root permissions, but sadly that won't help in our current situation :-( Ah, I hadn't realised that. That's a bit of a pain. Kind regards, Alastair. -- http://alastairs-place.net _______________________________________________ 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: Alternative to bootstrap_register?
by Uli Kusterer 20 Nov '09

20 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 18.11.2009, at 16:35, Alastair Houghton wrote: > Another alternative is just to use bootstrap_checkin() from the child process, and then you can use the port that that API returns. The parent would then use bootstrap_look_up(). > > (That's what my sample program does.) bootstrap_checkin() requires root privileges for the calling application (at least on 10.5 and earlier), which seems like overkill in our case. Apparently, your code uncovered that it works on 10.6 without root permissions, but sadly that won't help in our current situation :-( Cheers, -- Uli Kusterer "The witnesses of TeachText are everywhere..." _______________________________________________ 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: Alternative to bootstrap_register?
by Uli Kusterer 20 Nov '09

20 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 19.11.2009, at 18:31, Damien Sorresso wrote: > launchd doesn't provide support for multiple job instances currently. We are very aware of this need, however, so don't take this to mean that we see that use case as outside of our model. > > For now though, you can sort of achieve this by using the ServiceManagement API (in Snow Leopard) to submit unique-ified jobs to launchd on-the-fly. And on 10.4 and 10.5? Cheers, -- Uli Kusterer "The witnesses of TeachText are everywhere..." _______________________________________________ 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