diff options
Diffstat (limited to 'net/sunrpc/xprtrdma/svc_rdma_recvfrom.c')
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 891 |
1 files changed, 534 insertions, 357 deletions
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c index 828b149eaaef..e7e4a39ca6c6 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c @@ -89,12 +89,12 @@ * svc_rdma_recvfrom call returns. * * During the second svc_rdma_recvfrom call, RDMA Read sink pages - * are transferred from the svc_rdma_recv_ctxt to the second svc_rqst - * (see rdma_read_complete() below). + * are transferred from the svc_rdma_recv_ctxt to the second svc_rqst. */ +#include <linux/slab.h> #include <linux/spinlock.h> -#include <asm/unaligned.h> +#include <linux/unaligned.h> #include <rdma/ib_verbs.h> #include <rdma/rdma_cm.h> @@ -106,8 +106,6 @@ #include "xprt_rdma.h" #include <trace/events/rpcrdma.h> -#define RPCDBG_FACILITY RPCDBG_SVCXPRT - static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc); static inline struct svc_rdma_recv_ctxt * @@ -120,14 +118,19 @@ svc_rdma_next_recv_ctxt(struct list_head *list) static struct svc_rdma_recv_ctxt * svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma) { + int node = ibdev_to_node(rdma->sc_cm_id->device); struct svc_rdma_recv_ctxt *ctxt; + unsigned long pages; dma_addr_t addr; void *buffer; - ctxt = kmalloc(sizeof(*ctxt), GFP_KERNEL); + pages = svc_serv_maxpages(rdma->sc_xprt.xpt_server); + ctxt = kzalloc_node(struct_size(ctxt, rc_pages, pages), + GFP_KERNEL, node); if (!ctxt) goto fail0; - buffer = kmalloc(rdma->sc_max_req_size, GFP_KERNEL); + ctxt->rc_maxpages = pages; + buffer = kmalloc_node(rdma->sc_max_req_size, GFP_KERNEL, node); if (!buffer) goto fail1; addr = ib_dma_map_single(rdma->sc_pd->device, buffer, @@ -135,6 +138,12 @@ svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma) if (ib_dma_mapping_error(rdma->sc_pd->device, addr)) goto fail2; + svc_rdma_recv_cid_init(rdma, &ctxt->rc_cid); + pcl_init(&ctxt->rc_call_pcl); + pcl_init(&ctxt->rc_read_pcl); + pcl_init(&ctxt->rc_write_pcl); + pcl_init(&ctxt->rc_reply_pcl); + ctxt->rc_recv_wr.next = NULL; ctxt->rc_recv_wr.wr_cqe = &ctxt->rc_cqe; ctxt->rc_recv_wr.sg_list = &ctxt->rc_recv_sge; @@ -144,7 +153,7 @@ svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma) ctxt->rc_recv_sge.length = rdma->sc_max_req_size; ctxt->rc_recv_sge.lkey = rdma->sc_pd->local_dma_lkey; ctxt->rc_recv_buf = buffer; - ctxt->rc_temp = false; + svc_rdma_cc_init(rdma, &ctxt->rc_cc); return ctxt; fail2: @@ -172,36 +181,32 @@ static void svc_rdma_recv_ctxt_destroy(struct svcxprt_rdma *rdma, void svc_rdma_recv_ctxts_destroy(struct svcxprt_rdma *rdma) { struct svc_rdma_recv_ctxt *ctxt; + struct llist_node *node; - while ((ctxt = svc_rdma_next_recv_ctxt(&rdma->sc_recv_ctxts))) { - list_del(&ctxt->rc_list); + while ((node = llist_del_first(&rdma->sc_recv_ctxts))) { + ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node); svc_rdma_recv_ctxt_destroy(rdma, ctxt); } } -static struct svc_rdma_recv_ctxt * -svc_rdma_recv_ctxt_get(struct svcxprt_rdma *rdma) +/** + * svc_rdma_recv_ctxt_get - Allocate a recv_ctxt + * @rdma: controlling svcxprt_rdma + * + * Returns a recv_ctxt or (rarely) NULL if none are available. + */ +struct svc_rdma_recv_ctxt *svc_rdma_recv_ctxt_get(struct svcxprt_rdma *rdma) { struct svc_rdma_recv_ctxt *ctxt; + struct llist_node *node; - spin_lock(&rdma->sc_recv_lock); - ctxt = svc_rdma_next_recv_ctxt(&rdma->sc_recv_ctxts); - if (!ctxt) - goto out_empty; - list_del(&ctxt->rc_list); - spin_unlock(&rdma->sc_recv_lock); + node = llist_del_first(&rdma->sc_recv_ctxts); + if (!node) + return NULL; -out: + ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node); ctxt->rc_page_count = 0; return ctxt; - -out_empty: - spin_unlock(&rdma->sc_recv_lock); - - ctxt = svc_rdma_recv_ctxt_alloc(rdma); - if (!ctxt) - return NULL; - goto out; } /** @@ -213,72 +218,109 @@ out_empty: void svc_rdma_recv_ctxt_put(struct svcxprt_rdma *rdma, struct svc_rdma_recv_ctxt *ctxt) { - unsigned int i; + svc_rdma_cc_release(rdma, &ctxt->rc_cc, DMA_FROM_DEVICE); - for (i = 0; i < ctxt->rc_page_count; i++) - put_page(ctxt->rc_pages[i]); + /* @rc_page_count is normally zero here, but error flows + * can leave pages in @rc_pages. + */ + release_pages(ctxt->rc_pages, ctxt->rc_page_count); - if (!ctxt->rc_temp) { - spin_lock(&rdma->sc_recv_lock); - list_add(&ctxt->rc_list, &rdma->sc_recv_ctxts); - spin_unlock(&rdma->sc_recv_lock); - } else - svc_rdma_recv_ctxt_destroy(rdma, ctxt); + pcl_free(&ctxt->rc_call_pcl); + pcl_free(&ctxt->rc_read_pcl); + pcl_free(&ctxt->rc_write_pcl); + pcl_free(&ctxt->rc_reply_pcl); + + llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts); } -static int __svc_rdma_post_recv(struct svcxprt_rdma *rdma, - struct svc_rdma_recv_ctxt *ctxt) +/** + * svc_rdma_release_ctxt - Release transport-specific per-rqst resources + * @xprt: the transport which owned the context + * @vctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt + * + * Ensure that the recv_ctxt is released whether or not a Reply + * was sent. For example, the client could close the connection, + * or svc_process could drop an RPC, before the Reply is sent. + */ +void svc_rdma_release_ctxt(struct svc_xprt *xprt, void *vctxt) { - int ret; - - svc_xprt_get(&rdma->sc_xprt); - ret = ib_post_recv(rdma->sc_qp, &ctxt->rc_recv_wr, NULL); - trace_svcrdma_post_recv(&ctxt->rc_recv_wr, ret); - if (ret) - goto err_post; - return 0; + struct svc_rdma_recv_ctxt *ctxt = vctxt; + struct svcxprt_rdma *rdma = + container_of(xprt, struct svcxprt_rdma, sc_xprt); -err_post: - svc_rdma_recv_ctxt_put(rdma, ctxt); - svc_xprt_put(&rdma->sc_xprt); - return ret; + if (ctxt) + svc_rdma_recv_ctxt_put(rdma, ctxt); } -static int svc_rdma_post_recv(struct svcxprt_rdma *rdma) +static bool svc_rdma_refresh_recvs(struct svcxprt_rdma *rdma, + unsigned int wanted) { + const struct ib_recv_wr *bad_wr = NULL; struct svc_rdma_recv_ctxt *ctxt; + struct ib_recv_wr *recv_chain; + int ret; - ctxt = svc_rdma_recv_ctxt_get(rdma); - if (!ctxt) - return -ENOMEM; - return __svc_rdma_post_recv(rdma, ctxt); + if (test_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags)) + return false; + + recv_chain = NULL; + while (wanted--) { + ctxt = svc_rdma_recv_ctxt_get(rdma); + if (!ctxt) + break; + + trace_svcrdma_post_recv(&ctxt->rc_cid); + ctxt->rc_recv_wr.next = recv_chain; + recv_chain = &ctxt->rc_recv_wr; + rdma->sc_pending_recvs++; + } + if (!recv_chain) + return true; + + ret = ib_post_recv(rdma->sc_qp, recv_chain, &bad_wr); + if (ret) + goto err_free; + return true; + +err_free: + trace_svcrdma_rq_post_err(rdma, ret); + while (bad_wr) { + ctxt = container_of(bad_wr, struct svc_rdma_recv_ctxt, + rc_recv_wr); + bad_wr = bad_wr->next; + svc_rdma_recv_ctxt_put(rdma, ctxt); + } + /* Since we're destroying the xprt, no need to reset + * sc_pending_recvs. */ + return false; } /** * svc_rdma_post_recvs - Post initial set of Recv WRs * @rdma: fresh svcxprt_rdma * - * Returns true if successful, otherwise false. + * Return values: + * %true: Receive Queue initialization successful + * %false: memory allocation or DMA error */ bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma) { - struct svc_rdma_recv_ctxt *ctxt; - unsigned int i; - int ret; + unsigned int total; - for (i = 0; i < rdma->sc_max_requests; i++) { - ctxt = svc_rdma_recv_ctxt_get(rdma); + /* For each credit, allocate enough recv_ctxts for one + * posted Receive and one RPC in process. + */ + total = (rdma->sc_max_requests * 2) + rdma->sc_recv_batch; + while (total--) { + struct svc_rdma_recv_ctxt *ctxt; + + ctxt = svc_rdma_recv_ctxt_alloc(rdma); if (!ctxt) return false; - ctxt->rc_temp = true; - ret = __svc_rdma_post_recv(rdma, ctxt); - if (ret) { - pr_err("svcrdma: failure posting recv buffers: %d\n", - ret); - return false; - } + llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts); } - return true; + + return svc_rdma_refresh_recvs(rdma, rdma->sc_max_requests); } /** @@ -286,8 +328,6 @@ bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma) * @cq: Completion Queue context * @wc: Work Completion object * - * NB: The svc_xprt/svcxprt_rdma is pinned whenever it's possible that - * the Receive completion handler could be running. */ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc) { @@ -295,42 +335,48 @@ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc) struct ib_cqe *cqe = wc->wr_cqe; struct svc_rdma_recv_ctxt *ctxt; - trace_svcrdma_wc_receive(wc); + rdma->sc_pending_recvs--; /* WARNING: Only wc->wr_cqe and wc->status are reliable */ ctxt = container_of(cqe, struct svc_rdma_recv_ctxt, rc_cqe); if (wc->status != IB_WC_SUCCESS) goto flushed; - - if (svc_rdma_post_recv(rdma)) - goto post_err; + trace_svcrdma_wc_recv(wc, &ctxt->rc_cid); + + /* If receive posting fails, the connection is about to be + * lost anyway. The server will not be able to send a reply + * for this RPC, and the client will retransmit this RPC + * anyway when it reconnects. + * + * Therefore we drop the Receive, even if status was SUCCESS + * to reduce the likelihood of replayed requests once the + * client reconnects. + */ + if (rdma->sc_pending_recvs < rdma->sc_max_requests) + if (!svc_rdma_refresh_recvs(rdma, rdma->sc_recv_batch)) + goto dropped; /* All wc fields are now known to be valid */ ctxt->rc_byte_len = wc->byte_len; - ib_dma_sync_single_for_cpu(rdma->sc_pd->device, - ctxt->rc_recv_sge.addr, - wc->byte_len, DMA_FROM_DEVICE); spin_lock(&rdma->sc_rq_dto_lock); list_add_tail(&ctxt->rc_list, &rdma->sc_rq_dto_q); - spin_unlock(&rdma->sc_rq_dto_lock); + /* Note the unlock pairs with the smp_rmb in svc_xprt_ready: */ set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags); + spin_unlock(&rdma->sc_rq_dto_lock); if (!test_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags)) svc_xprt_enqueue(&rdma->sc_xprt); - goto out; + return; flushed: - if (wc->status != IB_WC_WR_FLUSH_ERR) - pr_err("svcrdma: Recv: %s (%u/0x%x)\n", - ib_wc_status_msg(wc->status), - wc->status, wc->vendor_err); -post_err: + if (wc->status == IB_WC_WR_FLUSH_ERR) + trace_svcrdma_wc_recv_flush(wc, &ctxt->rc_cid); + else + trace_svcrdma_wc_recv_err(wc, &ctxt->rc_cid); +dropped: svc_rdma_recv_ctxt_put(rdma, ctxt); - set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); - svc_xprt_enqueue(&rdma->sc_xprt); -out: - svc_xprt_put(&rdma->sc_xprt); + svc_xprt_deferred_close(&rdma->sc_xprt); } /** @@ -367,122 +413,189 @@ static void svc_rdma_build_arg_xdr(struct svc_rqst *rqstp, arg->len = ctxt->rc_byte_len; } -/* This accommodates the largest possible Write chunk, - * in one segment. +/** + * xdr_count_read_segments - Count number of Read segments in Read list + * @rctxt: Ingress receive context + * @p: Start of an un-decoded Read list + * + * Before allocating anything, ensure the ingress Read list is safe + * to use. + * + * The segment count is limited to how many segments can fit in the + * transport header without overflowing the buffer. That's about 40 + * Read segments for a 1KB inline threshold. + * + * Return values: + * %true: Read list is valid. @rctxt's xdr_stream is updated to point + * to the first byte past the Read list. rc_read_pcl and + * rc_call_pcl cl_count fields are set to the number of + * Read segments in the list. + * %false: Read list is corrupt. @rctxt's xdr_stream is left in an + * unknown state. */ -#define MAX_BYTES_WRITE_SEG ((u32)(RPCSVC_MAXPAGES << PAGE_SHIFT)) +static bool xdr_count_read_segments(struct svc_rdma_recv_ctxt *rctxt, __be32 *p) +{ + rctxt->rc_call_pcl.cl_count = 0; + rctxt->rc_read_pcl.cl_count = 0; + while (xdr_item_is_present(p)) { + u32 position, handle, length; + u64 offset; + + p = xdr_inline_decode(&rctxt->rc_stream, + rpcrdma_readseg_maxsz * sizeof(*p)); + if (!p) + return false; -/* This accommodates the largest possible Position-Zero - * Read chunk or Reply chunk, in one segment. - */ -#define MAX_BYTES_SPECIAL_SEG ((u32)((RPCSVC_MAXPAGES + 2) << PAGE_SHIFT)) + xdr_decode_read_segment(p, &position, &handle, + &length, &offset); + if (position) { + if (position & 3) + return false; + ++rctxt->rc_read_pcl.cl_count; + } else { + ++rctxt->rc_call_pcl.cl_count; + } + + p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); + if (!p) + return false; + } + return true; +} /* Sanity check the Read list. * - * Implementation limits: - * - This implementation supports only one Read chunk. - * * Sanity checks: - * - Read list does not overflow buffer. - * - Segment size limited by largest NFS data payload. - * - * The segment count is limited to how many segments can - * fit in the transport header without overflowing the - * buffer. That's about 40 Read segments for a 1KB inline - * threshold. - * - * Returns pointer to the following Write list. + * - Read list does not overflow Receive buffer. + * - Chunk size limited by largest NFS data payload. + * + * Return values: + * %true: Read list is valid. @rctxt's xdr_stream is updated + * to point to the first byte past the Read list. + * %false: Read list is corrupt. @rctxt's xdr_stream is left + * in an unknown state. */ -static __be32 *xdr_check_read_list(__be32 *p, const __be32 *end) +static bool xdr_check_read_list(struct svc_rdma_recv_ctxt *rctxt) { - u32 position; - bool first; - - first = true; - while (*p++ != xdr_zero) { - if (first) { - position = be32_to_cpup(p++); - first = false; - } else if (be32_to_cpup(p++) != position) { - return NULL; - } - p++; /* handle */ - if (be32_to_cpup(p++) > MAX_BYTES_SPECIAL_SEG) - return NULL; - p += 2; /* offset */ + __be32 *p; - if (p > end) - return NULL; - } - return p; + p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); + if (!p) + return false; + if (!xdr_count_read_segments(rctxt, p)) + return false; + if (!pcl_alloc_call(rctxt, p)) + return false; + return pcl_alloc_read(rctxt, p); } -/* The segment count is limited to how many segments can - * fit in the transport header without overflowing the - * buffer. That's about 60 Write segments for a 1KB inline - * threshold. - */ -static __be32 *xdr_check_write_chunk(__be32 *p, const __be32 *end, - u32 maxlen) +static bool xdr_check_write_chunk(struct svc_rdma_recv_ctxt *rctxt) { - u32 i, segcount; + u32 segcount; + __be32 *p; - segcount = be32_to_cpup(p++); - for (i = 0; i < segcount; i++) { - p++; /* handle */ - if (be32_to_cpup(p++) > maxlen) - return NULL; - p += 2; /* offset */ + if (xdr_stream_decode_u32(&rctxt->rc_stream, &segcount)) + return false; - if (p > end) - return NULL; - } + /* Before trusting the segcount value enough to use it in + * a computation, perform a simple range check. This is an + * arbitrary but sensible limit (ie, not architectural). + */ + if (unlikely(segcount > rctxt->rc_maxpages)) + return false; - return p; + p = xdr_inline_decode(&rctxt->rc_stream, + segcount * rpcrdma_segment_maxsz * sizeof(*p)); + return p != NULL; +} + +/** + * xdr_count_write_chunks - Count number of Write chunks in Write list + * @rctxt: Received header and decoding state + * @p: start of an un-decoded Write list + * + * Before allocating anything, ensure the ingress Write list is + * safe to use. + * + * Return values: + * %true: Write list is valid. @rctxt's xdr_stream is updated + * to point to the first byte past the Write list, and + * the number of Write chunks is in rc_write_pcl.cl_count. + * %false: Write list is corrupt. @rctxt's xdr_stream is left + * in an indeterminate state. + */ +static bool xdr_count_write_chunks(struct svc_rdma_recv_ctxt *rctxt, __be32 *p) +{ + rctxt->rc_write_pcl.cl_count = 0; + while (xdr_item_is_present(p)) { + if (!xdr_check_write_chunk(rctxt)) + return false; + ++rctxt->rc_write_pcl.cl_count; + p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); + if (!p) + return false; + } + return true; } /* Sanity check the Write list. * * Implementation limits: - * - This implementation supports only one Write chunk. + * - This implementation currently supports only one Write chunk. * * Sanity checks: - * - Write list does not overflow buffer. - * - Segment size limited by largest NFS data payload. - * - * Returns pointer to the following Reply chunk. + * - Write list does not overflow Receive buffer. + * - Chunk size limited by largest NFS data payload. + * + * Return values: + * %true: Write list is valid. @rctxt's xdr_stream is updated + * to point to the first byte past the Write list. + * %false: Write list is corrupt. @rctxt's xdr_stream is left + * in an unknown state. */ -static __be32 *xdr_check_write_list(__be32 *p, const __be32 *end) +static bool xdr_check_write_list(struct svc_rdma_recv_ctxt *rctxt) { - u32 chcount; + __be32 *p; - chcount = 0; - while (*p++ != xdr_zero) { - p = xdr_check_write_chunk(p, end, MAX_BYTES_WRITE_SEG); - if (!p) - return NULL; - if (chcount++ > 1) - return NULL; - } - return p; + p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); + if (!p) + return false; + if (!xdr_count_write_chunks(rctxt, p)) + return false; + if (!pcl_alloc_write(rctxt, &rctxt->rc_write_pcl, p)) + return false; + + rctxt->rc_cur_result_payload = pcl_first_chunk(&rctxt->rc_write_pcl); + return true; } /* Sanity check the Reply chunk. * * Sanity checks: - * - Reply chunk does not overflow buffer. - * - Segment size limited by largest NFS data payload. - * - * Returns pointer to the following RPC header. + * - Reply chunk does not overflow Receive buffer. + * - Chunk size limited by largest NFS data payload. + * + * Return values: + * %true: Reply chunk is valid. @rctxt's xdr_stream is updated + * to point to the first byte past the Reply chunk. + * %false: Reply chunk is corrupt. @rctxt's xdr_stream is left + * in an unknown state. */ -static __be32 *xdr_check_reply_chunk(__be32 *p, const __be32 *end) +static bool xdr_check_reply_chunk(struct svc_rdma_recv_ctxt *rctxt) { - if (*p++ != xdr_zero) { - p = xdr_check_write_chunk(p, end, MAX_BYTES_SPECIAL_SEG); - if (!p) - return NULL; - } - return p; + __be32 *p; + + p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); + if (!p) + return false; + + if (!xdr_item_is_present(p)) + return true; + if (!xdr_check_write_chunk(rctxt)) + return false; + + rctxt->rc_reply_pcl.cl_count = 1; + return pcl_alloc_write(rctxt, &rctxt->rc_reply_pcl, p); } /* RPC-over-RDMA Version One private extension: Remote Invalidation. @@ -491,206 +604,147 @@ static __be32 *xdr_check_reply_chunk(__be32 *p, const __be32 *end) * * If there is exactly one distinct R_key in the received transport * header, set rc_inv_rkey to that R_key. Otherwise, set it to zero. - * - * Perform this operation while the received transport header is - * still in the CPU cache. */ static void svc_rdma_get_inv_rkey(struct svcxprt_rdma *rdma, struct svc_rdma_recv_ctxt *ctxt) { - __be32 inv_rkey, *p; - u32 i, segcount; + struct svc_rdma_segment *segment; + struct svc_rdma_chunk *chunk; + u32 inv_rkey; ctxt->rc_inv_rkey = 0; if (!rdma->sc_snd_w_inv) return; - inv_rkey = xdr_zero; - p = ctxt->rc_recv_buf; - p += rpcrdma_fixed_maxsz; - - /* Read list */ - while (*p++ != xdr_zero) { - p++; /* position */ - if (inv_rkey == xdr_zero) - inv_rkey = *p; - else if (inv_rkey != *p) - return; - p += 4; + inv_rkey = 0; + pcl_for_each_chunk(chunk, &ctxt->rc_call_pcl) { + pcl_for_each_segment(segment, chunk) { + if (inv_rkey == 0) + inv_rkey = segment->rs_handle; + else if (inv_rkey != segment->rs_handle) + return; + } } - - /* Write list */ - while (*p++ != xdr_zero) { - segcount = be32_to_cpup(p++); - for (i = 0; i < segcount; i++) { - if (inv_rkey == xdr_zero) - inv_rkey = *p; - else if (inv_rkey != *p) + pcl_for_each_chunk(chunk, &ctxt->rc_read_pcl) { + pcl_for_each_segment(segment, chunk) { + if (inv_rkey == 0) + inv_rkey = segment->rs_handle; + else if (inv_rkey != segment->rs_handle) return; - p += 4; } } - - /* Reply chunk */ - if (*p++ != xdr_zero) { - segcount = be32_to_cpup(p++); - for (i = 0; i < segcount; i++) { - if (inv_rkey == xdr_zero) - inv_rkey = *p; - else if (inv_rkey != *p) + pcl_for_each_chunk(chunk, &ctxt->rc_write_pcl) { + pcl_for_each_segment(segment, chunk) { + if (inv_rkey == 0) + inv_rkey = segment->rs_handle; + else if (inv_rkey != segment->rs_handle) return; - p += 4; } } - - ctxt->rc_inv_rkey = be32_to_cpu(inv_rkey); + pcl_for_each_chunk(chunk, &ctxt->rc_reply_pcl) { + pcl_for_each_segment(segment, chunk) { + if (inv_rkey == 0) + inv_rkey = segment->rs_handle; + else if (inv_rkey != segment->rs_handle) + return; + } + } + ctxt->rc_inv_rkey = inv_rkey; } -/* On entry, xdr->head[0].iov_base points to first byte in the - * RPC-over-RDMA header. +/** + * svc_rdma_xdr_decode_req - Decode the transport header + * @rq_arg: xdr_buf containing ingress RPC/RDMA message + * @rctxt: state of decoding + * + * On entry, xdr->head[0].iov_base points to first byte of the + * RPC-over-RDMA transport header. * * On successful exit, head[0] points to first byte past the * RPC-over-RDMA header. For RDMA_MSG, this is the RPC message. + * * The length of the RPC-over-RDMA header is returned. * * Assumptions: * - The transport header is entirely contained in the head iovec. */ -static int svc_rdma_xdr_decode_req(struct xdr_buf *rq_arg) +static int svc_rdma_xdr_decode_req(struct xdr_buf *rq_arg, + struct svc_rdma_recv_ctxt *rctxt) { - __be32 *p, *end, *rdma_argp; + __be32 *p, *rdma_argp; unsigned int hdr_len; - /* Verify that there's enough bytes for header + something */ - if (rq_arg->len <= RPCRDMA_HDRLEN_ERR) - goto out_short; - rdma_argp = rq_arg->head[0].iov_base; - if (*(rdma_argp + 1) != rpcrdma_version) - goto out_version; + xdr_init_decode(&rctxt->rc_stream, rq_arg, rdma_argp, NULL); - switch (*(rdma_argp + 3)) { + p = xdr_inline_decode(&rctxt->rc_stream, + rpcrdma_fixed_maxsz * sizeof(*p)); + if (unlikely(!p)) + goto out_short; + p++; + if (*p != rpcrdma_version) + goto out_version; + p += 2; + rctxt->rc_msgtype = *p; + switch (rctxt->rc_msgtype) { case rdma_msg: break; case rdma_nomsg: break; - case rdma_done: goto out_drop; - case rdma_error: goto out_drop; - default: goto out_proc; } - end = (__be32 *)((unsigned long)rdma_argp + rq_arg->len); - p = xdr_check_read_list(rdma_argp + 4, end); - if (!p) + if (!xdr_check_read_list(rctxt)) goto out_inval; - p = xdr_check_write_list(p, end); - if (!p) - goto out_inval; - p = xdr_check_reply_chunk(p, end); - if (!p) + if (!xdr_check_write_list(rctxt)) goto out_inval; - if (p > end) + if (!xdr_check_reply_chunk(rctxt)) goto out_inval; - rq_arg->head[0].iov_base = p; - hdr_len = (unsigned long)p - (unsigned long)rdma_argp; + rq_arg->head[0].iov_base = rctxt->rc_stream.p; + hdr_len = xdr_stream_pos(&rctxt->rc_stream); rq_arg->head[0].iov_len -= hdr_len; rq_arg->len -= hdr_len; - trace_svcrdma_decode_rqst(rdma_argp, hdr_len); + trace_svcrdma_decode_rqst(rctxt, rdma_argp, hdr_len); return hdr_len; out_short: - trace_svcrdma_decode_short(rq_arg->len); + trace_svcrdma_decode_short_err(rctxt, rq_arg->len); return -EINVAL; out_version: - trace_svcrdma_decode_badvers(rdma_argp); + trace_svcrdma_decode_badvers_err(rctxt, rdma_argp); return -EPROTONOSUPPORT; out_drop: - trace_svcrdma_decode_drop(rdma_argp); + trace_svcrdma_decode_drop_err(rctxt, rdma_argp); return 0; out_proc: - trace_svcrdma_decode_badproc(rdma_argp); + trace_svcrdma_decode_badproc_err(rctxt, rdma_argp); return -EINVAL; out_inval: - trace_svcrdma_decode_parse(rdma_argp); + trace_svcrdma_decode_parse_err(rctxt, rdma_argp); return -EINVAL; } -static void rdma_read_complete(struct svc_rqst *rqstp, - struct svc_rdma_recv_ctxt *head) +static void svc_rdma_send_error(struct svcxprt_rdma *rdma, + struct svc_rdma_recv_ctxt *rctxt, + int status) { - int page_no; - - /* Move Read chunk pages to rqstp so that they will be released - * when svc_process is done with them. - */ - for (page_no = 0; page_no < head->rc_page_count; page_no++) { - put_page(rqstp->rq_pages[page_no]); - rqstp->rq_pages[page_no] = head->rc_pages[page_no]; - } - head->rc_page_count = 0; - - /* Point rq_arg.pages past header */ - rqstp->rq_arg.pages = &rqstp->rq_pages[head->rc_hdr_count]; - rqstp->rq_arg.page_len = head->rc_arg.page_len; - - /* rq_respages starts after the last arg page */ - rqstp->rq_respages = &rqstp->rq_pages[page_no]; - rqstp->rq_next_page = rqstp->rq_respages + 1; - - /* Rebuild rq_arg head and tail. */ - rqstp->rq_arg.head[0] = head->rc_arg.head[0]; - rqstp->rq_arg.tail[0] = head->rc_arg.tail[0]; - rqstp->rq_arg.len = head->rc_arg.len; - rqstp->rq_arg.buflen = head->rc_arg.buflen; -} + struct svc_rdma_send_ctxt *sctxt; -static void svc_rdma_send_error(struct svcxprt_rdma *xprt, - __be32 *rdma_argp, int status) -{ - struct svc_rdma_send_ctxt *ctxt; - unsigned int length; - __be32 *p; - int ret; - - ctxt = svc_rdma_send_ctxt_get(xprt); - if (!ctxt) + sctxt = svc_rdma_send_ctxt_get(rdma); + if (!sctxt) return; - - p = ctxt->sc_xprt_buf; - *p++ = *rdma_argp; - *p++ = *(rdma_argp + 1); - *p++ = xprt->sc_fc_credits; - *p++ = rdma_error; - switch (status) { - case -EPROTONOSUPPORT: - *p++ = err_vers; - *p++ = rpcrdma_version; - *p++ = rpcrdma_version; - trace_svcrdma_err_vers(*rdma_argp); - break; - default: - *p++ = err_chunk; - trace_svcrdma_err_chunk(*rdma_argp); - } - length = (unsigned long)p - (unsigned long)ctxt->sc_xprt_buf; - svc_rdma_sync_reply_hdr(xprt, ctxt, length); - - ctxt->sc_send_wr.opcode = IB_WR_SEND; - ret = svc_rdma_send(xprt, &ctxt->sc_send_wr); - if (ret) - svc_rdma_send_ctxt_put(xprt, ctxt); + svc_rdma_send_error_msg(rdma, sctxt, rctxt, status); } /* By convention, backchannel calls arrive via rdma_msg type @@ -698,35 +752,149 @@ static void svc_rdma_send_error(struct svcxprt_rdma *xprt, * the RPC/RDMA header small and fixed in size, so it is * straightforward to check the RPC header's direction field. */ -static bool svc_rdma_is_backchannel_reply(struct svc_xprt *xprt, - __be32 *rdma_resp) +static bool svc_rdma_is_reverse_direction_reply(struct svc_xprt *xprt, + struct svc_rdma_recv_ctxt *rctxt) { - __be32 *p; + __be32 *p = rctxt->rc_recv_buf; if (!xprt->xpt_bc_xprt) return false; - p = rdma_resp + 3; - if (*p++ != rdma_msg) + if (rctxt->rc_msgtype != rdma_msg) return false; - if (*p++ != xdr_zero) + if (!pcl_is_empty(&rctxt->rc_call_pcl)) return false; - if (*p++ != xdr_zero) + if (!pcl_is_empty(&rctxt->rc_read_pcl)) return false; - if (*p++ != xdr_zero) + if (!pcl_is_empty(&rctxt->rc_write_pcl)) return false; - - /* XID sanity */ - if (*p++ != *rdma_resp) + if (!pcl_is_empty(&rctxt->rc_reply_pcl)) return false; - /* call direction */ - if (*p == cpu_to_be32(RPC_CALL)) + + /* RPC call direction */ + if (*(p + 8) == cpu_to_be32(RPC_CALL)) return false; return true; } +/* Finish constructing the RPC Call message in rqstp::rq_arg. + * + * The incoming RPC/RDMA message is an RDMA_MSG type message + * with a single Read chunk (only the upper layer data payload + * was conveyed via RDMA Read). + */ +static void svc_rdma_read_complete_one(struct svc_rqst *rqstp, + struct svc_rdma_recv_ctxt *ctxt) +{ + struct svc_rdma_chunk *chunk = pcl_first_chunk(&ctxt->rc_read_pcl); + struct xdr_buf *buf = &rqstp->rq_arg; + unsigned int length; + + /* Split the Receive buffer between the head and tail + * buffers at Read chunk's position. XDR roundup of the + * chunk is not included in either the pagelist or in + * the tail. + */ + buf->tail[0].iov_base = buf->head[0].iov_base + chunk->ch_position; + buf->tail[0].iov_len = buf->head[0].iov_len - chunk->ch_position; + buf->head[0].iov_len = chunk->ch_position; + + /* Read chunk may need XDR roundup (see RFC 8166, s. 3.4.5.2). + * + * If the client already rounded up the chunk length, the + * length does not change. Otherwise, the length of the page + * list is increased to include XDR round-up. + * + * Currently these chunks always start at page offset 0, + * thus the rounded-up length never crosses a page boundary. + */ + buf->pages = &rqstp->rq_pages[0]; + length = xdr_align_size(chunk->ch_length); + buf->page_len = length; + buf->len += length; + buf->buflen += length; +} + +/* Finish constructing the RPC Call message in rqstp::rq_arg. + * + * The incoming RPC/RDMA message is an RDMA_MSG type message + * with payload in multiple Read chunks and no PZRC. + */ +static void svc_rdma_read_complete_multiple(struct svc_rqst *rqstp, + struct svc_rdma_recv_ctxt *ctxt) +{ + struct xdr_buf *buf = &rqstp->rq_arg; + + buf->len += ctxt->rc_readbytes; + buf->buflen += ctxt->rc_readbytes; + + buf->head[0].iov_base = page_address(rqstp->rq_pages[0]); + buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, ctxt->rc_readbytes); + buf->pages = &rqstp->rq_pages[1]; + buf->page_len = ctxt->rc_readbytes - buf->head[0].iov_len; +} + +/* Finish constructing the RPC Call message in rqstp::rq_arg. + * + * The incoming RPC/RDMA message is an RDMA_NOMSG type message + * (the RPC message body was conveyed via RDMA Read). + */ +static void svc_rdma_read_complete_pzrc(struct svc_rqst *rqstp, + struct svc_rdma_recv_ctxt *ctxt) +{ + struct xdr_buf *buf = &rqstp->rq_arg; + + buf->len += ctxt->rc_readbytes; + buf->buflen += ctxt->rc_readbytes; + + buf->head[0].iov_base = page_address(rqstp->rq_pages[0]); + buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, ctxt->rc_readbytes); + buf->pages = &rqstp->rq_pages[1]; + buf->page_len = ctxt->rc_readbytes - buf->head[0].iov_len; +} + +static noinline void svc_rdma_read_complete(struct svc_rqst *rqstp, + struct svc_rdma_recv_ctxt *ctxt) +{ + unsigned int i; + + /* Transfer the Read chunk pages into @rqstp.rq_pages, replacing + * the rq_pages that were already allocated for this rqstp. + */ + release_pages(rqstp->rq_respages, ctxt->rc_page_count); + for (i = 0; i < ctxt->rc_page_count; i++) + rqstp->rq_pages[i] = ctxt->rc_pages[i]; + + /* Update @rqstp's result send buffer to start after the + * last page in the RDMA Read payload. + */ + rqstp->rq_respages = &rqstp->rq_pages[ctxt->rc_page_count]; + rqstp->rq_next_page = rqstp->rq_respages + 1; + + /* Prevent svc_rdma_recv_ctxt_put() from releasing the + * pages in ctxt::rc_pages a second time. + */ + ctxt->rc_page_count = 0; + + /* Finish constructing the RPC Call message. The exact + * procedure for that depends on what kind of RPC/RDMA + * chunks were provided by the client. + */ + rqstp->rq_arg = ctxt->rc_saved_arg; + if (pcl_is_empty(&ctxt->rc_call_pcl)) { + if (ctxt->rc_read_pcl.cl_count == 1) + svc_rdma_read_complete_one(rqstp, ctxt); + else + svc_rdma_read_complete_multiple(rqstp, ctxt); + } else { + svc_rdma_read_complete_pzrc(rqstp, ctxt); + } + + trace_svcrdma_read_finished(&ctxt->rc_cid); +} + /** * svc_rdma_recvfrom - Receive an RPC call * @rqstp: request structure into which to receive an RPC Call @@ -744,9 +912,6 @@ static bool svc_rdma_is_backchannel_reply(struct svc_xprt *xprt, * * The next ctxt is removed from the "receive" lists. * - * - If the ctxt completes a Read, then finish assembling the Call - * message and return the number of bytes in the message. - * * - If the ctxt completes a Receive, then construct the Call * message from the contents of the Receive buffer. * @@ -755,7 +920,8 @@ static bool svc_rdma_is_backchannel_reply(struct svc_xprt *xprt, * in the message. * * - If there are Read chunks in this message, post Read WRs to - * pull that payload and return 0. + * pull that payload. When the Read WRs complete, build the + * full message and return the number of bytes in it. */ int svc_rdma_recvfrom(struct svc_rqst *rqstp) { @@ -763,80 +929,91 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp) struct svcxprt_rdma *rdma_xprt = container_of(xprt, struct svcxprt_rdma, sc_xprt); struct svc_rdma_recv_ctxt *ctxt; - __be32 *p; int ret; + /* Prevent svc_xprt_release() from releasing pages in rq_pages + * when returning 0 or an error. + */ + rqstp->rq_respages = rqstp->rq_pages; + rqstp->rq_next_page = rqstp->rq_respages; + + rqstp->rq_xprt_ctxt = NULL; + spin_lock(&rdma_xprt->sc_rq_dto_lock); ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_read_complete_q); if (ctxt) { list_del(&ctxt->rc_list); spin_unlock(&rdma_xprt->sc_rq_dto_lock); - rdma_read_complete(rqstp, ctxt); + svc_xprt_received(xprt); + svc_rdma_read_complete(rqstp, ctxt); goto complete; } ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_rq_dto_q); - if (!ctxt) { + if (ctxt) + list_del(&ctxt->rc_list); + else /* No new incoming requests, terminate the loop */ clear_bit(XPT_DATA, &xprt->xpt_flags); - spin_unlock(&rdma_xprt->sc_rq_dto_lock); - return 0; - } - list_del(&ctxt->rc_list); spin_unlock(&rdma_xprt->sc_rq_dto_lock); - atomic_inc(&rdma_stat_recv); + /* Unblock the transport for the next receive */ + svc_xprt_received(xprt); + if (!ctxt) + return 0; + percpu_counter_inc(&svcrdma_stat_recv); + ib_dma_sync_single_for_cpu(rdma_xprt->sc_pd->device, + ctxt->rc_recv_sge.addr, ctxt->rc_byte_len, + DMA_FROM_DEVICE); svc_rdma_build_arg_xdr(rqstp, ctxt); - /* Prevent svc_xprt_release from releasing pages in rq_pages - * if we return 0 or an error. - */ - rqstp->rq_respages = rqstp->rq_pages; - rqstp->rq_next_page = rqstp->rq_respages; - - p = (__be32 *)rqstp->rq_arg.head[0].iov_base; - ret = svc_rdma_xdr_decode_req(&rqstp->rq_arg); + ret = svc_rdma_xdr_decode_req(&rqstp->rq_arg, ctxt); if (ret < 0) goto out_err; if (ret == 0) goto out_drop; - rqstp->rq_xprt_hlen = ret; - if (svc_rdma_is_backchannel_reply(xprt, p)) { - ret = svc_rdma_handle_bc_reply(xprt->xpt_bc_xprt, p, - &rqstp->rq_arg); - svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); - return ret; - } + if (svc_rdma_is_reverse_direction_reply(xprt, ctxt)) + goto out_backchannel; + svc_rdma_get_inv_rkey(rdma_xprt, ctxt); - p += rpcrdma_fixed_maxsz; - if (*p != xdr_zero) - goto out_readchunk; + if (!pcl_is_empty(&ctxt->rc_read_pcl) || + !pcl_is_empty(&ctxt->rc_call_pcl)) + goto out_readlist; complete: rqstp->rq_xprt_ctxt = ctxt; rqstp->rq_prot = IPPROTO_MAX; svc_xprt_copy_addrs(rqstp, xprt); + set_bit(RQ_SECURE, &rqstp->rq_flags); return rqstp->rq_arg.len; -out_readchunk: - ret = svc_rdma_recv_read_chunk(rdma_xprt, rqstp, ctxt, p); - if (ret < 0) - goto out_postfail; - return 0; - out_err: - svc_rdma_send_error(rdma_xprt, p, ret); + svc_rdma_send_error(rdma_xprt, ctxt, ret); svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); return 0; -out_postfail: - if (ret == -EINVAL) - svc_rdma_send_error(rdma_xprt, p, ret); - svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); - return ret; +out_readlist: + /* This @rqstp is about to be recycled. Save the work + * already done constructing the Call message in rq_arg + * so it can be restored when the RDMA Reads have + * completed. + */ + ctxt->rc_saved_arg = rqstp->rq_arg; + + ret = svc_rdma_process_read_list(rdma_xprt, rqstp, ctxt); + if (ret < 0) { + if (ret == -EINVAL) + svc_rdma_send_error(rdma_xprt, ctxt, ret); + svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); + svc_xprt_deferred_close(xprt); + return ret; + } + return 0; +out_backchannel: + svc_rdma_handle_bc_reply(rqstp, ctxt); out_drop: svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); return 0; |