Date: Sun, 9 Aug 1998 21:37:18 +0000 (GMT) From: Terry Lambert <tlambert@primenet.com> To: dillon@backplane.com (Matthew Dillon) Cc: tlambert@primenet.com, wollman@khavrinen.lcs.mit.edu, freebsd-fs@FreeBSD.ORG Subject: Re: Filesystem locking during lookups Message-ID: <199808092137.OAA18972@usr04.primenet.com> In-Reply-To: <199808081748.KAA19458@apollo.backplane.com> from "Matthew Dillon" at Aug 8, 98 10:48:43 am
next in thread | previous in thread | raw e-mail | index | archive | help
> However, I am still seeing significant VM/fs corruption where > the mmaping of files being actively appended to (in this case, > the spool files) can cause physical corruption of the file. Whistle has a test bed capable of reproducing this and several (seemingly) related problems. Unfortunately, it's statistical, so it's not useful for diagnostics, only for verification. The "seemingly" comes from the fact that we have seen file corruption, but we can repeatably diagnose the missing page problems. The append related problems seem to be a seperate symptom of the same problem, but are not reliably diagnosible except over several days for us (so it looks like you have the best test-jig for that one). We're willing to test any proposed fixes, with a (comparatively) short turn around. Note that these problems are *not* specific to 3.0-current. In the diagnostic vein, can I ask if the appends are occurring on page or FS block boundries? If you could ensure that the file extension was always a page and on a page boundry, then this would guarantee to reliably diagnose whether or not this is related to partial page behaviour. Is your FS block size 8k? If the allocations were forced to 8k instead of one page, this would distinguish whether or not the problems were realted to the treatment of FS frags or not (presuming they still occur when the 4k page size boundary and extension is used). If the problems disappear in either one of these cases, the next thing to try is to set the FS frag size of an 8k for to 4k (yes, I realize this will cause fragmentation aut the wazoo); if the problem still occurs, then disable fragging altogether. If none of these (unacceptable; merely diagnostic) workarounds fix the problem, then that also eliminates a large amount of code from consideration. Meanwhile, I will coinue my instrumentation of the VM system at home, and start triggerring panics on alias references as soon as I'm able, and work at it from that angle. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-fs" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199808092137.OAA18972>