svcrdma: Change DMA mapping logic to avoid the page_address kernel API
authorTom Tucker <tom@ogc.us>
Tue, 12 Oct 2010 20:33:52 +0000 (15:33 -0500)
committerJ. Bruce Fields <bfields@redhat.com>
Mon, 18 Oct 2010 23:51:31 +0000 (19:51 -0400)
There was logic in the send path that assumed that a page containing data
to send to the client has a KVA. This is not always the case and can result
in data corruption when page_address returns zero and we end up DMA mapping
zero.

This patch changes the bus mapping logic to avoid page_address() where
necessary and converts all calls from ib_dma_map_single to ib_dma_map_page
in order to keep the map/unmap calls symmetric.

Signed-off-by: Tom Tucker <tom@ogc.us>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
net/sunrpc/xprtrdma/svc_rdma_sendto.c
net/sunrpc/xprtrdma/svc_rdma_transport.c

index 0194de8..926bdb4 100644 (file)
@@ -263,9 +263,9 @@ static int fast_reg_read_chunks(struct svcxprt_rdma *xprt,
        frmr->page_list_len = PAGE_ALIGN(byte_count) >> PAGE_SHIFT;
        for (page_no = 0; page_no < frmr->page_list_len; page_no++) {
                frmr->page_list->page_list[page_no] =
-                       ib_dma_map_single(xprt->sc_cm_id->device,
-                                         page_address(rqstp->rq_arg.pages[page_no]),
-                                         PAGE_SIZE, DMA_FROM_DEVICE);
+                       ib_dma_map_page(xprt->sc_cm_id->device,
+                                       rqstp->rq_arg.pages[page_no], 0,
+                                       PAGE_SIZE, DMA_FROM_DEVICE);
                if (ib_dma_mapping_error(xprt->sc_cm_id->device,
                                         frmr->page_list->page_list[page_no]))
                        goto fatal_err;
@@ -309,17 +309,21 @@ static int rdma_set_ctxt_sge(struct svcxprt_rdma *xprt,
                             int count)
 {
        int i;
+       unsigned long off;
 
        ctxt->count = count;
        ctxt->direction = DMA_FROM_DEVICE;
        for (i = 0; i < count; i++) {
                ctxt->sge[i].length = 0; /* in case map fails */
                if (!frmr) {
+                       BUG_ON(0 == virt_to_page(vec[i].iov_base));
+                       off = (unsigned long)vec[i].iov_base & ~PAGE_MASK;
                        ctxt->sge[i].addr =
-                               ib_dma_map_single(xprt->sc_cm_id->device,
-                                                 vec[i].iov_base,
-                                                 vec[i].iov_len,
-                                                 DMA_FROM_DEVICE);
+                               ib_dma_map_page(xprt->sc_cm_id->device,
+                                               virt_to_page(vec[i].iov_base),
+                                               off,
+                                               vec[i].iov_len,
+                                               DMA_FROM_DEVICE);
                        if (ib_dma_mapping_error(xprt->sc_cm_id->device,
                                                 ctxt->sge[i].addr))
                                return -EINVAL;
index b15e1eb..d4f5e0e 100644 (file)
@@ -70,8 +70,8 @@
  * on extra page for the RPCRMDA header.
  */
 static int fast_reg_xdr(struct svcxprt_rdma *xprt,
-                struct xdr_buf *xdr,
-                struct svc_rdma_req_map *vec)
+                       struct xdr_buf *xdr,
+                       struct svc_rdma_req_map *vec)
 {
        int sge_no;
        u32 sge_bytes;
@@ -96,21 +96,25 @@ static int fast_reg_xdr(struct svcxprt_rdma *xprt,
        vec->count = 2;
        sge_no++;
 
-       /* Build the FRMR */
+       /* Map the XDR head */
        frmr->kva = frva;
        frmr->direction = DMA_TO_DEVICE;
        frmr->access_flags = 0;
        frmr->map_len = PAGE_SIZE;
        frmr->page_list_len = 1;
+       page_off = (unsigned long)xdr->head[0].iov_base & ~PAGE_MASK;
        frmr->page_list->page_list[page_no] =
-               ib_dma_map_single(xprt->sc_cm_id->device,
-                                 (void *)xdr->head[0].iov_base,
-                                 PAGE_SIZE, DMA_TO_DEVICE);
+               ib_dma_map_page(xprt->sc_cm_id->device,
+                               virt_to_page(xdr->head[0].iov_base),
+                               page_off,
+                               PAGE_SIZE - page_off,
+                               DMA_TO_DEVICE);
        if (ib_dma_mapping_error(xprt->sc_cm_id->device,
                                 frmr->page_list->page_list[page_no]))
                goto fatal_err;
        atomic_inc(&xprt->sc_dma_used);
 
+       /* Map the XDR page list */
        page_off = xdr->page_base;
        page_bytes = xdr->page_len + page_off;
        if (!page_bytes)
@@ -128,9 +132,9 @@ static int fast_reg_xdr(struct svcxprt_rdma *xprt,
                page_bytes -= sge_bytes;
 
                frmr->page_list->page_list[page_no] =
-                       ib_dma_map_single(xprt->sc_cm_id->device,
-                                         page_address(page),
-                                         PAGE_SIZE, DMA_TO_DEVICE);
+                       ib_dma_map_page(xprt->sc_cm_id->device,
+                                       page, page_off,
+                                       sge_bytes, DMA_TO_DEVICE);
                if (ib_dma_mapping_error(xprt->sc_cm_id->device,
                                         frmr->page_list->page_list[page_no]))
                        goto fatal_err;
@@ -166,8 +170,10 @@ static int fast_reg_xdr(struct svcxprt_rdma *xprt,
                vec->sge[sge_no].iov_base = frva + frmr->map_len + page_off;
 
                frmr->page_list->page_list[page_no] =
-                       ib_dma_map_single(xprt->sc_cm_id->device, va, PAGE_SIZE,
-                                         DMA_TO_DEVICE);
+                   ib_dma_map_page(xprt->sc_cm_id->device, virt_to_page(va),
+                                   page_off,
+                                   PAGE_SIZE,
+                                   DMA_TO_DEVICE);
                if (ib_dma_mapping_error(xprt->sc_cm_id->device,
                                         frmr->page_list->page_list[page_no]))
                        goto fatal_err;
@@ -245,6 +251,35 @@ static int map_xdr(struct svcxprt_rdma *xprt,
        return 0;
 }
 
+static dma_addr_t dma_map_xdr(struct svcxprt_rdma *xprt,
+                             struct xdr_buf *xdr,
+                             u32 xdr_off, size_t len, int dir)
+{
+       struct page *page;
+       dma_addr_t dma_addr;
+       if (xdr_off < xdr->head[0].iov_len) {
+               /* This offset is in the head */
+               xdr_off += (unsigned long)xdr->head[0].iov_base & ~PAGE_MASK;
+               page = virt_to_page(xdr->head[0].iov_base);
+       } else {
+               xdr_off -= xdr->head[0].iov_len;
+               if (xdr_off < xdr->page_len) {
+                       /* This offset is in the page list */
+                       page = xdr->pages[xdr_off >> PAGE_SHIFT];
+                       xdr_off &= ~PAGE_MASK;
+               } else {
+                       /* This offset is in the tail */
+                       xdr_off -= xdr->page_len;
+                       xdr_off += (unsigned long)
+                               xdr->tail[0].iov_base & ~PAGE_MASK;
+                       page = virt_to_page(xdr->tail[0].iov_base);
+               }
+       }
+       dma_addr = ib_dma_map_page(xprt->sc_cm_id->device, page, xdr_off,
+                                  min_t(size_t, PAGE_SIZE, len), dir);
+       return dma_addr;
+}
+
 /* Assumptions:
  * - We are using FRMR
  *     - or -
@@ -293,10 +328,9 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
                sge[sge_no].length = sge_bytes;
                if (!vec->frmr) {
                        sge[sge_no].addr =
-                               ib_dma_map_single(xprt->sc_cm_id->device,
-                                                 (void *)
-                                                 vec->sge[xdr_sge_no].iov_base + sge_off,
-                                                 sge_bytes, DMA_TO_DEVICE);
+                               dma_map_xdr(xprt, &rqstp->rq_res, xdr_off,
+                                           sge_bytes, DMA_TO_DEVICE);
+                       xdr_off += sge_bytes;
                        if (ib_dma_mapping_error(xprt->sc_cm_id->device,
                                                 sge[sge_no].addr))
                                goto err;
@@ -494,7 +528,8 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
  * In all three cases, this function prepares the RPCRDMA header in
  * sge[0], the 'type' parameter indicates the type to place in the
  * RPCRDMA header, and the 'byte_count' field indicates how much of
- * the XDR to include in this RDMA_SEND.
+ * the XDR to include in this RDMA_SEND. NB: The offset of the payload
+ * to send is zero in the XDR.
  */
 static int send_reply(struct svcxprt_rdma *rdma,
                      struct svc_rqst *rqstp,
@@ -536,23 +571,24 @@ static int send_reply(struct svcxprt_rdma *rdma,
        ctxt->sge[0].lkey = rdma->sc_dma_lkey;
        ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp);
        ctxt->sge[0].addr =
-               ib_dma_map_single(rdma->sc_cm_id->device, page_address(page),
-                                 ctxt->sge[0].length, DMA_TO_DEVICE);
+           ib_dma_map_page(rdma->sc_cm_id->device, page, 0,
+                           ctxt->sge[0].length, DMA_TO_DEVICE);
        if (ib_dma_mapping_error(rdma->sc_cm_id->device, ctxt->sge[0].addr))
                goto err;
        atomic_inc(&rdma->sc_dma_used);
 
        ctxt->direction = DMA_TO_DEVICE;
 
-       /* Determine how many of our SGE are to be transmitted */
+       /* Map the payload indicated by 'byte_count' */
        for (sge_no = 1; byte_count && sge_no < vec->count; sge_no++) {
+               int xdr_off = 0;
                sge_bytes = min_t(size_t, vec->sge[sge_no].iov_len, byte_count);
                byte_count -= sge_bytes;
                if (!vec->frmr) {
                        ctxt->sge[sge_no].addr =
-                               ib_dma_map_single(rdma->sc_cm_id->device,
-                                                 vec->sge[sge_no].iov_base,
-                                                 sge_bytes, DMA_TO_DEVICE);
+                               dma_map_xdr(rdma, &rqstp->rq_res, xdr_off,
+                                           sge_bytes, DMA_TO_DEVICE);
+                       xdr_off += sge_bytes;
                        if (ib_dma_mapping_error(rdma->sc_cm_id->device,
                                                 ctxt->sge[sge_no].addr))
                                goto err;
index 950a206..e87e000 100644 (file)
@@ -121,7 +121,7 @@ void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
                 */
                if (ctxt->sge[i].lkey == xprt->sc_dma_lkey) {
                        atomic_dec(&xprt->sc_dma_used);
-                       ib_dma_unmap_single(xprt->sc_cm_id->device,
+                       ib_dma_unmap_page(xprt->sc_cm_id->device,
                                            ctxt->sge[i].addr,
                                            ctxt->sge[i].length,
                                            ctxt->direction);
@@ -503,8 +503,8 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt)
                BUG_ON(sge_no >= xprt->sc_max_sge);
                page = svc_rdma_get_page();
                ctxt->pages[sge_no] = page;
-               pa = ib_dma_map_single(xprt->sc_cm_id->device,
-                                    page_address(page), PAGE_SIZE,
+               pa = ib_dma_map_page(xprt->sc_cm_id->device,
+                                    page, 0, PAGE_SIZE,
                                     DMA_FROM_DEVICE);
                if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa))
                        goto err_put_ctxt;
@@ -800,8 +800,8 @@ static void frmr_unmap_dma(struct svcxprt_rdma *xprt,
                if (ib_dma_mapping_error(frmr->mr->device, addr))
                        continue;
                atomic_dec(&xprt->sc_dma_used);
-               ib_dma_unmap_single(frmr->mr->device, addr, PAGE_SIZE,
-                                   frmr->direction);
+               ib_dma_unmap_page(frmr->mr->device, addr, PAGE_SIZE,
+                                 frmr->direction);
        }
 }
 
@@ -1276,7 +1276,7 @@ int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr)
                                   atomic_read(&xprt->sc_sq_count) <
                                   xprt->sc_sq_depth);
                        if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
-                               return 0;
+                               return -ENOTCONN;
                        continue;
                }
                /* Take a transport ref for each WR posted */
@@ -1322,8 +1322,8 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
        length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va);
 
        /* Prepare SGE for local address */
-       sge.addr = ib_dma_map_single(xprt->sc_cm_id->device,
-                                  page_address(p), PAGE_SIZE, DMA_FROM_DEVICE);
+       sge.addr = ib_dma_map_page(xprt->sc_cm_id->device,
+                                  p, 0, PAGE_SIZE, DMA_FROM_DEVICE);
        if (ib_dma_mapping_error(xprt->sc_cm_id->device, sge.addr)) {
                put_page(p);
                return;
@@ -1350,7 +1350,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
        if (ret) {
                dprintk("svcrdma: Error %d posting send for protocol error\n",
                        ret);
-               ib_dma_unmap_single(xprt->sc_cm_id->device,
+               ib_dma_unmap_page(xprt->sc_cm_id->device,
                                  sge.addr, PAGE_SIZE,
                                  DMA_FROM_DEVICE);
                svc_rdma_put_context(ctxt, 1);