diff options
Diffstat (limited to 'net/sunrpc/xprtrdma/transport.c')
| -rw-r--r-- | net/sunrpc/xprtrdma/transport.c | 359 |
1 files changed, 136 insertions, 223 deletions
diff --git a/net/sunrpc/xprtrdma/transport.c b/net/sunrpc/xprtrdma/transport.c index fbc171ebfe91..9a8ce5df83ca 100644 --- a/net/sunrpc/xprtrdma/transport.c +++ b/net/sunrpc/xprtrdma/transport.c @@ -60,19 +60,16 @@ #include "xprt_rdma.h" #include <trace/events/rpcrdma.h> -#if IS_ENABLED(CONFIG_SUNRPC_DEBUG) -# define RPCDBG_FACILITY RPCDBG_TRANS -#endif - /* * tunables */ static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE; unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE; -static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE; +unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE; unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRWR; int xprt_rdma_pad_optimize; +static struct xprt_class xprt_rdma; #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) @@ -80,7 +77,6 @@ static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE; static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE; static unsigned int min_inline_size = RPCRDMA_MIN_INLINE; static unsigned int max_inline_size = RPCRDMA_MAX_INLINE; -static unsigned int zero; static unsigned int max_padding = PAGE_SIZE; static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS; static unsigned int max_memreg = RPCRDMA_LAST - 1; @@ -122,7 +118,7 @@ static struct ctl_table xr_tunables_table[] = { .maxlen = sizeof(unsigned int), .mode = 0644, .proc_handler = proc_dointvec_minmax, - .extra1 = &zero, + .extra1 = SYSCTL_ZERO, .extra2 = &max_padding, }, { @@ -141,16 +137,6 @@ static struct ctl_table xr_tunables_table[] = { .mode = 0644, .proc_handler = proc_dointvec, }, - { }, -}; - -static struct ctl_table sunrpc_table[] = { - { - .procname = "sunrpc", - .mode = 0555, - .child = xr_tunables_table - }, - { }, }; #endif @@ -239,29 +225,35 @@ xprt_rdma_connect_worker(struct work_struct *work) struct rpcrdma_xprt *r_xprt = container_of(work, struct rpcrdma_xprt, rx_connect_worker.work); struct rpc_xprt *xprt = &r_xprt->rx_xprt; + unsigned int pflags = current->flags; int rc; - rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia); + if (atomic_read(&xprt->swapper)) + current->flags |= PF_MEMALLOC; + rc = rpcrdma_xprt_connect(r_xprt); xprt_clear_connecting(xprt); - if (r_xprt->rx_ep.rep_connected > 0) { - if (!xprt_test_and_set_connected(xprt)) { - xprt->stat.connect_count++; - xprt->stat.connect_time += (long)jiffies - - xprt->stat.connect_start; - xprt_wake_pending_tasks(xprt, -EAGAIN); - } - } else { - if (xprt_test_and_clear_connected(xprt)) - xprt_wake_pending_tasks(xprt, rc); - } + if (!rc) { + xprt->connect_cookie++; + xprt->stat.connect_count++; + xprt->stat.connect_time += (long)jiffies - + xprt->stat.connect_start; + xprt_set_connected(xprt); + rc = -EAGAIN; + } else + rpcrdma_xprt_disconnect(r_xprt); + xprt_unlock_connect(xprt, r_xprt); + xprt_wake_pending_tasks(xprt, rc); + current_restore_flags(pflags, PF_MEMALLOC); } /** * xprt_rdma_inject_disconnect - inject a connection fault * @xprt: transport context * - * If @xprt is connected, disconnect it to simulate spurious connection - * loss. + * If @xprt is connected, disconnect it to simulate spurious + * connection loss. Caller must hold @xprt's send lock to + * ensure that data structures and hardware resources are + * stable during the rdma_disconnect() call. */ static void xprt_rdma_inject_disconnect(struct rpc_xprt *xprt) @@ -269,7 +261,7 @@ xprt_rdma_inject_disconnect(struct rpc_xprt *xprt) struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); trace_xprtrdma_op_inject_dsc(r_xprt); - rdma_disconnect(r_xprt->rx_ia.ri_id); + rdma_disconnect(r_xprt->rx_ep->re_id); } /** @@ -284,13 +276,10 @@ xprt_rdma_destroy(struct rpc_xprt *xprt) { struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); - trace_xprtrdma_op_destroy(r_xprt); - cancel_delayed_work_sync(&r_xprt->rx_connect_worker); - rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia); + rpcrdma_xprt_disconnect(r_xprt); rpcrdma_buffer_destroy(&r_xprt->rx_buf); - rpcrdma_ia_close(&r_xprt->rx_ia); xprt_rdma_free_addresses(xprt); xprt_free(xprt); @@ -298,6 +287,7 @@ xprt_rdma_destroy(struct rpc_xprt *xprt) module_put(THIS_MODULE); } +/* 60 second timeout, no retries */ static const struct rpc_timeout xprt_rdma_default_timeout = { .to_initval = 60 * HZ, .to_maxval = 60 * HZ, @@ -311,28 +301,32 @@ static const struct rpc_timeout xprt_rdma_default_timeout = { static struct rpc_xprt * xprt_setup_rdma(struct xprt_create *args) { - struct rpcrdma_create_data_internal cdata; struct rpc_xprt *xprt; struct rpcrdma_xprt *new_xprt; - struct rpcrdma_ep *new_ep; struct sockaddr *sap; int rc; if (args->addrlen > sizeof(xprt->addr)) return ERR_PTR(-EBADF); - xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt), 0, 0); - if (!xprt) + if (!try_module_get(THIS_MODULE)) + return ERR_PTR(-EIO); + + xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt), 0, + xprt_rdma_slot_table_entries); + if (!xprt) { + module_put(THIS_MODULE); return ERR_PTR(-ENOMEM); + } - /* 60 second timeout, no retries */ xprt->timeout = &xprt_rdma_default_timeout; + xprt->connect_timeout = xprt->timeout->to_initval; + xprt->max_reconnect_timeout = xprt->timeout->to_maxval; xprt->bind_timeout = RPCRDMA_BIND_TO; xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO; xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO; xprt->resvport = 0; /* privileged port not needed */ - xprt->tsh_size = 0; /* RPC-RDMA handles framing */ xprt->ops = &xprt_rdma_procs; /* @@ -343,6 +337,7 @@ xprt_setup_rdma(struct xprt_create *args) /* Ensure xprt->addr holds valid server TCP (not RDMA) * address, for any side protocols which peek at it */ xprt->prot = IPPROTO_TCP; + xprt->xprt_class = &xprt_rdma; xprt->addrlen = args->addrlen; memcpy(&xprt->addr, sap, xprt->addrlen); @@ -350,78 +345,21 @@ xprt_setup_rdma(struct xprt_create *args) xprt_set_bound(xprt); xprt_rdma_format_addresses(xprt, sap); - cdata.max_requests = xprt_rdma_slot_table_entries; - - cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */ - cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */ - - cdata.inline_wsize = xprt_rdma_max_inline_write; - if (cdata.inline_wsize > cdata.wsize) - cdata.inline_wsize = cdata.wsize; - - cdata.inline_rsize = xprt_rdma_max_inline_read; - if (cdata.inline_rsize > cdata.rsize) - cdata.inline_rsize = cdata.rsize; - - /* - * Create new transport instance, which includes initialized - * o ia - * o endpoint - * o buffers - */ - new_xprt = rpcx_to_rdmax(xprt); - - rc = rpcrdma_ia_open(new_xprt); - if (rc) - goto out1; - - /* - * initialize and create ep - */ - new_xprt->rx_data = cdata; - new_ep = &new_xprt->rx_ep; - - rc = rpcrdma_ep_create(&new_xprt->rx_ep, - &new_xprt->rx_ia, &new_xprt->rx_data); - if (rc) - goto out2; - rc = rpcrdma_buffer_create(new_xprt); - if (rc) - goto out3; + if (rc) { + xprt_rdma_free_addresses(xprt); + xprt_free(xprt); + module_put(THIS_MODULE); + return ERR_PTR(rc); + } INIT_DELAYED_WORK(&new_xprt->rx_connect_worker, xprt_rdma_connect_worker); - xprt->max_payload = frwr_maxpages(new_xprt); - if (xprt->max_payload == 0) - goto out4; - xprt->max_payload <<= PAGE_SHIFT; - dprintk("RPC: %s: transport data payload maximum: %zu bytes\n", - __func__, xprt->max_payload); - - if (!try_module_get(THIS_MODULE)) - goto out4; + xprt->max_payload = RPCRDMA_MAX_DATA_SEGS << PAGE_SHIFT; - dprintk("RPC: %s: %s:%s\n", __func__, - xprt->address_strings[RPC_DISPLAY_ADDR], - xprt->address_strings[RPC_DISPLAY_PORT]); - trace_xprtrdma_create(new_xprt); return xprt; - -out4: - rpcrdma_buffer_destroy(&new_xprt->rx_buf); - rc = -ENODEV; -out3: - rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia); -out2: - rpcrdma_ia_close(&new_xprt->rx_ia); -out1: - trace_xprtrdma_op_destroy(new_xprt); - xprt_rdma_free_addresses(xprt); - xprt_free(xprt); - return ERR_PTR(rc); } /** @@ -436,34 +374,10 @@ out1: void xprt_rdma_close(struct rpc_xprt *xprt) { struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); - struct rpcrdma_ep *ep = &r_xprt->rx_ep; - struct rpcrdma_ia *ia = &r_xprt->rx_ia; - - might_sleep(); - - trace_xprtrdma_op_close(r_xprt); - /* Prevent marshaling and sending of new requests */ - xprt_clear_connected(xprt); - - if (test_and_clear_bit(RPCRDMA_IAF_REMOVING, &ia->ri_flags)) { - rpcrdma_ia_remove(ia); - goto out; - } + rpcrdma_xprt_disconnect(r_xprt); - if (ep->rep_connected == -ENODEV) - return; - if (ep->rep_connected > 0) - xprt->reestablish_timeout = 0; - rpcrdma_ep_disconnect(ep, ia); - - /* Prepare @xprt for the next connection by reinitializing - * its credit grant to one (see RFC 8166, Section 3.3.3). - */ - r_xprt->rx_buf.rb_credits = 1; - xprt->cwnd = RPC_CWNDSHIFT; - -out: + xprt->reestablish_timeout = 0; ++xprt->connect_cookie; xprt_disconnect_done(xprt); } @@ -481,12 +395,6 @@ xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port) struct sockaddr *sap = (struct sockaddr *)&xprt->addr; char buf[8]; - dprintk("RPC: %s: setting port for xprt %p (%s:%s) to %u\n", - __func__, xprt, - xprt->address_strings[RPC_DISPLAY_ADDR], - xprt->address_strings[RPC_DISPLAY_PORT], - port); - rpc_set_port(sap, port); kfree(xprt->address_strings[RPC_DISPLAY_PORT]); @@ -518,31 +426,65 @@ xprt_rdma_timer(struct rpc_xprt *xprt, struct rpc_task *task) } /** - * xprt_rdma_connect - try to establish a transport connection + * xprt_rdma_set_connect_timeout - set timeouts for establishing a connection + * @xprt: controlling transport instance + * @connect_timeout: reconnect timeout after client disconnects + * @reconnect_timeout: reconnect timeout after server disconnects + * + */ +static void xprt_rdma_set_connect_timeout(struct rpc_xprt *xprt, + unsigned long connect_timeout, + unsigned long reconnect_timeout) +{ + struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); + + trace_xprtrdma_op_set_cto(r_xprt, connect_timeout, reconnect_timeout); + + spin_lock(&xprt->transport_lock); + + if (connect_timeout < xprt->connect_timeout) { + struct rpc_timeout to; + unsigned long initval; + + to = *xprt->timeout; + initval = connect_timeout; + if (initval < RPCRDMA_INIT_REEST_TO << 1) + initval = RPCRDMA_INIT_REEST_TO << 1; + to.to_initval = initval; + to.to_maxval = initval; + r_xprt->rx_timeout = to; + xprt->timeout = &r_xprt->rx_timeout; + xprt->connect_timeout = connect_timeout; + } + + if (reconnect_timeout < xprt->max_reconnect_timeout) + xprt->max_reconnect_timeout = reconnect_timeout; + + spin_unlock(&xprt->transport_lock); +} + +/** + * xprt_rdma_connect - schedule an attempt to reconnect * @xprt: transport state - * @task: RPC scheduler context + * @task: RPC scheduler context (unused) * */ static void xprt_rdma_connect(struct rpc_xprt *xprt, struct rpc_task *task) { struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); + struct rpcrdma_ep *ep = r_xprt->rx_ep; + unsigned long delay; + + WARN_ON_ONCE(!xprt_lock_connect(xprt, task, r_xprt)); - trace_xprtrdma_op_connect(r_xprt); - if (r_xprt->rx_ep.rep_connected != 0) { - /* Reconnect */ - schedule_delayed_work(&r_xprt->rx_connect_worker, - xprt->reestablish_timeout); - xprt->reestablish_timeout <<= 1; - if (xprt->reestablish_timeout > RPCRDMA_MAX_REEST_TO) - xprt->reestablish_timeout = RPCRDMA_MAX_REEST_TO; - else if (xprt->reestablish_timeout < RPCRDMA_INIT_REEST_TO) - xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO; - } else { - schedule_delayed_work(&r_xprt->rx_connect_worker, 0); - if (!RPC_IS_ASYNC(task)) - flush_delayed_work(&r_xprt->rx_connect_worker); + delay = 0; + if (ep && ep->re_connect_status != 0) { + delay = xprt_reconnect_delay(xprt); + xprt_reconnect_backoff(xprt, RPCRDMA_INIT_REEST_TO); } + trace_xprtrdma_op_connect(r_xprt, delay); + queue_delayed_work(system_long_wq, &r_xprt->rx_connect_worker, delay); } /** @@ -568,8 +510,8 @@ xprt_rdma_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task) return; out_sleep: - rpc_sleep_on(&xprt->backlog, task, NULL); - task->tk_status = -EAGAIN; + task->tk_status = -ENOMEM; + xprt_add_backlog(xprt, task); } /** @@ -581,57 +523,25 @@ out_sleep: static void xprt_rdma_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *rqst) { - memset(rqst, 0, sizeof(*rqst)); - rpcrdma_buffer_put(rpcr_to_rdmar(rqst)); - rpc_wake_up_next(&xprt->backlog); -} - -static bool -rpcrdma_get_sendbuf(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req, - size_t size, gfp_t flags) -{ - struct rpcrdma_regbuf *rb; - - if (req->rl_sendbuf && rdmab_length(req->rl_sendbuf) >= size) - return true; - - rb = rpcrdma_alloc_regbuf(size, DMA_TO_DEVICE, flags); - if (IS_ERR(rb)) - return false; + struct rpcrdma_xprt *r_xprt = + container_of(xprt, struct rpcrdma_xprt, rx_xprt); - rpcrdma_free_regbuf(req->rl_sendbuf); - r_xprt->rx_stats.hardway_register_count += size; - req->rl_sendbuf = rb; - return true; + rpcrdma_reply_put(&r_xprt->rx_buf, rpcr_to_rdmar(rqst)); + if (!xprt_wake_up_backlog(xprt, rqst)) { + memset(rqst, 0, sizeof(*rqst)); + rpcrdma_buffer_put(&r_xprt->rx_buf, rpcr_to_rdmar(rqst)); + } } -/* The rq_rcv_buf is used only if a Reply chunk is necessary. - * The decision to use a Reply chunk is made later in - * rpcrdma_marshal_req. This buffer is registered at that time. - * - * Otherwise, the associated RPC Reply arrives in a separate - * Receive buffer, arbitrarily chosen by the HCA. The buffer - * allocated here for the RPC Reply is not utilized in that - * case. See rpcrdma_inline_fixup. - * - * A regbuf is used here to remember the buffer size. - */ -static bool -rpcrdma_get_recvbuf(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req, - size_t size, gfp_t flags) +static bool rpcrdma_check_regbuf(struct rpcrdma_xprt *r_xprt, + struct rpcrdma_regbuf *rb, size_t size, + gfp_t flags) { - struct rpcrdma_regbuf *rb; - - if (req->rl_recvbuf && rdmab_length(req->rl_recvbuf) >= size) - return true; - - rb = rpcrdma_alloc_regbuf(size, DMA_NONE, flags); - if (IS_ERR(rb)) - return false; - - rpcrdma_free_regbuf(req->rl_recvbuf); - r_xprt->rx_stats.hardway_register_count += size; - req->rl_recvbuf = rb; + if (unlikely(rdmab_length(rb) < size)) { + if (!rpcrdma_regbuf_realloc(rb, size, flags)) + return false; + r_xprt->rx_stats.hardway_register_count += size; + } return true; } @@ -650,24 +560,20 @@ xprt_rdma_allocate(struct rpc_task *task) struct rpc_rqst *rqst = task->tk_rqstp; struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt); struct rpcrdma_req *req = rpcr_to_rdmar(rqst); - gfp_t flags; - - flags = RPCRDMA_DEF_GFP; - if (RPC_IS_SWAPPER(task)) - flags = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN; + gfp_t flags = rpc_task_gfp_mask(); - if (!rpcrdma_get_sendbuf(r_xprt, req, rqst->rq_callsize, flags)) + if (!rpcrdma_check_regbuf(r_xprt, req->rl_sendbuf, rqst->rq_callsize, + flags)) goto out_fail; - if (!rpcrdma_get_recvbuf(r_xprt, req, rqst->rq_rcvsize, flags)) + if (!rpcrdma_check_regbuf(r_xprt, req->rl_recvbuf, rqst->rq_rcvsize, + flags)) goto out_fail; - rqst->rq_buffer = req->rl_sendbuf->rg_base; - rqst->rq_rbuffer = req->rl_recvbuf->rg_base; - trace_xprtrdma_op_allocate(task, req); + rqst->rq_buffer = rdmab_data(req->rl_sendbuf); + rqst->rq_rbuffer = rdmab_data(req->rl_recvbuf); return 0; out_fail: - trace_xprtrdma_op_allocate(task, NULL); return -ENOMEM; } @@ -681,12 +587,18 @@ static void xprt_rdma_free(struct rpc_task *task) { struct rpc_rqst *rqst = task->tk_rqstp; - struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt); struct rpcrdma_req *req = rpcr_to_rdmar(rqst); - if (test_bit(RPCRDMA_REQ_F_PENDING, &req->rl_flags)) - rpcrdma_release_rqst(r_xprt, req); - trace_xprtrdma_op_free(task, req); + if (unlikely(!list_empty(&req->rl_registered))) { + trace_xprtrdma_mrs_zap(task); + frwr_unmap_sync(rpcx_to_rdmax(rqst->rq_xprt), req); + } + + /* XXX: If the RPC is completing because of a signal and + * not because a reply was received, we ought to ensure + * that the Send completion has fired, so that memory + * involved with the Send is not still visible to the NIC. + */ } /** @@ -733,12 +645,10 @@ xprt_rdma_send_request(struct rpc_rqst *rqst) goto drop_connection; rqst->rq_xtime = ktime_get(); - __set_bit(RPCRDMA_REQ_F_PENDING, &req->rl_flags); - if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req)) + if (frwr_send(r_xprt, req)) goto drop_connection; rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len; - rqst->rq_bytes_sent = 0; /* An RPC with no reply will throw off credit accounting, * so drop the connection to reset the credit grant. @@ -817,7 +727,7 @@ static const struct rpc_xprt_ops xprt_rdma_procs = { .alloc_slot = xprt_rdma_alloc_slot, .free_slot = xprt_rdma_free_slot, .release_request = xprt_release_rqst_cong, /* ditto */ - .set_retrans_timeout = xprt_set_retrans_timeout_def, /* ditto */ + .wait_for_reply_request = xprt_wait_for_reply_request_def, /* ditto */ .timer = xprt_rdma_timer, .rpcbind = rpcb_getport_async, /* sunrpc/rpcb_clnt.c */ .set_port = xprt_rdma_set_port, @@ -827,6 +737,7 @@ static const struct rpc_xprt_ops xprt_rdma_procs = { .send_request = xprt_rdma_send_request, .close = xprt_rdma_close, .destroy = xprt_rdma_destroy, + .set_connect_timeout = xprt_rdma_set_connect_timeout, .print_stats = xprt_rdma_print_stats, .enable_swap = xprt_rdma_enable_swap, .disable_swap = xprt_rdma_disable_swap, @@ -834,6 +745,7 @@ static const struct rpc_xprt_ops xprt_rdma_procs = { #if defined(CONFIG_SUNRPC_BACKCHANNEL) .bc_setup = xprt_rdma_bc_setup, .bc_maxpayload = xprt_rdma_bc_maxpayload, + .bc_num_slots = xprt_rdma_bc_max_slots, .bc_free_rqst = xprt_rdma_bc_free_rqst, .bc_destroy = xprt_rdma_bc_destroy, #endif @@ -845,6 +757,7 @@ static struct xprt_class xprt_rdma = { .owner = THIS_MODULE, .ident = XPRT_TRANSPORT_RDMA, .setup = xprt_setup_rdma, + .netid = { "rdma", "rdma6", "" }, }; void xprt_rdma_cleanup(void) @@ -876,7 +789,7 @@ int xprt_rdma_init(void) #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) if (!sunrpc_table_header) - sunrpc_table_header = register_sysctl_table(sunrpc_table); + sunrpc_table_header = register_sysctl("sunrpc", xr_tunables_table); #endif return 0; } |
