2 * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
3 * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the BSD-type
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
15 * Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
18 * Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials provided
21 * with the distribution.
23 * Neither the name of the Network Appliance, Inc. nor the names of
24 * its contributors may be used to endorse or promote products
25 * derived from this software without specific prior written
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 * Author: Tom Tucker <tom@opengridcomputing.com>
43 #include <linux/sunrpc/debug.h>
44 #include <linux/sunrpc/rpc_rdma.h>
45 #include <linux/spinlock.h>
46 #include <asm/unaligned.h>
47 #include <rdma/ib_verbs.h>
48 #include <rdma/rdma_cm.h>
49 #include <linux/sunrpc/svc_rdma.h>
51 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
54 * Replace the pages in the rq_argpages array with the pages from the SGE in
55 * the RDMA_RECV completion. The SGL should contain full pages up until the
58 static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
59 struct svc_rdma_op_ctxt *ctxt,
62 struct rpcrdma_msg *rmsgp;
67 /* Swap the page in the SGE with the page in argpages */
68 page = ctxt->pages[0];
69 put_page(rqstp->rq_pages[0]);
70 rqstp->rq_pages[0] = page;
72 /* Set up the XDR head */
73 rqstp->rq_arg.head[0].iov_base = page_address(page);
74 rqstp->rq_arg.head[0].iov_len =
75 min_t(size_t, byte_count, ctxt->sge[0].length);
76 rqstp->rq_arg.len = byte_count;
77 rqstp->rq_arg.buflen = byte_count;
79 /* Compute bytes past head in the SGL */
80 bc = byte_count - rqstp->rq_arg.head[0].iov_len;
82 /* If data remains, store it in the pagelist */
83 rqstp->rq_arg.page_len = bc;
84 rqstp->rq_arg.page_base = 0;
86 /* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */
87 rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
88 if (rmsgp->rm_type == rdma_nomsg)
89 rqstp->rq_arg.pages = &rqstp->rq_pages[0];
91 rqstp->rq_arg.pages = &rqstp->rq_pages[1];
94 while (bc && sge_no < ctxt->count) {
95 page = ctxt->pages[sge_no];
96 put_page(rqstp->rq_pages[sge_no]);
97 rqstp->rq_pages[sge_no] = page;
98 bc -= min_t(u32, bc, ctxt->sge[sge_no].length);
99 rqstp->rq_arg.buflen += ctxt->sge[sge_no].length;
102 rqstp->rq_respages = &rqstp->rq_pages[sge_no];
103 rqstp->rq_next_page = rqstp->rq_respages + 1;
105 /* If not all pages were used from the SGL, free the remaining ones */
107 while (sge_no < ctxt->count) {
108 page = ctxt->pages[sge_no++];
114 rqstp->rq_arg.tail[0].iov_base = NULL;
115 rqstp->rq_arg.tail[0].iov_len = 0;
118 static int rdma_read_max_sge(struct svcxprt_rdma *xprt, int sge_count)
120 if (!rdma_cap_read_multi_sge(xprt->sc_cm_id->device,
121 xprt->sc_cm_id->port_num))
124 return min_t(int, sge_count, xprt->sc_max_sge);
127 /* Issue an RDMA_READ using the local lkey to map the data sink */
128 int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
129 struct svc_rqst *rqstp,
130 struct svc_rdma_op_ctxt *head,
138 struct ib_send_wr read_wr;
139 int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
140 struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
142 u32 pg_off = *page_offset;
143 u32 pg_no = *page_no;
145 ctxt->direction = DMA_FROM_DEVICE;
146 ctxt->read_hdr = head;
148 min_t(int, pages_needed, rdma_read_max_sge(xprt, pages_needed));
149 read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
151 for (pno = 0; pno < pages_needed; pno++) {
152 int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
154 head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
155 head->arg.page_len += len;
156 head->arg.len += len;
159 rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
160 rqstp->rq_next_page = rqstp->rq_respages + 1;
161 ctxt->sge[pno].addr =
162 ib_dma_map_page(xprt->sc_cm_id->device,
163 head->arg.pages[pg_no], pg_off,
166 ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
167 ctxt->sge[pno].addr);
170 atomic_inc(&xprt->sc_dma_used);
172 /* The lkey here is either a local dma lkey or a dma_mr lkey */
173 ctxt->sge[pno].lkey = xprt->sc_dma_lkey;
174 ctxt->sge[pno].length = len;
177 /* adjust offset and wrap to next page if needed */
179 if (pg_off == PAGE_SIZE) {
186 if (last && rs_length == 0)
187 set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
189 clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
191 memset(&read_wr, 0, sizeof(read_wr));
192 read_wr.wr_id = (unsigned long)ctxt;
193 read_wr.opcode = IB_WR_RDMA_READ;
194 ctxt->wr_op = read_wr.opcode;
195 read_wr.send_flags = IB_SEND_SIGNALED;
196 read_wr.wr.rdma.rkey = rs_handle;
197 read_wr.wr.rdma.remote_addr = rs_offset;
198 read_wr.sg_list = ctxt->sge;
199 read_wr.num_sge = pages_needed;
201 ret = svc_rdma_send(xprt, &read_wr);
203 pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
204 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
208 /* return current location in page array */
210 *page_offset = pg_off;
212 atomic_inc(&rdma_stat_read);
215 svc_rdma_unmap_dma(ctxt);
216 svc_rdma_put_context(ctxt, 0);
220 /* Issue an RDMA_READ using an FRMR to map the data sink */
221 int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt,
222 struct svc_rqst *rqstp,
223 struct svc_rdma_op_ctxt *head,
231 struct ib_send_wr read_wr;
232 struct ib_send_wr inv_wr;
233 struct ib_send_wr fastreg_wr;
235 int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
236 struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
237 struct svc_rdma_fastreg_mr *frmr = svc_rdma_get_frmr(xprt);
239 u32 pg_off = *page_offset;
240 u32 pg_no = *page_no;
245 ctxt->direction = DMA_FROM_DEVICE;
247 pages_needed = min_t(int, pages_needed, xprt->sc_frmr_pg_list_len);
248 read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
250 frmr->kva = page_address(rqstp->rq_arg.pages[pg_no]);
251 frmr->direction = DMA_FROM_DEVICE;
252 frmr->access_flags = (IB_ACCESS_LOCAL_WRITE|IB_ACCESS_REMOTE_WRITE);
253 frmr->map_len = pages_needed << PAGE_SHIFT;
254 frmr->page_list_len = pages_needed;
256 for (pno = 0; pno < pages_needed; pno++) {
257 int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
259 head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
260 head->arg.page_len += len;
261 head->arg.len += len;
264 rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
265 rqstp->rq_next_page = rqstp->rq_respages + 1;
266 frmr->page_list->page_list[pno] =
267 ib_dma_map_page(xprt->sc_cm_id->device,
268 head->arg.pages[pg_no], 0,
269 PAGE_SIZE, DMA_FROM_DEVICE);
270 ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
271 frmr->page_list->page_list[pno]);
274 atomic_inc(&xprt->sc_dma_used);
276 /* adjust offset and wrap to next page if needed */
278 if (pg_off == PAGE_SIZE) {
285 if (last && rs_length == 0)
286 set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
288 clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
291 key = (u8)(frmr->mr->lkey & 0x000000FF);
292 ib_update_fast_reg_key(frmr->mr, ++key);
294 ctxt->sge[0].addr = (unsigned long)frmr->kva + *page_offset;
295 ctxt->sge[0].lkey = frmr->mr->lkey;
296 ctxt->sge[0].length = read;
298 ctxt->read_hdr = head;
300 /* Prepare FASTREG WR */
301 memset(&fastreg_wr, 0, sizeof(fastreg_wr));
302 fastreg_wr.opcode = IB_WR_FAST_REG_MR;
303 fastreg_wr.send_flags = IB_SEND_SIGNALED;
304 fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
305 fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
306 fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
307 fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
308 fastreg_wr.wr.fast_reg.length = frmr->map_len;
309 fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
310 fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
311 fastreg_wr.next = &read_wr;
313 /* Prepare RDMA_READ */
314 memset(&read_wr, 0, sizeof(read_wr));
315 read_wr.send_flags = IB_SEND_SIGNALED;
316 read_wr.wr.rdma.rkey = rs_handle;
317 read_wr.wr.rdma.remote_addr = rs_offset;
318 read_wr.sg_list = ctxt->sge;
320 if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_READ_W_INV) {
321 read_wr.opcode = IB_WR_RDMA_READ_WITH_INV;
322 read_wr.wr_id = (unsigned long)ctxt;
323 read_wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey;
325 read_wr.opcode = IB_WR_RDMA_READ;
326 read_wr.next = &inv_wr;
327 /* Prepare invalidate */
328 memset(&inv_wr, 0, sizeof(inv_wr));
329 inv_wr.wr_id = (unsigned long)ctxt;
330 inv_wr.opcode = IB_WR_LOCAL_INV;
331 inv_wr.send_flags = IB_SEND_SIGNALED | IB_SEND_FENCE;
332 inv_wr.ex.invalidate_rkey = frmr->mr->lkey;
334 ctxt->wr_op = read_wr.opcode;
337 ret = svc_rdma_send(xprt, &fastreg_wr);
339 pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
340 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
344 /* return current location in page array */
346 *page_offset = pg_off;
348 atomic_inc(&rdma_stat_read);
351 svc_rdma_unmap_dma(ctxt);
352 svc_rdma_put_context(ctxt, 0);
353 svc_rdma_put_frmr(xprt, frmr);
358 rdma_rcl_chunk_count(struct rpcrdma_read_chunk *ch)
362 for (count = 0; ch->rc_discrim != xdr_zero; ch++)
367 /* If there was additional inline content, append it to the end of arg.pages.
368 * Tail copy has to be done after the reader function has determined how many
369 * pages are needed for RDMA READ.
372 rdma_copy_tail(struct svc_rqst *rqstp, struct svc_rdma_op_ctxt *head,
373 u32 position, u32 byte_count, u32 page_offset, int page_no)
379 srcp = head->arg.head[0].iov_base + position;
380 byte_count = head->arg.head[0].iov_len - position;
381 if (byte_count > PAGE_SIZE) {
382 dprintk("svcrdma: large tail unsupported\n");
386 /* Fit as much of the tail on the current page as possible */
387 if (page_offset != PAGE_SIZE) {
388 destp = page_address(rqstp->rq_arg.pages[page_no]);
389 destp += page_offset;
390 while (byte_count--) {
393 if (page_offset == PAGE_SIZE && byte_count)
400 /* Fit the rest on the next page */
402 destp = page_address(rqstp->rq_arg.pages[page_no]);
406 rqstp->rq_respages = &rqstp->rq_arg.pages[page_no+1];
407 rqstp->rq_next_page = rqstp->rq_respages + 1;
410 byte_count = head->arg.head[0].iov_len - position;
411 head->arg.page_len += byte_count;
412 head->arg.len += byte_count;
413 head->arg.buflen += byte_count;
417 static int rdma_read_chunks(struct svcxprt_rdma *xprt,
418 struct rpcrdma_msg *rmsgp,
419 struct svc_rqst *rqstp,
420 struct svc_rdma_op_ctxt *head)
423 struct rpcrdma_read_chunk *ch;
424 u32 handle, page_offset, byte_count;
429 /* If no read list is present, return 0 */
430 ch = svc_rdma_get_read_chunk(rmsgp);
434 if (rdma_rcl_chunk_count(ch) > RPCSVC_MAXPAGES)
437 /* The request is completed when the RDMA_READs complete. The
438 * head context keeps all the pages that comprise the
441 head->arg.head[0] = rqstp->rq_arg.head[0];
442 head->arg.tail[0] = rqstp->rq_arg.tail[0];
443 head->hdr_count = head->count;
444 head->arg.page_base = 0;
445 head->arg.page_len = 0;
446 head->arg.len = rqstp->rq_arg.len;
447 head->arg.buflen = rqstp->rq_arg.buflen;
449 ch = (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
450 position = be32_to_cpu(ch->rc_position);
452 /* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */
454 head->arg.pages = &head->pages[0];
455 page_offset = head->byte_len;
457 head->arg.pages = &head->pages[head->count];
463 for (; ch->rc_discrim != xdr_zero; ch++) {
464 if (be32_to_cpu(ch->rc_position) != position)
467 handle = be32_to_cpu(ch->rc_target.rs_handle),
468 byte_count = be32_to_cpu(ch->rc_target.rs_length);
469 xdr_decode_hyper((__be32 *)&ch->rc_target.rs_offset,
472 while (byte_count > 0) {
473 last = (ch + 1)->rc_discrim == xdr_zero;
474 ret = xprt->sc_reader(xprt, rqstp, head,
475 &page_no, &page_offset,
482 head->arg.buflen += ret;
486 /* Read list may need XDR round-up (see RFC 5666, s. 3.7) */
487 if (page_offset & 3) {
488 u32 pad = 4 - (page_offset & 3);
490 head->arg.page_len += pad;
491 head->arg.len += pad;
492 head->arg.buflen += pad;
497 if (position && position < head->arg.head[0].iov_len)
498 ret = rdma_copy_tail(rqstp, head, position,
499 byte_count, page_offset, page_no);
500 head->arg.head[0].iov_len = position;
501 head->position = position;
504 /* Detach arg pages. svc_recv will replenish them */
506 &rqstp->rq_pages[page_no] < rqstp->rq_respages; page_no++)
507 rqstp->rq_pages[page_no] = NULL;
512 static int rdma_read_complete(struct svc_rqst *rqstp,
513 struct svc_rdma_op_ctxt *head)
519 for (page_no = 0; page_no < head->count; page_no++) {
520 put_page(rqstp->rq_pages[page_no]);
521 rqstp->rq_pages[page_no] = head->pages[page_no];
524 /* Adjustments made for RDMA_NOMSG type requests */
525 if (head->position == 0) {
526 if (head->arg.len <= head->sge[0].length) {
527 head->arg.head[0].iov_len = head->arg.len -
529 head->arg.page_len = 0;
531 head->arg.head[0].iov_len = head->sge[0].length -
533 head->arg.page_len = head->arg.len -
538 /* Point rq_arg.pages past header */
539 rqstp->rq_arg.pages = &rqstp->rq_pages[head->hdr_count];
540 rqstp->rq_arg.page_len = head->arg.page_len;
541 rqstp->rq_arg.page_base = head->arg.page_base;
543 /* rq_respages starts after the last arg page */
544 rqstp->rq_respages = &rqstp->rq_arg.pages[page_no];
545 rqstp->rq_next_page = rqstp->rq_respages + 1;
547 /* Rebuild rq_arg head and tail. */
548 rqstp->rq_arg.head[0] = head->arg.head[0];
549 rqstp->rq_arg.tail[0] = head->arg.tail[0];
550 rqstp->rq_arg.len = head->arg.len;
551 rqstp->rq_arg.buflen = head->arg.buflen;
553 /* Free the context */
554 svc_rdma_put_context(head, 0);
556 /* XXX: What should this be? */
557 rqstp->rq_prot = IPPROTO_MAX;
558 svc_xprt_copy_addrs(rqstp, rqstp->rq_xprt);
560 ret = rqstp->rq_arg.head[0].iov_len
561 + rqstp->rq_arg.page_len
562 + rqstp->rq_arg.tail[0].iov_len;
563 dprintk("svcrdma: deferred read ret=%d, rq_arg.len=%u, "
564 "rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zu\n",
565 ret, rqstp->rq_arg.len, rqstp->rq_arg.head[0].iov_base,
566 rqstp->rq_arg.head[0].iov_len);
572 * Set up the rqstp thread context to point to the RQ buffer. If
573 * necessary, pull additional data from the client with an RDMA_READ
576 int svc_rdma_recvfrom(struct svc_rqst *rqstp)
578 struct svc_xprt *xprt = rqstp->rq_xprt;
579 struct svcxprt_rdma *rdma_xprt =
580 container_of(xprt, struct svcxprt_rdma, sc_xprt);
581 struct svc_rdma_op_ctxt *ctxt = NULL;
582 struct rpcrdma_msg *rmsgp;
586 dprintk("svcrdma: rqstp=%p\n", rqstp);
588 spin_lock_bh(&rdma_xprt->sc_rq_dto_lock);
589 if (!list_empty(&rdma_xprt->sc_read_complete_q)) {
590 ctxt = list_entry(rdma_xprt->sc_read_complete_q.next,
591 struct svc_rdma_op_ctxt,
593 list_del_init(&ctxt->dto_q);
594 spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
595 return rdma_read_complete(rqstp, ctxt);
596 } else if (!list_empty(&rdma_xprt->sc_rq_dto_q)) {
597 ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next,
598 struct svc_rdma_op_ctxt,
600 list_del_init(&ctxt->dto_q);
602 atomic_inc(&rdma_stat_rq_starve);
603 clear_bit(XPT_DATA, &xprt->xpt_flags);
606 spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
608 /* This is the EAGAIN path. The svc_recv routine will
609 * return -EAGAIN, the nfsd thread will go to call into
610 * svc_recv again and we shouldn't be on the active
613 if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
618 dprintk("svcrdma: processing ctxt=%p on xprt=%p, rqstp=%p, status=%d\n",
619 ctxt, rdma_xprt, rqstp, ctxt->wc_status);
620 atomic_inc(&rdma_stat_recv);
622 /* Build up the XDR from the receive buffers. */
623 rdma_build_arg_xdr(rqstp, ctxt, ctxt->byte_len);
625 /* Decode the RDMA header. */
626 len = svc_rdma_xdr_decode_req(&rmsgp, rqstp);
627 rqstp->rq_xprt_hlen = len;
629 /* If the request is invalid, reply with an error */
632 svc_rdma_send_error(rdma_xprt, rmsgp, ERR_VERS);
636 /* Read read-list data. */
637 ret = rdma_read_chunks(rdma_xprt, rmsgp, rqstp, ctxt);
639 /* read-list posted, defer until data received from client. */
641 } else if (ret < 0) {
642 /* Post of read-list failed, free context. */
643 svc_rdma_put_context(ctxt, 1);
647 ret = rqstp->rq_arg.head[0].iov_len
648 + rqstp->rq_arg.page_len
649 + rqstp->rq_arg.tail[0].iov_len;
650 svc_rdma_put_context(ctxt, 0);
652 dprintk("svcrdma: ret=%d, rq_arg.len=%u, "
653 "rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zd\n",
654 ret, rqstp->rq_arg.len,
655 rqstp->rq_arg.head[0].iov_base,
656 rqstp->rq_arg.head[0].iov_len);
657 rqstp->rq_prot = IPPROTO_MAX;
658 svc_xprt_copy_addrs(rqstp, xprt);
663 svc_rdma_put_context(ctxt, 1);
664 dprintk("svcrdma: transport %p is closing\n", xprt);
666 * Set the close bit and enqueue it. svc_recv will see the
667 * close bit and call svc_xprt_delete
669 set_bit(XPT_CLOSE, &xprt->xpt_flags);