On Wed, Sep 2, 2015 at 12:14 PM, Yann Ylavic <ylavic....@gmail.com> wrote: > On Wed, Sep 2, 2015 at 11:48 AM, Plüm, Rüdiger, Vodafone Group > <ruediger.pl...@vodafone.com> wrote: >> >> From: Yann Ylavic [mailto:ylavic....@gmail.com] >>> >>> BTW, even on Unixes there is a race on these files between the main >>> process and the children, or the children themselves from different >>> "generation". >> >> Really? I would think that these are files with the same name but different >> inodes and hence different. > > Yes, but I was thinking of eg. an old-generation child removing the > file between shm_create (post_config's slotmem_create) and shm_attach > (child_init's slotmem_attach) from the new-generation.
Hm, that seems to be guarded by the "globallistmem" cache in mod_slotmem_shm... So I am probably wrong about this race. On Windows, "globallistmem" is not inherited (forked), hence the shm_attach() path is taken, so we can't use anonymous SHMs anyway... How about the generation in the filename? Windows only?