[CRYPTO] api: scatterwalk_copychunks() fails to advance through scatterlist
authorJ. Bruce Fields <bfields@citi.umich.edu>
Tue, 20 Mar 2007 21:50:12 +0000 (08:50 +1100)
committerHerbert Xu <herbert@gondor.apana.org.au>
Tue, 20 Mar 2007 21:50:12 +0000 (08:50 +1100)
In the loop in scatterwalk_copychunks(), if walk->offset is zero,
then scatterwalk_pagedone rounds that up to the nearest page boundary:

walk->offset += PAGE_SIZE - 1;
walk->offset &= PAGE_MASK;

which is a no-op in this case, so we don't advance to the next element
of the scatterlist array:

if (walk->offset >= walk->sg->offset + walk->sg->length)
scatterwalk_start(walk, sg_next(walk->sg));

and we end up copying the same data twice.

It appears that other callers of scatterwalk_{page}done first advance
walk->offset, so I believe that's the correct thing to do here.

This caused a bug in NFS when run with krb5p security, which would
cause some writes to fail with permissions errors--for example, writes
of less than 8 bytes (the des blocksize) at the start of a file.

A git-bisect shows the bug was originally introduced by
5c64097aa0f6dc4f27718ef47ca9a12538d62860, first in 2.6.19-rc1.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
crypto/scatterwalk.c

index 35172d3f043b47f02ef5c8e20a0ffaf7976bac5b..a66423121773a8f2419e359b3812dd641640fdd3 100644 (file)
@@ -91,6 +91,8 @@ void scatterwalk_copychunks(void *buf, struct scatter_walk *walk,
                memcpy_dir(buf, vaddr, len_this_page, out);
                scatterwalk_unmap(vaddr, out);
 
+               scatterwalk_advance(walk, nbytes);
+
                if (nbytes == len_this_page)
                        break;
 
@@ -99,7 +101,5 @@ void scatterwalk_copychunks(void *buf, struct scatter_walk *walk,
 
                scatterwalk_pagedone(walk, out, 1);
        }
-
-       scatterwalk_advance(walk, nbytes);
 }
 EXPORT_SYMBOL_GPL(scatterwalk_copychunks);