lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <20241025120948.916e8c099fee7200e32ae617@kernel.org>
Date: Fri, 25 Oct 2024 12:09:48 +0900
From: Masami Hiramatsu (Google) <mhiramat@...nel.org>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: Leo Yan <leo.yan@....com>, Mikel Rychliski <mikel@...elr.com>, Viktor
 Malik <vmalik@...hat.com>, Steven Rostedt <rostedt@...dmis.org>,
 linux-kernel@...r.kernel.org
Subject: Re: [GIT PULL] probes: Fixes for v6.12-rc4-2

On Thu, 24 Oct 2024 14:05:08 -0700
Linus Torvalds <torvalds@...ux-foundation.org> wrote:

> On Wed, 23 Oct 2024 at 07:36, Masami Hiramatsu <mhiramat@...nel.org> wrote:
> >
> > - objpool: Fix choosing allocation for percpu slots
> >
> >   Fixes to allocate objpool's percpu slots correctly according to the
> >   GFP flag. It checks whether "any bit" in GFP_ATOMIC is set to choose
> >   the vmalloc source, but it should check "all bits" in GFP_ATOMIC flag
> >   is set, because GFP_ATOMIC is a combined flag.
> 
> So the old code was buggy, but I don't think the new code is wonderful either.

Agreed.

> 
> For example, it does not recognized GFP_NOWAIT, which has very similar
> characteristics to GFP_ATOMIC (it's basically the same thing as
> GFP_ATOMIC, but without the "try to allocate urgently", and with an
> added "don't warn on failure" as a result).
> 
> So what I think that code *should* test for is "can I sleep".
> 
> Which is indicated by __GFP_IO or __GFP_FS (and, I guess also
> __GFP_DIRECT_RECLAIM).

This code is for switching __vmalloc_node() to kmalloc_node() based on
GFP_ATOMIC, because __vmalloc_node() does not support GFP_ATOMIC.

---
/**
 * __vmalloc_node_range - allocate virtually contiguous memory
...
 *
 * Allocate enough pages to cover @size from the page level
 * allocator with @gfp_mask flags. Please note that the full set of gfp
 * flags are not supported. GFP_KERNEL, GFP_NOFS and GFP_NOIO are all
 * supported.
---

(BTW, the above kerneldoc is a bit outdated, because the function name
has been changed to `__vmalloc_node_range_noprof()`)

As above said, it supports GFP_KENEL & GFP_NOFS & GFP_NOIO, which
means __GFP_RECLAIM (= ___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM) is
required.

Since GFP_ATOMIC is __GFP_HIGH | __GFP_KSWAPD_RECLAIM, the difference
is ___GFP_DIRECT_RECLAIM is not set if GFP_ATOMIC is specified.

But I don't want to touch such bitflags deeper. What about simply
masking with GFP_ATOMIC | GFP_KERNEL? I mean

if ((gfp & (GFP_ATOMIC | GFP_KERNEL)) == GFP_ATOMIC)
	/* use kmalloc_node() because vmalloc does not support GFP_ATOMIC */
	slot = kmalloc_node(...)
else
	slot = __vmalloc_node(...)

Or, simply failback to kmalloc_node() if __vmalloc_node() fails.
(I think this is more robust.)

slot = __vmalloc_node(...)
if (!slot)
	slot = kmalloc_node(...)

(Note that slot is freed by kvfree() so either works)

Thank you,

-- 
Masami Hiramatsu (Google) <mhiramat@...nel.org>

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ