Return to BSD News archive
Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!agate!howland.reston.ans.net!gatech!udel!newsserv.cs.sunysb.edu!stark.UUCP!cs.sunysb.edu!newsserv!stark!gene
From: newsserv!stark!gene@cs.sunysb.edu (Gene Stark)
Newsgroups: comp.os.386bsd.bugs
Subject: "strip -x 386bsd" hangs over NFS (FIX)
Date: 27 Nov 93 17:56:14
Organization: Gene Stark's home system
Lines: 54
Distribution: world
Message-ID: <NEWSSERV!STARK!GENE.93Nov27175614@stark.uucp>
NNTP-Posting-Host: stark.uucp
I sent the following bug report to "FreeBSD-bugfiler@freefall.cdrom.com",
but am posting it here in case anyone wants it fast.
- Gene Stark
------------------------------------------------------
Subject: strip -x 386bsd hangs over NFS
Index: sys/nfs/nfs_bio.c FreeBSD-1.0.2
Description:
Running "strip -x 386bsd" (the last step in building a kernel)
over NFS hangs in uninterruptible sleep.
Repeat-By:
Copy an unstripped kernel to an NFS-mounted directory and
run "strip -x" on it.
Fix:
Apply the patch below to /sys/nfs/nfs_bio.c. The problem is
that vnode_pager_io sets up a uio structure with a null
process pointer, and then nfs_write uses that pointer to
check whether the file size resource limit is exceeded.
If so, nfs_write returns EFBIG, which is ultimately voided
by the call to vm_pager_put in vm_object_page_clean.
The process thus hangs forever waiting for pageout to complete.
The code for nfs_write was evidently cribbed from ufs_write
in ufs_vnops.c. The check for vp->v_type == VREG is actually
already done earlier in nfs_write, but why not be a little
defensive? The problem evidently does not exist in isofs
(no writing), or pcfs (done correctly).
*** /sys/nfs/nfs_bio.c Wed Oct 6 07:06:20 1993
--- nfs_bio.c Wed Nov 24 20:00:56 1993
***************
*** 244,250 ****
* Maybe this should be above the vnode op call, but so long as
* file servers have no limits, i don't think it matters
*/
! if (uio->uio_offset + uio->uio_resid >
p->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
psignal(p, SIGXFSZ);
return (EFBIG);
--- 244,251 ----
* Maybe this should be above the vnode op call, but so long as
* file servers have no limits, i don't think it matters
*/
! if (vp->v_type == VREG && p &&
! uio->uio_offset + uio->uio_resid >
p->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
psignal(p, SIGXFSZ);
return (EFBIG);
--