On Fri, 29 Apr 2005 10:47:54 -0400 Mike Edwards <[EMAIL PROTECTED]> wrote:
> Apr 29 08:45:34 psychology kernel: __alloc_pages: 2-order allocation failed > (gfp=0x20/0) > Apr 29 08:45:34 psychology kernel: __alloc_pages: task(spamc) pid(25917) > caller(000000000046edc0) > Apr 29 08:45:34 psychology kernel: DEBUG: __get_free_pages() order 2 failed > called by 000000000046aaa8 ... > 000000000046a9c0 t kmem_cache_grow > 000000000046ad00 T kmem_cache_alloc_batch Does spamc use AF_UNIX sockets to communicate with other processes? If so, someone should audit it to make sure it checks properly for return values on write() calls. If spamc uses non-blocking AF_UNIX sockets, -ENOBUFS can happen under normal circumstances. If the sockets were blocking then the kernel could sleep to swap out some pages to free up some space and then the call could succeed. A quick perusal of the spamc sources seems to indicate that it does use non-blocking AF_UNIX sockets as the transport to talk to the spamassasindaemon. I'm on vacation so cannot investigate this more deeply. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
