xprtrdma: Fix a NULL dereference in frwr_unmap_sync()
authorChuck Lever <chuck.lever@oracle.com>
Sat, 1 May 2021 19:38:02 +0000 (15:38 -0400)
committerTrond Myklebust <trond.myklebust@hammerspace.com>
Sat, 1 May 2021 23:42:22 +0000 (19:42 -0400)
commit9e895cd9649abe4392c59d14e31b0f5667d082d2
tree65806dfb3557a48362d1548ab031ff1498345d32
parentf8f7e0fb22b2e75be55f2f0c13e229e75b0eac07
xprtrdma: Fix a NULL dereference in frwr_unmap_sync()

The normal mechanism that invalidates and unmaps MRs is
frwr_unmap_async(). frwr_unmap_sync() is used only when an RPC
Reply bearing Write or Reply chunks has been lost (ie, almost
never).

Coverity found that after commit 9a301cafc861 ("xprtrdma: Move
fr_linv_done field to struct rpcrdma_mr"), the while() loop in
frwr_unmap_sync() exits only once @mr is NULL, unconditionally
causing subsequent dereferences of @mr to Oops.

I've tested this fix by creating a client that skips invoking
frwr_unmap_async() when RPC Replies complete. That forces all
invalidation tasks to fall upon frwr_unmap_sync(). Simple workloads
with this fix applied to the adulterated client work as designed.

Reported-by: coverity-bot <keescook+coverity-bot@chromium.org>
Addresses-Coverity-ID: 1504556 ("Null pointer dereferences")
Fixes: 9a301cafc861 ("xprtrdma: Move fr_linv_done field to struct rpcrdma_mr")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
net/sunrpc/xprtrdma/frwr_ops.c