[00:15:14] --- dev-zero@jabber.org has left [01:51:09] --- dev-zero@jabber.org has become available [01:51:27] --- dev-zero@jabber.org has left [01:51:39] --- dev-zero@jabber.org has become available [01:52:07] --- dev-zero@jabber.org has left: offline [02:19:32] --- abo has left [06:00:25] --- Jeffrey Altman has left: Replaced by new connection [07:46:53] --- pod has become available [08:38:27] What does Nov 21 01:17:14 surgeon kernel: WARNING: afs_ufswr vcp=ffff8101728debc0, exOrW=0 (printed from afs_UFSWrite()) mean? [08:40:05] This is 1.4.11 on amd64 kernel 2.6.18-92.1.22.el5.centos.plus [09:48:45] It means that UFSWrite is seeing a write on a file that it doesn't think is open for writing. [11:36:00] My suspicion is that we've still got pages in flight when release is being called, and so the close is racing with pdflush. [13:24:37] --- Russ has become available [14:22:54] The problem is that sys_close() does flush, then release, and I just can't come up with a scenario where a process can both be the last writer, and dirty a page between those two happening. [15:24:39] --- jaltman has become available [15:30:35] --- haba has left: Lost connection [18:05:43] --- mdionne has become available [19:22:28] what about a page incorrectly left marked as dirty in the mapping's radix tree, for instance from failing to call set_page_writeback. [19:22:45] --- mdionne has left