summary is pretty uninformative, but i can't think of a better one. Distribution: slackware 11.0 Hardware Environment: Supermicro X6DH8-G Software Environment: 2.6.19.2 using ext3 and nfs. Problem Description: following messages were printed (inbetween "cut here" marks). kernel BUG at fs/inode.c:1117! invalid opcode: 0000 [#1] SMP Modules linked in: CPU: 1 EIP: 0060:[<c016e651>] Not tainted VLI EFLAGS: 00010246 (2.6.19.2 #5) EIP is at iput+0x5b/0x65 eax: c04935a0 ebx: c87b20a8 ecx: 00000000 edx: 00000000 esi: 00000000 edi: f786e2c0 ebp: 00000000 esp: dbf69f20 ds: 007b es: 007b ss: 0068 Process nfsd (pid: 25285, ti=dbf68000 task=f7b58a70 task.ti=dbf68000) Stack: f74d6000 c03d5c62 f74d6000 00000000 c03d5ca1 f786e2c0 f74d6000 00000000 00000042 c03d75c9 f786e2f8 f786e2c0 c03d7702 00000000 f75298f4 f79fd080 f780e3c0 00057e40 00000000 f7b58a70 c011727a 00000000 00000000 0000003d Call Trace: [<c03d5c62>] svc_sock_release+0xde/0x149 [<c03d5ca1>] svc_sock_release+0x11d/0x149 [<c03d75c9>] svc_recv+0x39a/0x415 [<c03d7702>] svc_send+0x8f/0x11b [<c011727a>] default_wake_function+0x0/0xc [<c011727a>] default_wake_function+0x0/0xc [<c01f5439>] nfsd+0xc5/0x280 [<c01f5374>] nfsd+0x0/0x280 [<c010380f>] kernel_thread_helper+0x7/0x10 ======================= Code: 00 85 c0 74 1e 8b 83 98 00 00 00 ba e5 e5 16 c0 8b 40 20 85 c0 74 08 8b 40 18 85 c0 0f 45 d0 89 d8 ff d2 5b c3 89 d8 ff d2 eb c9 <0f> 0b 5d 04 80 9d 40 c0 eb b4 83 ec 10 89 7c 24 08 89 6c 24 0c EIP: [<c016e651>] iput+0x5b/0x65 SS:ESP 0068:dbf69f20 <5>rpc-srv/tcp: nfsd: got error -104 when sending 4228 bytes - shutting down socket
What version of nfs-utils (showmount -V)? Were you starting or stopping things at the time, or did it just die while in normal use? Has this happened more than once (i.e. is it repeatable?) Did any of the clients crash or do something unusual? Thanks. NeilBrown
showmount complained about uppercase V, but gave up with lowercase v :) 1.0.10 the problem happened during normal system use (though system was having somewhat high load, mostly from nfs). it is not happening regularly, but it seems to have happened twice, with about a minute inbetween. hard to tell about clients, most are not under my control.
Created attachment 10282 [details] Patch which might fix the problem I think this patch might fix the problem.
Are you in a position to test the patch? Has it happened against since the two times you report? Thanks.
patch results for 2.6.19.2 : Hunk #1 succeeded at 499 (offset -100 lines). Hunk #2 succeeded at 1243 (offset -52 lines). Hunk #3 succeeded at 1664 with fuzz 1 (offset -108 lines). patch results for 2.6.20 : Hunk #1 succeeded at 529 (offset -70 lines). Hunk #2 succeeded at 1247 (offset -48 lines). Hunk #3 succeeded at 1698 with fuzz 1 (offset -74 lines). so i applied the patch to 2.6.20 and that's what the machine is running now. the problem has not occurred again, but there also has not been such a prolonged period of high load since then.
Created attachment 10322 [details] Revised patch That patch had a bug - when you shut down the NFS server it will not close the sockets properly and a reboot will be required. That probably won't be a problem for you as people rarely shut down the NFS server except when they want to reboot. The attached patch is against 2.6.20 and fixes this bug. I will submit it for 2.6.20.1 Thanks.
indeed, nfs server shutdown is not something that occurs too often. thanks for the fast response and fix, i suppose the report can be set to 'resolved' now :) ah. i can do that myself. please, correct the status if i get it wrong.