On Fri, 2022-01-21 at 14:58 +0800, Ian Kent wrote: > On Thu, 2022-01-20 at 14:55 -0300, Andreas Hasenack wrote: > > Hi again, > > > > On Thu, Jan 20, 2022 at 10:16 AM Andreas Hasenack > > <andreas@xxxxxxxxxxxxx> wrote: > > > Something still feels wrong, though, and behaves differently from > > > 5.1.7. > > > > > > I applied all your patches for 5.1.8, even unrelated ones, and > > > tried this: > > > # this would trigger the segfault in plain 5.1.8: > > > root@j-s390x:~# time ls -la /cifs/localhost/private > > > total 4 > > > drwxr-xr-x 2 root root 0 Dec 27 20:15 . > > > drwxr-xr-x 5 root root 0 Jan 20 13:03 .. > > > -rwxr-xr-x 1 root root 64 Jan 19 13:02 hello-private.txt > > > > > > real 0m0.130s > > > user 0m0.000s > > > sys 0m0.001s > > > root@j-s390x:~# time ls -la /cifs/localhost/pub > > > total 4 > > > drwxr-xr-x 2 root root 0 Dec 27 20:15 . > > > drwxr-xr-x 5 root root 0 Jan 20 13:03 .. > > > -rwxr-xr-x 1 root root 63 Jan 19 13:02 hello-public.txt > > > > > > real 0m0.014s > > > user 0m0.001s > > > sys 0m0.001s > > > > > > # This is expected to fail, because I only have > > > /etc/creds/localhost, > > > not /etc/creds/127.0.0.1 > > > root@j-s390x:~# time ls -la /cifs/127.0.0.1/private > > > ls: cannot access '/cifs/127.0.0.1/private': No such file or > > > directory > > > > > > real 0m0.075s > > > user 0m0.000s > > > sys 0m0.001s > > > > > > # but this hangs, and ls process goes into D state. With 5.1.7 it > > > works. > > > root@j-s390x:~# time ls -la /cifs/127.0.0.1/pub > > > <hangs> > > > > The above hang happens for me when patch 5 is applied: [PATCH 05/19] > > autofs-5.1.8 - fix root offset error handling > > > > In summary, for my s390x case: > > - [PATCH 05/19] autofs-5.1.8 - fix root offset error handling: > > introduces the hang in the above scenario > > - [PATCH 08/19] autofs-5.1.8 - fix loop under run in > > cache_get_offset_parent(): fixes the s390x segfault, and failure to > > mount the private cifs share > > Thanks for you efforts on this problem. > The patch 5 introducing a hang is interesting, I'll have to have a > look at that. I think this patch might help with that hang. I may merge that into the original patch but it's separate for now to facilitate testing. autofs-5.1.8 - fix fix root offset error handling From: Ian Kent <raven@xxxxxxxxxx> The change to fix root offset error handlling is missing a cache read lock prior to the key lookup, the following unmatched unlock then causes a hang. Signed-off-by: Ian Kent <raven@xxxxxxxxxx> --- CHANGELOG | 1 + daemon/direct.c | 1 + 2 files changed, 2 insertions(+) diff --git a/CHANGELOG b/CHANGELOG index 6f18a0bb..f81b0259 100644 --- a/CHANGELOG +++ b/CHANGELOG @@ -3,6 +3,7 @@ - fix set open file limit. - improve descriptor open error reporting. - fix root offset error handling. +- fix fix root offset error handling. 19/10/2021 autofs-5.1.8 - add xdr_exports(). diff --git a/daemon/direct.c b/daemon/direct.c index 8810900c..cf3f24d7 100644 --- a/daemon/direct.c +++ b/daemon/direct.c @@ -1275,6 +1275,7 @@ static void *do_mount_direct(void *arg) /* If this is a multi-mount subtree mount failure * ensure the tree continues to expire. */ + cache_readlock(mt.mc); me = cache_lookup_distinct(mt.mc, mt.name); if (me && IS_MM(me) && !IS_MM_ROOT(me)) conditional_alarm_add(ap, ap->exp_runfreq);
