[DannyBee]

A. There is no real standard here B. Apple doesn't bend, period. They don't actually care whether the use is appropriate or inappropriate. C. Blaming this on homebrew is hilarious

Remind me again why i should need to reboot into recovery mode to create a directory?

[tptacek]

You need to boot into recovery mode to create /usr/local (or to disable SIP altogether) because /usr is immutable, so that code executing under an admin account can't inject itself into binaries (or into PATHs) under /usr.

I use Homebrew and will probably keep doing that because Homebrew is really convenient and just generally works better than MacPorts for me, but what Homebrew is demanding --- ownership over /usr/local --- is not in fact reasonable. It makes a lot of sense to me that OSX will want to keep perms on /usr/local (which is in everyone's PATH) locked down, and so it seems extra hacky that they're going to add a 'brew doctor' check to see if Software Update has "broken" those privileges.

The commenter upthread that suggested Homebrew should use ~/.homebrew or something by default is right.

[mzs]

And I contend that homebrew did not want to used ~/.homebrew because that would mean modifying files to add it to PATH and be additional testing for people that write recipes.

[tptacek]

And that's totally fine, and I let Homebrew own /usr/local too. What I'm not going to do is freak out when a totally sane and useful security feature breaks that install mode, especially when I could, if I cared that much about convenience, just take 30 seconds to disable the security feature.

[mzs]

I just want to make clear I'm not freaking out (and I know you're not either).

I would hope though that out of this someone would think, "Hmm... oh drat... yeah I guess this is just one of those things that nobody predicted at first. Shot I think it was a bad idea to have a user own /usr/local/. But you know, it's a good time to improve things... hmm... I know let's build and stage as nobody and then copy under /usr/local/! We still can have binary packages always work and nobody has to mess with PATH, man pages, or dynamic linker trivia. We might have to write something though that asks for a username and password, but whoa... OSX makes that surprisingly easy... I should do that!"

It's not going to be me cause I don't use homebrew, but that would be way better than these hacks that will ammount to people being encouraged to just disable SIP cause of the headache after ever software update.

[s73v3r]

Keep in mind that you can re-enable SIP right after you create the directory, if you had to create it at all.

[quesera]

A. /usr/local has never ever been writable by a nonprivileged user. Every standard that exists, including the principle of least surprise, is emphatic on that point. The security implications of "any other way" are massive.

B. If Linus or the FHS changed their ways in this, there would be outrage. It's inappropriate and just plain dumb. [edit: I shouldn't have said "dumb", please replace with "in violation of all guarantees of system integrity".]

C. See above.

You apparently have issues with SIP. That's a separate issue from the current thread, but OK. You should reboot to securely change nvram settings that will allow you to manipulate the protected zone of the file system (including /usr) because this concession to inconvenience saves you from privilege escalation attacks.

The real question is why you deleted /usr/local in the first place. It's standard OSX, and always has been. If you deleted it before SIP, but object to its absence now that SIP applies to the directory where it lives...I'm not sure how to help.

[DannyBee]

A. Single user systems are not the same as server systems, and plenty of single user systems have had the equivalent of /usr/local as writable. If you want to argue this security is worth it, that's a different argument. I suspect most people who just want to get shit done are going to find it a huge pain in the ass.

You can also argue these people are stupid.

As for standards, vendors follow FHS and friends exactly as far as it helps them justify whatever they want to do.

If you really want to argue from the perspective that "this is standard and that's the reason it's done", that seems awfully silly to me given the layout and other permissions of apple systems.

It's pretty non-FHS/etc to have /Applications be writable, for example.

B. let's be clear: there is outrage no matter what is changed and why, and the argument is always "it is inappropriate and dumb". So this statement is fairly independent of this change.

[quesera]

A. OSX is not a single user system. It inherits from various Unixes, and no Unix has ever had nonprivileged writability on any directory under /usr.

FHS does not apply to OSX in any way. Why do people think so?

Also, /Applications is not writable by nonprivileged users either!

Agreed on ~"changes yield outrage", but in this case, it's just a software application vendor violating obvious historical and specific guidelines from software OS vendor, and people blaming OS vendor when things break. This perplexes me.

[tptacek]

You might just as productively argue that there is no such thing as a modern single-user system, since Windows isn't that either. His point is pretty clear.

OTOH: The debate here is a little confusing. What SIP is doing w/r/t/ /usr/local isn't unprecedented; securelevels and the immutable flag could have the same effect on OpenBSD all the way back in the 1990s.

[quesera]

> You might just as productively argue that there is no such thing as a modern single-user system, since Windows isn't that either. His point is pretty clear.

Correct, and nonprivileged users can't install software globally on Windows either. Because it's now a modern OS, and that's how things are done. For very good reasons.

His point, as far as I can tell, amounts to: applications should be able to trash permissions on convenient directories and not have the OS complain or repair their damage.

Perhaps I'm being ungenerous here? I apologize if so.

On re read, I realize that tptacek is responding only to the SIP part of the discussion. OK, SIP is surprising to desktop computer users. But it's a good feature that requires a little adjustment to work with. But really not much, and none for the vast majority of users who don't muck with system directories.

[tptacek]

He's arguing that users should be able to elevate their privileges by supplying root's credentials and then create a simple directory without having to reboot.

I more agree with you than with him, but his argument seems pretty straightforward.

[tptacek]

The "single user systems aren't the same as server systems" argument cuts both ways. It is also not a big deal to reboot a single-user systems.

[mzs]

I see what you are saying but please step back and take a wider view for a moment.

Even assuming OSX is a single user system (which is arguable), it is 2015 and should that single user be able to accidentally damage /usr/local? I think not, that by today an OS should help protect you from broken postinstall script or what not.

Moreover please think about this: homebrew is not wanting /usr/local/ root:admin but instead /usr/local/ `/usr/bin/whoami`:admin.

Further even on our laptops where there is just one user, I use a non-admin user. I have no problems using FUS, GUI prompted temporary admin permissions, or /usr/bin/login -p and sudo day-to-day, but homebrew has a problem with sudo.

Apple SIP has no problem with /usr/local/ 0:0 but homebrew does. Because of shortcuts homebrew has forced a number of poor decisions on it's users.

[alextgordon]

Where in FHS does it specify that system and user files should be put in

    /Applications
    /Library
    /Network
    /System
    /User Information
    /Users
    /Volumes

?

[quesera]

FHS does not apply to OSX. Never has, never will.

[tptacek]

He's asking because you brought FHS up.

[quesera]

Under B, as an example of standards by another vendor who would be pilloried if they allowed the same lackadaisical approach to preserving system permissions integrity. In response to the comment that Apple was being unreasonable.

Apologies if that was not clear.

[theptip]

There is a standard: http://news.stanford.edu/pr/2015/pr-worms-digest-plastics-09....

This is from RedHat. Not implemented by everyone, but modern OSes (e.g. CoreOS) seem to follow the patterns. CoreOS also do everything in their power to prevent you from mounting /usr as writeable.

Perhaps that doesn't pass your bar for 'real' though since it's not globally implemented ;)

[duskwuff]

I think that's the wrong link? (It's to a press release from Stanford's engineering department about plastic-eating worms...)