diff options
| author |   gluk <gluk@openbsd.org> | 2001-05-28 00:23:02 +0000 |
|---|---|---|
| committer | gluk <gluk@openbsd.org> | 2001-05-28 00:23:02 +0000 |
| commit | c4b5909bbca9b1333a9e8f3cf4024949d178ba15 (patch) | |
| tree | d4b84a82f5dedd933de41bef5d87dd85f88a3c23 /sys/kern/vfs_cluster.c | |
| parent | Fix "(seeUVM(9))" (diff) | |
| download | wireguard-openbsd-c4b5909bbca9b1333a9e8f3cf4024949d178ba15.tar.xz wireguard-openbsd-c4b5909bbca9b1333a9e8f3cf4024949d178ba15.zip | |
cluster_rbuild() have a race between incore and getblk. incore() returns
zero indicating that buffer is not in a cache, but getblk() going to sleep:
getblk->getnewbuf->tsleep. When getnewbuf() returns after a sleep, getblk()
may find B_DONE buffer in hash and return it. When io operation finishes
biodone() calls cluster_callback() which moves pages from one big cluster
buffer into several component buffers and calls biodone() for every component
buffer. Since there are a component buffer with B_DONE already set, biodone()
panices: "biodone already".
costa@ ok.
Diffstat (limited to 'sys/kern/vfs_cluster.c')
| -rw-r--r-- | sys/kern/vfs_cluster.c | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/sys/kern/vfs_cluster.c b/sys/kern/vfs_cluster.c index a0dbb42b9a7..bd1d70d8b05 100644 --- a/sys/kern/vfs_cluster.c +++ b/sys/kern/vfs_cluster.c @@ -1,4 +1,4 @@ -/* $OpenBSD: vfs_cluster.c,v 1.23 2001/05/20 22:18:10 gluk Exp $ */ +/* $OpenBSD: vfs_cluster.c,v 1.24 2001/05/28 00:23:02 gluk Exp $ */ /* $NetBSD: vfs_cluster.c,v 1.12 1996/04/22 01:39:05 christos Exp $ */ /*- @@ -353,6 +353,7 @@ cluster_rbuild(vp, filesize, bp, lbn, blkno, size, run, flags) pagemove(bdata, bdata + tbp->b_bufsize, size); } tbp->b_blkno = bn; + tbp->b_flags &= ~(B_DONE | B_ERROR); tbp->b_flags |= flags | B_READ | B_ASYNC; b_save->bs_children[b_save->bs_nchildren++] = tbp; } |