Skip to contentSkip to conversation

mm/highmem: VM_BUG_ON() if offset + len > PAGE_SIZE

PATCH Applied Fabio M. De Francesco 2 participants 5 messages Apr 26, 2022 MM core

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm/highmem: VM_BUG_ON() if offset + len > PAGE_SIZE

Add VM_BUG_ON() bounds checking to make sure that, if "offset + len> PAGE_SIZE", memset() does not corrupt data in adjacent pages.

---
 include/linux/highmem.h | 2 ++
 1 file changed, 2 insertions(+)
include/linux/highmem.h +2 -0
index 6b2d59e025c5..d54dbaae9a5e 100644
@@ -380,6 +380,8 @@ static inline void memcpy_to_page(struct page *page, size_t offset,
 static inline void memzero_page(struct page *page, size_t offset, size_t len)
 {
 	char *addr = kmap_local_page(page);
+
+	VM_BUG_ON(offset + len > PAGE_SIZE);
 	memset(addr + offset, 0, len);
 	flush_dcache_page(page);
 	kunmap_local(addr);
Signature · 1 line
2.34.1
maintainer
#1 lore raw
Quoted 3 lines · On Tue, 26 Apr 2022 21:30:20 +0200 "Fabio M. De Francesco"…

Add VM_BUG_ON() bounds checking to make sure that, if "offset + len> PAGE_SIZE", memset() does not corrupt data in adjacent pages.

hm, why? To match all the other functions in there?

I suppose that's logical. Or we could just delete all the other VM_BUG_ON()s. Have any of them proven to be at all useful?

#2 lore raw
Quoted 1 line · On martedì 26 aprile 2022 21:34:12 CEST Andrew Morton wrote:

On Tue, 26 Apr 2022 21:30:20 +0200 "Fabio M. De Francesco"

Quoted 10 lines · <[email protected]> wrote:
Quoted 4 lines

Add VM_BUG_ON() bounds checking to make sure that, if "offset + len> PAGE_SIZE", memset() does not corrupt data in adjacent pages.

hm, why? To match all the other functions in there?

I suppose that's logical. Or we could just delete all the other VM_BUG_ON()s. Have any of them proven to be at all useful?

I am not so sure about it being so useful. I just noted that memzero_page() is the only function of that family that is implemented with no VM_BUG_ON(). I have no actual proofs of usefulness :(

This is why yesterday I sent an "RFC Patch" (please see
https://lore.kernel.org/lkml/[email protected]/

Soon after sending it I thought that VM_WARN_ON_ONCE() could have been better suited, but Ira Weiny wrote to use VM_BUG_ON() for consistency.

Now I could either delete all other VM_BUG_ON() or replace them with VM_WARN_ON_ONCE() (or some other macro).

Ah, a third solution might be to leave highmem.h as it is now :)

What do you prefer?

Thanks,

Fabio

maintainer
#3 lore raw
Quoted 23 lines · On Tue, 26 Apr 2022 22:19:57 +0200 "Fabio M. De Francesco"…
Quoted 1 line · On martedì 26 aprile 2022 21:34:12 CEST Andrew Morton wrote:

On Tue, 26 Apr 2022 21:30:20 +0200 "Fabio M. De Francesco"

Quoted 10 lines · <[email protected]> wrote:
Quoted 4 lines

Add VM_BUG_ON() bounds checking to make sure that, if "offset + len> PAGE_SIZE", memset() does not corrupt data in adjacent pages.

hm, why? To match all the other functions in there?

I suppose that's logical. Or we could just delete all the other VM_BUG_ON()s. Have any of them proven to be at all useful?

I am not so sure about it being so useful. I just noted that memzero_page() is the only function of that family that is implemented with no VM_BUG_ON(). I have no actual proofs of usefulness :(

This is why yesterday I sent an "RFC Patch" (please see
https://lore.kernel.org/lkml/[email protected]/

Soon after sending it I thought that VM_WARN_ON_ONCE() could have been better suited, but Ira Weiny wrote to use VM_BUG_ON() for consistency.

Quoted 6 lines

Now I could either delete all other VM_BUG_ON() or replace them with VM_WARN_ON_ONCE() (or some other macro).

Ah, a third solution might be to leave highmem.h as it is now :)

What do you prefer?

Merge this patch as-is, I guess. Going through and removing unuseful VM_BUG_ON()s is a separable activity.

#4 lore raw
Quoted 1 line · On martedì 26 aprile 2022 22:48:11 CEST Andrew Morton wrote:

On Tue, 26 Apr 2022 22:19:57 +0200 "Fabio M. De Francesco"

Quoted 6 lines · <[email protected]> wrote:
Quoted 6 lines · On martedì 26 aprile 2022 21:34:12 CEST Andrew Morton wrote:
Quoted 1 line

On Tue, 26 Apr 2022 21:30:20 +0200 "Fabio M. De Francesco"

Quoted 2 lines · <[email protected]> wrote:
Quoted 2 lines

Add VM_BUG_ON() bounds checking to make sure that, if "offset +

len>

Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
Quoted 2 lines

PAGE_SIZE", memset() does not corrupt data in adjacent pages.

hm, why? To match all the other functions in there?

I suppose that's logical. Or we could just delete all the other VM_BUG_ON()s. Have any of them proven to be at all useful?

I am not so sure about it being so useful. I just noted that

memzero_page()

Quoted 12 lines
Quoted 12 lines

is the only function of that family that is implemented with no VM_BUG_ON(). I have no actual proofs of usefulness :(

This is why yesterday I sent an "RFC Patch" (please see
https://lore.kernel.org/lkml/[email protected]/

Soon after sending it I thought that VM_WARN_ON_ONCE() could have been better suited, but Ira Weiny wrote to use VM_BUG_ON() for consistency.

Now I could either delete all other VM_BUG_ON() or replace them with VM_WARN_ON_ONCE() (or some other macro).

Quoted 7 lines
Quoted 3 lines

Ah, a third solution might be to leave highmem.h as it is now :)

What do you prefer?

Merge this patch as-is, I guess. Going through and removing unuseful VM_BUG_ON()s is a separable activity.

Thanks!

While at this, I've just noted that you sent a confirmation which lists all the patches of mine that are currently in your tree.

I see that you took v1 of "Extend and reorganize Highmem's documentation". I suppose that you missed the v2 of that series at
https://lore.kernel.org/lkml/[email protected]/

Can you please check it?

Thanks,

Fabio