Re: stack smashing detected with 'nvme sanitize-log /dev/nvme0'

[Christoph Hellwig <hch@lst.de>]

On Wed, Jul 26, 2023 at 01:52:04PM +0200, Daniel Wagner wrote:
> FYI, I got a a bug report [1] with a 'stack smashing detected' when running
> 'nvme sanitize-log /dev/nvme0' on Debian. Originally, it was reported against
> udisk. udisk recently added libnvme which does now a sanitize-log call, so this
> problem might exists for a while.
> 
> We figured out that an older kernel such as 4.19.289 work but newer not (it's a
> bit hard for the reporter to test all combinations on his setup due to compiler
> changes etc.).
> 
> There was a bit of refactoring in v5.2 which could be the cause of the stack
> smash, because saw this recent fix:
> 
>  b8f6446b6853 ("nvme-pci: fix DMA direction of unmapping integrity data")
> 
> [1] https://github.com/storaged-project/udisks/issues/1152

If you think it is related to DMA, there are good ways to check for:

  1) force that an IOMMU is used for this device
  2) hack nvme or the blk-map code that we never do the direct mapping
     to user space but do the copy based version, and then enable
     all kernel memory debugging helpers, most importantly KASAN