Date: Fri, 05 Jan 2001 17:44:02 +0100 From: Tor.Egge@fast.no To: blk@skynet.be Cc: grog@lemis.com, andy.depetter@ops.skynet.be, freebsd-stable@FreeBSD.ORG Subject: Re: Problems with corrupted vinum devices... Message-ID: <200101051644.RAA17887@midten.fast.no> In-Reply-To: Your message of "Fri, 5 Jan 2001 17:04:47 %2B0100" References: <v0422080ab67b97176227@[172.17.1.121]>
next in thread | previous in thread | raw e-mail | index | archive | help
> However, at some point within the next couple of hours (we're not
> quite sure when), the machine crashed, and vinum failed to come up on
> the reboot. We had to comment out the /etc/fstab entries that
> referred to the vinum volumes, and since then have been trying to
> debug the problems as to why vinum can't find or start any devices at
> all.
I suggest increasing INITIAL_DRIVES in vinumvar.h to avoid array
resize and the associated race conditions. When I tried to configure
vinum to use 14 disks yesterday, the machine immediately crashed with
a trap 12 in response to 'vinum create'. I bumped INITIAL_DRIVES to
16 to avoid the drive array resize that caused the problem. I bumped
INITIAL_SUBDISKS_IN_PLEX too, just to be safe.
To avoid similar races with RAID-5 under high load or with
softupdates, I had to bump INITIAL_LOCKS to avoid a fatal range lock
array resize.
For 4.2-STABLE, I'm currently using the following patch to implement
the workarounds:
diff -ru /tmp/vinum/vinumlock.c ./vinumlock.c
--- /tmp/vinum/vinumlock.c Mon May 22 18:21:37 2000
+++ ./vinumlock.c Thu Jan 4 20:07:06 2001
@@ -249,7 +249,7 @@
#endif
plex->lockwaits++; /* waited one more time */
tsleep((void *) lock->stripe, PRIBIO, "vrlock", 2 * hz);
- lock = plex->lock; /* start again */
+ lock = plex->lock - 1; /* start again */
foundlocks = 0;
pos = NULL;
}
@@ -288,6 +288,12 @@
unlockrange(int plexno, struct rangelock *lock)
{
daddr_t lockaddr;
+ struct plex *plex;
+
+ plex = &PLEX[plexno];
+ if (lock < &plex->lock[0] || lock >= &plex->lock[plex->alloclocks])
+ panic("vinum: rangelock %p on plex %d invalid, not between %p and %p",
+ lock, plexno, &plex->lock[0], &plex->lock[plex->alloclocks]);
#ifdef VINUMDEBUG
if (debug & DEBUG_LASTREQS)
diff -ru /tmp/vinum/vinumvar.h ./vinumvar.h
--- /tmp/vinum/vinumvar.h Mon May 22 18:21:37 2000
+++ ./vinumvar.h Thu Jan 4 19:12:01 2001
@@ -158,15 +158,15 @@
* probably too small.
*/
- INITIAL_DRIVES = 4,
+ INITIAL_DRIVES = 16,
INITIAL_VOLUMES = 4,
INITIAL_PLEXES = 8,
INITIAL_SUBDISKS = 16,
- INITIAL_SUBDISKS_IN_PLEX = 4, /* number of subdisks to allocate to a plex */
+ INITIAL_SUBDISKS_IN_PLEX = 16, /* number of subdisks to allocate to a plex */
INITIAL_SUBDISKS_IN_DRIVE = 4, /* number of subdisks to allocate to a drive */
INITIAL_DRIVE_FREELIST = 16, /* number of entries in drive freelist */
PLEX_REGION_TABLE_SIZE = 8, /* number of entries in plex region tables */
- INITIAL_LOCKS = 256, /* number of locks to allocate to a plex */
+ INITIAL_LOCKS = 4096, /* number of locks to allocate to a plex */
MAX_REVIVE_BLOCKSIZE = MAXPHYS, /* maximum revive block size */
DEFAULT_REVIVE_BLOCKSIZE = 65536, /* default revive block size */
VINUMHOSTNAMELEN = 32, /* host name field in label */
- Tor Egge
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200101051644.RAA17887>