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[]; 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(struct_size(ctxt, rw_first_sgl, SG_CHUNK_SIZE), 68 GFP_KERNEL); 69 if (!ctxt) 70 goto out; 71 INIT_LIST_HEAD(&ctxt->rw_list); 72 } 73 74 ctxt->rw_sg_table.sgl = ctxt->rw_first_sgl; 75 if (sg_alloc_table_chained(&ctxt->rw_sg_table, sges, 76 ctxt->rw_sg_table.sgl, 77 SG_CHUNK_SIZE)) { 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, SG_CHUNK_SIZE); 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); 218 219 svc_rdma_write_info_free(info); 220 } 221 222 /* State for pulling a Read chunk. 223 */ 224 struct svc_rdma_read_info { 225 struct svc_rdma_recv_ctxt *ri_readctxt; 226 unsigned int ri_position; 227 unsigned int ri_pageno; 228 unsigned int ri_pageoff; 229 unsigned int ri_chunklen; 230 231 struct svc_rdma_chunk_ctxt ri_cc; 232 }; 233 234 static struct svc_rdma_read_info * 235 svc_rdma_read_info_alloc(struct svcxprt_rdma *rdma) 236 { 237 struct svc_rdma_read_info *info; 238 239 info = kmalloc(sizeof(*info), GFP_KERNEL); 240 if (!info) 241 return info; 242 243 svc_rdma_cc_init(rdma, &info->ri_cc); 244 info->ri_cc.cc_cqe.done = svc_rdma_wc_read_done; 245 return info; 246 } 247 248 static void svc_rdma_read_info_free(struct svc_rdma_read_info *info) 249 { 250 svc_rdma_cc_release(&info->ri_cc, DMA_FROM_DEVICE); 251 kfree(info); 252 } 253 254 /** 255 * svc_rdma_wc_read_done - Handle completion of an RDMA Read ctx 256 * @cq: controlling Completion Queue 257 * @wc: Work Completion 258 * 259 */ 260 static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc) 261 { 262 struct ib_cqe *cqe = wc->wr_cqe; 263 struct svc_rdma_chunk_ctxt *cc = 264 container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe); 265 struct svcxprt_rdma *rdma = cc->cc_rdma; 266 struct svc_rdma_read_info *info = 267 container_of(cc, struct svc_rdma_read_info, ri_cc); 268 269 trace_svcrdma_wc_read(wc); 270 271 atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail); 272 wake_up(&rdma->sc_send_wait); 273 274 if (unlikely(wc->status != IB_WC_SUCCESS)) { 275 set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); 276 svc_rdma_recv_ctxt_put(rdma, info->ri_readctxt); 277 } else { 278 spin_lock(&rdma->sc_rq_dto_lock); 279 list_add_tail(&info->ri_readctxt->rc_list, 280 &rdma->sc_read_complete_q); 281 /* Note the unlock pairs with the smp_rmb in svc_xprt_ready: */ 282 set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags); 283 spin_unlock(&rdma->sc_rq_dto_lock); 284 285 svc_xprt_enqueue(&rdma->sc_xprt); 286 } 287 288 svc_rdma_read_info_free(info); 289 } 290 291 /* This function sleeps when the transport's Send Queue is congested. 292 * 293 * Assumptions: 294 * - If ib_post_send() succeeds, only one completion is expected, 295 * even if one or more WRs are flushed. This is true when posting 296 * an rdma_rw_ctx or when posting a single signaled WR. 297 */ 298 static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc) 299 { 300 struct svcxprt_rdma *rdma = cc->cc_rdma; 301 struct svc_xprt *xprt = &rdma->sc_xprt; 302 struct ib_send_wr *first_wr; 303 const struct ib_send_wr *bad_wr; 304 struct list_head *tmp; 305 struct ib_cqe *cqe; 306 int ret; 307 308 if (cc->cc_sqecount > rdma->sc_sq_depth) 309 return -EINVAL; 310 311 first_wr = NULL; 312 cqe = &cc->cc_cqe; 313 list_for_each(tmp, &cc->cc_rwctxts) { 314 struct svc_rdma_rw_ctxt *ctxt; 315 316 ctxt = list_entry(tmp, struct svc_rdma_rw_ctxt, rw_list); 317 first_wr = rdma_rw_ctx_wrs(&ctxt->rw_ctx, rdma->sc_qp, 318 rdma->sc_port_num, cqe, first_wr); 319 cqe = NULL; 320 } 321 322 do { 323 if (atomic_sub_return(cc->cc_sqecount, 324 &rdma->sc_sq_avail) > 0) { 325 ret = ib_post_send(rdma->sc_qp, first_wr, &bad_wr); 326 trace_svcrdma_post_rw(&cc->cc_cqe, 327 cc->cc_sqecount, ret); 328 if (ret) 329 break; 330 return 0; 331 } 332 333 trace_svcrdma_sq_full(rdma); 334 atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail); 335 wait_event(rdma->sc_send_wait, 336 atomic_read(&rdma->sc_sq_avail) > cc->cc_sqecount); 337 trace_svcrdma_sq_retry(rdma); 338 } while (1); 339 340 set_bit(XPT_CLOSE, &xprt->xpt_flags); 341 342 /* If even one was posted, there will be a completion. */ 343 if (bad_wr != first_wr) 344 return 0; 345 346 atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail); 347 wake_up(&rdma->sc_send_wait); 348 return -ENOTCONN; 349 } 350 351 /* Build and DMA-map an SGL that covers one kvec in an xdr_buf 352 */ 353 static void svc_rdma_vec_to_sg(struct svc_rdma_write_info *info, 354 unsigned int len, 355 struct svc_rdma_rw_ctxt *ctxt) 356 { 357 struct scatterlist *sg = ctxt->rw_sg_table.sgl; 358 359 sg_set_buf(&sg[0], info->wi_base, len); 360 info->wi_base += len; 361 362 ctxt->rw_nents = 1; 363 } 364 365 /* Build and DMA-map an SGL that covers part of an xdr_buf's pagelist. 366 */ 367 static void svc_rdma_pagelist_to_sg(struct svc_rdma_write_info *info, 368 unsigned int remaining, 369 struct svc_rdma_rw_ctxt *ctxt) 370 { 371 unsigned int sge_no, sge_bytes, page_off, page_no; 372 struct xdr_buf *xdr = info->wi_xdr; 373 struct scatterlist *sg; 374 struct page **page; 375 376 page_off = info->wi_next_off + xdr->page_base; 377 page_no = page_off >> PAGE_SHIFT; 378 page_off = offset_in_page(page_off); 379 page = xdr->pages + page_no; 380 info->wi_next_off += remaining; 381 sg = ctxt->rw_sg_table.sgl; 382 sge_no = 0; 383 do { 384 sge_bytes = min_t(unsigned int, remaining, 385 PAGE_SIZE - page_off); 386 sg_set_page(sg, *page, sge_bytes, page_off); 387 388 remaining -= sge_bytes; 389 sg = sg_next(sg); 390 page_off = 0; 391 sge_no++; 392 page++; 393 } while (remaining); 394 395 ctxt->rw_nents = sge_no; 396 } 397 398 /* Construct RDMA Write WRs to send a portion of an xdr_buf containing 399 * an RPC Reply. 400 */ 401 static int 402 svc_rdma_build_writes(struct svc_rdma_write_info *info, 403 void (*constructor)(struct svc_rdma_write_info *info, 404 unsigned int len, 405 struct svc_rdma_rw_ctxt *ctxt), 406 unsigned int remaining) 407 { 408 struct svc_rdma_chunk_ctxt *cc = &info->wi_cc; 409 struct svcxprt_rdma *rdma = cc->cc_rdma; 410 struct svc_rdma_rw_ctxt *ctxt; 411 __be32 *seg; 412 int ret; 413 414 seg = info->wi_segs + info->wi_seg_no * rpcrdma_segment_maxsz; 415 do { 416 unsigned int write_len; 417 u32 seg_length, seg_handle; 418 u64 seg_offset; 419 420 if (info->wi_seg_no >= info->wi_nsegs) 421 goto out_overflow; 422 423 seg_handle = be32_to_cpup(seg); 424 seg_length = be32_to_cpup(seg + 1); 425 xdr_decode_hyper(seg + 2, &seg_offset); 426 seg_offset += info->wi_seg_off; 427 428 write_len = min(remaining, seg_length - info->wi_seg_off); 429 ctxt = svc_rdma_get_rw_ctxt(rdma, 430 (write_len >> PAGE_SHIFT) + 2); 431 if (!ctxt) 432 goto out_noctx; 433 434 constructor(info, write_len, ctxt); 435 ret = rdma_rw_ctx_init(&ctxt->rw_ctx, rdma->sc_qp, 436 rdma->sc_port_num, ctxt->rw_sg_table.sgl, 437 ctxt->rw_nents, 0, seg_offset, 438 seg_handle, DMA_TO_DEVICE); 439 if (ret < 0) 440 goto out_initerr; 441 442 trace_svcrdma_send_wseg(seg_handle, write_len, seg_offset); 443 444 list_add(&ctxt->rw_list, &cc->cc_rwctxts); 445 cc->cc_sqecount += ret; 446 if (write_len == seg_length - info->wi_seg_off) { 447 seg += 4; 448 info->wi_seg_no++; 449 info->wi_seg_off = 0; 450 } else { 451 info->wi_seg_off += write_len; 452 } 453 remaining -= write_len; 454 } while (remaining); 455 456 return 0; 457 458 out_overflow: 459 dprintk("svcrdma: inadequate space in Write chunk (%u)\n", 460 info->wi_nsegs); 461 return -E2BIG; 462 463 out_noctx: 464 dprintk("svcrdma: no R/W ctxs available\n"); 465 return -ENOMEM; 466 467 out_initerr: 468 svc_rdma_put_rw_ctxt(rdma, ctxt); 469 trace_svcrdma_dma_map_rwctx(rdma, ret); 470 return -EIO; 471 } 472 473 /* Send one of an xdr_buf's kvecs by itself. To send a Reply 474 * chunk, the whole RPC Reply is written back to the client. 475 * This function writes either the head or tail of the xdr_buf 476 * containing the Reply. 477 */ 478 static int svc_rdma_send_xdr_kvec(struct svc_rdma_write_info *info, 479 struct kvec *vec) 480 { 481 info->wi_base = vec->iov_base; 482 return svc_rdma_build_writes(info, svc_rdma_vec_to_sg, 483 vec->iov_len); 484 } 485 486 /* Send an xdr_buf's page list by itself. A Write chunk is just 487 * the page list. A Reply chunk is @xdr's head, page list, and 488 * tail. This function is shared between the two types of chunk. 489 */ 490 static int svc_rdma_send_xdr_pagelist(struct svc_rdma_write_info *info, 491 struct xdr_buf *xdr, 492 unsigned int offset, 493 unsigned long length) 494 { 495 info->wi_xdr = xdr; 496 info->wi_next_off = offset - xdr->head[0].iov_len; 497 return svc_rdma_build_writes(info, svc_rdma_pagelist_to_sg, 498 length); 499 } 500 501 /** 502 * svc_rdma_send_write_chunk - Write all segments in a Write chunk 503 * @rdma: controlling RDMA transport 504 * @wr_ch: Write chunk provided by client 505 * @xdr: xdr_buf containing the data payload 506 * @offset: payload's byte offset in @xdr 507 * @length: size of payload, in bytes 508 * 509 * Returns a non-negative number of bytes the chunk consumed, or 510 * %-E2BIG if the payload was larger than the Write chunk, 511 * %-EINVAL if client provided too many segments, 512 * %-ENOMEM if rdma_rw context pool was exhausted, 513 * %-ENOTCONN if posting failed (connection is lost), 514 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 515 */ 516 int svc_rdma_send_write_chunk(struct svcxprt_rdma *rdma, __be32 *wr_ch, 517 struct xdr_buf *xdr, 518 unsigned int offset, unsigned long length) 519 { 520 struct svc_rdma_write_info *info; 521 int ret; 522 523 if (!length) 524 return 0; 525 526 info = svc_rdma_write_info_alloc(rdma, wr_ch); 527 if (!info) 528 return -ENOMEM; 529 530 ret = svc_rdma_send_xdr_pagelist(info, xdr, offset, length); 531 if (ret < 0) 532 goto out_err; 533 534 ret = svc_rdma_post_chunk_ctxt(&info->wi_cc); 535 if (ret < 0) 536 goto out_err; 537 538 trace_svcrdma_send_write_chunk(xdr->page_len); 539 return length; 540 541 out_err: 542 svc_rdma_write_info_free(info); 543 return ret; 544 } 545 546 /** 547 * svc_rdma_send_reply_chunk - Write all segments in the Reply chunk 548 * @rdma: controlling RDMA transport 549 * @rctxt: Write and Reply chunks from client 550 * @xdr: xdr_buf containing an RPC Reply 551 * 552 * Returns a non-negative number of bytes the chunk consumed, or 553 * %-E2BIG if the payload was larger than the Reply chunk, 554 * %-EINVAL if client provided too many segments, 555 * %-ENOMEM if rdma_rw context pool was exhausted, 556 * %-ENOTCONN if posting failed (connection is lost), 557 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 558 */ 559 int svc_rdma_send_reply_chunk(struct svcxprt_rdma *rdma, 560 const struct svc_rdma_recv_ctxt *rctxt, 561 struct xdr_buf *xdr) 562 { 563 struct svc_rdma_write_info *info; 564 int consumed, ret; 565 566 info = svc_rdma_write_info_alloc(rdma, rctxt->rc_reply_chunk); 567 if (!info) 568 return -ENOMEM; 569 570 ret = svc_rdma_send_xdr_kvec(info, &xdr->head[0]); 571 if (ret < 0) 572 goto out_err; 573 consumed = xdr->head[0].iov_len; 574 575 /* Send the page list in the Reply chunk only if the 576 * client did not provide Write chunks. 577 */ 578 if (!rctxt->rc_write_list && xdr->page_len) { 579 ret = svc_rdma_send_xdr_pagelist(info, xdr, 580 xdr->head[0].iov_len, 581 xdr->page_len); 582 if (ret < 0) 583 goto out_err; 584 consumed += xdr->page_len; 585 } 586 587 if (xdr->tail[0].iov_len) { 588 ret = svc_rdma_send_xdr_kvec(info, &xdr->tail[0]); 589 if (ret < 0) 590 goto out_err; 591 consumed += xdr->tail[0].iov_len; 592 } 593 594 ret = svc_rdma_post_chunk_ctxt(&info->wi_cc); 595 if (ret < 0) 596 goto out_err; 597 598 trace_svcrdma_send_reply_chunk(consumed); 599 return consumed; 600 601 out_err: 602 svc_rdma_write_info_free(info); 603 return ret; 604 } 605 606 static int svc_rdma_build_read_segment(struct svc_rdma_read_info *info, 607 struct svc_rqst *rqstp, 608 u32 rkey, u32 len, u64 offset) 609 { 610 struct svc_rdma_recv_ctxt *head = info->ri_readctxt; 611 struct svc_rdma_chunk_ctxt *cc = &info->ri_cc; 612 struct svc_rdma_rw_ctxt *ctxt; 613 unsigned int sge_no, seg_len; 614 struct scatterlist *sg; 615 int ret; 616 617 sge_no = PAGE_ALIGN(info->ri_pageoff + len) >> PAGE_SHIFT; 618 ctxt = svc_rdma_get_rw_ctxt(cc->cc_rdma, sge_no); 619 if (!ctxt) 620 goto out_noctx; 621 ctxt->rw_nents = sge_no; 622 623 sg = ctxt->rw_sg_table.sgl; 624 for (sge_no = 0; sge_no < ctxt->rw_nents; sge_no++) { 625 seg_len = min_t(unsigned int, len, 626 PAGE_SIZE - info->ri_pageoff); 627 628 head->rc_arg.pages[info->ri_pageno] = 629 rqstp->rq_pages[info->ri_pageno]; 630 if (!info->ri_pageoff) 631 head->rc_page_count++; 632 633 sg_set_page(sg, rqstp->rq_pages[info->ri_pageno], 634 seg_len, info->ri_pageoff); 635 sg = sg_next(sg); 636 637 info->ri_pageoff += seg_len; 638 if (info->ri_pageoff == PAGE_SIZE) { 639 info->ri_pageno++; 640 info->ri_pageoff = 0; 641 } 642 len -= seg_len; 643 644 /* Safety check */ 645 if (len && 646 &rqstp->rq_pages[info->ri_pageno + 1] > rqstp->rq_page_end) 647 goto out_overrun; 648 } 649 650 ret = rdma_rw_ctx_init(&ctxt->rw_ctx, cc->cc_rdma->sc_qp, 651 cc->cc_rdma->sc_port_num, 652 ctxt->rw_sg_table.sgl, ctxt->rw_nents, 653 0, offset, rkey, DMA_FROM_DEVICE); 654 if (ret < 0) 655 goto out_initerr; 656 657 list_add(&ctxt->rw_list, &cc->cc_rwctxts); 658 cc->cc_sqecount += ret; 659 return 0; 660 661 out_noctx: 662 dprintk("svcrdma: no R/W ctxs available\n"); 663 return -ENOMEM; 664 665 out_overrun: 666 dprintk("svcrdma: request overruns rq_pages\n"); 667 return -EINVAL; 668 669 out_initerr: 670 trace_svcrdma_dma_map_rwctx(cc->cc_rdma, ret); 671 svc_rdma_put_rw_ctxt(cc->cc_rdma, ctxt); 672 return -EIO; 673 } 674 675 /* Walk the segments in the Read chunk starting at @p and construct 676 * RDMA Read operations to pull the chunk to the server. 677 */ 678 static int svc_rdma_build_read_chunk(struct svc_rqst *rqstp, 679 struct svc_rdma_read_info *info, 680 __be32 *p) 681 { 682 unsigned int i; 683 int ret; 684 685 ret = -EINVAL; 686 info->ri_chunklen = 0; 687 while (*p++ != xdr_zero && be32_to_cpup(p++) == info->ri_position) { 688 u32 rs_handle, rs_length; 689 u64 rs_offset; 690 691 rs_handle = be32_to_cpup(p++); 692 rs_length = be32_to_cpup(p++); 693 p = xdr_decode_hyper(p, &rs_offset); 694 695 ret = svc_rdma_build_read_segment(info, rqstp, 696 rs_handle, rs_length, 697 rs_offset); 698 if (ret < 0) 699 break; 700 701 trace_svcrdma_send_rseg(rs_handle, rs_length, rs_offset); 702 info->ri_chunklen += rs_length; 703 } 704 705 /* Pages under I/O have been copied to head->rc_pages. 706 * Prevent their premature release by svc_xprt_release() . 707 */ 708 for (i = 0; i < info->ri_readctxt->rc_page_count; i++) 709 rqstp->rq_pages[i] = NULL; 710 711 return ret; 712 } 713 714 /* Construct RDMA Reads to pull over a normal Read chunk. The chunk 715 * data lands in the page list of head->rc_arg.pages. 716 * 717 * Currently NFSD does not look at the head->rc_arg.tail[0] iovec. 718 * Therefore, XDR round-up of the Read chunk and trailing 719 * inline content must both be added at the end of the pagelist. 720 */ 721 static int svc_rdma_build_normal_read_chunk(struct svc_rqst *rqstp, 722 struct svc_rdma_read_info *info, 723 __be32 *p) 724 { 725 struct svc_rdma_recv_ctxt *head = info->ri_readctxt; 726 int ret; 727 728 ret = svc_rdma_build_read_chunk(rqstp, info, p); 729 if (ret < 0) 730 goto out; 731 732 trace_svcrdma_send_read_chunk(info->ri_chunklen, info->ri_position); 733 734 head->rc_hdr_count = 0; 735 736 /* Split the Receive buffer between the head and tail 737 * buffers at Read chunk's position. XDR roundup of the 738 * chunk is not included in either the pagelist or in 739 * the tail. 740 */ 741 head->rc_arg.tail[0].iov_base = 742 head->rc_arg.head[0].iov_base + info->ri_position; 743 head->rc_arg.tail[0].iov_len = 744 head->rc_arg.head[0].iov_len - info->ri_position; 745 head->rc_arg.head[0].iov_len = info->ri_position; 746 747 /* Read chunk may need XDR roundup (see RFC 8166, s. 3.4.5.2). 748 * 749 * If the client already rounded up the chunk length, the 750 * length does not change. Otherwise, the length of the page 751 * list is increased to include XDR round-up. 752 * 753 * Currently these chunks always start at page offset 0, 754 * thus the rounded-up length never crosses a page boundary. 755 */ 756 info->ri_chunklen = XDR_QUADLEN(info->ri_chunklen) << 2; 757 758 head->rc_arg.page_len = info->ri_chunklen; 759 head->rc_arg.len += info->ri_chunklen; 760 head->rc_arg.buflen += info->ri_chunklen; 761 762 out: 763 return ret; 764 } 765 766 /* Construct RDMA Reads to pull over a Position Zero Read chunk. 767 * The start of the data lands in the first page just after 768 * the Transport header, and the rest lands in the page list of 769 * head->rc_arg.pages. 770 * 771 * Assumptions: 772 * - A PZRC has an XDR-aligned length (no implicit round-up). 773 * - There can be no trailing inline content (IOW, we assume 774 * a PZRC is never sent in an RDMA_MSG message, though it's 775 * allowed by spec). 776 */ 777 static int svc_rdma_build_pz_read_chunk(struct svc_rqst *rqstp, 778 struct svc_rdma_read_info *info, 779 __be32 *p) 780 { 781 struct svc_rdma_recv_ctxt *head = info->ri_readctxt; 782 int ret; 783 784 ret = svc_rdma_build_read_chunk(rqstp, info, p); 785 if (ret < 0) 786 goto out; 787 788 trace_svcrdma_send_pzr(info->ri_chunklen); 789 790 head->rc_arg.len += info->ri_chunklen; 791 head->rc_arg.buflen += info->ri_chunklen; 792 793 head->rc_hdr_count = 1; 794 head->rc_arg.head[0].iov_base = page_address(head->rc_pages[0]); 795 head->rc_arg.head[0].iov_len = min_t(size_t, PAGE_SIZE, 796 info->ri_chunklen); 797 798 head->rc_arg.page_len = info->ri_chunklen - 799 head->rc_arg.head[0].iov_len; 800 801 out: 802 return ret; 803 } 804 805 /** 806 * svc_rdma_recv_read_chunk - Pull a Read chunk from the client 807 * @rdma: controlling RDMA transport 808 * @rqstp: set of pages to use as Read sink buffers 809 * @head: pages under I/O collect here 810 * @p: pointer to start of Read chunk 811 * 812 * Returns: 813 * %0 if all needed RDMA Reads were posted successfully, 814 * %-EINVAL if client provided too many segments, 815 * %-ENOMEM if rdma_rw context pool was exhausted, 816 * %-ENOTCONN if posting failed (connection is lost), 817 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 818 * 819 * Assumptions: 820 * - All Read segments in @p have the same Position value. 821 */ 822 int svc_rdma_recv_read_chunk(struct svcxprt_rdma *rdma, struct svc_rqst *rqstp, 823 struct svc_rdma_recv_ctxt *head, __be32 *p) 824 { 825 struct svc_rdma_read_info *info; 826 int ret; 827 828 /* The request (with page list) is constructed in 829 * head->rc_arg. Pages involved with RDMA Read I/O are 830 * transferred there. 831 */ 832 head->rc_arg.head[0] = rqstp->rq_arg.head[0]; 833 head->rc_arg.tail[0] = rqstp->rq_arg.tail[0]; 834 head->rc_arg.pages = head->rc_pages; 835 head->rc_arg.page_base = 0; 836 head->rc_arg.page_len = 0; 837 head->rc_arg.len = rqstp->rq_arg.len; 838 head->rc_arg.buflen = rqstp->rq_arg.buflen; 839 840 info = svc_rdma_read_info_alloc(rdma); 841 if (!info) 842 return -ENOMEM; 843 info->ri_readctxt = head; 844 info->ri_pageno = 0; 845 info->ri_pageoff = 0; 846 847 info->ri_position = be32_to_cpup(p + 1); 848 if (info->ri_position) 849 ret = svc_rdma_build_normal_read_chunk(rqstp, info, p); 850 else 851 ret = svc_rdma_build_pz_read_chunk(rqstp, info, p); 852 if (ret < 0) 853 goto out_err; 854 855 ret = svc_rdma_post_chunk_ctxt(&info->ri_cc); 856 if (ret < 0) 857 goto out_err; 858 return 0; 859 860 out_err: 861 svc_rdma_read_info_free(info); 862 return ret; 863 } 864