Cyril Roelandt, le Mon 17 Dec 2012 00:51:29 +0100, a écrit : > * libdiskfs/dir-renamed.c (diskfs_rename_dir): remove a redundant call to > pthread_mutex_unlock. > > Signed-off-by: Cyril Roelandt <tipec...@gmail.com> > --- > libdiskfs/dir-renamed.c | 2 -- > 1 file changed, 2 deletions(-) > > diff --git a/libdiskfs/dir-renamed.c b/libdiskfs/dir-renamed.c > index 3990b13..7829373 100644 > --- a/libdiskfs/dir-renamed.c > +++ b/libdiskfs/dir-renamed.c > @@ -228,8 +228,6 @@ diskfs_rename_dir (struct node *fdp, struct node *fnp, > const char *fromname, > diskfs_nput (tnp); > if (fdp && fdp != tdp) > pthread_mutex_unlock (&fdp->lock); > - if (fnp) > - pthread_mutex_unlock (&fnp->lock);
I believe it's not so simple: AIUI, diskfs_lookup actually returns fnp in tmpnp (see assert (!tmpnp || tmpnp == fnp);), locked, so an unlock is actually needed. The !tmpnp case however needs to be further investigated, since in that case fnp would not be locked. I guess this testcase could be triggered by running a lot of concurrent renamers for the same directory. So nacked, but worth further investigation. Samuel
