Masaru OKI wrote: > Module Name: src > Committed By: oki > Date: Fri Mar 12 13:33:19 UTC 2010 > > Modified Files: > src/sys/netinet: in.c > > Log Message: > Fixed a number of race conditions in the case of receiving ipv4 packet. > found by iij seil team. > > > To generate a diff of this commit: > cvs rdiff -u -r1.136 -r1.137 src/sys/netinet/in.c > > Please note that diffs are not public domain; they are subject to the > copyright notices on the relevant files.
Hi, Some questions and remarks: * Can you detail a little bit what are the exactly race conditions this patch is covering ? * I think the locking is too exhaustive and the lock paths are too long. Also, some code rearranging can be done in order to shorten these lock paths. * you bothered to splsoftnet() before in_ifinit that does splnet() anyway * I don't think you need to splsoftnet for pfil_run_hooks * splsoftnet for if_ioctl ? I think if_ioctls should be fixed instead - although all I've seen until now are doing splnet for critical paths. * probably only rtinit blocks - in in_control() and in_scrubprefix() - and tail operations on ifaddrs should be locked. * panics when trying to set an IP address for a gre tunnel - rip_usrreq wants to splsoftnet, too * it's been pulled-up in netbsd-5. At least the latter panic should be solved before anything else. -- Mihai
