1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2016-2018 Oracle. All rights reserved. 4 * 5 * Use the core R/W API to move RPC-over-RDMA Read and Write chunks. 6 */ 7 8 #include <rdma/rw.h> 9 10 #include <linux/sunrpc/rpc_rdma.h> 11 #include <linux/sunrpc/svc_rdma.h> 12 #include <linux/sunrpc/debug.h> 13 14 #include "xprt_rdma.h" 15 #include <trace/events/rpcrdma.h> 16 17 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 18 19 static void svc_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc); 20 static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc); 21 22 /* Each R/W context contains state for one chain of RDMA Read or 23 * Write Work Requests. 24 * 25 * Each WR chain handles a single contiguous server-side buffer, 26 * because scatterlist entries after the first have to start on 27 * page alignment. xdr_buf iovecs cannot guarantee alignment. 28 * 29 * Each WR chain handles only one R_key. Each RPC-over-RDMA segment 30 * from a client may contain a unique R_key, so each WR chain moves 31 * up to one segment at a time. 32 * 33 * The scatterlist makes this data structure over 4KB in size. To 34 * make it less likely to fail, and to handle the allocation for 35 * smaller I/O requests without disabling bottom-halves, these 36 * contexts are created on demand, but cached and reused until the 37 * controlling svcxprt_rdma is destroyed. 38 */ 39 struct svc_rdma_rw_ctxt { 40 struct list_head rw_list; 41 struct rdma_rw_ctx rw_ctx; 42 int rw_nents; 43 struct sg_table rw_sg_table; 44 struct scatterlist rw_first_sgl[0]; 45 }; 46 47 static inline struct svc_rdma_rw_ctxt * 48 svc_rdma_next_ctxt(struct list_head *list) 49 { 50 return list_first_entry_or_null(list, struct svc_rdma_rw_ctxt, 51 rw_list); 52 } 53 54 static struct svc_rdma_rw_ctxt * 55 svc_rdma_get_rw_ctxt(struct svcxprt_rdma *rdma, unsigned int sges) 56 { 57 struct svc_rdma_rw_ctxt *ctxt; 58 59 spin_lock(&rdma->sc_rw_ctxt_lock); 60 61 ctxt = svc_rdma_next_ctxt(&rdma->sc_rw_ctxts); 62 if (ctxt) { 63 list_del(&ctxt->rw_list); 64 spin_unlock(&rdma->sc_rw_ctxt_lock); 65 } else { 66 spin_unlock(&rdma->sc_rw_ctxt_lock); 67 ctxt = kmalloc(sizeof(*ctxt) + 68 SG_CHUNK_SIZE * sizeof(struct scatterlist), 69 GFP_KERNEL); 70 if (!ctxt) 71 goto out; 72 INIT_LIST_HEAD(&ctxt->rw_list); 73 } 74 75 ctxt->rw_sg_table.sgl = ctxt->rw_first_sgl; 76 if (sg_alloc_table_chained(&ctxt->rw_sg_table, sges, 77 ctxt->rw_sg_table.sgl)) { 78 kfree(ctxt); 79 ctxt = NULL; 80 } 81 out: 82 return ctxt; 83 } 84 85 static void svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma, 86 struct svc_rdma_rw_ctxt *ctxt) 87 { 88 sg_free_table_chained(&ctxt->rw_sg_table, true); 89 90 spin_lock(&rdma->sc_rw_ctxt_lock); 91 list_add(&ctxt->rw_list, &rdma->sc_rw_ctxts); 92 spin_unlock(&rdma->sc_rw_ctxt_lock); 93 } 94 95 /** 96 * svc_rdma_destroy_rw_ctxts - Free accumulated R/W contexts 97 * @rdma: transport about to be destroyed 98 * 99 */ 100 void svc_rdma_destroy_rw_ctxts(struct svcxprt_rdma *rdma) 101 { 102 struct svc_rdma_rw_ctxt *ctxt; 103 104 while ((ctxt = svc_rdma_next_ctxt(&rdma->sc_rw_ctxts)) != NULL) { 105 list_del(&ctxt->rw_list); 106 kfree(ctxt); 107 } 108 } 109 110 /* A chunk context tracks all I/O for moving one Read or Write 111 * chunk. This is a a set of rdma_rw's that handle data movement 112 * for all segments of one chunk. 113 * 114 * These are small, acquired with a single allocator call, and 115 * no more than one is needed per chunk. They are allocated on 116 * demand, and not cached. 117 */ 118 struct svc_rdma_chunk_ctxt { 119 struct ib_cqe cc_cqe; 120 struct svcxprt_rdma *cc_rdma; 121 struct list_head cc_rwctxts; 122 int cc_sqecount; 123 }; 124 125 static void svc_rdma_cc_init(struct svcxprt_rdma *rdma, 126 struct svc_rdma_chunk_ctxt *cc) 127 { 128 cc->cc_rdma = rdma; 129 svc_xprt_get(&rdma->sc_xprt); 130 131 INIT_LIST_HEAD(&cc->cc_rwctxts); 132 cc->cc_sqecount = 0; 133 } 134 135 static void svc_rdma_cc_release(struct svc_rdma_chunk_ctxt *cc, 136 enum dma_data_direction dir) 137 { 138 struct svcxprt_rdma *rdma = cc->cc_rdma; 139 struct svc_rdma_rw_ctxt *ctxt; 140 141 while ((ctxt = svc_rdma_next_ctxt(&cc->cc_rwctxts)) != NULL) { 142 list_del(&ctxt->rw_list); 143 144 rdma_rw_ctx_destroy(&ctxt->rw_ctx, rdma->sc_qp, 145 rdma->sc_port_num, ctxt->rw_sg_table.sgl, 146 ctxt->rw_nents, dir); 147 svc_rdma_put_rw_ctxt(rdma, ctxt); 148 } 149 svc_xprt_put(&rdma->sc_xprt); 150 } 151 152 /* State for sending a Write or Reply chunk. 153 * - Tracks progress of writing one chunk over all its segments 154 * - Stores arguments for the SGL constructor functions 155 */ 156 struct svc_rdma_write_info { 157 /* write state of this chunk */ 158 unsigned int wi_seg_off; 159 unsigned int wi_seg_no; 160 unsigned int wi_nsegs; 161 __be32 *wi_segs; 162 163 /* SGL constructor arguments */ 164 struct xdr_buf *wi_xdr; 165 unsigned char *wi_base; 166 unsigned int wi_next_off; 167 168 struct svc_rdma_chunk_ctxt wi_cc; 169 }; 170 171 static struct svc_rdma_write_info * 172 svc_rdma_write_info_alloc(struct svcxprt_rdma *rdma, __be32 *chunk) 173 { 174 struct svc_rdma_write_info *info; 175 176 info = kmalloc(sizeof(*info), GFP_KERNEL); 177 if (!info) 178 return info; 179 180 info->wi_seg_off = 0; 181 info->wi_seg_no = 0; 182 info->wi_nsegs = be32_to_cpup(++chunk); 183 info->wi_segs = ++chunk; 184 svc_rdma_cc_init(rdma, &info->wi_cc); 185 info->wi_cc.cc_cqe.done = svc_rdma_write_done; 186 return info; 187 } 188 189 static void svc_rdma_write_info_free(struct svc_rdma_write_info *info) 190 { 191 svc_rdma_cc_release(&info->wi_cc, DMA_TO_DEVICE); 192 kfree(info); 193 } 194 195 /** 196 * svc_rdma_write_done - Write chunk completion 197 * @cq: controlling Completion Queue 198 * @wc: Work Completion 199 * 200 * Pages under I/O are freed by a subsequent Send completion. 201 */ 202 static void svc_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc) 203 { 204 struct ib_cqe *cqe = wc->wr_cqe; 205 struct svc_rdma_chunk_ctxt *cc = 206 container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe); 207 struct svcxprt_rdma *rdma = cc->cc_rdma; 208 struct svc_rdma_write_info *info = 209 container_of(cc, struct svc_rdma_write_info, wi_cc); 210 211 trace_svcrdma_wc_write(wc); 212 213 atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail); 214 wake_up(&rdma->sc_send_wait); 215 216 if (unlikely(wc->status != IB_WC_SUCCESS)) { 217 set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); 218 if (wc->status != IB_WC_WR_FLUSH_ERR) 219 pr_err("svcrdma: write ctx: %s (%u/0x%x)\n", 220 ib_wc_status_msg(wc->status), 221 wc->status, wc->vendor_err); 222 } 223 224 svc_rdma_write_info_free(info); 225 } 226 227 /* State for pulling a Read chunk. 228 */ 229 struct svc_rdma_read_info { 230 struct svc_rdma_recv_ctxt *ri_readctxt; 231 unsigned int ri_position; 232 unsigned int ri_pageno; 233 unsigned int ri_pageoff; 234 unsigned int ri_chunklen; 235 236 struct svc_rdma_chunk_ctxt ri_cc; 237 }; 238 239 static struct svc_rdma_read_info * 240 svc_rdma_read_info_alloc(struct svcxprt_rdma *rdma) 241 { 242 struct svc_rdma_read_info *info; 243 244 info = kmalloc(sizeof(*info), GFP_KERNEL); 245 if (!info) 246 return info; 247 248 svc_rdma_cc_init(rdma, &info->ri_cc); 249 info->ri_cc.cc_cqe.done = svc_rdma_wc_read_done; 250 return info; 251 } 252 253 static void svc_rdma_read_info_free(struct svc_rdma_read_info *info) 254 { 255 svc_rdma_cc_release(&info->ri_cc, DMA_FROM_DEVICE); 256 kfree(info); 257 } 258 259 /** 260 * svc_rdma_wc_read_done - Handle completion of an RDMA Read ctx 261 * @cq: controlling Completion Queue 262 * @wc: Work Completion 263 * 264 */ 265 static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc) 266 { 267 struct ib_cqe *cqe = wc->wr_cqe; 268 struct svc_rdma_chunk_ctxt *cc = 269 container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe); 270 struct svcxprt_rdma *rdma = cc->cc_rdma; 271 struct svc_rdma_read_info *info = 272 container_of(cc, struct svc_rdma_read_info, ri_cc); 273 274 trace_svcrdma_wc_read(wc); 275 276 atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail); 277 wake_up(&rdma->sc_send_wait); 278 279 if (unlikely(wc->status != IB_WC_SUCCESS)) { 280 set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); 281 if (wc->status != IB_WC_WR_FLUSH_ERR) 282 pr_err("svcrdma: read ctx: %s (%u/0x%x)\n", 283 ib_wc_status_msg(wc->status), 284 wc->status, wc->vendor_err); 285 svc_rdma_recv_ctxt_put(rdma, info->ri_readctxt); 286 } else { 287 spin_lock(&rdma->sc_rq_dto_lock); 288 list_add_tail(&info->ri_readctxt->rc_list, 289 &rdma->sc_read_complete_q); 290 spin_unlock(&rdma->sc_rq_dto_lock); 291 292 set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags); 293 svc_xprt_enqueue(&rdma->sc_xprt); 294 } 295 296 svc_rdma_read_info_free(info); 297 } 298 299 /* This function sleeps when the transport's Send Queue is congested. 300 * 301 * Assumptions: 302 * - If ib_post_send() succeeds, only one completion is expected, 303 * even if one or more WRs are flushed. This is true when posting 304 * an rdma_rw_ctx or when posting a single signaled WR. 305 */ 306 static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc) 307 { 308 struct svcxprt_rdma *rdma = cc->cc_rdma; 309 struct svc_xprt *xprt = &rdma->sc_xprt; 310 struct ib_send_wr *first_wr; 311 const struct ib_send_wr *bad_wr; 312 struct list_head *tmp; 313 struct ib_cqe *cqe; 314 int ret; 315 316 if (cc->cc_sqecount > rdma->sc_sq_depth) 317 return -EINVAL; 318 319 first_wr = NULL; 320 cqe = &cc->cc_cqe; 321 list_for_each(tmp, &cc->cc_rwctxts) { 322 struct svc_rdma_rw_ctxt *ctxt; 323 324 ctxt = list_entry(tmp, struct svc_rdma_rw_ctxt, rw_list); 325 first_wr = rdma_rw_ctx_wrs(&ctxt->rw_ctx, rdma->sc_qp, 326 rdma->sc_port_num, cqe, first_wr); 327 cqe = NULL; 328 } 329 330 do { 331 if (atomic_sub_return(cc->cc_sqecount, 332 &rdma->sc_sq_avail) > 0) { 333 ret = ib_post_send(rdma->sc_qp, first_wr, &bad_wr); 334 trace_svcrdma_post_rw(&cc->cc_cqe, 335 cc->cc_sqecount, ret); 336 if (ret) 337 break; 338 return 0; 339 } 340 341 trace_svcrdma_sq_full(rdma); 342 atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail); 343 wait_event(rdma->sc_send_wait, 344 atomic_read(&rdma->sc_sq_avail) > cc->cc_sqecount); 345 trace_svcrdma_sq_retry(rdma); 346 } while (1); 347 348 set_bit(XPT_CLOSE, &xprt->xpt_flags); 349 350 /* If even one was posted, there will be a completion. */ 351 if (bad_wr != first_wr) 352 return 0; 353 354 atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail); 355 wake_up(&rdma->sc_send_wait); 356 return -ENOTCONN; 357 } 358 359 /* Build and DMA-map an SGL that covers one kvec in an xdr_buf 360 */ 361 static void svc_rdma_vec_to_sg(struct svc_rdma_write_info *info, 362 unsigned int len, 363 struct svc_rdma_rw_ctxt *ctxt) 364 { 365 struct scatterlist *sg = ctxt->rw_sg_table.sgl; 366 367 sg_set_buf(&sg[0], info->wi_base, len); 368 info->wi_base += len; 369 370 ctxt->rw_nents = 1; 371 } 372 373 /* Build and DMA-map an SGL that covers part of an xdr_buf's pagelist. 374 */ 375 static void svc_rdma_pagelist_to_sg(struct svc_rdma_write_info *info, 376 unsigned int remaining, 377 struct svc_rdma_rw_ctxt *ctxt) 378 { 379 unsigned int sge_no, sge_bytes, page_off, page_no; 380 struct xdr_buf *xdr = info->wi_xdr; 381 struct scatterlist *sg; 382 struct page **page; 383 384 page_off = info->wi_next_off + xdr->page_base; 385 page_no = page_off >> PAGE_SHIFT; 386 page_off = offset_in_page(page_off); 387 page = xdr->pages + page_no; 388 info->wi_next_off += remaining; 389 sg = ctxt->rw_sg_table.sgl; 390 sge_no = 0; 391 do { 392 sge_bytes = min_t(unsigned int, remaining, 393 PAGE_SIZE - page_off); 394 sg_set_page(sg, *page, sge_bytes, page_off); 395 396 remaining -= sge_bytes; 397 sg = sg_next(sg); 398 page_off = 0; 399 sge_no++; 400 page++; 401 } while (remaining); 402 403 ctxt->rw_nents = sge_no; 404 } 405 406 /* Construct RDMA Write WRs to send a portion of an xdr_buf containing 407 * an RPC Reply. 408 */ 409 static int 410 svc_rdma_build_writes(struct svc_rdma_write_info *info, 411 void (*constructor)(struct svc_rdma_write_info *info, 412 unsigned int len, 413 struct svc_rdma_rw_ctxt *ctxt), 414 unsigned int remaining) 415 { 416 struct svc_rdma_chunk_ctxt *cc = &info->wi_cc; 417 struct svcxprt_rdma *rdma = cc->cc_rdma; 418 struct svc_rdma_rw_ctxt *ctxt; 419 __be32 *seg; 420 int ret; 421 422 seg = info->wi_segs + info->wi_seg_no * rpcrdma_segment_maxsz; 423 do { 424 unsigned int write_len; 425 u32 seg_length, seg_handle; 426 u64 seg_offset; 427 428 if (info->wi_seg_no >= info->wi_nsegs) 429 goto out_overflow; 430 431 seg_handle = be32_to_cpup(seg); 432 seg_length = be32_to_cpup(seg + 1); 433 xdr_decode_hyper(seg + 2, &seg_offset); 434 seg_offset += info->wi_seg_off; 435 436 write_len = min(remaining, seg_length - info->wi_seg_off); 437 ctxt = svc_rdma_get_rw_ctxt(rdma, 438 (write_len >> PAGE_SHIFT) + 2); 439 if (!ctxt) 440 goto out_noctx; 441 442 constructor(info, write_len, ctxt); 443 ret = rdma_rw_ctx_init(&ctxt->rw_ctx, rdma->sc_qp, 444 rdma->sc_port_num, ctxt->rw_sg_table.sgl, 445 ctxt->rw_nents, 0, seg_offset, 446 seg_handle, DMA_TO_DEVICE); 447 if (ret < 0) 448 goto out_initerr; 449 450 trace_svcrdma_encode_wseg(seg_handle, write_len, seg_offset); 451 list_add(&ctxt->rw_list, &cc->cc_rwctxts); 452 cc->cc_sqecount += ret; 453 if (write_len == seg_length - info->wi_seg_off) { 454 seg += 4; 455 info->wi_seg_no++; 456 info->wi_seg_off = 0; 457 } else { 458 info->wi_seg_off += write_len; 459 } 460 remaining -= write_len; 461 } while (remaining); 462 463 return 0; 464 465 out_overflow: 466 dprintk("svcrdma: inadequate space in Write chunk (%u)\n", 467 info->wi_nsegs); 468 return -E2BIG; 469 470 out_noctx: 471 dprintk("svcrdma: no R/W ctxs available\n"); 472 return -ENOMEM; 473 474 out_initerr: 475 svc_rdma_put_rw_ctxt(rdma, ctxt); 476 trace_svcrdma_dma_map_rwctx(rdma, ret); 477 return -EIO; 478 } 479 480 /* Send one of an xdr_buf's kvecs by itself. To send a Reply 481 * chunk, the whole RPC Reply is written back to the client. 482 * This function writes either the head or tail of the xdr_buf 483 * containing the Reply. 484 */ 485 static int svc_rdma_send_xdr_kvec(struct svc_rdma_write_info *info, 486 struct kvec *vec) 487 { 488 info->wi_base = vec->iov_base; 489 return svc_rdma_build_writes(info, svc_rdma_vec_to_sg, 490 vec->iov_len); 491 } 492 493 /* Send an xdr_buf's page list by itself. A Write chunk is 494 * just the page list. a Reply chunk is the head, page list, 495 * and tail. This function is shared between the two types 496 * of chunk. 497 */ 498 static int svc_rdma_send_xdr_pagelist(struct svc_rdma_write_info *info, 499 struct xdr_buf *xdr) 500 { 501 info->wi_xdr = xdr; 502 info->wi_next_off = 0; 503 return svc_rdma_build_writes(info, svc_rdma_pagelist_to_sg, 504 xdr->page_len); 505 } 506 507 /** 508 * svc_rdma_send_write_chunk - Write all segments in a Write chunk 509 * @rdma: controlling RDMA transport 510 * @wr_ch: Write chunk provided by client 511 * @xdr: xdr_buf containing the data payload 512 * 513 * Returns a non-negative number of bytes the chunk consumed, or 514 * %-E2BIG if the payload was larger than the Write chunk, 515 * %-EINVAL if client provided too many segments, 516 * %-ENOMEM if rdma_rw context pool was exhausted, 517 * %-ENOTCONN if posting failed (connection is lost), 518 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 519 */ 520 int svc_rdma_send_write_chunk(struct svcxprt_rdma *rdma, __be32 *wr_ch, 521 struct xdr_buf *xdr) 522 { 523 struct svc_rdma_write_info *info; 524 int ret; 525 526 if (!xdr->page_len) 527 return 0; 528 529 info = svc_rdma_write_info_alloc(rdma, wr_ch); 530 if (!info) 531 return -ENOMEM; 532 533 ret = svc_rdma_send_xdr_pagelist(info, xdr); 534 if (ret < 0) 535 goto out_err; 536 537 ret = svc_rdma_post_chunk_ctxt(&info->wi_cc); 538 if (ret < 0) 539 goto out_err; 540 541 trace_svcrdma_encode_write(xdr->page_len); 542 return xdr->page_len; 543 544 out_err: 545 svc_rdma_write_info_free(info); 546 return ret; 547 } 548 549 /** 550 * svc_rdma_send_reply_chunk - Write all segments in the Reply chunk 551 * @rdma: controlling RDMA transport 552 * @rp_ch: Reply chunk provided by client 553 * @writelist: true if client provided a Write list 554 * @xdr: xdr_buf containing an RPC Reply 555 * 556 * Returns a non-negative number of bytes the chunk consumed, or 557 * %-E2BIG if the payload was larger than the Reply chunk, 558 * %-EINVAL if client provided too many segments, 559 * %-ENOMEM if rdma_rw context pool was exhausted, 560 * %-ENOTCONN if posting failed (connection is lost), 561 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 562 */ 563 int svc_rdma_send_reply_chunk(struct svcxprt_rdma *rdma, __be32 *rp_ch, 564 bool writelist, struct xdr_buf *xdr) 565 { 566 struct svc_rdma_write_info *info; 567 int consumed, ret; 568 569 info = svc_rdma_write_info_alloc(rdma, rp_ch); 570 if (!info) 571 return -ENOMEM; 572 573 ret = svc_rdma_send_xdr_kvec(info, &xdr->head[0]); 574 if (ret < 0) 575 goto out_err; 576 consumed = xdr->head[0].iov_len; 577 578 /* Send the page list in the Reply chunk only if the 579 * client did not provide Write chunks. 580 */ 581 if (!writelist && xdr->page_len) { 582 ret = svc_rdma_send_xdr_pagelist(info, xdr); 583 if (ret < 0) 584 goto out_err; 585 consumed += xdr->page_len; 586 } 587 588 if (xdr->tail[0].iov_len) { 589 ret = svc_rdma_send_xdr_kvec(info, &xdr->tail[0]); 590 if (ret < 0) 591 goto out_err; 592 consumed += xdr->tail[0].iov_len; 593 } 594 595 ret = svc_rdma_post_chunk_ctxt(&info->wi_cc); 596 if (ret < 0) 597 goto out_err; 598 599 trace_svcrdma_encode_reply(consumed); 600 return consumed; 601 602 out_err: 603 svc_rdma_write_info_free(info); 604 return ret; 605 } 606 607 static int svc_rdma_build_read_segment(struct svc_rdma_read_info *info, 608 struct svc_rqst *rqstp, 609 u32 rkey, u32 len, u64 offset) 610 { 611 struct svc_rdma_recv_ctxt *head = info->ri_readctxt; 612 struct svc_rdma_chunk_ctxt *cc = &info->ri_cc; 613 struct svc_rdma_rw_ctxt *ctxt; 614 unsigned int sge_no, seg_len; 615 struct scatterlist *sg; 616 int ret; 617 618 sge_no = PAGE_ALIGN(info->ri_pageoff + len) >> PAGE_SHIFT; 619 ctxt = svc_rdma_get_rw_ctxt(cc->cc_rdma, sge_no); 620 if (!ctxt) 621 goto out_noctx; 622 ctxt->rw_nents = sge_no; 623 624 sg = ctxt->rw_sg_table.sgl; 625 for (sge_no = 0; sge_no < ctxt->rw_nents; sge_no++) { 626 seg_len = min_t(unsigned int, len, 627 PAGE_SIZE - info->ri_pageoff); 628 629 head->rc_arg.pages[info->ri_pageno] = 630 rqstp->rq_pages[info->ri_pageno]; 631 if (!info->ri_pageoff) 632 head->rc_page_count++; 633 634 sg_set_page(sg, rqstp->rq_pages[info->ri_pageno], 635 seg_len, info->ri_pageoff); 636 sg = sg_next(sg); 637 638 info->ri_pageoff += seg_len; 639 if (info->ri_pageoff == PAGE_SIZE) { 640 info->ri_pageno++; 641 info->ri_pageoff = 0; 642 } 643 len -= seg_len; 644 645 /* Safety check */ 646 if (len && 647 &rqstp->rq_pages[info->ri_pageno + 1] > rqstp->rq_page_end) 648 goto out_overrun; 649 } 650 651 ret = rdma_rw_ctx_init(&ctxt->rw_ctx, cc->cc_rdma->sc_qp, 652 cc->cc_rdma->sc_port_num, 653 ctxt->rw_sg_table.sgl, ctxt->rw_nents, 654 0, offset, rkey, DMA_FROM_DEVICE); 655 if (ret < 0) 656 goto out_initerr; 657 658 list_add(&ctxt->rw_list, &cc->cc_rwctxts); 659 cc->cc_sqecount += ret; 660 return 0; 661 662 out_noctx: 663 dprintk("svcrdma: no R/W ctxs available\n"); 664 return -ENOMEM; 665 666 out_overrun: 667 dprintk("svcrdma: request overruns rq_pages\n"); 668 return -EINVAL; 669 670 out_initerr: 671 trace_svcrdma_dma_map_rwctx(cc->cc_rdma, ret); 672 svc_rdma_put_rw_ctxt(cc->cc_rdma, ctxt); 673 return -EIO; 674 } 675 676 /* Walk the segments in the Read chunk starting at @p and construct 677 * RDMA Read operations to pull the chunk to the server. 678 */ 679 static int svc_rdma_build_read_chunk(struct svc_rqst *rqstp, 680 struct svc_rdma_read_info *info, 681 __be32 *p) 682 { 683 unsigned int i; 684 int ret; 685 686 ret = -EINVAL; 687 info->ri_chunklen = 0; 688 while (*p++ != xdr_zero && be32_to_cpup(p++) == info->ri_position) { 689 u32 rs_handle, rs_length; 690 u64 rs_offset; 691 692 rs_handle = be32_to_cpup(p++); 693 rs_length = be32_to_cpup(p++); 694 p = xdr_decode_hyper(p, &rs_offset); 695 696 ret = svc_rdma_build_read_segment(info, rqstp, 697 rs_handle, rs_length, 698 rs_offset); 699 if (ret < 0) 700 break; 701 702 trace_svcrdma_encode_rseg(rs_handle, rs_length, rs_offset); 703 info->ri_chunklen += rs_length; 704 } 705 706 /* Pages under I/O have been copied to head->rc_pages. 707 * Prevent their premature release by svc_xprt_release() . 708 */ 709 for (i = 0; i < info->ri_readctxt->rc_page_count; i++) 710 rqstp->rq_pages[i] = NULL; 711 712 return ret; 713 } 714 715 /* Construct RDMA Reads to pull over a normal Read chunk. The chunk 716 * data lands in the page list of head->rc_arg.pages. 717 * 718 * Currently NFSD does not look at the head->rc_arg.tail[0] iovec. 719 * Therefore, XDR round-up of the Read chunk and trailing 720 * inline content must both be added at the end of the pagelist. 721 */ 722 static int svc_rdma_build_normal_read_chunk(struct svc_rqst *rqstp, 723 struct svc_rdma_read_info *info, 724 __be32 *p) 725 { 726 struct svc_rdma_recv_ctxt *head = info->ri_readctxt; 727 int ret; 728 729 ret = svc_rdma_build_read_chunk(rqstp, info, p); 730 if (ret < 0) 731 goto out; 732 733 trace_svcrdma_encode_read(info->ri_chunklen, info->ri_position); 734 735 head->rc_hdr_count = 0; 736 737 /* Split the Receive buffer between the head and tail 738 * buffers at Read chunk's position. XDR roundup of the 739 * chunk is not included in either the pagelist or in 740 * the tail. 741 */ 742 head->rc_arg.tail[0].iov_base = 743 head->rc_arg.head[0].iov_base + info->ri_position; 744 head->rc_arg.tail[0].iov_len = 745 head->rc_arg.head[0].iov_len - info->ri_position; 746 head->rc_arg.head[0].iov_len = info->ri_position; 747 748 /* Read chunk may need XDR roundup (see RFC 8166, s. 3.4.5.2). 749 * 750 * If the client already rounded up the chunk length, the 751 * length does not change. Otherwise, the length of the page 752 * list is increased to include XDR round-up. 753 * 754 * Currently these chunks always start at page offset 0, 755 * thus the rounded-up length never crosses a page boundary. 756 */ 757 info->ri_chunklen = XDR_QUADLEN(info->ri_chunklen) << 2; 758 759 head->rc_arg.page_len = info->ri_chunklen; 760 head->rc_arg.len += info->ri_chunklen; 761 head->rc_arg.buflen += info->ri_chunklen; 762 763 out: 764 return ret; 765 } 766 767 /* Construct RDMA Reads to pull over a Position Zero Read chunk. 768 * The start of the data lands in the first page just after 769 * the Transport header, and the rest lands in the page list of 770 * head->rc_arg.pages. 771 * 772 * Assumptions: 773 * - A PZRC has an XDR-aligned length (no implicit round-up). 774 * - There can be no trailing inline content (IOW, we assume 775 * a PZRC is never sent in an RDMA_MSG message, though it's 776 * allowed by spec). 777 */ 778 static int svc_rdma_build_pz_read_chunk(struct svc_rqst *rqstp, 779 struct svc_rdma_read_info *info, 780 __be32 *p) 781 { 782 struct svc_rdma_recv_ctxt *head = info->ri_readctxt; 783 int ret; 784 785 ret = svc_rdma_build_read_chunk(rqstp, info, p); 786 if (ret < 0) 787 goto out; 788 789 trace_svcrdma_encode_pzr(info->ri_chunklen); 790 791 head->rc_arg.len += info->ri_chunklen; 792 head->rc_arg.buflen += info->ri_chunklen; 793 794 head->rc_hdr_count = 1; 795 head->rc_arg.head[0].iov_base = page_address(head->rc_pages[0]); 796 head->rc_arg.head[0].iov_len = min_t(size_t, PAGE_SIZE, 797 info->ri_chunklen); 798 799 head->rc_arg.page_len = info->ri_chunklen - 800 head->rc_arg.head[0].iov_len; 801 802 out: 803 return ret; 804 } 805 806 /** 807 * svc_rdma_recv_read_chunk - Pull a Read chunk from the client 808 * @rdma: controlling RDMA transport 809 * @rqstp: set of pages to use as Read sink buffers 810 * @head: pages under I/O collect here 811 * @p: pointer to start of Read chunk 812 * 813 * Returns: 814 * %0 if all needed RDMA Reads were posted successfully, 815 * %-EINVAL if client provided too many segments, 816 * %-ENOMEM if rdma_rw context pool was exhausted, 817 * %-ENOTCONN if posting failed (connection is lost), 818 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 819 * 820 * Assumptions: 821 * - All Read segments in @p have the same Position value. 822 */ 823 int svc_rdma_recv_read_chunk(struct svcxprt_rdma *rdma, struct svc_rqst *rqstp, 824 struct svc_rdma_recv_ctxt *head, __be32 *p) 825 { 826 struct svc_rdma_read_info *info; 827 int ret; 828 829 /* The request (with page list) is constructed in 830 * head->rc_arg. Pages involved with RDMA Read I/O are 831 * transferred there. 832 */ 833 head->rc_arg.head[0] = rqstp->rq_arg.head[0]; 834 head->rc_arg.tail[0] = rqstp->rq_arg.tail[0]; 835 head->rc_arg.pages = head->rc_pages; 836 head->rc_arg.page_base = 0; 837 head->rc_arg.page_len = 0; 838 head->rc_arg.len = rqstp->rq_arg.len; 839 head->rc_arg.buflen = rqstp->rq_arg.buflen; 840 841 info = svc_rdma_read_info_alloc(rdma); 842 if (!info) 843 return -ENOMEM; 844 info->ri_readctxt = head; 845 info->ri_pageno = 0; 846 info->ri_pageoff = 0; 847 848 info->ri_position = be32_to_cpup(p + 1); 849 if (info->ri_position) 850 ret = svc_rdma_build_normal_read_chunk(rqstp, info, p); 851 else 852 ret = svc_rdma_build_pz_read_chunk(rqstp, info, p); 853 if (ret < 0) 854 goto out_err; 855 856 ret = svc_rdma_post_chunk_ctxt(&info->ri_cc); 857 if (ret < 0) 858 goto out_err; 859 return 0; 860 861 out_err: 862 svc_rdma_read_info_free(info); 863 return ret; 864 } 865