LXR linux/net/sunrpc/xprtrdma/svc_rdma

   1/*
   2 * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the BSD-type
   8 * license below:
   9 *
  10 * Redistribution and use in source and binary forms, with or without
  11 * modification, are permitted provided that the following conditions
  12 * are met:
  13 *
  14 *      Redistributions of source code must retain the above copyright
  15 *      notice, this list of conditions and the following disclaimer.
  16 *
  17 *      Redistributions in binary form must reproduce the above
  18 *      copyright notice, this list of conditions and the following
  19 *      disclaimer in the documentation and/or other materials provided
  20 *      with the distribution.
  21 *
  22 *      Neither the name of the Network Appliance, Inc. nor the names of
  23 *      its contributors may be used to endorse or promote products
  24 *      derived from this software without specific prior written
  25 *      permission.
  26 *
  27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  30 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  31 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  32 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  33 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  34 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  35 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  36 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  37 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  38 *
  39 * Author: Tom Tucker <tom@opengridcomputing.com>
  40 */
  41
  42#include <linux/sunrpc/xdr.h>
  43#include <linux/sunrpc/debug.h>
  44#include <asm/unaligned.h>
  45#include <linux/sunrpc/rpc_rdma.h>
  46#include <linux/sunrpc/svc_rdma.h>
  47
  48#define RPCDBG_FACILITY RPCDBG_SVCXPRT
  49
  50/*
  51 * Decodes a read chunk list. The expected format is as follows:
  52 *    descrim  : xdr_one
  53 *    position : u32 offset into XDR stream
  54 *    handle   : u32 RKEY
  55 *    . . .
  56 *  end-of-list: xdr_zero
  57 */
  58static u32 *decode_read_list(u32 *va, u32 *vaend)
  59{
  60        struct rpcrdma_read_chunk *ch = (struct rpcrdma_read_chunk *)va;
  61
  62        while (ch->rc_discrim != xdr_zero) {
  63                if (((unsigned long)ch + sizeof(struct rpcrdma_read_chunk)) >
  64                    (unsigned long)vaend) {
  65                        dprintk("svcrdma: vaend=%p, ch=%p\n", vaend, ch);
  66                        return NULL;
  67                }
  68                ch++;
  69        }
  70        return (u32 *)&ch->rc_position;
  71}
  72
  73/*
  74 * Determine number of chunks and total bytes in chunk list. The chunk
  75 * list has already been verified to fit within the RPCRDMA header.
  76 */
  77void svc_rdma_rcl_chunk_counts(struct rpcrdma_read_chunk *ch,
  78                               int *ch_count, int *byte_count)
  79{
  80        /* compute the number of bytes represented by read chunks */
  81        *byte_count = 0;
  82        *ch_count = 0;
  83        for (; ch->rc_discrim != 0; ch++) {
  84                *byte_count = *byte_count + ntohl(ch->rc_target.rs_length);
  85                *ch_count = *ch_count + 1;
  86        }
  87}
  88
  89/*
  90 * Decodes a write chunk list. The expected format is as follows:
  91 *    descrim  : xdr_one
  92 *    nchunks  : <count>
  93 *       handle   : u32 RKEY              ---+
  94 *       length   : u32 <len of segment>     |
  95 *       offset   : remove va                + <count>
  96 *       . . .                               |
  97 *                                        ---+
  98 */
  99static u32 *decode_write_list(u32 *va, u32 *vaend)
 100{
 101        int nchunks;
 102
 103        struct rpcrdma_write_array *ary =
 104                (struct rpcrdma_write_array *)va;
 105
 106        /* Check for not write-array */
 107        if (ary->wc_discrim == xdr_zero)
 108                return (u32 *)&ary->wc_nchunks;
 109
 110        if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) >
 111            (unsigned long)vaend) {
 112                dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend);
 113                return NULL;
 114        }
 115        nchunks = ntohl(ary->wc_nchunks);
 116        if (((unsigned long)&ary->wc_array[0] +
 117             (sizeof(struct rpcrdma_write_chunk) * nchunks)) >
 118            (unsigned long)vaend) {
 119                dprintk("svcrdma: ary=%p, wc_nchunks=%d, vaend=%p\n",
 120                        ary, nchunks, vaend);
 121                return NULL;
 122        }
 123        /*
 124         * rs_length is the 2nd 4B field in wc_target and taking its
 125         * address skips the list terminator
 126         */
 127        return (u32 *)&ary->wc_array[nchunks].wc_target.rs_length;
 128}
 129
 130static u32 *decode_reply_array(u32 *va, u32 *vaend)
 131{
 132        int nchunks;
 133        struct rpcrdma_write_array *ary =
 134                (struct rpcrdma_write_array *)va;
 135
 136        /* Check for no reply-array */
 137        if (ary->wc_discrim == xdr_zero)
 138                return (u32 *)&ary->wc_nchunks;
 139
 140        if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) >
 141            (unsigned long)vaend) {
 142                dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend);
 143                return NULL;
 144        }
 145        nchunks = ntohl(ary->wc_nchunks);
 146        if (((unsigned long)&ary->wc_array[0] +
 147             (sizeof(struct rpcrdma_write_chunk) * nchunks)) >
 148            (unsigned long)vaend) {
 149                dprintk("svcrdma: ary=%p, wc_nchunks=%d, vaend=%p\n",
 150                        ary, nchunks, vaend);
 151                return NULL;
 152        }
 153        return (u32 *)&ary->wc_array[nchunks];
 154}
 155
 156int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req,
 157                            struct svc_rqst *rqstp)
 158{
 159        struct rpcrdma_msg *rmsgp = NULL;
 160        u32 *va;
 161        u32 *vaend;
 162        u32 hdr_len;
 163
 164        rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
 165
 166        /* Verify that there's enough bytes for header + something */
 167        if (rqstp->rq_arg.len <= RPCRDMA_HDRLEN_MIN) {
 168                dprintk("svcrdma: header too short = %d\n",
 169                        rqstp->rq_arg.len);
 170                return -EINVAL;
 171        }
 172
 173        /* Decode the header */
 174        rmsgp->rm_xid = ntohl(rmsgp->rm_xid);
 175        rmsgp->rm_vers = ntohl(rmsgp->rm_vers);
 176        rmsgp->rm_credit = ntohl(rmsgp->rm_credit);
 177        rmsgp->rm_type = ntohl(rmsgp->rm_type);
 178
 179        if (rmsgp->rm_vers != RPCRDMA_VERSION)
 180                return -ENOSYS;
 181
 182        /* Pull in the extra for the padded case and bump our pointer */
 183        if (rmsgp->rm_type == RDMA_MSGP) {
 184                int hdrlen;
 185                rmsgp->rm_body.rm_padded.rm_align =
 186                        ntohl(rmsgp->rm_body.rm_padded.rm_align);
 187                rmsgp->rm_body.rm_padded.rm_thresh =
 188                        ntohl(rmsgp->rm_body.rm_padded.rm_thresh);
 189
 190                va = &rmsgp->rm_body.rm_padded.rm_pempty[4];
 191                rqstp->rq_arg.head[0].iov_base = va;
 192                hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp);
 193                rqstp->rq_arg.head[0].iov_len -= hdrlen;
 194                if (hdrlen > rqstp->rq_arg.len)
 195                        return -EINVAL;
 196                return hdrlen;
 197        }
 198
 199        /* The chunk list may contain either a read chunk list or a write
 200         * chunk list and a reply chunk list.
 201         */
 202        va = &rmsgp->rm_body.rm_chunks[0];
 203        vaend = (u32 *)((unsigned long)rmsgp + rqstp->rq_arg.len);
 204        va = decode_read_list(va, vaend);
 205        if (!va)
 206                return -EINVAL;
 207        va = decode_write_list(va, vaend);
 208        if (!va)
 209                return -EINVAL;
 210        va = decode_reply_array(va, vaend);
 211        if (!va)
 212                return -EINVAL;
 213
 214        rqstp->rq_arg.head[0].iov_base = va;
 215        hdr_len = (unsigned long)va - (unsigned long)rmsgp;
 216        rqstp->rq_arg.head[0].iov_len -= hdr_len;
 217
 218        *rdma_req = rmsgp;
 219        return hdr_len;
 220}
 221
 222int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *rqstp)
 223{
 224        struct rpcrdma_msg *rmsgp = NULL;
 225        struct rpcrdma_read_chunk *ch;
 226        struct rpcrdma_write_array *ary;
 227        u32 *va;
 228        u32 hdrlen;
 229
 230        dprintk("svcrdma: processing deferred RDMA header on rqstp=%p\n",
 231                rqstp);
 232        rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
 233
 234        /* Pull in the extra for the padded case and bump our pointer */
 235        if (rmsgp->rm_type == RDMA_MSGP) {
 236                va = &rmsgp->rm_body.rm_padded.rm_pempty[4];
 237                rqstp->rq_arg.head[0].iov_base = va;
 238                hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp);
 239                rqstp->rq_arg.head[0].iov_len -= hdrlen;
 240                return hdrlen;
 241        }
 242
 243        /*
 244         * Skip all chunks to find RPC msg. These were previously processed
 245         */
 246        va = &rmsgp->rm_body.rm_chunks[0];
 247
 248        /* Skip read-list */
 249        for (ch = (struct rpcrdma_read_chunk *)va;
 250             ch->rc_discrim != xdr_zero; ch++);
 251        va = (u32 *)&ch->rc_position;
 252
 253        /* Skip write-list */
 254        ary = (struct rpcrdma_write_array *)va;
 255        if (ary->wc_discrim == xdr_zero)
 256                va = (u32 *)&ary->wc_nchunks;
 257        else
 258                /*
 259                 * rs_length is the 2nd 4B field in wc_target and taking its
 260                 * address skips the list terminator
 261                 */
 262                va = (u32 *)&ary->wc_array[ary->wc_nchunks].wc_target.rs_length;
 263
 264        /* Skip reply-array */
 265        ary = (struct rpcrdma_write_array *)va;
 266        if (ary->wc_discrim == xdr_zero)
 267                va = (u32 *)&ary->wc_nchunks;
 268        else
 269                va = (u32 *)&ary->wc_array[ary->wc_nchunks];
 270
 271        rqstp->rq_arg.head[0].iov_base = va;
 272        hdrlen = (unsigned long)va - (unsigned long)rmsgp;
 273        rqstp->rq_arg.head[0].iov_len -= hdrlen;
 274
 275        return hdrlen;
 276}
 277
 278int svc_rdma_xdr_encode_error(struct svcxprt_rdma *xprt,
 279                              struct rpcrdma_msg *rmsgp,
 280                              enum rpcrdma_errcode err, u32 *va)
 281{
 282        u32 *startp = va;
 283
 284        *va++ = htonl(rmsgp->rm_xid);
 285        *va++ = htonl(rmsgp->rm_vers);
 286        *va++ = htonl(xprt->sc_max_requests);
 287        *va++ = htonl(RDMA_ERROR);
 288        *va++ = htonl(err);
 289        if (err == ERR_VERS) {
 290                *va++ = htonl(RPCRDMA_VERSION);
 291                *va++ = htonl(RPCRDMA_VERSION);
 292        }
 293
 294        return (int)((unsigned long)va - (unsigned long)startp);
 295}
 296
 297int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *rmsgp)
 298{
 299        struct rpcrdma_write_array *wr_ary;
 300
 301        /* There is no read-list in a reply */
 302
 303        /* skip write list */
 304        wr_ary = (struct rpcrdma_write_array *)
 305                &rmsgp->rm_body.rm_chunks[1];
 306        if (wr_ary->wc_discrim)
 307                wr_ary = (struct rpcrdma_write_array *)
 308                        &wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)].
 309                        wc_target.rs_length;
 310        else
 311                wr_ary = (struct rpcrdma_write_array *)
 312                        &wr_ary->wc_nchunks;
 313
 314        /* skip reply array */
 315        if (wr_ary->wc_discrim)
 316                wr_ary = (struct rpcrdma_write_array *)
 317                        &wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)];
 318        else
 319                wr_ary = (struct rpcrdma_write_array *)
 320                        &wr_ary->wc_nchunks;
 321
 322        return (unsigned long) wr_ary - (unsigned long) rmsgp;
 323}
 324
 325void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *rmsgp, int chunks)
 326{
 327        struct rpcrdma_write_array *ary;
 328
 329        /* no read-list */
 330        rmsgp->rm_body.rm_chunks[0] = xdr_zero;
 331
 332        /* write-array discrim */
 333        ary = (struct rpcrdma_write_array *)
 334                &rmsgp->rm_body.rm_chunks[1];
 335        ary->wc_discrim = xdr_one;
 336        ary->wc_nchunks = htonl(chunks);
 337
 338        /* write-list terminator */
 339        ary->wc_array[chunks].wc_target.rs_handle = xdr_zero;
 340
 341        /* reply-array discriminator */
 342        ary->wc_array[chunks].wc_target.rs_length = xdr_zero;
 343}
 344
 345void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *ary,
 346                                 int chunks)
 347{
 348        ary->wc_discrim = xdr_one;
 349        ary->wc_nchunks = htonl(chunks);
 350}
 351
 352void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *ary,
 353                                     int chunk_no,
 354                                     __be32 rs_handle,
 355                                     __be64 rs_offset,
 356                                     u32 write_len)
 357{
 358        struct rpcrdma_segment *seg = &ary->wc_array[chunk_no].wc_target;
 359        seg->rs_handle = rs_handle;
 360        seg->rs_offset = rs_offset;
 361        seg->rs_length = htonl(write_len);
 362}
 363
 364void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *xprt,
 365                                  struct rpcrdma_msg *rdma_argp,
 366                                  struct rpcrdma_msg *rdma_resp,
 367                                  enum rpcrdma_proc rdma_type)
 368{
 369        rdma_resp->rm_xid = htonl(rdma_argp->rm_xid);
 370        rdma_resp->rm_vers = htonl(rdma_argp->rm_vers);
 371        rdma_resp->rm_credit = htonl(xprt->sc_max_requests);
 372        rdma_resp->rm_type = htonl(rdma_type);
 373
 374        /* Encode <nul> chunks lists */
 375        rdma_resp->rm_body.rm_chunks[0] = xdr_zero;
 376        rdma_resp->rm_body.rm_chunks[1] = xdr_zero;
 377        rdma_resp->rm_body.rm_chunks[2] = xdr_zero;
 378}
 379