On Mon, 2018-10-01 at 10:12 -0700, Nick Desaulniers wrote: +AD4 On Mon, Oct 1, 2018 at 9:16 AM Evan Green +ADw-evgreen+AEA-chromium.org+AD4 wrote: +AD4 +AD4 +AD4 +AD4 From: Robb Glasser +ADw-rglasser+AEA-google.com+AD4 +AD4 +AD4 +AD4 +AD4 sg+AF8-ioctl could be spammed by requests, leading to a double free in +AD4 +AD4 +AF8AXw-free+AF8-pages. This protects the entry points of sg+AF8-ioctl where the +AD4 +AD4 memory could be corrupted by a double call to +AF8AXw-free+AF8-pages if multiple +AD4 +AD4 requests are happening concurrently. +AD4 +AD4 +AD4 +AD4 Signed-off-by: Robb Glasser +ADw-rglasser+AEA-google.com+AD4 +AD4 +AD4 Signed-off-by: Nick Desaulniers +ADw-ndesaulniers+AEA-google.com+AD4 +AD4 +AD4 Signed-off-by: Evan Green +ADw-evgreen+AEA-chromium.org+AD4 +AD4 +AD4 Cc: stable+AEA-vger.kernel.org +AD4 +AD4 +AD4 +AD4 --- +AD4 +AD4 Reposting this patch from last summer, as it looks like it fell in between +AD4 +AD4 the cracks. +AD4 +AD4 Christoph, do you still feel strongly about: https://lkml.org/lkml/2017/8/5/75 ? I don't know how Christoph feels about it, but serializing all SG I/O seems like a regression to me. If one sg command hangs I usually try to send another sg command to the same SCSI device from another shell to get more information about the nature of the hang. Serializing all SG I/O would make that impossible. Bart.
