Linux 6.1 Lands Code To Warn About W+X Mappings By Default

[phoronix]

Phoronix: Linux 6.1 Lands Code To Warn About W+X Mappings By Default

In addition to MGLRU and Maple Tree having been merged overnight, Linus Torvalds also picked up the x86/mm changes for Linux 6.1 that will now by default warn about W+X mappings...

https://www.phoronix.com/news/Linux-6.1-x86-MM-Lands

[sinepgib]

Someone can bring some more specifics about the systemd+eBPF issues? It sounds interesting enough to merit some links

[johncall]

Linux 6.1 is shaping up to be a damn fine kernel release

Agreed! Thanks for all of your articles letting us know about the goodness going into this version!

[fitzie]

Someone can bring some more specifics about the systemd+eBPF issues? It sounds interesting enough to merit some links

here are some links:

https://lore.kernel.org/lkml/[email protected]/

https://marc.info/?l=linux-kernel&m=166483125325327&w=2

https://lore.kernel.org/all/CAMj1kXHcF_iK_g0OZSkSv56Wmr=eQGQwNstcNjLEfS=mm7a06w@mail.gmail.com/

https://lore.kernel.org/bpf/[email protected]/T/#u

https://lore.kernel.org/bpf/[email protected]/

[intelfx]

Someone can bring some more specifics about the systemd+eBPF issues? It sounds interesting enough to merit some links

There's nothing about systemd in it. It simply looks like systemd is the only (or the only visible) user of this functionality

[sinepgib]

here are some links:

https://lore.kernel.org/lkml/[email protected]/

https://marc.info/?l=linux-kernel&m=166483125325327&w=2

https://lore.kernel.org/all/CAMj1kXH...ail.gmail.com/
https://lore.kernel.org/bpf/c84cc27c...ernel.org/T/#u
https://lore.kernel.org/bpf/[email protected]/

There's nothing about systemd in it. It simply looks like systemd is the only (or the only visible) user of this functionality

Thank you both. Indeed, the last two links show clearly that the bug is not really dependent upon the loaded programs. I guess mentioning systemd pays in clicks.

[Weasel]

Wait, so what about dynamically generated code? Like JITs?

[sinepgib]

Wait, so what about dynamically generated code? Like JITs?

First you write, then you mark read only and executable for the rest of its life. W^X is about never being both at the same time and never going back to writable once it's executable. You always need your page to start out as writable after all, how will you copy the code otherwise?

[Weasel]

First you write, then you mark read only and executable for the rest of its life. W^X is about never being both at the same time and never going back to writable once it's executable. You always need your page to start out as writable after all, how will you copy the code otherwise?

Normal code is just mapped by the kernel in most applications. It's not even copied. And even if it was, the kernel obviously has power to ignore its own rules here, if it has to.

But yeah, if this still works for user-land apps to set the executable bit after writing something, it's not a problem. It would also break compressors like UPX otherwise.

[sinepgib]

Normal code is just mapped by the kernel in most applications. It's not even copied. And even if it was, the kernel obviously has power to ignore its own rules here, if it has to.

"Mapped" is a way of saying the page is already in memory and you're assigning it a virtual address, but at some point you copy from disk to memory.

But yeah, if this still works for user-land apps to set the executable bit after writing something, it's not a problem. It would also break compressors like UPX otherwise.

It should work AFAIK.