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

Re: Alternative to bootstrap_register?
by Damien Sorresso 18 Nov '09

18 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 18, 2009, at 7:02 AM, Uli Kusterer wrote: > On 17.11.2009, at 20:27, Damien Sorresso wrote: >>> Can you expand on what you're trying to do? > > We have a 32 bit application that needs to create a 64 bit process to do processor intensive tasks and communicate with the main application typically hundreds of times per second, depending on the number of tasks. Passing shared memory objects and rights is also involved. Performance is critical, AppleEvents, distributed objects etc. seem too slow. > >> The recommended replacement is to have two launchd jobs, one which advertises a service with a well-known name in the bootstrap namespace. > > We don't have a well-known name, that would have been the case for suggestion #2 in bootstrap.h, however this is not the case. In this case the name is dynamic and has no general purpose use for the system, in other words scenario #3 applies. Just because a service doesn't have general purpose use for the system doesn't mean you should not register it with a well-known name. Privileged helper tools that are tied to specific applications also don't have general use for the system, but their MachService names (if they have them) are still registered with launchd and bootstrap_check_in(). Registering services with well-known names is part of launchd's on-demand model. > To rephrase the original question for clarification: > > Is recommendation #3 in bootstrap.h regarding "sending a Mach send-right directly" technically accurate and valid? Yes. > If yes, how does it envision sending a Mach send-right directly given the very fundamental principles and security constraints of Mach IPC? What "fundamental principles and security constraints" are you referring to? In Mach, holding a right to a port also means that you can transmit that right to another process if you want. This is fundamental to Mach and how Mac OS X works. We fling rights around the system all the time; that's what bootstrap_check_in() and bootstrap_look_up() are for. -- 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 Alastair Houghton 18 Nov '09

18 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=p40Dr8j5CFb4hSVlk6ZDJ/DWzvc=; b=HfBD0AhA+aW+xTm0WrJvrH UBjLreIFsumXANBqwfvRRibHD63uHNC8bxssWiPpEo+itmVoNPS6iOwOOs/BHkcJ V3rJFkGeTudXpkW2rSf17DHtcgHDiHsgJOjaVfdQStceuNc+40z3e9trZlWY4uyu zbLHfVm6zoS+kCXgnfqfM= On 18 Nov 2009, at 15:58, Jean-Daniel Dupas wrote: > Yes, but your server has to checkin before the client can lookup, else the lookup will failed. It work great in your example but this cannot easily be done when the client process is the one that launches the server process. > That's where bootstrap_register was really useful has using this function, a client was able to create the service and lookup without having to wait the server readiness. Sure, though even in that case it's fairly easy to arrange for the child to indicate its readiness through some other mechanism (distributed notifications via notify(3), for instance). I'm slightly wary of the NSMachBootstrapService method because of its reliance on the bootstrap_register() deprecated API/SPI, and launchd is only easy to use that way on 10.6, right? Also, I'm not sure launchd supports Mach ports on 10.4, if that's an issue. Anyway, I think we've probably covered every possibility here. 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 Jean-Daniel Dupas 18 Nov '09

18 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Le 18 nov. 2009 à 16:35, Alastair Houghton a écrit : > On 18 Nov 2009, at 15:22, Jean-Daniel Dupas wrote: > > [snip] > >> First case: >> Parent process create a receive right using [NSMachPort port]. >> It register it using an UUID string as the name for example using -[NSMachBootstrapServer registerPort:name:]. >> It starts a child process passing it the service name as argument (or using env variable). >> The child process get a send right using the -[NSMachBootstrapServer portForName:] method >> >> Your Mach channel is up. > > [snip] > >> Second case. You want to have a send right on a child process. >> >> Parent process register a service using launchd API (and pass an UUID string as service name). >> Parent query a send right using bootstrap_lookup. -> launchd start the child process. >> Child process retrieves the receive right using the launchd API. >> >> Your Mach channel is up. >> >> If you need this configuration (child has receive right, parent has send right), an other way is to use method 1 and then the child process can create a port and send a send right to the parent using the first mach channel. > > 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.) > Yes, but your server has to checkin before the client can lookup, else the lookup will failed. It work great in your example but this cannot easily be done when the client process is the one that launches the server process. That's where bootstrap_register was really useful has using this function, a client was able to create the service and lookup without having to wait the server readiness. > Kind regards, > > Alastair. > > -- > http://alastairs-place.net > > > > -- Jean-Daniel _______________________________________________ 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 18 Nov '09

18 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=bKG99CxCq0/xcE9u3MCi+9jmLQw=; b=UU4ZGxgLfUlKE2mCmvZ0KB ArimpuJOveNAbNHIV5HQ4ruXBkWd0KfvXB7fXHXcSmOwqWXtZ6kIJ2a19qVJSL05 lG+amtD25cUIgOV6JUJhuKyByupMJavdu9azhBMpowhZ5Xe4J6gUgKFs5An7ELc8 o7XQniaE96dGFfnboDf6o= On 18 Nov 2009, at 15:22, Jean-Daniel Dupas wrote: [snip] > First case: > Parent process create a receive right using [NSMachPort port]. > It register it using an UUID string as the name for example using -[NSMachBootstrapServer registerPort:name:]. > It starts a child process passing it the service name as argument (or using env variable). > The child process get a send right using the -[NSMachBootstrapServer portForName:] method > > Your Mach channel is up. [snip] > Second case. You want to have a send right on a child process. > > Parent process register a service using launchd API (and pass an UUID string as service name). > Parent query a send right using bootstrap_lookup. -> launchd start the child process. > Child process retrieves the receive right using the launchd API. > > Your Mach channel is up. > > If you need this configuration (child has receive right, parent has send right), an other way is to use method 1 and then the child process can create a port and send a send right to the parent using the first mach channel. 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.) 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 Alastair Houghton 18 Nov '09

18 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=NW4mi6XkfsrALMMuw2daZ028Pno=; b=HXFR/CAJOplcnV0JCAPyDg vtgbU2BOuSTYlKnbVFxrJx6+0O3/ZFq0k67sMC6fAlu1AKaXW9H6pV3SlrXVDeaH A8LVg6uPmZl3jsvrSDJ5AkVyguvG+hWhByvJwdzE3IJOm9BoPa4QjAj4Jf5OECnZ t9tKCyUvkINrYWRjfUIkk= On 18 Nov 2009, at 15:13, Uli Kusterer wrote: > Well, that'd mean we'd be writing potentially security-unsafe root-daemon code (in addition to our other two executables). Not necessarily. You can write a Mach server that doesn't need to run as root (that's how the sample program I wrote works; it runs just fine as the current user). Even if you're using launchd, you can tell it to execute your server as some user besides root, though---depending on what exactly you're trying to do--- > I'd rather trust Apple wrote their bootstrap server correctly (and it gets exercised a lot more than we alone could do it). > > Can you elaborate what you mean by "putting random ports in the bootstrap namespace" seeming to be "evil"? It seems to me it was made for exactly that? I should qualify my statement---it really depends on how many you plan to put in there, and what they're there for. In the previous named-semaphore case, it seems sensible for them to have their own separate namespace and it would be evil to put potentially large numbers of named semaphores in that namespace because of the potential for namespace collisions with other things. Personally I would tend to stick named server ports in there (for which you *don't* need bootstrap_register()), though you're right that you could make an exception for the odd one-off. This is a moot point though, because bootstrap_register() is deprecated, and you'd need to use it to do that... for servers you *don't* need bootstrap_register() because bootstrap_checkin() is sufficient and will create the server port for you. 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 Jean-Daniel Dupas 18 Nov '09

18 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.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… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Alternative to bootstrap_register?
by Uli Kusterer 18 Nov '09

18 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 17.11.2009, at 20:05, Alastair Houghton wrote: > That is, use a server to register and name your send rights. > > As Tyler Sample pointed out to me today, you *could* use the bootstrap server for that purpose, but it's a little evil because you need to use bootstrap_register() (or something similar), and putting random ports in the bootstrap namespace seems a little evil to me. I prefer the separate server approach. Well, that'd mean we'd be writing potentially security-unsafe root-daemon code (in addition to our other two executables). I'd rather trust Apple wrote their bootstrap server correctly (and it gets exercised a lot more than we alone could do it). Can you elaborate what you mean by "putting random ports in the bootstrap namespace" seeming to be "evil"? It seems to me it was made for exactly that? 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 18 Nov '09

18 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.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… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Alternative to bootstrap_register?
by Uli Kusterer 18 Nov '09

18 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 17.11.2009, at 20:27, Damien Sorresso wrote: >> Can you expand on what you're trying to do? We have a 32 bit application that needs to create a 64 bit process to do processor intensive tasks and communicate with the main application typically hundreds of times per second, depending on the number of tasks. Passing shared memory objects and rights is also involved. Performance is critical, AppleEvents, distributed objects etc. seem too slow. > The recommended replacement is to have two launchd jobs, one which advertises a service with a well-known name in the bootstrap namespace. We don't have a well-known name, that would have been the case for suggestion #2 in bootstrap.h, however this is not the case. In this case the name is dynamic and has no general purpose use for the system, in other words scenario #3 applies. To rephrase the original question for clarification: Is recommendation #3 in bootstrap.h regarding "sending a Mach send-right directly" technically accurate and valid? If yes, how does it envision sending a Mach send-right directly given the very fundamental principles and security constraints of Mach IPC? 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 Damien Sorresso 17 Nov '09

17 Nov '09

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 17, 2009, at 9:27 AM, Uli Kusterer wrote: > In <servers/bootstrap.h> there is the following note about the bootstrap_register() call: > > * This API is deprecated. Old scenarios and recommendations: > * > * 1) Code that used to call bootstrap_check_in() and then bootstrap_register() > * can now always call bootstrap_check_in(). > * > * 2) If the code was registering a well known name, please switch to launchd. > * > * 3) If the code was registering a dynamically generated string and passing > * the string to other applications, please rewrite the code to send a Mach > * send-right directly. > > In case of suggestion #3, the application that used to call bootstrap_register only has the pid of the target application to which to "send a Mach send-right directly". > > In order to send a Mach send right however one needs a send right itself to a port to which the target application has receive rights. In order to get that send right, the target application itself also needs to send this application a Mach send right, however in order send that send right it also needs a send right to send it and the loop never ends. > > Does anyone know what paragraph 3) means to achieve and how, possibly with a short example? The recommended replacement is to have two launchd jobs, one which advertises a service with a well-known name in the bootstrap namespace. The other (the client) looks up that service and sends a request to it. Part of the response to that request is a send right to whatever port the client needs a send right to. Can you expand on what you're trying to do? > To be noted, the open source CFMessagePortCreateLocal() still uses bootstrap_register years later after the API has been marked as deprecated. Yes, we're aware. ;) Feel free to file a Radar. -- 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