I am trying to automate the conversion of linux kernel-generated coredumps to minidumps (using breakpad tools) on an embedded linux system running on an aarch64 board with 1Gb RAM.
The main problem I have is that the coredumps generated are too big to be stored and have to be gzipped when generated.
So I tried passing the core file stream directly to the breakpad tool used for conversion :
# echo '| xargs -I {} /usr/bin/core2md {} /proc /tmp/dmp.%e.%p' >/proc/sys/kernel/core_pattern
But it appears that core2md is expecting a file on disk (it mmaps the entire file then retrieves information using offsets).
I would like to know if there is a way to patch the core2md source code to :
* map the entire stream into memory without having to write a file on disk,
* or extract all necessary informations by directly reading the stream,
* or gunzip the file into memory
