On Thu, Jul 18, 2024 at 11:27:23AM -0700, Roman Kisel wrote: > A powerful way to diagnose crashes is to analyze the core dump produced upon > the failure. Missing or malformed core dump files hinder these investigations. > I'd like to propose changes that add logging as to why the kernel would not > finish writing out the core dump file. > > To help in diagnosing the user mode helper not writing out the entire coredump > contents, the changes also log short statistics on the dump collection. I'd > advocate for keeping this at the info level on these grounds. > > For validation, I built the kernel and a simple user space to exercize the new > code. > > [V3] > - Standartized the existing logging to report TGID and comm consistently > - Fixed compiler warnings for the 32-bit systems (used %zd in the format strings) > > [V2] > https://lore.kernel.org/all/20240712215223.605363-1-romank@xxxxxxxxxxxxxxxxxxx/ > - Used _ratelimited to avoid spamming the system log > - Added comm and PID to the log messages > - Added logging to the failure paths in dump_interrupted, dump_skip, and dump_emit > - Fixed compiler warnings produced when CONFIG_COREDUMP is disabled > > [V1] > https://lore.kernel.org/all/20240617234133.1167523-1-romank@xxxxxxxxxxxxxxxxxxx/ > > Roman Kisel (2): > coredump: Standartize and fix logging > binfmt_elf, coredump: Log the reason of the failed core dumps > > fs/binfmt_elf.c | 48 +++++++++---- > fs/coredump.c | 150 +++++++++++++++++++++++++++------------ > include/linux/coredump.h | 30 +++++++- > kernel/signal.c | 21 +++++- > 4 files changed, 188 insertions(+), 61 deletions(-) This looks good to me! I'll put this in -next once the merge window closes. Thanks! -Kees -- Kees Cook
