Re: [SSI-devel] Re: 1.9 rmtfb assert bug solved?
Brought to you by:
brucewalker,
rogertsang
From: Roger T. <rog...@gm...> - 2005-08-25 00:58:42
|
Alright I'll add a SSI_XXX comment. I don't see any immediate fixes besides rewriting rmtfb that would alleviate the problem of client/server having different ino's.=20 Changing rmtfb_gethandle() might help, but will involve reop_export_path(). -Roger On 8/24/05, Brian J. Watson <Bri...@hp...> wrote: > Hi Roger, >=20 > The rmtfb_tryagain_handle() function is fine as a short-term hack to get > things working. You should also add an SSI_XXX comment that the > underlying bug should be fixed sometime, which is that the inos should > not be allowed to be different on the rmtfb client and server nodes. >=20 > Thanks, >=20 > Brian >=20 >=20 > John Byrne wrote: > > Roger Tsang wrote: > > > >> The problem is the rmtfb server stores the local inode when opening > >> /dev/tty. It is the controlling tty as understood. The client works > >> with it's own /dev/tty inode without realizing that the server has > >> opened the controlling tty instead. So rmtfb_get_cmn() tests fail > >> because the server is trying to find the rfb_id for the client's > >> inode. For some reason, perhaps when the force is strong, some > >> applications ignore the -ERFB_TRYAGAIN errors and are able to > >> continue. This bug started to surface when we upgraded to a more > >> modern kernel (2.4 -> 2.6). There seems to be a pattern. The > >> problem, if not fixed, is even more pronounced after porting OpenSSI > >> to kernel-2.6.11. > >> > >> This patch fixes three main things. Time to check into the trunk? I > >> also wonder if this patch should be backported. > >> > >> 1. remote /dev/tty > >> 2. local open /dev/pts > >> 3. false as_do_vma() assert failures > >> > >> -Roger > >> > > > > Brian will have to comment on the rmtfb changes; the other changes > > appear to be okay. > > > > John > > > > > > >=20 > |