Re: [tip:x86/boot] x86/KASLR: Fix boot crash with certain memory configurations

[Baoquan He]

On 07/10/16 at 01:24pm, Ingo Molnar wrote:
> 
> * Baoquan He  wrote:
> 
> > Hi Ingo,
> > 
> > I am sorry the previous post didn't contain formal patch log. I made a new 
> > one 
> > as below. The boot crash could not only happen with certain memory. Because 
> > of 
> > this code bug the regions which need be avoided like the zipped kernel with 
> > its 
> > unzipping running code, initrd, kernel command line could be corrupted if 
> > mem_avoid_overlap() can't find the overlap region with the lowest address. 
> > But 
> > it's very lucky that Xiaolong's system which has only 300M memory can 
> > always 
> > reproduce it. I checked the boog log and found on his system no any other 
> > slot 
> > can be chosen except for the original one. If we have a system with large 
> > memory 
> > it may not be easy to hit it, at least with low probability since there are 
> > many 
> > candidate slots.
> 
> So if it's just a better changelog that what I wrote I wouldn't rebase the 
> commit: 
> there's other commits on top of the fix meanwhile. Should a rebase become 
> necessary at a later point I'll update it with your extended changelog.

Yes, it's just a changelog update, no need to rebase it now.

Thanks
Baoquan

Re: [tip:x86/boot] x86/KASLR: Fix boot crash with certain memory configurations

[Baoquan He]

On 07/10/16 at 01:24pm, Ingo Molnar wrote:
> 
> * Baoquan He  wrote:
> 
> > Hi Ingo,
> > 
> > I am sorry the previous post didn't contain formal patch log. I made a new 
> > one 
> > as below. The boot crash could not only happen with certain memory. Because 
> > of 
> > this code bug the regions which need be avoided like the zipped kernel with 
> > its 
> > unzipping running code, initrd, kernel command line could be corrupted if 
> > mem_avoid_overlap() can't find the overlap region with the lowest address. 
> > But 
> > it's very lucky that Xiaolong's system which has only 300M memory can 
> > always 
> > reproduce it. I checked the boog log and found on his system no any other 
> > slot 
> > can be chosen except for the original one. If we have a system with large 
> > memory 
> > it may not be easy to hit it, at least with low probability since there are 
> > many 
> > candidate slots.
> 
> So if it's just a better changelog that what I wrote I wouldn't rebase the 
> commit: 
> there's other commits on top of the fix meanwhile. Should a rebase become 
> necessary at a later point I'll update it with your extended changelog.

Yes, it's just a changelog update, no need to rebase it now.

Thanks
Baoquan

Re: [tip:x86/boot] x86/KASLR: Fix boot crash with certain memory configurations

[Ingo Molnar]

* Baoquan He  wrote:

> Hi Ingo,
> 
> I am sorry the previous post didn't contain formal patch log. I made a new 
> one 
> as below. The boot crash could not only happen with certain memory. Because 
> of 
> this code bug the regions which need be avoided like the zipped kernel with 
> its 
> unzipping running code, initrd, kernel command line could be corrupted if 
> mem_avoid_overlap() can't find the overlap region with the lowest address. 
> But 
> it's very lucky that Xiaolong's system which has only 300M memory can always 
> reproduce it. I checked the boog log and found on his system no any other 
> slot 
> can be chosen except for the original one. If we have a system with large 
> memory 
> it may not be easy to hit it, at least with low probability since there are 
> many 
> candidate slots.

So if it's just a better changelog that what I wrote I wouldn't rebase the 
commit: 
there's other commits on top of the fix meanwhile. Should a rebase become 
necessary at a later point I'll update it with your extended changelog.

Thanks,

Ingo

Re: [tip:x86/boot] x86/KASLR: Fix boot crash with certain memory configurations

[Ingo Molnar]

* Baoquan He  wrote:

> Hi Ingo,
> 
> I am sorry the previous post didn't contain formal patch log. I made a new 
> one 
> as below. The boot crash could not only happen with certain memory. Because 
> of 
> this code bug the regions which need be avoided like the zipped kernel with 
> its 
> unzipping running code, initrd, kernel command line could be corrupted if 
> mem_avoid_overlap() can't find the overlap region with the lowest address. 
> But 
> it's very lucky that Xiaolong's system which has only 300M memory can always 
> reproduce it. I checked the boog log and found on his system no any other 
> slot 
> can be chosen except for the original one. If we have a system with large 
> memory 
> it may not be easy to hit it, at least with low probability since there are 
> many 
> candidate slots.

So if it's just a better changelog that what I wrote I wouldn't rebase the 
commit: 
there's other commits on top of the fix meanwhile. Should a rebase become 
necessary at a later point I'll update it with your extended changelog.

Thanks,

Ingo

Re: [tip:x86/boot] x86/KASLR: Fix boot crash with certain memory configurations

[Baoquan He]

Hi Ingo,

I am sorry the previous post didn't contain formal patch log. I made a
new one as below. The boot crash could not only happen with certain
memory. Because of this code bug the regions which need be avoided like
the zipped kernel with its unzipping running code, initrd, kernel
command line could be corrupted if mem_avoid_overlap() can't find the
overlap region with the lowest address. But it's very lucky that
Xiaolong's system which has only 300M memory can always reproduce it.
I checked the boog log and found on his system no any other slot can be
chosen except for the original one. If we have a system with large memory
it may not be easy to hit it, at least with low probability since there
are many candidate slots.



>From 8f48aa39f3e49f9c1a9bb8ee61547dda7c2c05c3 Mon Sep 17 00:00:00 2001
From: Baoquan He 
Date: Fri, 1 Jul 2016 15:34:40 +0800
Subject: [PATCH] x86/KASLR: Fix boot crash caused by wrongly chosen kernel
 physical address

System halted with the separate randomization code applied. With debug printing
we got the reason that the chosen kernel physical address randomly is overlapped
with input dada. So input data and its running space must be corrupted during
decompressing kernel, then boot crash happened.

The root cause is that in function mem_avoid_overlap() local variable 'earliest'
is not updated correctly. Function mem_avoid_overlap is used to find the overlap
region with the lowest address, and 'earliest' is used to track the lowest 
address.
Decompressing kernel could step into those regions which need be avoided if we
didn't handle these overlap region correctly. So fix the code bug now.

Signed-off-by: Baoquan He 
---
 arch/x86/boot/compressed/kaslr.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/x86/boot/compressed/kaslr.c b/arch/x86/boot/compressed/kaslr.c
index 304c5c3..8e1fdf7 100644
--- a/arch/x86/boot/compressed/kaslr.c
+++ b/arch/x86/boot/compressed/kaslr.c
@@ -285,6 +285,7 @@ static bool mem_avoid_overlap(struct mem_vector *img,
if (mem_overlaps(img, _avoid[i]) &&
mem_avoid[i].start < earliest) {
*overlap = mem_avoid[i];
+   earliest = overlap->start;
is_overlapping = true;
}
}
@@ -299,6 +300,7 @@ static bool mem_avoid_overlap(struct mem_vector *img,
 
if (mem_overlaps(img, ) && (avoid.start < earliest)) {
*overlap = avoid;
+   earliest = overlap->start;
is_overlapping = true;
}
 
-- 
2.5.5

Re: [tip:x86/boot] x86/KASLR: Fix boot crash with certain memory configurations

[Baoquan He]

Hi Ingo,

I am sorry the previous post didn't contain formal patch log. I made a
new one as below. The boot crash could not only happen with certain
memory. Because of this code bug the regions which need be avoided like
the zipped kernel with its unzipping running code, initrd, kernel
command line could be corrupted if mem_avoid_overlap() can't find the
overlap region with the lowest address. But it's very lucky that
Xiaolong's system which has only 300M memory can always reproduce it.
I checked the boog log and found on his system no any other slot can be
chosen except for the original one. If we have a system with large memory
it may not be easy to hit it, at least with low probability since there
are many candidate slots.



>From 8f48aa39f3e49f9c1a9bb8ee61547dda7c2c05c3 Mon Sep 17 00:00:00 2001
From: Baoquan He 
Date: Fri, 1 Jul 2016 15:34:40 +0800
Subject: [PATCH] x86/KASLR: Fix boot crash caused by wrongly chosen kernel
 physical address

System halted with the separate randomization code applied. With debug printing
we got the reason that the chosen kernel physical address randomly is overlapped
with input dada. So input data and its running space must be corrupted during
decompressing kernel, then boot crash happened.

The root cause is that in function mem_avoid_overlap() local variable 'earliest'
is not updated correctly. Function mem_avoid_overlap is used to find the overlap
region with the lowest address, and 'earliest' is used to track the lowest 
address.
Decompressing kernel could step into those regions which need be avoided if we
didn't handle these overlap region correctly. So fix the code bug now.

Signed-off-by: Baoquan He 
---
 arch/x86/boot/compressed/kaslr.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/x86/boot/compressed/kaslr.c b/arch/x86/boot/compressed/kaslr.c
index 304c5c3..8e1fdf7 100644
--- a/arch/x86/boot/compressed/kaslr.c
+++ b/arch/x86/boot/compressed/kaslr.c
@@ -285,6 +285,7 @@ static bool mem_avoid_overlap(struct mem_vector *img,
if (mem_overlaps(img, _avoid[i]) &&
mem_avoid[i].start < earliest) {
*overlap = mem_avoid[i];
+   earliest = overlap->start;
is_overlapping = true;
}
}
@@ -299,6 +300,7 @@ static bool mem_avoid_overlap(struct mem_vector *img,
 
if (mem_overlaps(img, ) && (avoid.start < earliest)) {
*overlap = avoid;
+   earliest = overlap->start;
is_overlapping = true;
}
 
-- 
2.5.5