Skip to contentSkip to conversation

IA64: virt_to_page() can be called with NULL arg

PATCH In review [email protected] 4 participants 10 messages Dec 19, 2006 – Dec 20, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 3/3 IA64: virt_to_page() can be called with NULL arg

From: Kirill Korotaev <[email protected]>

It does not return NULL when arg is NULL.

---

 include/asm-ia64/pgalloc.h |    3 ++-
 1 files changed, 2 insertions(+), 1 deletion(-)
include/asm-ia64/pgalloc.h renamed from include/asm-ia64/pgalloc.h~ia64-virt_to_page-can-be-called-with-null-arg +2 -1
diff -puN include/asm-ia64/pgalloc.h~ia64-virt_to_page-can-be-called-with-null-arg include/asm-ia64/pgalloc.h
@@ -137,7 +137,8 @@ pmd_populate_kernel(struct mm_struct *mm
 static inline struct page *pte_alloc_one(struct mm_struct *mm,
 					 unsigned long addr)
 {
-	return virt_to_page(pgtable_quicklist_alloc());
+	void *pg = pgtable_quicklist_alloc();
+	return pg ? virt_to_page(pg) : NULL;
 }
 
 static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm,

_

#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"akpm" = akpm <[email protected]> writes:

akpm> From: Kirill Korotaev <[email protected]> It does not return NULL akpm> when arg is NULL.

Shouldn't the real fix be to track down who calls virt_to_page() with a NULL pointer? IMHO it is bogus to do so.

Cheers,
Jes

#2 lore raw
In reply to [email protected]
Quoted 9 lines · Kirill Korotaev
Quoted 7 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"akpm" = akpm <[email protected]> writes:

akpm> From: Kirill Korotaev <[email protected]> It does not return NULL akpm> when arg is NULL.

Shouldn't the real fix be to track down who calls virt_to_page() with a NULL pointer? IMHO it is bogus to do so.

what do you propose? to insert BUG_ON(!kaddr) into virt_to_page()? in this case caller in question should be still fixed.

If you hit this, yes I'd insert the BUG_ON in your test kernel and fix the code. Maybe add the BUG_ON in upstream for CONFIG_DEBUG or something.

Which callers did you see cause this? If it was a common problem I would expect a lot of data corruption or crashes on ia64 systems which I haven't heard of.

Cheers,
Jes

#3 lore raw
In reply to [email protected]
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"akpm" = akpm <[email protected]> writes:

akpm> From: Kirill Korotaev <[email protected]> It does not return NULL akpm> when arg is NULL.

Shouldn't the real fix be to track down who calls virt_to_page() with a NULL pointer? IMHO it is bogus to do so.

what do you propose? to insert BUG_ON(!kaddr) into virt_to_page()? in this case caller in question should be still fixed.

Thanks,
Kirill

#4 lore raw
In reply to [email protected]
Quoted 7 lines · Kirill Korotaev
Quoted 3 lines · Jes Sorensen

If you hit this, yes I'd insert the BUG_ON in your test kernel and fix the code. Maybe add the BUG_ON in upstream for CONFIG_DEBUG or something.

I guess then all the platforms should be analyzed/patched carefully or all the callers of virt_to_page().
Care to create debug patch?

Well you suggested a patch which just hides the problem. I suggest you change it to have the BUG_ON().

Quoted 11 lines
Quoted 3 lines

Which callers did you see cause this? If it was a common problem I would expect a lot of data corruption or crashes on ia64 systems which I haven't heard of.

from the patch:
pte_alloc_one() calls pgtable_quicklist_alloc() which can return NULL in case of allocation failure.

It was hit on OpenVZ where kernel memory is accounted and limited on per-container basis (it is possible to DoS using page tables allocations). In mainstream the bug can be hit if OOM killer
kills the process and __get_free_page() returns NULL which is rare, but still possible.

I see, since you have it tracked down, it would be good to fix it and push a patch upstream. Unless of course Andrew or Linus thinks this is the wrong approach.

Cheers,
Jes

#5 lore raw
In reply to [email protected]
Quoted 10 lines · Jes Sorensen
Quoted 8 lines · Kirill Korotaev
Quoted 7 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"akpm" = akpm <[email protected]> writes:

akpm> From: Kirill Korotaev <[email protected]> It does not return NULL akpm> when arg is NULL.

Shouldn't the real fix be to track down who calls virt_to_page() with a NULL pointer? IMHO it is bogus to do so.

Quoted 7 lines
Quoted 2 lines

what do you propose? to insert BUG_ON(!kaddr) into virt_to_page()? in this case caller in question should be still fixed.

If you hit this, yes I'd insert the BUG_ON in your test kernel and fix the code. Maybe add the BUG_ON in upstream for CONFIG_DEBUG or something.

I guess then all the platforms should be analyzed/patched carefully or all the callers of virt_to_page().
Care to create debug patch?

Quoted 3 lines

Which callers did you see cause this? If it was a common problem I would expect a lot of data corruption or crashes on ia64 systems which I haven't heard of.

from the patch:
pte_alloc_one() calls pgtable_quicklist_alloc() which can return NULL in case of allocation failure.

It was hit on OpenVZ where kernel memory is accounted and limited on per-container basis (it is possible to DoS using page tables allocations). In mainstream the bug can be hit if OOM killer
kills the process and __get_free_page() returns NULL which is rare, but still possible.

Thanks,
Kirill

#6 lore raw
In reply to [email protected]
Quoted 5 lines · On Wed, 20 Dec 2006 11:14:53 +0100 Jes Sorensen <[email protected]> wrote:
Quoted 4 lines · Kirill Korotaev
Quoted 3 lines · Jes Sorensen

If you hit this, yes I'd insert the BUG_ON in your test kernel and fix the code. Maybe add the BUG_ON in upstream for CONFIG_DEBUG or something.

Quoted 7 lines
Quoted 3 lines

I guess then all the platforms should be analyzed/patched carefully or all the callers of virt_to_page().
Care to create debug patch?

Well you suggested a patch which just hides the problem. I suggest you change it to have the BUG_ON().

The patch doesn't hide any problems:

include/asm-ia64/pgalloc.h renamed from include/asm-ia64/pgalloc.h~ia64-virt_to_page-can-be-called-with-null-arg +2 -1
@@ -137,7 +137,8 @@ pmd_populate_kernel(struct mm_struct *mm
 static inline struct page *pte_alloc_one(struct mm_struct *mm,
 					 unsigned long addr)
 {
-	return virt_to_page(pgtable_quicklist_alloc());
+	void *pg = pgtable_quicklist_alloc();
+	return pg ? virt_to_page(pg) : NULL;
 }
 
 

if the memory allocation function fails, we propagate that failure back to callers. All very good.

From a quick look it seems that the pte_alloc() callers are all nicely handling the failures.

The patch looks good to me?

#7 lore raw
In reply to [email protected]
Quoted 1 line · Andrew Morton

The patch doesn't hide any problems:

[snip]

Quoted 2 lines

if the memory allocation function fails, we propagate that failure back to callers. All very good.

ARGH!

I somehow put it in my head that the patch was modifying virt_to_page() rather than the pte function.

My apologies, this was written by a stupid person who can't read :-(

Cheers,
Jes

#8 lore raw
In reply to [email protected]

Jes,

Quoted 2 lines

Well you suggested a patch which just hides the problem. I suggest you change it to have the BUG_ON().

IMHO you are wrong.
the suggested patch *fixes* one particular place, which can be triggered on mainstream IA64 by a standard user and is actually a *SECURITY* bug which can be potentially exploited (when OOM killer is enabled).
It doesn't hide anything, It just doesn't help to catch other places.

Quoted 14 lines
Quoted 4 lines
Quoted 3 lines

Which callers did you see cause this? If it was a common problem I would expect a lot of data corruption or crashes on ia64 systems which I haven't heard of.

Quoted 8 lines

from the patch:
pte_alloc_one() calls pgtable_quicklist_alloc() which can return NULL in case of allocation failure.

It was hit on OpenVZ where kernel memory is accounted and limited on per-container basis (it is possible to DoS using page tables allocations). In mainstream the bug can be hit if OOM killer
kills the process and __get_free_page() returns NULL which is rare, but still possible.

Quoted 3 lines

I see, since you have it tracked down, it would be good to fix it and push a patch upstream. Unless of course Andrew or Linus thinks this is the wrong approach.

Maybe the fact that I came without an exploit to crash IA64 makes you think it should not be commited, ok, you can leave it as is then.
NOTE: I don't mind against the debug you proposed. It is quite a good idea.

Thanks,
Kirill

#9 lore raw
In reply to [email protected]
Quoted 9 lines · Kirill Korotaev

Jes,

Quoted 2 lines

Well you suggested a patch which just hides the problem. I suggest you change it to have the BUG_ON().

IMHO you are wrong.
the suggested patch *fixes* one particular place, which can be triggered on mainstream IA64 by a standard user and is actually a *SECURITY* bug which can be potentially exploited (when OOM killer is enabled).
It doesn't hide anything, It just doesn't help to catch other places.

Hi Kirill,

See my response to Andrew, you are indeed right.

I'll go back and hide in my corner in shame :-(

Cheers,
Jes