cachefiles: Fix race between read_waiter and read_copier
Get series with b4
b4 am [email protected]Versions newest first
- v1 Latest1 patch 0 repliesNew
Patches in v1 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/1 | cachefiles: Fix race between read_waiter and read_copier involving op->to_dofs/ | +1 | 0 replies |
Reviews trailers per patch
No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.
Cover letter [PATCH 0/1] cachefiles: Fix race between read_waiter and read_copier
This patch was originally posted by Lei Xue in 2018:
https://lore.kernel.org/patchwork/patch/889373/
The responses on the above thread ended up fixing a separate, but related problem in this code path, but the last portion of the commit message indicated the original problem was thought to have been fixed as well.
commit 934140ab028713a61de8bca58c05332416d037d1
Author: Kiran Kumar Modukuri <[email protected]>
Date: 2018-07-25 15:04:25 +0100
cachefiles: Fix refcounting bug in backing-file read monitoring
However, the original problem reported by Lei still remains and is fairly easy to reproduce. My reproducer details are below and I could reproduce within a few minutes. I ended up at the same patch after tracing of the problem and proving this race still exists, then testing with this patch applied. The work is detailed in the following bug:
https://bugzilla.redhat.com/show_bug.cgi?id=1829662
I am re-submitting this with Lei as author as I've only rebased the patch on dhowells fscache-fixes branch, changed the subject line, and cleaned up the patch header.
Reproducer
==========
# NFS server setup / config
# uname -r 3.10.0-1127.el7.x86_64 # cat /etc/exports /data *(rw,sec=sys:krb5) # df -h /data Filesystem Size Used Avail Use% Mounted on /dev/vdb 16G 16G 340M 98% /data # time for i in $(seq 1 4000); do dd if=/dev/zero of=/data/file$i.bin bs=1M count=4; done &
# NFS client config and test
# uname -r 3.10.0-1062.23.1.el7.x86_64 # _or_ # From https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/commit/?h=fscache-fixes&id=1836833f67ab49363d221aeb120956448ca5be4f # uname -r 5.7.0-rc3-1836833f67ab # df /var/cache/fscache Filesystem Size Used Avail Use% Mounted on /dev/md0 7.8G 36M 7.3G 1% /var/cache/fscache # make sure cachefilesd is running systemctl status cachefilesd # To run the test on the client: # mount -overs=3,fsc nfs-server:/data /mnt/nfs # cat dd-ioload.sh NFS_MNT=/mnt/nfs echo 3 > /proc/sys/vm/drop_caches for i in $(seq 1 2000); do dd if=$NFS_MNT/file$i.bin of=/dev/null bs=28k >/dev/null 2>&1 & done wait # while true; do date; time ./dd-ioload.sh; done &