Takahiro Yasui [tyasui@xxxxxxxxxx] wrote: > On 11/18/09 07:11, Mikulas Patocka wrote: > > @@ -760,6 +775,7 @@ static void do_mirror(struct work_struct > > bio_list_init(&ms->reads); > > bio_list_init(&ms->writes); > > bio_list_init(&ms->failures); > > + bio_list_init(&ms->hold); > > spin_unlock_irqrestore(&ms->lock, flags); > > Initializing the hold list in do_mirror() is a problem. Some bios might > already in it and they will never be processed. This makes device-mapper > "stuck" when the device goes suspend. The hold list should be inizialized > only when the mirror device is created in alloc_context(). > > alloc_context() > spin_lock_init(&ms->lock); > bio_list_init(&ms->reads); > bio_list_init(&ms->writes); > bio_list_init(&ms->failures); > + bio_list_init(&ms->hold); > > Thanks, > Taka Good catch! -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
