4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 ksocknal_lib_get_conn_addrs(ksock_conn_t *conn)
42 int rc = lnet_sock_getaddr(conn->ksnc_sock, 1, &conn->ksnc_ipaddr,
45 /* Didn't need the {get,put}connsock dance to deref ksnc_sock... */
46 LASSERT(!conn->ksnc_closing);
49 CERROR("Error %d getting sock peer IP\n", rc);
53 rc = lnet_sock_getaddr(conn->ksnc_sock, 0, &conn->ksnc_myipaddr, NULL);
55 CERROR("Error %d getting sock local IP\n", rc);
63 ksocknal_lib_zc_capable(ksock_conn_t *conn)
65 int caps = conn->ksnc_sock->sk->sk_route_caps;
67 if (conn->ksnc_proto == &ksocknal_protocol_v1x)
70 /* ZC if the socket supports scatter/gather and doesn't need software
72 return ((caps & NETIF_F_SG) != 0 && (caps & NETIF_F_ALL_CSUM) != 0);
76 ksocknal_lib_send_iov(ksock_conn_t *conn, ksock_tx_t *tx)
78 struct socket *sock = conn->ksnc_sock;
82 if (*ksocknal_tunables.ksnd_enable_csum && /* checksum enabled */
83 conn->ksnc_proto == &ksocknal_protocol_v2x && /* V2.x connection */
84 tx->tx_nob == tx->tx_resid && /* frist sending */
85 tx->tx_msg.ksm_csum == 0) /* not checksummed */
86 ksocknal_lib_csum_tx(tx);
88 /* NB we can't trust socket ops to either consume our iovs
89 * or leave them alone. */
92 #if SOCKNAL_SINGLE_FRAG_TX
94 struct kvec *scratchiov = &scratch;
95 unsigned int niov = 1;
97 struct kvec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
98 unsigned int niov = tx->tx_niov;
100 struct msghdr msg = {.msg_flags = MSG_DONTWAIT};
103 for (nob = i = 0; i < niov; i++) {
104 scratchiov[i] = tx->tx_iov[i];
105 nob += scratchiov[i].iov_len;
108 if (!list_empty(&conn->ksnc_tx_queue) ||
110 msg.msg_flags |= MSG_MORE;
112 rc = kernel_sendmsg(sock, &msg, scratchiov, niov, nob);
118 ksocknal_lib_send_kiov(ksock_conn_t *conn, ksock_tx_t *tx)
120 struct socket *sock = conn->ksnc_sock;
121 lnet_kiov_t *kiov = tx->tx_kiov;
125 /* Not NOOP message */
126 LASSERT(tx->tx_lnetmsg != NULL);
128 /* NB we can't trust socket ops to either consume our iovs
129 * or leave them alone. */
130 if (tx->tx_msg.ksm_zc_cookies[0] != 0) {
131 /* Zero copy is enabled */
132 struct sock *sk = sock->sk;
133 struct page *page = kiov->kiov_page;
134 int offset = kiov->kiov_offset;
135 int fragsize = kiov->kiov_len;
136 int msgflg = MSG_DONTWAIT;
138 CDEBUG(D_NET, "page %p + offset %x for %d\n",
139 page, offset, kiov->kiov_len);
141 if (!list_empty(&conn->ksnc_tx_queue) ||
142 fragsize < tx->tx_resid)
145 if (sk->sk_prot->sendpage != NULL) {
146 rc = sk->sk_prot->sendpage(sk, page,
147 offset, fragsize, msgflg);
149 rc = tcp_sendpage(sk, page, offset, fragsize, msgflg);
152 #if SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_RISK_KMAP_DEADLOCK
154 struct kvec *scratchiov = &scratch;
155 unsigned int niov = 1;
157 #ifdef CONFIG_HIGHMEM
158 #warning "XXX risk of kmap deadlock on multiple frags..."
160 struct kvec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
161 unsigned int niov = tx->tx_nkiov;
163 struct msghdr msg = {.msg_flags = MSG_DONTWAIT};
166 for (nob = i = 0; i < niov; i++) {
167 scratchiov[i].iov_base = kmap(kiov[i].kiov_page) +
169 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
172 if (!list_empty(&conn->ksnc_tx_queue) ||
174 msg.msg_flags |= MSG_MORE;
176 rc = kernel_sendmsg(sock, &msg, (struct kvec *)scratchiov, niov, nob);
178 for (i = 0; i < niov; i++)
179 kunmap(kiov[i].kiov_page);
185 ksocknal_lib_eager_ack(ksock_conn_t *conn)
188 struct socket *sock = conn->ksnc_sock;
190 /* Remind the socket to ACK eagerly. If I don't, the socket might
191 * think I'm about to send something it could piggy-back the ACK
192 * on, introducing delay in completing zero-copy sends in my
195 kernel_setsockopt(sock, SOL_TCP, TCP_QUICKACK,
196 (char *)&opt, sizeof(opt));
200 ksocknal_lib_recv_iov(ksock_conn_t *conn)
202 #if SOCKNAL_SINGLE_FRAG_RX
204 struct kvec *scratchiov = &scratch;
205 unsigned int niov = 1;
207 struct kvec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
208 unsigned int niov = conn->ksnc_rx_niov;
210 struct kvec *iov = conn->ksnc_rx_iov;
211 struct msghdr msg = {
221 /* NB we can't trust socket ops to either consume our iovs
222 * or leave them alone. */
225 for (nob = i = 0; i < niov; i++) {
226 scratchiov[i] = iov[i];
227 nob += scratchiov[i].iov_len;
229 LASSERT(nob <= conn->ksnc_rx_nob_wanted);
231 rc = kernel_recvmsg(conn->ksnc_sock, &msg,
232 scratchiov, niov, nob, MSG_DONTWAIT);
235 if (conn->ksnc_proto == &ksocknal_protocol_v2x) {
236 saved_csum = conn->ksnc_msg.ksm_csum;
237 conn->ksnc_msg.ksm_csum = 0;
240 if (saved_csum != 0) {
241 /* accumulate checksum */
242 for (i = 0, sum = rc; sum > 0; i++, sum -= fragnob) {
245 fragnob = iov[i].iov_len;
249 conn->ksnc_rx_csum = ksocknal_csum(conn->ksnc_rx_csum,
250 iov[i].iov_base, fragnob);
252 conn->ksnc_msg.ksm_csum = saved_csum;
259 ksocknal_lib_kiov_vunmap(void *addr)
268 ksocknal_lib_kiov_vmap(lnet_kiov_t *kiov, int niov,
269 struct kvec *iov, struct page **pages)
275 if (!*ksocknal_tunables.ksnd_zc_recv || pages == NULL)
278 LASSERT(niov <= LNET_MAX_IOV);
281 niov < *ksocknal_tunables.ksnd_zc_recv_min_nfrags)
284 for (nob = i = 0; i < niov; i++) {
285 if ((kiov[i].kiov_offset != 0 && i > 0) ||
286 (kiov[i].kiov_offset + kiov[i].kiov_len != PAGE_CACHE_SIZE && i < niov - 1))
289 pages[i] = kiov[i].kiov_page;
290 nob += kiov[i].kiov_len;
293 addr = vmap(pages, niov, VM_MAP, PAGE_KERNEL);
297 iov->iov_base = addr + kiov[0].kiov_offset;
304 ksocknal_lib_recv_kiov(ksock_conn_t *conn)
306 #if SOCKNAL_SINGLE_FRAG_RX || !SOCKNAL_RISK_KMAP_DEADLOCK
308 struct kvec *scratchiov = &scratch;
309 struct page **pages = NULL;
310 unsigned int niov = 1;
312 #ifdef CONFIG_HIGHMEM
313 #warning "XXX risk of kmap deadlock on multiple frags..."
315 struct kvec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
316 struct page **pages = conn->ksnc_scheduler->kss_rx_scratch_pgs;
317 unsigned int niov = conn->ksnc_rx_nkiov;
319 lnet_kiov_t *kiov = conn->ksnc_rx_kiov;
320 struct msghdr msg = {
332 /* NB we can't trust socket ops to either consume our iovs
333 * or leave them alone. */
334 addr = ksocknal_lib_kiov_vmap(kiov, niov, scratchiov, pages);
336 nob = scratchiov[0].iov_len;
340 for (nob = i = 0; i < niov; i++) {
341 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
342 scratchiov[i].iov_base = kmap(kiov[i].kiov_page) +
348 LASSERT(nob <= conn->ksnc_rx_nob_wanted);
350 rc = kernel_recvmsg(conn->ksnc_sock, &msg,
351 (struct kvec *)scratchiov, n, nob, MSG_DONTWAIT);
353 if (conn->ksnc_msg.ksm_csum != 0) {
354 for (i = 0, sum = rc; sum > 0; i++, sum -= fragnob) {
357 /* Dang! have to kmap again because I have nowhere to
358 * stash the mapped address. But by doing it while the
359 * page is still mapped, the kernel just bumps the map
360 * count and returns me the address it stashed. */
361 base = kmap(kiov[i].kiov_page) + kiov[i].kiov_offset;
362 fragnob = kiov[i].kiov_len;
366 conn->ksnc_rx_csum = ksocknal_csum(conn->ksnc_rx_csum,
369 kunmap(kiov[i].kiov_page);
374 ksocknal_lib_kiov_vunmap(addr);
376 for (i = 0; i < niov; i++)
377 kunmap(kiov[i].kiov_page);
384 ksocknal_lib_csum_tx(ksock_tx_t *tx)
390 LASSERT(tx->tx_iov[0].iov_base == &tx->tx_msg);
391 LASSERT(tx->tx_conn != NULL);
392 LASSERT(tx->tx_conn->ksnc_proto == &ksocknal_protocol_v2x);
394 tx->tx_msg.ksm_csum = 0;
396 csum = ksocknal_csum(~0, tx->tx_iov[0].iov_base,
397 tx->tx_iov[0].iov_len);
399 if (tx->tx_kiov != NULL) {
400 for (i = 0; i < tx->tx_nkiov; i++) {
401 base = kmap(tx->tx_kiov[i].kiov_page) +
402 tx->tx_kiov[i].kiov_offset;
404 csum = ksocknal_csum(csum, base, tx->tx_kiov[i].kiov_len);
406 kunmap(tx->tx_kiov[i].kiov_page);
409 for (i = 1; i < tx->tx_niov; i++)
410 csum = ksocknal_csum(csum, tx->tx_iov[i].iov_base,
411 tx->tx_iov[i].iov_len);
414 if (*ksocknal_tunables.ksnd_inject_csum_error) {
416 *ksocknal_tunables.ksnd_inject_csum_error = 0;
419 tx->tx_msg.ksm_csum = csum;
423 ksocknal_lib_get_conn_tunables(ksock_conn_t *conn, int *txmem, int *rxmem, int *nagle)
425 struct socket *sock = conn->ksnc_sock;
429 rc = ksocknal_connsock_addref(conn);
431 LASSERT(conn->ksnc_closing);
432 *txmem = *rxmem = *nagle = 0;
436 rc = lnet_sock_getbuf(sock, txmem, rxmem);
438 len = sizeof(*nagle);
439 rc = kernel_getsockopt(sock, SOL_TCP, TCP_NODELAY,
440 (char *)nagle, &len);
443 ksocknal_connsock_decref(conn);
448 *txmem = *rxmem = *nagle = 0;
454 ksocknal_lib_setup_sock(struct socket *sock)
462 struct linger linger;
464 sock->sk->sk_allocation = GFP_NOFS;
466 /* Ensure this socket aborts active sends immediately when we close
472 rc = kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
473 (char *)&linger, sizeof(linger));
475 CERROR("Can't set SO_LINGER: %d\n", rc);
480 rc = kernel_setsockopt(sock, SOL_TCP, TCP_LINGER2,
481 (char *)&option, sizeof(option));
483 CERROR("Can't set SO_LINGER2: %d\n", rc);
487 if (!*ksocknal_tunables.ksnd_nagle) {
490 rc = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
491 (char *)&option, sizeof(option));
493 CERROR("Can't disable nagle: %d\n", rc);
498 rc = lnet_sock_setbuf(sock, *ksocknal_tunables.ksnd_tx_buffer_size,
499 *ksocknal_tunables.ksnd_rx_buffer_size);
501 CERROR("Can't set buffer tx %d, rx %d buffers: %d\n",
502 *ksocknal_tunables.ksnd_tx_buffer_size,
503 *ksocknal_tunables.ksnd_rx_buffer_size, rc);
507 /* TCP_BACKOFF_* sockopt tunables unsupported in stock kernels */
509 /* snapshot tunables */
510 keep_idle = *ksocknal_tunables.ksnd_keepalive_idle;
511 keep_count = *ksocknal_tunables.ksnd_keepalive_count;
512 keep_intvl = *ksocknal_tunables.ksnd_keepalive_intvl;
514 do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
516 option = (do_keepalive ? 1 : 0);
517 rc = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
518 (char *)&option, sizeof(option));
520 CERROR("Can't set SO_KEEPALIVE: %d\n", rc);
527 rc = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
528 (char *)&keep_idle, sizeof(keep_idle));
530 CERROR("Can't set TCP_KEEPIDLE: %d\n", rc);
534 rc = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
535 (char *)&keep_intvl, sizeof(keep_intvl));
537 CERROR("Can't set TCP_KEEPINTVL: %d\n", rc);
541 rc = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
542 (char *)&keep_count, sizeof(keep_count));
544 CERROR("Can't set TCP_KEEPCNT: %d\n", rc);
552 ksocknal_lib_push_conn(ksock_conn_t *conn)
560 rc = ksocknal_connsock_addref(conn);
561 if (rc != 0) /* being shut down */
564 sk = conn->ksnc_sock->sk;
568 nonagle = tp->nonagle;
572 rc = kernel_setsockopt(conn->ksnc_sock, SOL_TCP, TCP_NODELAY,
573 (char *)&val, sizeof(val));
577 tp->nonagle = nonagle;
580 ksocknal_connsock_decref(conn);
583 extern void ksocknal_read_callback(ksock_conn_t *conn);
584 extern void ksocknal_write_callback(ksock_conn_t *conn);
586 * socket call back in Linux
589 ksocknal_data_ready(struct sock *sk)
593 /* interleave correctly with closing sockets... */
595 read_lock(&ksocknal_data.ksnd_global_lock);
597 conn = sk->sk_user_data;
598 if (conn == NULL) { /* raced with ksocknal_terminate_conn */
599 LASSERT(sk->sk_data_ready != &ksocknal_data_ready);
600 sk->sk_data_ready(sk);
602 ksocknal_read_callback(conn);
604 read_unlock(&ksocknal_data.ksnd_global_lock);
608 ksocknal_write_space(struct sock *sk)
614 /* interleave correctly with closing sockets... */
616 read_lock(&ksocknal_data.ksnd_global_lock);
618 conn = sk->sk_user_data;
619 wspace = sk_stream_wspace(sk);
620 min_wpace = sk_stream_min_wspace(sk);
622 CDEBUG(D_NET, "sk %p wspace %d low water %d conn %p%s%s%s\n",
623 sk, wspace, min_wpace, conn,
624 (conn == NULL) ? "" : (conn->ksnc_tx_ready ?
625 " ready" : " blocked"),
626 (conn == NULL) ? "" : (conn->ksnc_tx_scheduled ?
627 " scheduled" : " idle"),
628 (conn == NULL) ? "" : (list_empty(&conn->ksnc_tx_queue) ?
629 " empty" : " queued"));
631 if (conn == NULL) { /* raced with ksocknal_terminate_conn */
632 LASSERT(sk->sk_write_space != &ksocknal_write_space);
633 sk->sk_write_space(sk);
635 read_unlock(&ksocknal_data.ksnd_global_lock);
639 if (wspace >= min_wpace) { /* got enough space */
640 ksocknal_write_callback(conn);
642 /* Clear SOCK_NOSPACE _after_ ksocknal_write_callback so the
643 * ENOMEM check in ksocknal_transmit is race-free (think about
646 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
649 read_unlock(&ksocknal_data.ksnd_global_lock);
653 ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn)
655 conn->ksnc_saved_data_ready = sock->sk->sk_data_ready;
656 conn->ksnc_saved_write_space = sock->sk->sk_write_space;
660 ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn)
662 sock->sk->sk_user_data = conn;
663 sock->sk->sk_data_ready = ksocknal_data_ready;
664 sock->sk->sk_write_space = ksocknal_write_space;
669 ksocknal_lib_reset_callback(struct socket *sock, ksock_conn_t *conn)
671 /* Remove conn's network callbacks.
672 * NB I _have_ to restore the callback, rather than storing a noop,
673 * since the socket could survive past this module being unloaded!! */
674 sock->sk->sk_data_ready = conn->ksnc_saved_data_ready;
675 sock->sk->sk_write_space = conn->ksnc_saved_write_space;
677 /* A callback could be in progress already; they hold a read lock
678 * on ksnd_global_lock (to serialise with me) and NOOP if
679 * sk_user_data is NULL. */
680 sock->sk->sk_user_data = NULL;
686 ksocknal_lib_memory_pressure(ksock_conn_t *conn)
689 ksock_sched_t *sched;
691 sched = conn->ksnc_scheduler;
692 spin_lock_bh(&sched->kss_lock);
694 if (!test_bit(SOCK_NOSPACE, &conn->ksnc_sock->flags) &&
695 !conn->ksnc_tx_ready) {
696 /* SOCK_NOSPACE is set when the socket fills
697 * and cleared in the write_space callback
698 * (which also sets ksnc_tx_ready). If
699 * SOCK_NOSPACE and ksnc_tx_ready are BOTH
700 * zero, I didn't fill the socket and
701 * write_space won't reschedule me, so I
702 * return -ENOMEM to get my caller to retry
707 spin_unlock_bh(&sched->kss_lock);