Tom Tucker | ef1eac0 | 2007-12-12 16:13:27 -0600 | [diff] [blame] | 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 | */ |
| 58 | static 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 | u64 ch_offset; |
| 64 | |
| 65 | if (((unsigned long)ch + sizeof(struct rpcrdma_read_chunk)) > |
| 66 | (unsigned long)vaend) { |
| 67 | dprintk("svcrdma: vaend=%p, ch=%p\n", vaend, ch); |
| 68 | return NULL; |
| 69 | } |
| 70 | |
| 71 | ch->rc_discrim = ntohl(ch->rc_discrim); |
| 72 | ch->rc_position = ntohl(ch->rc_position); |
| 73 | ch->rc_target.rs_handle = ntohl(ch->rc_target.rs_handle); |
| 74 | ch->rc_target.rs_length = ntohl(ch->rc_target.rs_length); |
| 75 | va = (u32 *)&ch->rc_target.rs_offset; |
| 76 | xdr_decode_hyper(va, &ch_offset); |
| 77 | put_unaligned(ch_offset, (u64 *)va); |
| 78 | ch++; |
| 79 | } |
| 80 | return (u32 *)&ch->rc_position; |
| 81 | } |
| 82 | |
| 83 | /* |
| 84 | * Determine number of chunks and total bytes in chunk list. The chunk |
| 85 | * list has already been verified to fit within the RPCRDMA header. |
| 86 | */ |
| 87 | void svc_rdma_rcl_chunk_counts(struct rpcrdma_read_chunk *ch, |
| 88 | int *ch_count, int *byte_count) |
| 89 | { |
| 90 | /* compute the number of bytes represented by read chunks */ |
| 91 | *byte_count = 0; |
| 92 | *ch_count = 0; |
| 93 | for (; ch->rc_discrim != 0; ch++) { |
| 94 | *byte_count = *byte_count + ch->rc_target.rs_length; |
| 95 | *ch_count = *ch_count + 1; |
| 96 | } |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * Decodes a write chunk list. The expected format is as follows: |
| 101 | * descrim : xdr_one |
| 102 | * nchunks : <count> |
| 103 | * handle : u32 RKEY ---+ |
| 104 | * length : u32 <len of segment> | |
| 105 | * offset : remove va + <count> |
| 106 | * . . . | |
| 107 | * ---+ |
| 108 | */ |
| 109 | static u32 *decode_write_list(u32 *va, u32 *vaend) |
| 110 | { |
| 111 | int ch_no; |
| 112 | struct rpcrdma_write_array *ary = |
| 113 | (struct rpcrdma_write_array *)va; |
| 114 | |
| 115 | /* Check for not write-array */ |
| 116 | if (ary->wc_discrim == xdr_zero) |
| 117 | return (u32 *)&ary->wc_nchunks; |
| 118 | |
| 119 | if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) > |
| 120 | (unsigned long)vaend) { |
| 121 | dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend); |
| 122 | return NULL; |
| 123 | } |
| 124 | ary->wc_discrim = ntohl(ary->wc_discrim); |
| 125 | ary->wc_nchunks = ntohl(ary->wc_nchunks); |
| 126 | if (((unsigned long)&ary->wc_array[0] + |
| 127 | (sizeof(struct rpcrdma_write_chunk) * ary->wc_nchunks)) > |
| 128 | (unsigned long)vaend) { |
| 129 | dprintk("svcrdma: ary=%p, wc_nchunks=%d, vaend=%p\n", |
| 130 | ary, ary->wc_nchunks, vaend); |
| 131 | return NULL; |
| 132 | } |
| 133 | for (ch_no = 0; ch_no < ary->wc_nchunks; ch_no++) { |
| 134 | u64 ch_offset; |
| 135 | |
| 136 | ary->wc_array[ch_no].wc_target.rs_handle = |
| 137 | ntohl(ary->wc_array[ch_no].wc_target.rs_handle); |
| 138 | ary->wc_array[ch_no].wc_target.rs_length = |
| 139 | ntohl(ary->wc_array[ch_no].wc_target.rs_length); |
| 140 | va = (u32 *)&ary->wc_array[ch_no].wc_target.rs_offset; |
| 141 | xdr_decode_hyper(va, &ch_offset); |
| 142 | put_unaligned(ch_offset, (u64 *)va); |
| 143 | } |
| 144 | |
| 145 | /* |
| 146 | * rs_length is the 2nd 4B field in wc_target and taking its |
| 147 | * address skips the list terminator |
| 148 | */ |
| 149 | return (u32 *)&ary->wc_array[ch_no].wc_target.rs_length; |
| 150 | } |
| 151 | |
| 152 | static u32 *decode_reply_array(u32 *va, u32 *vaend) |
| 153 | { |
| 154 | int ch_no; |
| 155 | struct rpcrdma_write_array *ary = |
| 156 | (struct rpcrdma_write_array *)va; |
| 157 | |
| 158 | /* Check for no reply-array */ |
| 159 | if (ary->wc_discrim == xdr_zero) |
| 160 | return (u32 *)&ary->wc_nchunks; |
| 161 | |
| 162 | if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) > |
| 163 | (unsigned long)vaend) { |
| 164 | dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend); |
| 165 | return NULL; |
| 166 | } |
| 167 | ary->wc_discrim = ntohl(ary->wc_discrim); |
| 168 | ary->wc_nchunks = ntohl(ary->wc_nchunks); |
| 169 | if (((unsigned long)&ary->wc_array[0] + |
| 170 | (sizeof(struct rpcrdma_write_chunk) * ary->wc_nchunks)) > |
| 171 | (unsigned long)vaend) { |
| 172 | dprintk("svcrdma: ary=%p, wc_nchunks=%d, vaend=%p\n", |
| 173 | ary, ary->wc_nchunks, vaend); |
| 174 | return NULL; |
| 175 | } |
| 176 | for (ch_no = 0; ch_no < ary->wc_nchunks; ch_no++) { |
| 177 | u64 ch_offset; |
| 178 | |
| 179 | ary->wc_array[ch_no].wc_target.rs_handle = |
| 180 | ntohl(ary->wc_array[ch_no].wc_target.rs_handle); |
| 181 | ary->wc_array[ch_no].wc_target.rs_length = |
| 182 | ntohl(ary->wc_array[ch_no].wc_target.rs_length); |
| 183 | va = (u32 *)&ary->wc_array[ch_no].wc_target.rs_offset; |
| 184 | xdr_decode_hyper(va, &ch_offset); |
| 185 | put_unaligned(ch_offset, (u64 *)va); |
| 186 | } |
| 187 | |
| 188 | return (u32 *)&ary->wc_array[ch_no]; |
| 189 | } |
| 190 | |
| 191 | int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req, |
| 192 | struct svc_rqst *rqstp) |
| 193 | { |
| 194 | struct rpcrdma_msg *rmsgp = NULL; |
| 195 | u32 *va; |
| 196 | u32 *vaend; |
| 197 | u32 hdr_len; |
| 198 | |
| 199 | rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base; |
| 200 | |
| 201 | /* Verify that there's enough bytes for header + something */ |
| 202 | if (rqstp->rq_arg.len <= RPCRDMA_HDRLEN_MIN) { |
| 203 | dprintk("svcrdma: header too short = %d\n", |
| 204 | rqstp->rq_arg.len); |
| 205 | return -EINVAL; |
| 206 | } |
| 207 | |
| 208 | /* Decode the header */ |
| 209 | rmsgp->rm_xid = ntohl(rmsgp->rm_xid); |
| 210 | rmsgp->rm_vers = ntohl(rmsgp->rm_vers); |
| 211 | rmsgp->rm_credit = ntohl(rmsgp->rm_credit); |
| 212 | rmsgp->rm_type = ntohl(rmsgp->rm_type); |
| 213 | |
| 214 | if (rmsgp->rm_vers != RPCRDMA_VERSION) |
| 215 | return -ENOSYS; |
| 216 | |
| 217 | /* Pull in the extra for the padded case and bump our pointer */ |
| 218 | if (rmsgp->rm_type == RDMA_MSGP) { |
| 219 | int hdrlen; |
| 220 | rmsgp->rm_body.rm_padded.rm_align = |
| 221 | ntohl(rmsgp->rm_body.rm_padded.rm_align); |
| 222 | rmsgp->rm_body.rm_padded.rm_thresh = |
| 223 | ntohl(rmsgp->rm_body.rm_padded.rm_thresh); |
| 224 | |
| 225 | va = &rmsgp->rm_body.rm_padded.rm_pempty[4]; |
| 226 | rqstp->rq_arg.head[0].iov_base = va; |
| 227 | hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp); |
| 228 | rqstp->rq_arg.head[0].iov_len -= hdrlen; |
| 229 | if (hdrlen > rqstp->rq_arg.len) |
| 230 | return -EINVAL; |
| 231 | return hdrlen; |
| 232 | } |
| 233 | |
| 234 | /* The chunk list may contain either a read chunk list or a write |
| 235 | * chunk list and a reply chunk list. |
| 236 | */ |
| 237 | va = &rmsgp->rm_body.rm_chunks[0]; |
| 238 | vaend = (u32 *)((unsigned long)rmsgp + rqstp->rq_arg.len); |
| 239 | va = decode_read_list(va, vaend); |
| 240 | if (!va) |
| 241 | return -EINVAL; |
| 242 | va = decode_write_list(va, vaend); |
| 243 | if (!va) |
| 244 | return -EINVAL; |
| 245 | va = decode_reply_array(va, vaend); |
| 246 | if (!va) |
| 247 | return -EINVAL; |
| 248 | |
| 249 | rqstp->rq_arg.head[0].iov_base = va; |
| 250 | hdr_len = (unsigned long)va - (unsigned long)rmsgp; |
| 251 | rqstp->rq_arg.head[0].iov_len -= hdr_len; |
| 252 | |
| 253 | *rdma_req = rmsgp; |
| 254 | return hdr_len; |
| 255 | } |
| 256 | |
| 257 | int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *rqstp) |
| 258 | { |
| 259 | struct rpcrdma_msg *rmsgp = NULL; |
| 260 | struct rpcrdma_read_chunk *ch; |
| 261 | struct rpcrdma_write_array *ary; |
| 262 | u32 *va; |
| 263 | u32 hdrlen; |
| 264 | |
| 265 | dprintk("svcrdma: processing deferred RDMA header on rqstp=%p\n", |
| 266 | rqstp); |
| 267 | rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base; |
| 268 | |
| 269 | /* Pull in the extra for the padded case and bump our pointer */ |
| 270 | if (rmsgp->rm_type == RDMA_MSGP) { |
| 271 | va = &rmsgp->rm_body.rm_padded.rm_pempty[4]; |
| 272 | rqstp->rq_arg.head[0].iov_base = va; |
| 273 | hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp); |
| 274 | rqstp->rq_arg.head[0].iov_len -= hdrlen; |
| 275 | return hdrlen; |
| 276 | } |
| 277 | |
| 278 | /* |
| 279 | * Skip all chunks to find RPC msg. These were previously processed |
| 280 | */ |
| 281 | va = &rmsgp->rm_body.rm_chunks[0]; |
| 282 | |
| 283 | /* Skip read-list */ |
| 284 | for (ch = (struct rpcrdma_read_chunk *)va; |
| 285 | ch->rc_discrim != xdr_zero; ch++); |
| 286 | va = (u32 *)&ch->rc_position; |
| 287 | |
| 288 | /* Skip write-list */ |
| 289 | ary = (struct rpcrdma_write_array *)va; |
| 290 | if (ary->wc_discrim == xdr_zero) |
| 291 | va = (u32 *)&ary->wc_nchunks; |
| 292 | else |
| 293 | /* |
| 294 | * rs_length is the 2nd 4B field in wc_target and taking its |
| 295 | * address skips the list terminator |
| 296 | */ |
| 297 | va = (u32 *)&ary->wc_array[ary->wc_nchunks].wc_target.rs_length; |
| 298 | |
| 299 | /* Skip reply-array */ |
| 300 | ary = (struct rpcrdma_write_array *)va; |
| 301 | if (ary->wc_discrim == xdr_zero) |
| 302 | va = (u32 *)&ary->wc_nchunks; |
| 303 | else |
| 304 | va = (u32 *)&ary->wc_array[ary->wc_nchunks]; |
| 305 | |
| 306 | rqstp->rq_arg.head[0].iov_base = va; |
| 307 | hdrlen = (unsigned long)va - (unsigned long)rmsgp; |
| 308 | rqstp->rq_arg.head[0].iov_len -= hdrlen; |
| 309 | |
| 310 | return hdrlen; |
| 311 | } |
| 312 | |
| 313 | int svc_rdma_xdr_encode_error(struct svcxprt_rdma *xprt, |
| 314 | struct rpcrdma_msg *rmsgp, |
| 315 | enum rpcrdma_errcode err, u32 *va) |
| 316 | { |
| 317 | u32 *startp = va; |
| 318 | |
| 319 | *va++ = htonl(rmsgp->rm_xid); |
| 320 | *va++ = htonl(rmsgp->rm_vers); |
| 321 | *va++ = htonl(xprt->sc_max_requests); |
| 322 | *va++ = htonl(RDMA_ERROR); |
| 323 | *va++ = htonl(err); |
| 324 | if (err == ERR_VERS) { |
| 325 | *va++ = htonl(RPCRDMA_VERSION); |
| 326 | *va++ = htonl(RPCRDMA_VERSION); |
| 327 | } |
| 328 | |
| 329 | return (int)((unsigned long)va - (unsigned long)startp); |
| 330 | } |
| 331 | |
| 332 | int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *rmsgp) |
| 333 | { |
| 334 | struct rpcrdma_write_array *wr_ary; |
| 335 | |
| 336 | /* There is no read-list in a reply */ |
| 337 | |
| 338 | /* skip write list */ |
| 339 | wr_ary = (struct rpcrdma_write_array *) |
| 340 | &rmsgp->rm_body.rm_chunks[1]; |
| 341 | if (wr_ary->wc_discrim) |
| 342 | wr_ary = (struct rpcrdma_write_array *) |
| 343 | &wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)]. |
| 344 | wc_target.rs_length; |
| 345 | else |
| 346 | wr_ary = (struct rpcrdma_write_array *) |
| 347 | &wr_ary->wc_nchunks; |
| 348 | |
| 349 | /* skip reply array */ |
| 350 | if (wr_ary->wc_discrim) |
| 351 | wr_ary = (struct rpcrdma_write_array *) |
| 352 | &wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)]; |
| 353 | else |
| 354 | wr_ary = (struct rpcrdma_write_array *) |
| 355 | &wr_ary->wc_nchunks; |
| 356 | |
| 357 | return (unsigned long) wr_ary - (unsigned long) rmsgp; |
| 358 | } |
| 359 | |
| 360 | void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *rmsgp, int chunks) |
| 361 | { |
| 362 | struct rpcrdma_write_array *ary; |
| 363 | |
| 364 | /* no read-list */ |
| 365 | rmsgp->rm_body.rm_chunks[0] = xdr_zero; |
| 366 | |
| 367 | /* write-array discrim */ |
| 368 | ary = (struct rpcrdma_write_array *) |
| 369 | &rmsgp->rm_body.rm_chunks[1]; |
| 370 | ary->wc_discrim = xdr_one; |
| 371 | ary->wc_nchunks = htonl(chunks); |
| 372 | |
| 373 | /* write-list terminator */ |
| 374 | ary->wc_array[chunks].wc_target.rs_handle = xdr_zero; |
| 375 | |
| 376 | /* reply-array discriminator */ |
| 377 | ary->wc_array[chunks].wc_target.rs_length = xdr_zero; |
| 378 | } |
| 379 | |
| 380 | void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *ary, |
| 381 | int chunks) |
| 382 | { |
| 383 | ary->wc_discrim = xdr_one; |
| 384 | ary->wc_nchunks = htonl(chunks); |
| 385 | } |
| 386 | |
| 387 | void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *ary, |
| 388 | int chunk_no, |
| 389 | u32 rs_handle, u64 rs_offset, |
| 390 | u32 write_len) |
| 391 | { |
| 392 | struct rpcrdma_segment *seg = &ary->wc_array[chunk_no].wc_target; |
| 393 | seg->rs_handle = htonl(rs_handle); |
| 394 | seg->rs_length = htonl(write_len); |
| 395 | xdr_encode_hyper((u32 *) &seg->rs_offset, rs_offset); |
| 396 | } |
| 397 | |
| 398 | void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *xprt, |
| 399 | struct rpcrdma_msg *rdma_argp, |
| 400 | struct rpcrdma_msg *rdma_resp, |
| 401 | enum rpcrdma_proc rdma_type) |
| 402 | { |
| 403 | rdma_resp->rm_xid = htonl(rdma_argp->rm_xid); |
| 404 | rdma_resp->rm_vers = htonl(rdma_argp->rm_vers); |
| 405 | rdma_resp->rm_credit = htonl(xprt->sc_max_requests); |
| 406 | rdma_resp->rm_type = htonl(rdma_type); |
| 407 | |
| 408 | /* Encode <nul> chunks lists */ |
| 409 | rdma_resp->rm_body.rm_chunks[0] = xdr_zero; |
| 410 | rdma_resp->rm_body.rm_chunks[1] = xdr_zero; |
| 411 | rdma_resp->rm_body.rm_chunks[2] = xdr_zero; |
| 412 | } |