On Sun, 13 Sep 2026 09:13:21 +0000, [email protected] wrote:
> - [High] blk_trace_synthesize_old_trace unconditionally parses ring buffer
> entries as 64-byte blk_io_trace2 structures, resulting in an out-of-bounds
> read and info leak when processing 48-byte v1 entries because the required
> dependency patch is missing.

Correct, and as the patch says, Adriano Cordova's patch has to be applied
first.

The read side cannot know the layout: magic and sequence are the ftrace
trace_entry header and neither writer fills them, and the sizes overlap,
since a 48-byte v1 record carrying a 16-byte BLK_TA_REMAP PDU is also 64.
His patch fixes it on the write side, by not recording a v1 entry at all.
With that applied this patch is correct -- measured on a BLKTRACESETUP
trace with the blk tracer, reading a partition for that remap.

Reply via email to