* nfs NLM locks greater than 8 bytes
@ 2003-11-04 0:52 Bill Akers
2003-11-05 4:19 ` Trond Myklebust
0 siblings, 1 reply; 2+ messages in thread
From: Bill Akers @ 2003-11-04 0:52 UTC (permalink / raw)
To: linux-kernel
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Has anyone looked at the NFS NLM byte lock size issue between the Linux kernel
based NFS server and BSD 5 clients? BSD 5 uses a large cookie for locking
(16 or even 20 bytes). Linux automatically rejects any lock attempts with a
cookie over 8 bytes. This issue has become important for us because of
Apple's release of OS X 10.3 (Panther), which is based on BSD 5.
Jonathan Lennox summed it up pretty well on the Freebsd bug list:
> Linux's implementation of NFS NLM locks is buggy: it doesn't support lock
> cookies longer than 8 bytes in size. See the comment in
> <http://lxr.linux.no/source/include/linux/lockd/xdr.h?v=2.6.0-test2> on the
> definition of 'struct nlm_cookie': "NLM cookies. Technically they can be 1K,
> Nobody uses over 8 bytes however."
>
> Unfortunately, this is actually "nobody" except FreeBSD 5.x, which uses
> 16-byte cookies. As a result, any attempt by a FreeBSD client to lock an
> NFS-mounted file from a Linux server results in the process on the FreeBSD
> client hanging, unkillably.
His full original post is:
http://lists.freebsd.org/pipermail/freebsd-bugs/2003-September/003069.html
I have checked out every recent version of the kernel on kernel.org (2.4 and
2.6) and haven't seen any modifications to xdr.h or xdr.c that might address
this issue. Nor have I been able to find any posts on Linux lists that
discuss the issue. There is a BSD patch to dumb down their client code and
make the cookie size smaller. The fear being that it will take a long time
to get a linux patch into the kernel. But if no one has even brought up the
issue here, it will certainly take forever. :-)
I can probably dust off enough of my old coding skills to up the check from 8
bytes to some higher number (16, 20, 1024 bytes?). But I certainly don't
know enough about the Linux kernel to know what ramifications that will have.
Thanks,
- --
Bill Akers
Director of Network Operations
UCLA Graduate School of Education and Information Studies
akers@gseis.ucla.edu
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.7 (GNU/Linux)
iD8DBQE/pvhUHuBhMoY9pyIRAn5iAKCpWedBMVY4c41vz1/dNicIx1WJNACcClN2
jf6yR5yyOxTD0i1kx91Tyiw=
=+5DB
-----END PGP SIGNATURE-----
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: nfs NLM locks greater than 8 bytes
2003-11-04 0:52 nfs NLM locks greater than 8 bytes Bill Akers
@ 2003-11-05 4:19 ` Trond Myklebust
0 siblings, 0 replies; 2+ messages in thread
From: Trond Myklebust @ 2003-11-05 4:19 UTC (permalink / raw)
To: Bill Akers; +Cc: linux-kernel
>>>>> " " == Bill Akers <akers@gseis.ucla.edu> writes:
>> Unfortunately, this is actually "nobody" except FreeBSD 5.x,
>> which uses 16-byte cookies. As a result, any attempt by a
>> FreeBSD client to lock an NFS-mounted file from a Linux server
>> results in the process on the FreeBSD client hanging,
>> unkillably.
> His full original post is:
> http://lists.freebsd.org/pipermail/freebsd-bugs/2003-September/003069.html
"A proper fix would be either to somehow compress all three
pieces of information -- pid, pid_start, and msg_seq -- into
eight bytes (difficult); maintain an in-kernel table mapping an
eight-byte sequence number to lockd_msg_ident; or find some
other, smaller way of defending against pid recycling."
That's a laugh... They need 16-20 bytes to defend against pid
recycling??? For exactly how many lifetimes of the universe are they
planning on waiting on a lock?
Anyhow, Greg Banks of SGI sent me this patch in March last year. I
haven't been too crazy about it because the debugging code gets a bit
obscurantist.
It will solve your problem though...
Cheers,
Trond
-------------------------
This patch makes lockd handle NLM cookies larger than 8 bytes, which
is needed for MacOS X interoperability.
diff -ruN --exclude-from=excludes linux-2.4.19.orig/fs/lockd/svclock.c
linux-2.4.19/fs/lockd/svclock.c
--- linux-2.4.19.orig/fs/lockd/svclock.c Tue Mar 4 19:26:53 2003
+++ linux-2.4.19/fs/lockd/svclock.c Tue Mar 4 20:32:10 2003
@@ -106,6 +106,7 @@
{
struct nlm_block **head, *block;
struct file_lock *fl;
+ NLMDBG_COOKIE_BUF
dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
file, lock->fl.fl_pid,
@@ -113,11 +114,11 @@
(long long)lock->fl.fl_end, lock->fl.fl_type);
for (head = &nlm_blocked; (block = *head); head = &block->b_next) {
fl = &block->b_call.a_args.lock.fl;
- dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%x\n",
+ dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
block->b_file, fl->fl_pid,
(long long)fl->fl_start,
(long long)fl->fl_end, fl->fl_type,
- *(unsigned int*)(block->b_call.a_args.cookie.data));
+ nlmdbg_cookie2a(&block->b_call.a_args.cookie));
if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
if (remove) {
*head = block->b_next;
@@ -572,12 +573,13 @@
struct nlm_rqst *call = (struct nlm_rqst *) task->tk_calldata;
struct nlm_block *block;
unsigned long timeout;
+ NLMDBG_COOKIE_BUF
dprintk("lockd: GRANT_MSG RPC callback\n");
- dprintk("callback: looking for cookie %x \n",
- *(unsigned int *)(call->a_args.cookie.data));
+ dprintk("callback: looking for cookie %s \n",
+ nlmdbg_cookie2a(&call->a_args.cookie));
if (!(block = nlmsvc_find_block(&call->a_args.cookie))) {
- dprintk("lockd: no block for cookie %x\n", *(u32 *)(call->a_args.cookie.data));
+ dprintk("lockd: no block for cookie %s\n", nlmdbg_cookie2a(&call->a_args.cookie));
return;
}
diff -ruN --exclude-from=excludes linux-2.4.19.orig/fs/lockd/xdr.c
linux-2.4.19/fs/lockd/xdr.c
--- linux-2.4.19.orig/fs/lockd/xdr.c Tue Mar 4 19:26:53 2003
+++ linux-2.4.19/fs/lockd/xdr.c Tue Mar 4 21:11:12 2003
@@ -55,7 +55,7 @@
c->len=4;
memset(c->data, 0, 4); /* hockeypux brain damage */
}
- else if(len<=8)
+ else if(len<=NLM_MAXCOOKIELEN)
{
c->len=len;
memcpy(c->data, p, len);
@@ -64,7 +64,7 @@
else
{
printk(KERN_NOTICE
- "lockd: bad cookie size %d (only cookies under 8 bytes are supported.)\n", len);
+ "lockd: bad cookie size %d (only cookies under %d bytes are supported.)\n", len,
NLM_MAXCOOKIELEN);
return NULL;
}
return p;
@@ -86,7 +86,7 @@
if ((len = ntohl(*p++)) != NFS2_FHSIZE) {
printk(KERN_NOTICE
- "lockd: bad fhandle size %x (should be %d)\n",
+ "lockd: bad fhandle size %d (should be %d)\n",
len, NFS2_FHSIZE);
return NULL;
}
@@ -534,7 +534,7 @@
* Buffer requirements for NLM
*/
#define NLM_void_sz 0
-#define NLM_cookie_sz 3 /* 1 len , 2 data */
+#define NLM_cookie_sz 1+QUADLEN(NLM_MAXCOOKIELEN)
#define NLM_caller_sz 1+QUADLEN(sizeof(system_utsname.nodename))
#define NLM_netobj_sz 1+QUADLEN(XDR_MAX_NETOBJ)
/* #define NLM_owner_sz 1+QUADLEN(NLM_MAXOWNER) */
@@ -644,3 +644,23 @@
}
#endif
+#ifdef RPC_DEBUG
+const char *__nlmdbg_cookie2a(const struct nlm_cookie *cookie, char *buf, int len)
+{
+ int i;
+
+ len--; /* allow for trailing \0 */
+ if (len < 3)
+ return "???";
+ for (i = 0 ; i < cookie->len ; i++) {
+ if (len < 2) {
+ strcpy(buf-3, "...");
+ break;
+ }
+ sprintf(buf, "%02x", cookie->data[i]);
+ buf += 2;
+ len -= 2;
+ }
+ *buf = '\0';
+}
+#endif
diff -ruN --exclude-from=excludes linux-2.4.19.orig/fs/lockd/xdr4.c
linux-2.4.19/fs/lockd/xdr4.c
--- linux-2.4.19.orig/fs/lockd/xdr4.c Tue Mar 4 19:26:53 2003
+++ linux-2.4.19/fs/lockd/xdr4.c Tue Mar 4 20:37:59 2003
@@ -56,7 +56,7 @@
c->len=4;
memset(c->data, 0, 4); /* hockeypux brain damage */
}
- else if(len<=8)
+ else if(len<=NLM_MAXCOOKIELEN)
{
c->len=len;
memcpy(c->data, p, len);
@@ -65,7 +65,7 @@
else
{
printk(KERN_NOTICE
- "lockd: bad cookie size %d (only cookies under 8 bytes are supported.)\n", len);
+ "lockd: bad cookie size %d (only cookies under %d bytes are supported.)\n", len,
NLM_MAXCOOKIELEN);
return NULL;
}
return p;
@@ -540,7 +540,7 @@
* Buffer requirements for NLM
*/
#define NLM4_void_sz 0
-#define NLM4_cookie_sz 3 /* 1 len , 2 data */
+#define NLM4_cookie_sz 1+XDR_QUADLEN(NLM_MAXCOOKIELEN)
#define NLM4_caller_sz 1+XDR_QUADLEN(NLM_MAXSTRLEN)
#define NLM4_netobj_sz 1+XDR_QUADLEN(XDR_MAX_NETOBJ)
/* #define NLM4_owner_sz 1+XDR_QUADLEN(NLM4_MAXOWNER) */
diff -ruN --exclude-from=excludes linux-2.4.19.orig/include/linux/lockd/debug.h
linux-2.4.19/include/linux/lockd/debug.h
--- linux-2.4.19.orig/include/linux/lockd/debug.h Tue Mar 4 19:26:40 2003
+++ linux-2.4.19/include/linux/lockd/debug.h Tue Mar 4 20:32:10 2003
@@ -49,4 +49,17 @@
#define NLMDBG_ALL 0x7fff
+/*
+ * Support for printing NLM cookies in dprintk()
+ */
+#ifdef RPC_DEBUG
+struct nlm_cookie;
+extern const char *__nlmdbg_cookie2a(const struct nlm_cookie *, char *, int);
+#define NLMDBG_COOKIE_BUF char __nlmdbg_cookie_buf[2*NLM_MAXCOOKIELEN+1];
+#define nlmdbg_cookie2a(c) \
+ __nlmdbg_cookie2a(__nlmdbg_cookie_buf, sizeof(__nlmdbg_cookie_buf), (c))
+#else
+#define NLMDBG_COOKIE_BUF
+#endif
+
#endif /* LINUX_LOCKD_DEBUG_H */
diff -ruN --exclude-from=excludes linux-2.4.19.orig/include/linux/lockd/xdr.h
linux-2.4.19/include/linux/lockd/xdr.h
--- linux-2.4.19.orig/include/linux/lockd/xdr.h Tue Mar 4 19:26:40 2003
+++ linux-2.4.19/include/linux/lockd/xdr.h Tue Mar 4 20:32:10 2003
@@ -33,14 +33,15 @@
};
/*
- * NLM cookies. Technically they can be 1K, Nobody uses over 8 bytes
- * however.
+ * NLM cookies. Technically they can be 1K, Few people use over 8 bytes,
+ * FreeBSD uses 16, Apple Mac OS-X 10.3 uses 20.
*/
struct nlm_cookie
{
- unsigned char data[8];
unsigned int len;
+#define NLM_MAXCOOKIELEN 32
+ unsigned char data[NLM_MAXCOOKIELEN];
};
/*
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2003-11-05 4:20 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2003-11-04 0:52 nfs NLM locks greater than 8 bytes Bill Akers
2003-11-05 4:19 ` Trond Myklebust
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.