Skip to contentSkip to conversation

remove xdr_kmap()

sunrpc William Lee Irwin III 3 participants 5 messages Dec 28, 2004 – Dec 29, 2004

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

I was auditing nfs for missing flush_dcache_page() calls. They appear to be done almost nowhere in nfs and so likely nowhere near enough, and that makes nfs one of the primary suspects for breaking various machines I have.

In this process, I stumbled over a blatant kmap() deadlock in xdr_kmap(), which fortunately is never called.

This patch removes these dangerous nowhere-called functions. vs. 2.6.10-bk1

-- wli

include/linux/sunrpc/xdr.h +0 -2
Index: wli-2.6.10-bk1/include/linux/sunrpc/xdr.h
===================================================================
@@ -145,8 +145,6 @@
 /*
  * XDR buffer helper functions
  */
-extern int xdr_kmap(struct kvec *, struct xdr_buf *, size_t);
-extern void xdr_kunmap(struct xdr_buf *, size_t);
 extern void xdr_shift_buf(struct xdr_buf *, size_t);
 extern void _copy_from_pages(char *, struct page **, size_t, size_t);
 extern void xdr_buf_from_iov(struct kvec *, struct xdr_buf *);
net/sunrpc/xdr.c +0 -87

93 lines not on this page · raw message

maintainer
#1 lore raw
Quoted 2 lines · On Tue, 28 Dec 2004 15:04:16 -0800 William Lee Irwin III <[email protected]>…

In this process, I stumbled over a blatant kmap() deadlock in xdr_kmap(), which fortunately is never called.

This got zapped by a cleanup patch by Adrian Bunk which I applied yesterdat. Linus just hasn't pulled from my tree yet.

#2 lore raw
Quoted 2 lines · On Tue, 28 Dec 2004 15:04:16 -0800 William Lee Irwin III <[email protected]>…
Quoted 2 lines

In this process, I stumbled over a blatant kmap() deadlock in xdr_kmap(), which fortunately is never called.

Quoted 3 lines · David S. Miller

This got zapped by a cleanup patch by Adrian Bunk which I applied yesterdat. Linus just hasn't pulled from my tree yet.

Sounds good. I only missed it because it was in the middle of a larger set of changes. Now I just have to find where in nfs the missing flush_dcache_page() calls need to be so I can boot 2.6 on a bunch of boxen.

-- wli

maintainer
#3 lore raw
Quoted 5 lines · On Tue, 28 Dec 2004 18:09:38 -0800 William Lee Irwin III <[email protected]>…
Quoted 2 lines · On Tue, 28 Dec 2004 15:04:16 -0800 William Lee Irwin III <[email protected]>…
Quoted 2 lines

In this process, I stumbled over a blatant kmap() deadlock in xdr_kmap(), which fortunately is never called.

On Tue, Dec 28, 2004 at 05:12:46PM -0800, David S. Miller wrote:

Quoted 8 lines
Quoted 3 lines

This got zapped by a cleanup patch by Adrian Bunk which I applied yesterdat. Linus just hasn't pulled from my tree yet.

Sounds good. I only missed it because it was in the middle of a larger set of changes. Now I just have to find where in nfs the missing flush_dcache_page() calls need to be so I can boot 2.6 on a bunch of boxen.

I remember adding the calls ages ago, wonder what happened.

I see a bunch of memclear_highpage_flush() calls, but flush_dcache_page() calls. I guess these are done at the sunrpc/xdr layer.

There is a flush_dcache_page() call for xdr_partial_copy_from_skb() but calls are also needed in _copy_to_pages() and _shift_data_right_pages(). The rest which access pages are copying from pages, not into them, so those should be ok.

#4 lore raw

ty den 28.12.2004 Klokka 19:49 (-0800) skreiv David S. Miller:

Quoted 3 lines

There is a flush_dcache_page() call for xdr_partial_copy_from_skb() but calls are also needed in _copy_to_pages() and _shift_data_right_pages().

Will do.

Cheers,
Trond

Signature · 1 line
Trond Myklebust <[email protected]>