Merge tag 'iommu-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
[linux/fpc-iii.git] / net / sunrpc / xprtsock.c
blobc56a66cdf4ac80d8cab6cd72c90fd86155013e3b
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/net/sunrpc/xprtsock.c
5 * Client-side transport implementation for sockets.
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
30 #include <linux/in.h>
31 #include <linux/net.h>
32 #include <linux/mm.h>
33 #include <linux/un.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
44 #endif
46 #include <net/sock.h>
47 #include <net/checksum.h>
48 #include <net/udp.h>
49 #include <net/tcp.h>
50 #include <linux/bvec.h>
51 #include <linux/highmem.h>
52 #include <linux/uio.h>
53 #include <linux/sched/mm.h>
55 #include <trace/events/sunrpc.h>
57 #include "socklib.h"
58 #include "sunrpc.h"
60 static void xs_close(struct rpc_xprt *xprt);
61 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
62 struct socket *sock);
65 * xprtsock tunables
67 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
71 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
74 #define XS_TCP_LINGER_TO (15U * HZ)
75 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
86 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
92 static struct ctl_table_header *sunrpc_table_header;
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
98 static struct ctl_table xs_tunables_table[] = {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_jiffies,
151 { },
154 static struct ctl_table sunrpc_table[] = {
156 .procname = "sunrpc",
157 .mode = 0555,
158 .child = xs_tunables_table
160 { },
164 * Wait duration for a reply from the RPC portmapper.
166 #define XS_BIND_TO (60U * HZ)
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
172 #define XS_UDP_REEST_TO (2U * HZ)
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
183 #define XS_TCP_INIT_REEST_TO (3U * HZ)
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
190 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
192 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193 # undef RPC_DEBUG_DATA
194 # define RPCDBG_FACILITY RPCDBG_TRANS
195 #endif
197 #ifdef RPC_DEBUG_DATA
198 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
200 u8 *buf = (u8 *) packet;
201 int j;
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
205 if (!(j & 31)) {
206 if (j)
207 dprintk("\n");
208 dprintk("0x%04x ", j);
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 dprintk("\n");
215 #else
216 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
218 /* NOP */
220 #endif
222 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
224 return (struct rpc_xprt *) sk->sk_user_data;
227 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
229 return (struct sockaddr *) &xprt->addr;
232 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
234 return (struct sockaddr_un *) &xprt->addr;
237 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
239 return (struct sockaddr_in *) &xprt->addr;
242 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
244 return (struct sockaddr_in6 *) &xprt->addr;
247 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
253 char buf[128];
255 switch (sap->sa_family) {
256 case AF_LOCAL:
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
261 break;
262 case AF_INET:
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
268 break;
269 case AF_INET6:
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
275 break;
276 default:
277 BUG();
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
283 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
285 struct sockaddr *sap = xs_addr(xprt);
286 char buf[128];
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
295 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
297 const char *netid)
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
305 static void xs_update_peer_port(struct rpc_xprt *xprt)
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
310 xs_format_common_peer_ports(xprt);
313 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
315 unsigned int i;
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 switch (i) {
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
321 continue;
322 default:
323 kfree(xprt->address_strings[i]);
327 static size_t
328 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
330 size_t i,n;
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 return want;
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
336 if (buf->pages[i])
337 continue;
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
340 i *= PAGE_SIZE;
341 return i > buf->page_base ? i - buf->page_base : 0;
344 return want;
347 static ssize_t
348 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
350 ssize_t ret;
351 if (seek != 0)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
357 static ssize_t
358 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
365 static ssize_t
366 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
368 size_t seek)
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
374 static ssize_t
375 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
376 size_t count)
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
382 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383 static void
384 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
386 struct bvec_iter bi = {
387 .bi_size = count,
389 struct bio_vec bv;
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
395 #else
396 static inline void
397 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
400 #endif
402 static ssize_t
403 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
406 size_t want, seek_init = seek, offset = 0;
407 ssize_t ret;
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 if (seek < want) {
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
412 if (ret <= 0)
413 goto sock_err;
414 offset += ret;
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
416 goto out;
417 if (ret != want)
418 goto out;
419 seek = 0;
420 } else {
421 seek -= want;
422 offset += want;
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
427 GFP_KERNEL);
428 if (seek < want) {
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
433 if (ret <= 0)
434 goto sock_err;
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 ret -= buf->page_base;
437 offset += ret;
438 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
439 goto out;
440 if (ret != want)
441 goto out;
442 seek = 0;
443 } else {
444 seek -= want;
445 offset += want;
448 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
449 if (seek < want) {
450 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
451 if (ret <= 0)
452 goto sock_err;
453 offset += ret;
454 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
455 goto out;
456 if (ret != want)
457 goto out;
458 } else if (offset < seek_init)
459 offset = seek_init;
460 ret = -EMSGSIZE;
461 out:
462 *read = offset - seek_init;
463 return ret;
464 sock_err:
465 offset += seek;
466 goto out;
469 static void
470 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
472 if (!transport->recv.copied) {
473 if (buf->head[0].iov_len >= transport->recv.offset)
474 memcpy(buf->head[0].iov_base,
475 &transport->recv.xid,
476 transport->recv.offset);
477 transport->recv.copied = transport->recv.offset;
481 static bool
482 xs_read_stream_request_done(struct sock_xprt *transport)
484 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
487 static void
488 xs_read_stream_check_eor(struct sock_xprt *transport,
489 struct msghdr *msg)
491 if (xs_read_stream_request_done(transport))
492 msg->msg_flags |= MSG_EOR;
495 static ssize_t
496 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
497 int flags, struct rpc_rqst *req)
499 struct xdr_buf *buf = &req->rq_private_buf;
500 size_t want, read;
501 ssize_t ret;
503 xs_read_header(transport, buf);
505 want = transport->recv.len - transport->recv.offset;
506 if (want != 0) {
507 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
508 transport->recv.copied + want,
509 transport->recv.copied,
510 &read);
511 transport->recv.offset += read;
512 transport->recv.copied += read;
515 if (transport->recv.offset == transport->recv.len)
516 xs_read_stream_check_eor(transport, msg);
518 if (want == 0)
519 return 0;
521 switch (ret) {
522 default:
523 break;
524 case -EFAULT:
525 case -EMSGSIZE:
526 msg->msg_flags |= MSG_TRUNC;
527 return read;
528 case 0:
529 return -ESHUTDOWN;
531 return ret < 0 ? ret : read;
534 static size_t
535 xs_read_stream_headersize(bool isfrag)
537 if (isfrag)
538 return sizeof(__be32);
539 return 3 * sizeof(__be32);
542 static ssize_t
543 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
544 int flags, size_t want, size_t seek)
546 struct kvec kvec = {
547 .iov_base = &transport->recv.fraghdr,
548 .iov_len = want,
550 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
553 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
554 static ssize_t
555 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
557 struct rpc_xprt *xprt = &transport->xprt;
558 struct rpc_rqst *req;
559 ssize_t ret;
561 /* Look up and lock the request corresponding to the given XID */
562 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
563 if (!req) {
564 printk(KERN_WARNING "Callback slot table overflowed\n");
565 return -ESHUTDOWN;
567 if (transport->recv.copied && !req->rq_private_buf.len)
568 return -ESHUTDOWN;
570 ret = xs_read_stream_request(transport, msg, flags, req);
571 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
572 xprt_complete_bc_request(req, transport->recv.copied);
573 else
574 req->rq_private_buf.len = transport->recv.copied;
576 return ret;
578 #else /* CONFIG_SUNRPC_BACKCHANNEL */
579 static ssize_t
580 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
582 return -ESHUTDOWN;
584 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
586 static ssize_t
587 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
589 struct rpc_xprt *xprt = &transport->xprt;
590 struct rpc_rqst *req;
591 ssize_t ret = 0;
593 /* Look up and lock the request corresponding to the given XID */
594 spin_lock(&xprt->queue_lock);
595 req = xprt_lookup_rqst(xprt, transport->recv.xid);
596 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
597 msg->msg_flags |= MSG_TRUNC;
598 goto out;
600 xprt_pin_rqst(req);
601 spin_unlock(&xprt->queue_lock);
603 ret = xs_read_stream_request(transport, msg, flags, req);
605 spin_lock(&xprt->queue_lock);
606 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
607 xprt_complete_rqst(req->rq_task, transport->recv.copied);
608 else
609 req->rq_private_buf.len = transport->recv.copied;
610 xprt_unpin_rqst(req);
611 out:
612 spin_unlock(&xprt->queue_lock);
613 return ret;
616 static ssize_t
617 xs_read_stream(struct sock_xprt *transport, int flags)
619 struct msghdr msg = { 0 };
620 size_t want, read = 0;
621 ssize_t ret = 0;
623 if (transport->recv.len == 0) {
624 want = xs_read_stream_headersize(transport->recv.copied != 0);
625 ret = xs_read_stream_header(transport, &msg, flags, want,
626 transport->recv.offset);
627 if (ret <= 0)
628 goto out_err;
629 transport->recv.offset = ret;
630 if (transport->recv.offset != want)
631 return transport->recv.offset;
632 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
633 RPC_FRAGMENT_SIZE_MASK;
634 transport->recv.offset -= sizeof(transport->recv.fraghdr);
635 read = ret;
638 switch (be32_to_cpu(transport->recv.calldir)) {
639 default:
640 msg.msg_flags |= MSG_TRUNC;
641 break;
642 case RPC_CALL:
643 ret = xs_read_stream_call(transport, &msg, flags);
644 break;
645 case RPC_REPLY:
646 ret = xs_read_stream_reply(transport, &msg, flags);
648 if (msg.msg_flags & MSG_TRUNC) {
649 transport->recv.calldir = cpu_to_be32(-1);
650 transport->recv.copied = -1;
652 if (ret < 0)
653 goto out_err;
654 read += ret;
655 if (transport->recv.offset < transport->recv.len) {
656 if (!(msg.msg_flags & MSG_TRUNC))
657 return read;
658 msg.msg_flags = 0;
659 ret = xs_read_discard(transport->sock, &msg, flags,
660 transport->recv.len - transport->recv.offset);
661 if (ret <= 0)
662 goto out_err;
663 transport->recv.offset += ret;
664 read += ret;
665 if (transport->recv.offset != transport->recv.len)
666 return read;
668 if (xs_read_stream_request_done(transport)) {
669 trace_xs_stream_read_request(transport);
670 transport->recv.copied = 0;
672 transport->recv.offset = 0;
673 transport->recv.len = 0;
674 return read;
675 out_err:
676 return ret != 0 ? ret : -ESHUTDOWN;
679 static __poll_t xs_poll_socket(struct sock_xprt *transport)
681 return transport->sock->ops->poll(transport->file, transport->sock,
682 NULL);
685 static bool xs_poll_socket_readable(struct sock_xprt *transport)
687 __poll_t events = xs_poll_socket(transport);
689 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
692 static void xs_poll_check_readable(struct sock_xprt *transport)
695 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
696 if (!xs_poll_socket_readable(transport))
697 return;
698 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
699 queue_work(xprtiod_workqueue, &transport->recv_worker);
702 static void xs_stream_data_receive(struct sock_xprt *transport)
704 size_t read = 0;
705 ssize_t ret = 0;
707 mutex_lock(&transport->recv_mutex);
708 if (transport->sock == NULL)
709 goto out;
710 for (;;) {
711 ret = xs_read_stream(transport, MSG_DONTWAIT);
712 if (ret < 0)
713 break;
714 read += ret;
715 cond_resched();
717 if (ret == -ESHUTDOWN)
718 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
719 else
720 xs_poll_check_readable(transport);
721 out:
722 mutex_unlock(&transport->recv_mutex);
723 trace_xs_stream_read_data(&transport->xprt, ret, read);
726 static void xs_stream_data_receive_workfn(struct work_struct *work)
728 struct sock_xprt *transport =
729 container_of(work, struct sock_xprt, recv_worker);
730 unsigned int pflags = memalloc_nofs_save();
732 xs_stream_data_receive(transport);
733 memalloc_nofs_restore(pflags);
736 static void
737 xs_stream_reset_connect(struct sock_xprt *transport)
739 transport->recv.offset = 0;
740 transport->recv.len = 0;
741 transport->recv.copied = 0;
742 transport->xmit.offset = 0;
745 static void
746 xs_stream_start_connect(struct sock_xprt *transport)
748 transport->xprt.stat.connect_count++;
749 transport->xprt.stat.connect_start = jiffies;
752 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
755 * xs_nospace - handle transmit was incomplete
756 * @req: pointer to RPC request
759 static int xs_nospace(struct rpc_rqst *req)
761 struct rpc_xprt *xprt = req->rq_xprt;
762 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
763 struct sock *sk = transport->inet;
764 int ret = -EAGAIN;
766 trace_rpc_socket_nospace(req, transport);
768 /* Protect against races with write_space */
769 spin_lock(&xprt->transport_lock);
771 /* Don't race with disconnect */
772 if (xprt_connected(xprt)) {
773 /* wait for more buffer space */
774 sk->sk_write_pending++;
775 xprt_wait_for_buffer_space(xprt);
776 } else
777 ret = -ENOTCONN;
779 spin_unlock(&xprt->transport_lock);
781 /* Race breaker in case memory is freed before above code is called */
782 if (ret == -EAGAIN) {
783 struct socket_wq *wq;
785 rcu_read_lock();
786 wq = rcu_dereference(sk->sk_wq);
787 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
788 rcu_read_unlock();
790 sk->sk_write_space(sk);
792 return ret;
795 static void
796 xs_stream_prepare_request(struct rpc_rqst *req)
798 xdr_free_bvec(&req->rq_rcv_buf);
799 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
803 * Determine if the previous message in the stream was aborted before it
804 * could complete transmission.
806 static bool
807 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
809 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
813 * Return the stream record marker field for a record of length < 2^31-1
815 static rpc_fraghdr
816 xs_stream_record_marker(struct xdr_buf *xdr)
818 if (!xdr->len)
819 return 0;
820 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
824 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
825 * @req: pointer to RPC request
827 * Return values:
828 * 0: The request has been sent
829 * EAGAIN: The socket was blocked, please call again later to
830 * complete the request
831 * ENOTCONN: Caller needs to invoke connect logic then call again
832 * other: Some other error occured, the request was not sent
834 static int xs_local_send_request(struct rpc_rqst *req)
836 struct rpc_xprt *xprt = req->rq_xprt;
837 struct sock_xprt *transport =
838 container_of(xprt, struct sock_xprt, xprt);
839 struct xdr_buf *xdr = &req->rq_snd_buf;
840 rpc_fraghdr rm = xs_stream_record_marker(xdr);
841 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
842 struct msghdr msg = {
843 .msg_flags = XS_SENDMSG_FLAGS,
845 unsigned int sent;
846 int status;
848 /* Close the stream if the previous transmission was incomplete */
849 if (xs_send_request_was_aborted(transport, req)) {
850 xs_close(xprt);
851 return -ENOTCONN;
854 xs_pktdump("packet data:",
855 req->rq_svec->iov_base, req->rq_svec->iov_len);
857 req->rq_xtime = ktime_get();
858 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
859 transport->xmit.offset, rm, &sent);
860 dprintk("RPC: %s(%u) = %d\n",
861 __func__, xdr->len - transport->xmit.offset, status);
863 if (status == -EAGAIN && sock_writeable(transport->inet))
864 status = -ENOBUFS;
866 if (likely(sent > 0) || status == 0) {
867 transport->xmit.offset += sent;
868 req->rq_bytes_sent = transport->xmit.offset;
869 if (likely(req->rq_bytes_sent >= msglen)) {
870 req->rq_xmit_bytes_sent += transport->xmit.offset;
871 transport->xmit.offset = 0;
872 return 0;
874 status = -EAGAIN;
877 switch (status) {
878 case -ENOBUFS:
879 break;
880 case -EAGAIN:
881 status = xs_nospace(req);
882 break;
883 default:
884 dprintk("RPC: sendmsg returned unrecognized error %d\n",
885 -status);
886 fallthrough;
887 case -EPIPE:
888 xs_close(xprt);
889 status = -ENOTCONN;
892 return status;
896 * xs_udp_send_request - write an RPC request to a UDP socket
897 * @req: pointer to RPC request
899 * Return values:
900 * 0: The request has been sent
901 * EAGAIN: The socket was blocked, please call again later to
902 * complete the request
903 * ENOTCONN: Caller needs to invoke connect logic then call again
904 * other: Some other error occurred, the request was not sent
906 static int xs_udp_send_request(struct rpc_rqst *req)
908 struct rpc_xprt *xprt = req->rq_xprt;
909 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
910 struct xdr_buf *xdr = &req->rq_snd_buf;
911 struct msghdr msg = {
912 .msg_name = xs_addr(xprt),
913 .msg_namelen = xprt->addrlen,
914 .msg_flags = XS_SENDMSG_FLAGS,
916 unsigned int sent;
917 int status;
919 xs_pktdump("packet data:",
920 req->rq_svec->iov_base,
921 req->rq_svec->iov_len);
923 if (!xprt_bound(xprt))
924 return -ENOTCONN;
926 if (!xprt_request_get_cong(xprt, req))
927 return -EBADSLT;
929 req->rq_xtime = ktime_get();
930 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
932 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
933 xdr->len, status);
935 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
936 if (status == -EPERM)
937 goto process_status;
939 if (status == -EAGAIN && sock_writeable(transport->inet))
940 status = -ENOBUFS;
942 if (sent > 0 || status == 0) {
943 req->rq_xmit_bytes_sent += sent;
944 if (sent >= req->rq_slen)
945 return 0;
946 /* Still some bytes left; set up for a retry later. */
947 status = -EAGAIN;
950 process_status:
951 switch (status) {
952 case -ENOTSOCK:
953 status = -ENOTCONN;
954 /* Should we call xs_close() here? */
955 break;
956 case -EAGAIN:
957 status = xs_nospace(req);
958 break;
959 case -ENETUNREACH:
960 case -ENOBUFS:
961 case -EPIPE:
962 case -ECONNREFUSED:
963 case -EPERM:
964 /* When the server has died, an ICMP port unreachable message
965 * prompts ECONNREFUSED. */
966 break;
967 default:
968 dprintk("RPC: sendmsg returned unrecognized error %d\n",
969 -status);
972 return status;
976 * xs_tcp_send_request - write an RPC request to a TCP socket
977 * @req: pointer to RPC request
979 * Return values:
980 * 0: The request has been sent
981 * EAGAIN: The socket was blocked, please call again later to
982 * complete the request
983 * ENOTCONN: Caller needs to invoke connect logic then call again
984 * other: Some other error occurred, the request was not sent
986 * XXX: In the case of soft timeouts, should we eventually give up
987 * if sendmsg is not able to make progress?
989 static int xs_tcp_send_request(struct rpc_rqst *req)
991 struct rpc_xprt *xprt = req->rq_xprt;
992 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
993 struct xdr_buf *xdr = &req->rq_snd_buf;
994 rpc_fraghdr rm = xs_stream_record_marker(xdr);
995 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
996 struct msghdr msg = {
997 .msg_flags = XS_SENDMSG_FLAGS,
999 bool vm_wait = false;
1000 unsigned int sent;
1001 int status;
1003 /* Close the stream if the previous transmission was incomplete */
1004 if (xs_send_request_was_aborted(transport, req)) {
1005 if (transport->sock != NULL)
1006 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1007 return -ENOTCONN;
1010 xs_pktdump("packet data:",
1011 req->rq_svec->iov_base,
1012 req->rq_svec->iov_len);
1014 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1015 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1017 /* Continue transmitting the packet/record. We must be careful
1018 * to cope with writespace callbacks arriving _after_ we have
1019 * called sendmsg(). */
1020 req->rq_xtime = ktime_get();
1021 while (1) {
1022 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1023 transport->xmit.offset, rm, &sent);
1025 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1026 xdr->len - transport->xmit.offset, status);
1028 /* If we've sent the entire packet, immediately
1029 * reset the count of bytes sent. */
1030 transport->xmit.offset += sent;
1031 req->rq_bytes_sent = transport->xmit.offset;
1032 if (likely(req->rq_bytes_sent >= msglen)) {
1033 req->rq_xmit_bytes_sent += transport->xmit.offset;
1034 transport->xmit.offset = 0;
1035 return 0;
1038 WARN_ON_ONCE(sent == 0 && status == 0);
1040 if (status == -EAGAIN ) {
1042 * Return EAGAIN if we're sure we're hitting the
1043 * socket send buffer limits.
1045 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1046 break;
1048 * Did we hit a memory allocation failure?
1050 if (sent == 0) {
1051 status = -ENOBUFS;
1052 if (vm_wait)
1053 break;
1054 /* Retry, knowing now that we're below the
1055 * socket send buffer limit
1057 vm_wait = true;
1059 continue;
1061 if (status < 0)
1062 break;
1063 vm_wait = false;
1066 switch (status) {
1067 case -ENOTSOCK:
1068 status = -ENOTCONN;
1069 /* Should we call xs_close() here? */
1070 break;
1071 case -EAGAIN:
1072 status = xs_nospace(req);
1073 break;
1074 case -ECONNRESET:
1075 case -ECONNREFUSED:
1076 case -ENOTCONN:
1077 case -EADDRINUSE:
1078 case -ENOBUFS:
1079 case -EPIPE:
1080 break;
1081 default:
1082 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1083 -status);
1086 return status;
1089 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1091 transport->old_data_ready = sk->sk_data_ready;
1092 transport->old_state_change = sk->sk_state_change;
1093 transport->old_write_space = sk->sk_write_space;
1094 transport->old_error_report = sk->sk_error_report;
1097 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1099 sk->sk_data_ready = transport->old_data_ready;
1100 sk->sk_state_change = transport->old_state_change;
1101 sk->sk_write_space = transport->old_write_space;
1102 sk->sk_error_report = transport->old_error_report;
1105 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1107 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1109 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1110 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1111 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1112 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1115 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1117 set_bit(nr, &transport->sock_state);
1118 queue_work(xprtiod_workqueue, &transport->error_worker);
1121 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1123 smp_mb__before_atomic();
1124 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1125 clear_bit(XPRT_CLOSING, &xprt->state);
1126 xs_sock_reset_state_flags(xprt);
1127 smp_mb__after_atomic();
1131 * xs_error_report - callback to handle TCP socket state errors
1132 * @sk: socket
1134 * Note: we don't call sock_error() since there may be a rpc_task
1135 * using the socket, and so we don't want to clear sk->sk_err.
1137 static void xs_error_report(struct sock *sk)
1139 struct sock_xprt *transport;
1140 struct rpc_xprt *xprt;
1142 read_lock_bh(&sk->sk_callback_lock);
1143 if (!(xprt = xprt_from_sock(sk)))
1144 goto out;
1146 transport = container_of(xprt, struct sock_xprt, xprt);
1147 transport->xprt_err = -sk->sk_err;
1148 if (transport->xprt_err == 0)
1149 goto out;
1150 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1151 xprt, -transport->xprt_err);
1152 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1154 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1155 smp_mb__before_atomic();
1156 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1157 out:
1158 read_unlock_bh(&sk->sk_callback_lock);
1161 static void xs_reset_transport(struct sock_xprt *transport)
1163 struct socket *sock = transport->sock;
1164 struct sock *sk = transport->inet;
1165 struct rpc_xprt *xprt = &transport->xprt;
1166 struct file *filp = transport->file;
1168 if (sk == NULL)
1169 return;
1171 if (atomic_read(&transport->xprt.swapper))
1172 sk_clear_memalloc(sk);
1174 kernel_sock_shutdown(sock, SHUT_RDWR);
1176 mutex_lock(&transport->recv_mutex);
1177 write_lock_bh(&sk->sk_callback_lock);
1178 transport->inet = NULL;
1179 transport->sock = NULL;
1180 transport->file = NULL;
1182 sk->sk_user_data = NULL;
1184 xs_restore_old_callbacks(transport, sk);
1185 xprt_clear_connected(xprt);
1186 write_unlock_bh(&sk->sk_callback_lock);
1187 xs_sock_reset_connection_flags(xprt);
1188 /* Reset stream record info */
1189 xs_stream_reset_connect(transport);
1190 mutex_unlock(&transport->recv_mutex);
1192 trace_rpc_socket_close(xprt, sock);
1193 fput(filp);
1195 xprt_disconnect_done(xprt);
1199 * xs_close - close a socket
1200 * @xprt: transport
1202 * This is used when all requests are complete; ie, no DRC state remains
1203 * on the server we want to save.
1205 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1206 * xs_reset_transport() zeroing the socket from underneath a writer.
1208 static void xs_close(struct rpc_xprt *xprt)
1210 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1212 dprintk("RPC: xs_close xprt %p\n", xprt);
1214 xs_reset_transport(transport);
1215 xprt->reestablish_timeout = 0;
1218 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1220 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1221 xprt);
1222 xprt_disconnect_done(xprt);
1225 static void xs_xprt_free(struct rpc_xprt *xprt)
1227 xs_free_peer_addresses(xprt);
1228 xprt_free(xprt);
1232 * xs_destroy - prepare to shutdown a transport
1233 * @xprt: doomed transport
1236 static void xs_destroy(struct rpc_xprt *xprt)
1238 struct sock_xprt *transport = container_of(xprt,
1239 struct sock_xprt, xprt);
1240 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1242 cancel_delayed_work_sync(&transport->connect_worker);
1243 xs_close(xprt);
1244 cancel_work_sync(&transport->recv_worker);
1245 cancel_work_sync(&transport->error_worker);
1246 xs_xprt_free(xprt);
1247 module_put(THIS_MODULE);
1251 * xs_udp_data_read_skb - receive callback for UDP sockets
1252 * @xprt: transport
1253 * @sk: socket
1254 * @skb: skbuff
1257 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1258 struct sock *sk,
1259 struct sk_buff *skb)
1261 struct rpc_task *task;
1262 struct rpc_rqst *rovr;
1263 int repsize, copied;
1264 u32 _xid;
1265 __be32 *xp;
1267 repsize = skb->len;
1268 if (repsize < 4) {
1269 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1270 return;
1273 /* Copy the XID from the skb... */
1274 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1275 if (xp == NULL)
1276 return;
1278 /* Look up and lock the request corresponding to the given XID */
1279 spin_lock(&xprt->queue_lock);
1280 rovr = xprt_lookup_rqst(xprt, *xp);
1281 if (!rovr)
1282 goto out_unlock;
1283 xprt_pin_rqst(rovr);
1284 xprt_update_rtt(rovr->rq_task);
1285 spin_unlock(&xprt->queue_lock);
1286 task = rovr->rq_task;
1288 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1289 copied = repsize;
1291 /* Suck it into the iovec, verify checksum if not done by hw. */
1292 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1293 spin_lock(&xprt->queue_lock);
1294 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1295 goto out_unpin;
1299 spin_lock(&xprt->transport_lock);
1300 xprt_adjust_cwnd(xprt, task, copied);
1301 spin_unlock(&xprt->transport_lock);
1302 spin_lock(&xprt->queue_lock);
1303 xprt_complete_rqst(task, copied);
1304 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1305 out_unpin:
1306 xprt_unpin_rqst(rovr);
1307 out_unlock:
1308 spin_unlock(&xprt->queue_lock);
1311 static void xs_udp_data_receive(struct sock_xprt *transport)
1313 struct sk_buff *skb;
1314 struct sock *sk;
1315 int err;
1317 mutex_lock(&transport->recv_mutex);
1318 sk = transport->inet;
1319 if (sk == NULL)
1320 goto out;
1321 for (;;) {
1322 skb = skb_recv_udp(sk, 0, 1, &err);
1323 if (skb == NULL)
1324 break;
1325 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1326 consume_skb(skb);
1327 cond_resched();
1329 xs_poll_check_readable(transport);
1330 out:
1331 mutex_unlock(&transport->recv_mutex);
1334 static void xs_udp_data_receive_workfn(struct work_struct *work)
1336 struct sock_xprt *transport =
1337 container_of(work, struct sock_xprt, recv_worker);
1338 unsigned int pflags = memalloc_nofs_save();
1340 xs_udp_data_receive(transport);
1341 memalloc_nofs_restore(pflags);
1345 * xs_data_ready - "data ready" callback for UDP sockets
1346 * @sk: socket with data to read
1349 static void xs_data_ready(struct sock *sk)
1351 struct rpc_xprt *xprt;
1353 read_lock_bh(&sk->sk_callback_lock);
1354 dprintk("RPC: xs_data_ready...\n");
1355 xprt = xprt_from_sock(sk);
1356 if (xprt != NULL) {
1357 struct sock_xprt *transport = container_of(xprt,
1358 struct sock_xprt, xprt);
1359 transport->old_data_ready(sk);
1360 /* Any data means we had a useful conversation, so
1361 * then we don't need to delay the next reconnect
1363 if (xprt->reestablish_timeout)
1364 xprt->reestablish_timeout = 0;
1365 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1366 queue_work(xprtiod_workqueue, &transport->recv_worker);
1368 read_unlock_bh(&sk->sk_callback_lock);
1372 * Helper function to force a TCP close if the server is sending
1373 * junk and/or it has put us in CLOSE_WAIT
1375 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1377 xprt_force_disconnect(xprt);
1380 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1381 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1383 return PAGE_SIZE;
1385 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1388 * xs_tcp_state_change - callback to handle TCP socket state changes
1389 * @sk: socket whose state has changed
1392 static void xs_tcp_state_change(struct sock *sk)
1394 struct rpc_xprt *xprt;
1395 struct sock_xprt *transport;
1397 read_lock_bh(&sk->sk_callback_lock);
1398 if (!(xprt = xprt_from_sock(sk)))
1399 goto out;
1400 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1401 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1402 sk->sk_state, xprt_connected(xprt),
1403 sock_flag(sk, SOCK_DEAD),
1404 sock_flag(sk, SOCK_ZAPPED),
1405 sk->sk_shutdown);
1407 transport = container_of(xprt, struct sock_xprt, xprt);
1408 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1409 switch (sk->sk_state) {
1410 case TCP_ESTABLISHED:
1411 if (!xprt_test_and_set_connected(xprt)) {
1412 xprt->connect_cookie++;
1413 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1414 xprt_clear_connecting(xprt);
1416 xprt->stat.connect_count++;
1417 xprt->stat.connect_time += (long)jiffies -
1418 xprt->stat.connect_start;
1419 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1421 break;
1422 case TCP_FIN_WAIT1:
1423 /* The client initiated a shutdown of the socket */
1424 xprt->connect_cookie++;
1425 xprt->reestablish_timeout = 0;
1426 set_bit(XPRT_CLOSING, &xprt->state);
1427 smp_mb__before_atomic();
1428 clear_bit(XPRT_CONNECTED, &xprt->state);
1429 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1430 smp_mb__after_atomic();
1431 break;
1432 case TCP_CLOSE_WAIT:
1433 /* The server initiated a shutdown of the socket */
1434 xprt->connect_cookie++;
1435 clear_bit(XPRT_CONNECTED, &xprt->state);
1436 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1437 fallthrough;
1438 case TCP_CLOSING:
1440 * If the server closed down the connection, make sure that
1441 * we back off before reconnecting
1443 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1444 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1445 break;
1446 case TCP_LAST_ACK:
1447 set_bit(XPRT_CLOSING, &xprt->state);
1448 smp_mb__before_atomic();
1449 clear_bit(XPRT_CONNECTED, &xprt->state);
1450 smp_mb__after_atomic();
1451 break;
1452 case TCP_CLOSE:
1453 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1454 &transport->sock_state))
1455 xprt_clear_connecting(xprt);
1456 clear_bit(XPRT_CLOSING, &xprt->state);
1457 /* Trigger the socket release */
1458 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1460 out:
1461 read_unlock_bh(&sk->sk_callback_lock);
1464 static void xs_write_space(struct sock *sk)
1466 struct socket_wq *wq;
1467 struct sock_xprt *transport;
1468 struct rpc_xprt *xprt;
1470 if (!sk->sk_socket)
1471 return;
1472 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1474 if (unlikely(!(xprt = xprt_from_sock(sk))))
1475 return;
1476 transport = container_of(xprt, struct sock_xprt, xprt);
1477 rcu_read_lock();
1478 wq = rcu_dereference(sk->sk_wq);
1479 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1480 goto out;
1482 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1483 sk->sk_write_pending--;
1484 out:
1485 rcu_read_unlock();
1489 * xs_udp_write_space - callback invoked when socket buffer space
1490 * becomes available
1491 * @sk: socket whose state has changed
1493 * Called when more output buffer space is available for this socket.
1494 * We try not to wake our writers until they can make "significant"
1495 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1496 * with a bunch of small requests.
1498 static void xs_udp_write_space(struct sock *sk)
1500 read_lock_bh(&sk->sk_callback_lock);
1502 /* from net/core/sock.c:sock_def_write_space */
1503 if (sock_writeable(sk))
1504 xs_write_space(sk);
1506 read_unlock_bh(&sk->sk_callback_lock);
1510 * xs_tcp_write_space - callback invoked when socket buffer space
1511 * becomes available
1512 * @sk: socket whose state has changed
1514 * Called when more output buffer space is available for this socket.
1515 * We try not to wake our writers until they can make "significant"
1516 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1517 * with a bunch of small requests.
1519 static void xs_tcp_write_space(struct sock *sk)
1521 read_lock_bh(&sk->sk_callback_lock);
1523 /* from net/core/stream.c:sk_stream_write_space */
1524 if (sk_stream_is_writeable(sk))
1525 xs_write_space(sk);
1527 read_unlock_bh(&sk->sk_callback_lock);
1530 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1532 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1533 struct sock *sk = transport->inet;
1535 if (transport->rcvsize) {
1536 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1537 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1539 if (transport->sndsize) {
1540 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1541 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1542 sk->sk_write_space(sk);
1547 * xs_udp_set_buffer_size - set send and receive limits
1548 * @xprt: generic transport
1549 * @sndsize: requested size of send buffer, in bytes
1550 * @rcvsize: requested size of receive buffer, in bytes
1552 * Set socket send and receive buffer size limits.
1554 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1556 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1558 transport->sndsize = 0;
1559 if (sndsize)
1560 transport->sndsize = sndsize + 1024;
1561 transport->rcvsize = 0;
1562 if (rcvsize)
1563 transport->rcvsize = rcvsize + 1024;
1565 xs_udp_do_set_buffer_size(xprt);
1569 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1570 * @xprt: controlling transport
1571 * @task: task that timed out
1573 * Adjust the congestion window after a retransmit timeout has occurred.
1575 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1577 spin_lock(&xprt->transport_lock);
1578 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1579 spin_unlock(&xprt->transport_lock);
1582 static int xs_get_random_port(void)
1584 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1585 unsigned short range;
1586 unsigned short rand;
1588 if (max < min)
1589 return -EADDRINUSE;
1590 range = max - min + 1;
1591 rand = (unsigned short) prandom_u32() % range;
1592 return rand + min;
1595 static unsigned short xs_sock_getport(struct socket *sock)
1597 struct sockaddr_storage buf;
1598 unsigned short port = 0;
1600 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1601 goto out;
1602 switch (buf.ss_family) {
1603 case AF_INET6:
1604 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1605 break;
1606 case AF_INET:
1607 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1609 out:
1610 return port;
1614 * xs_set_port - reset the port number in the remote endpoint address
1615 * @xprt: generic transport
1616 * @port: new port number
1619 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1621 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1623 rpc_set_port(xs_addr(xprt), port);
1624 xs_update_peer_port(xprt);
1627 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1629 if (transport->srcport == 0 && transport->xprt.reuseport)
1630 transport->srcport = xs_sock_getport(sock);
1633 static int xs_get_srcport(struct sock_xprt *transport)
1635 int port = transport->srcport;
1637 if (port == 0 && transport->xprt.resvport)
1638 port = xs_get_random_port();
1639 return port;
1642 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1644 if (transport->srcport != 0)
1645 transport->srcport = 0;
1646 if (!transport->xprt.resvport)
1647 return 0;
1648 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1649 return xprt_max_resvport;
1650 return --port;
1652 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1654 struct sockaddr_storage myaddr;
1655 int err, nloop = 0;
1656 int port = xs_get_srcport(transport);
1657 unsigned short last;
1660 * If we are asking for any ephemeral port (i.e. port == 0 &&
1661 * transport->xprt.resvport == 0), don't bind. Let the local
1662 * port selection happen implicitly when the socket is used
1663 * (for example at connect time).
1665 * This ensures that we can continue to establish TCP
1666 * connections even when all local ephemeral ports are already
1667 * a part of some TCP connection. This makes no difference
1668 * for UDP sockets, but also doens't harm them.
1670 * If we're asking for any reserved port (i.e. port == 0 &&
1671 * transport->xprt.resvport == 1) xs_get_srcport above will
1672 * ensure that port is non-zero and we will bind as needed.
1674 if (port <= 0)
1675 return port;
1677 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1678 do {
1679 rpc_set_port((struct sockaddr *)&myaddr, port);
1680 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1681 transport->xprt.addrlen);
1682 if (err == 0) {
1683 transport->srcport = port;
1684 break;
1686 last = port;
1687 port = xs_next_srcport(transport, port);
1688 if (port > last)
1689 nloop++;
1690 } while (err == -EADDRINUSE && nloop != 2);
1692 if (myaddr.ss_family == AF_INET)
1693 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1694 &((struct sockaddr_in *)&myaddr)->sin_addr,
1695 port, err ? "failed" : "ok", err);
1696 else
1697 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1698 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1699 port, err ? "failed" : "ok", err);
1700 return err;
1704 * We don't support autobind on AF_LOCAL sockets
1706 static void xs_local_rpcbind(struct rpc_task *task)
1708 xprt_set_bound(task->tk_xprt);
1711 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1715 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1716 static struct lock_class_key xs_key[2];
1717 static struct lock_class_key xs_slock_key[2];
1719 static inline void xs_reclassify_socketu(struct socket *sock)
1721 struct sock *sk = sock->sk;
1723 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1724 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1727 static inline void xs_reclassify_socket4(struct socket *sock)
1729 struct sock *sk = sock->sk;
1731 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1732 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1735 static inline void xs_reclassify_socket6(struct socket *sock)
1737 struct sock *sk = sock->sk;
1739 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1740 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1743 static inline void xs_reclassify_socket(int family, struct socket *sock)
1745 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1746 return;
1748 switch (family) {
1749 case AF_LOCAL:
1750 xs_reclassify_socketu(sock);
1751 break;
1752 case AF_INET:
1753 xs_reclassify_socket4(sock);
1754 break;
1755 case AF_INET6:
1756 xs_reclassify_socket6(sock);
1757 break;
1760 #else
1761 static inline void xs_reclassify_socket(int family, struct socket *sock)
1764 #endif
1766 static void xs_dummy_setup_socket(struct work_struct *work)
1770 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1771 struct sock_xprt *transport, int family, int type,
1772 int protocol, bool reuseport)
1774 struct file *filp;
1775 struct socket *sock;
1776 int err;
1778 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1779 if (err < 0) {
1780 dprintk("RPC: can't create %d transport socket (%d).\n",
1781 protocol, -err);
1782 goto out;
1784 xs_reclassify_socket(family, sock);
1786 if (reuseport)
1787 sock_set_reuseport(sock->sk);
1789 err = xs_bind(transport, sock);
1790 if (err) {
1791 sock_release(sock);
1792 goto out;
1795 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1796 if (IS_ERR(filp))
1797 return ERR_CAST(filp);
1798 transport->file = filp;
1800 return sock;
1801 out:
1802 return ERR_PTR(err);
1805 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1806 struct socket *sock)
1808 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1809 xprt);
1811 if (!transport->inet) {
1812 struct sock *sk = sock->sk;
1814 write_lock_bh(&sk->sk_callback_lock);
1816 xs_save_old_callbacks(transport, sk);
1818 sk->sk_user_data = xprt;
1819 sk->sk_data_ready = xs_data_ready;
1820 sk->sk_write_space = xs_udp_write_space;
1821 sock_set_flag(sk, SOCK_FASYNC);
1822 sk->sk_error_report = xs_error_report;
1824 xprt_clear_connected(xprt);
1826 /* Reset to new socket */
1827 transport->sock = sock;
1828 transport->inet = sk;
1830 write_unlock_bh(&sk->sk_callback_lock);
1833 xs_stream_start_connect(transport);
1835 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1839 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1840 * @transport: socket transport to connect
1842 static int xs_local_setup_socket(struct sock_xprt *transport)
1844 struct rpc_xprt *xprt = &transport->xprt;
1845 struct file *filp;
1846 struct socket *sock;
1847 int status;
1849 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1850 SOCK_STREAM, 0, &sock, 1);
1851 if (status < 0) {
1852 dprintk("RPC: can't create AF_LOCAL "
1853 "transport socket (%d).\n", -status);
1854 goto out;
1856 xs_reclassify_socket(AF_LOCAL, sock);
1858 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1859 if (IS_ERR(filp)) {
1860 status = PTR_ERR(filp);
1861 goto out;
1863 transport->file = filp;
1865 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1866 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1868 status = xs_local_finish_connecting(xprt, sock);
1869 trace_rpc_socket_connect(xprt, sock, status);
1870 switch (status) {
1871 case 0:
1872 dprintk("RPC: xprt %p connected to %s\n",
1873 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1874 xprt->stat.connect_count++;
1875 xprt->stat.connect_time += (long)jiffies -
1876 xprt->stat.connect_start;
1877 xprt_set_connected(xprt);
1878 case -ENOBUFS:
1879 break;
1880 case -ENOENT:
1881 dprintk("RPC: xprt %p: socket %s does not exist\n",
1882 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1883 break;
1884 case -ECONNREFUSED:
1885 dprintk("RPC: xprt %p: connection refused for %s\n",
1886 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1887 break;
1888 default:
1889 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1890 __func__, -status,
1891 xprt->address_strings[RPC_DISPLAY_ADDR]);
1894 out:
1895 xprt_clear_connecting(xprt);
1896 xprt_wake_pending_tasks(xprt, status);
1897 return status;
1900 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1902 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1903 int ret;
1905 if (RPC_IS_ASYNC(task)) {
1907 * We want the AF_LOCAL connect to be resolved in the
1908 * filesystem namespace of the process making the rpc
1909 * call. Thus we connect synchronously.
1911 * If we want to support asynchronous AF_LOCAL calls,
1912 * we'll need to figure out how to pass a namespace to
1913 * connect.
1915 task->tk_rpc_status = -ENOTCONN;
1916 rpc_exit(task, -ENOTCONN);
1917 return;
1919 ret = xs_local_setup_socket(transport);
1920 if (ret && !RPC_IS_SOFTCONN(task))
1921 msleep_interruptible(15000);
1924 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1926 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1927 * know that we have exclusive access to the socket), to guard against
1928 * races with xs_reset_transport.
1930 static void xs_set_memalloc(struct rpc_xprt *xprt)
1932 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1933 xprt);
1936 * If there's no sock, then we have nothing to set. The
1937 * reconnecting process will get it for us.
1939 if (!transport->inet)
1940 return;
1941 if (atomic_read(&xprt->swapper))
1942 sk_set_memalloc(transport->inet);
1946 * xs_enable_swap - Tag this transport as being used for swap.
1947 * @xprt: transport to tag
1949 * Take a reference to this transport on behalf of the rpc_clnt, and
1950 * optionally mark it for swapping if it wasn't already.
1952 static int
1953 xs_enable_swap(struct rpc_xprt *xprt)
1955 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1957 if (atomic_inc_return(&xprt->swapper) != 1)
1958 return 0;
1959 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1960 return -ERESTARTSYS;
1961 if (xs->inet)
1962 sk_set_memalloc(xs->inet);
1963 xprt_release_xprt(xprt, NULL);
1964 return 0;
1968 * xs_disable_swap - Untag this transport as being used for swap.
1969 * @xprt: transport to tag
1971 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1972 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1974 static void
1975 xs_disable_swap(struct rpc_xprt *xprt)
1977 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1979 if (!atomic_dec_and_test(&xprt->swapper))
1980 return;
1981 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1982 return;
1983 if (xs->inet)
1984 sk_clear_memalloc(xs->inet);
1985 xprt_release_xprt(xprt, NULL);
1987 #else
1988 static void xs_set_memalloc(struct rpc_xprt *xprt)
1992 static int
1993 xs_enable_swap(struct rpc_xprt *xprt)
1995 return -EINVAL;
1998 static void
1999 xs_disable_swap(struct rpc_xprt *xprt)
2002 #endif
2004 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2006 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2008 if (!transport->inet) {
2009 struct sock *sk = sock->sk;
2011 write_lock_bh(&sk->sk_callback_lock);
2013 xs_save_old_callbacks(transport, sk);
2015 sk->sk_user_data = xprt;
2016 sk->sk_data_ready = xs_data_ready;
2017 sk->sk_write_space = xs_udp_write_space;
2018 sock_set_flag(sk, SOCK_FASYNC);
2020 xprt_set_connected(xprt);
2022 /* Reset to new socket */
2023 transport->sock = sock;
2024 transport->inet = sk;
2026 xs_set_memalloc(xprt);
2028 write_unlock_bh(&sk->sk_callback_lock);
2030 xs_udp_do_set_buffer_size(xprt);
2032 xprt->stat.connect_start = jiffies;
2035 static void xs_udp_setup_socket(struct work_struct *work)
2037 struct sock_xprt *transport =
2038 container_of(work, struct sock_xprt, connect_worker.work);
2039 struct rpc_xprt *xprt = &transport->xprt;
2040 struct socket *sock;
2041 int status = -EIO;
2043 sock = xs_create_sock(xprt, transport,
2044 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2045 IPPROTO_UDP, false);
2046 if (IS_ERR(sock))
2047 goto out;
2049 dprintk("RPC: worker connecting xprt %p via %s to "
2050 "%s (port %s)\n", xprt,
2051 xprt->address_strings[RPC_DISPLAY_PROTO],
2052 xprt->address_strings[RPC_DISPLAY_ADDR],
2053 xprt->address_strings[RPC_DISPLAY_PORT]);
2055 xs_udp_finish_connecting(xprt, sock);
2056 trace_rpc_socket_connect(xprt, sock, 0);
2057 status = 0;
2058 out:
2059 xprt_clear_connecting(xprt);
2060 xprt_unlock_connect(xprt, transport);
2061 xprt_wake_pending_tasks(xprt, status);
2065 * xs_tcp_shutdown - gracefully shut down a TCP socket
2066 * @xprt: transport
2068 * Initiates a graceful shutdown of the TCP socket by calling the
2069 * equivalent of shutdown(SHUT_RDWR);
2071 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2073 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2074 struct socket *sock = transport->sock;
2075 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2077 if (sock == NULL)
2078 return;
2079 switch (skst) {
2080 default:
2081 kernel_sock_shutdown(sock, SHUT_RDWR);
2082 trace_rpc_socket_shutdown(xprt, sock);
2083 break;
2084 case TCP_CLOSE:
2085 case TCP_TIME_WAIT:
2086 xs_reset_transport(transport);
2090 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2091 struct socket *sock)
2093 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2094 unsigned int keepidle;
2095 unsigned int keepcnt;
2096 unsigned int timeo;
2098 spin_lock(&xprt->transport_lock);
2099 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2100 keepcnt = xprt->timeout->to_retries + 1;
2101 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2102 (xprt->timeout->to_retries + 1);
2103 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2104 spin_unlock(&xprt->transport_lock);
2106 /* TCP Keepalive options */
2107 sock_set_keepalive(sock->sk);
2108 tcp_sock_set_keepidle(sock->sk, keepidle);
2109 tcp_sock_set_keepintvl(sock->sk, keepidle);
2110 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2112 /* TCP user timeout (see RFC5482) */
2113 tcp_sock_set_user_timeout(sock->sk, timeo);
2116 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2117 unsigned long connect_timeout,
2118 unsigned long reconnect_timeout)
2120 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2121 struct rpc_timeout to;
2122 unsigned long initval;
2124 spin_lock(&xprt->transport_lock);
2125 if (reconnect_timeout < xprt->max_reconnect_timeout)
2126 xprt->max_reconnect_timeout = reconnect_timeout;
2127 if (connect_timeout < xprt->connect_timeout) {
2128 memcpy(&to, xprt->timeout, sizeof(to));
2129 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2130 /* Arbitrary lower limit */
2131 if (initval < XS_TCP_INIT_REEST_TO << 1)
2132 initval = XS_TCP_INIT_REEST_TO << 1;
2133 to.to_initval = initval;
2134 to.to_maxval = initval;
2135 memcpy(&transport->tcp_timeout, &to,
2136 sizeof(transport->tcp_timeout));
2137 xprt->timeout = &transport->tcp_timeout;
2138 xprt->connect_timeout = connect_timeout;
2140 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2141 spin_unlock(&xprt->transport_lock);
2144 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2146 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2147 int ret = -ENOTCONN;
2149 if (!transport->inet) {
2150 struct sock *sk = sock->sk;
2152 /* Avoid temporary address, they are bad for long-lived
2153 * connections such as NFS mounts.
2154 * RFC4941, section 3.6 suggests that:
2155 * Individual applications, which have specific
2156 * knowledge about the normal duration of connections,
2157 * MAY override this as appropriate.
2159 if (xs_addr(xprt)->sa_family == PF_INET6) {
2160 ip6_sock_set_addr_preferences(sk,
2161 IPV6_PREFER_SRC_PUBLIC);
2164 xs_tcp_set_socket_timeouts(xprt, sock);
2166 write_lock_bh(&sk->sk_callback_lock);
2168 xs_save_old_callbacks(transport, sk);
2170 sk->sk_user_data = xprt;
2171 sk->sk_data_ready = xs_data_ready;
2172 sk->sk_state_change = xs_tcp_state_change;
2173 sk->sk_write_space = xs_tcp_write_space;
2174 sock_set_flag(sk, SOCK_FASYNC);
2175 sk->sk_error_report = xs_error_report;
2177 /* socket options */
2178 sock_reset_flag(sk, SOCK_LINGER);
2179 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2181 xprt_clear_connected(xprt);
2183 /* Reset to new socket */
2184 transport->sock = sock;
2185 transport->inet = sk;
2187 write_unlock_bh(&sk->sk_callback_lock);
2190 if (!xprt_bound(xprt))
2191 goto out;
2193 xs_set_memalloc(xprt);
2195 xs_stream_start_connect(transport);
2197 /* Tell the socket layer to start connecting... */
2198 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2199 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2200 switch (ret) {
2201 case 0:
2202 xs_set_srcport(transport, sock);
2203 fallthrough;
2204 case -EINPROGRESS:
2205 /* SYN_SENT! */
2206 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2207 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2208 break;
2209 case -EADDRNOTAVAIL:
2210 /* Source port number is unavailable. Try a new one! */
2211 transport->srcport = 0;
2213 out:
2214 return ret;
2218 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2219 * @work: queued work item
2221 * Invoked by a work queue tasklet.
2223 static void xs_tcp_setup_socket(struct work_struct *work)
2225 struct sock_xprt *transport =
2226 container_of(work, struct sock_xprt, connect_worker.work);
2227 struct socket *sock = transport->sock;
2228 struct rpc_xprt *xprt = &transport->xprt;
2229 int status = -EIO;
2231 if (!sock) {
2232 sock = xs_create_sock(xprt, transport,
2233 xs_addr(xprt)->sa_family, SOCK_STREAM,
2234 IPPROTO_TCP, true);
2235 if (IS_ERR(sock)) {
2236 status = PTR_ERR(sock);
2237 goto out;
2241 dprintk("RPC: worker connecting xprt %p via %s to "
2242 "%s (port %s)\n", xprt,
2243 xprt->address_strings[RPC_DISPLAY_PROTO],
2244 xprt->address_strings[RPC_DISPLAY_ADDR],
2245 xprt->address_strings[RPC_DISPLAY_PORT]);
2247 status = xs_tcp_finish_connecting(xprt, sock);
2248 trace_rpc_socket_connect(xprt, sock, status);
2249 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2250 xprt, -status, xprt_connected(xprt),
2251 sock->sk->sk_state);
2252 switch (status) {
2253 default:
2254 printk("%s: connect returned unhandled error %d\n",
2255 __func__, status);
2256 fallthrough;
2257 case -EADDRNOTAVAIL:
2258 /* We're probably in TIME_WAIT. Get rid of existing socket,
2259 * and retry
2261 xs_tcp_force_close(xprt);
2262 break;
2263 case 0:
2264 case -EINPROGRESS:
2265 case -EALREADY:
2266 xprt_unlock_connect(xprt, transport);
2267 return;
2268 case -EINVAL:
2269 /* Happens, for instance, if the user specified a link
2270 * local IPv6 address without a scope-id.
2272 case -ECONNREFUSED:
2273 case -ECONNRESET:
2274 case -ENETDOWN:
2275 case -ENETUNREACH:
2276 case -EHOSTUNREACH:
2277 case -EADDRINUSE:
2278 case -ENOBUFS:
2280 * xs_tcp_force_close() wakes tasks with -EIO.
2281 * We need to wake them first to ensure the
2282 * correct error code.
2284 xprt_wake_pending_tasks(xprt, status);
2285 xs_tcp_force_close(xprt);
2286 goto out;
2288 status = -EAGAIN;
2289 out:
2290 xprt_clear_connecting(xprt);
2291 xprt_unlock_connect(xprt, transport);
2292 xprt_wake_pending_tasks(xprt, status);
2296 * xs_connect - connect a socket to a remote endpoint
2297 * @xprt: pointer to transport structure
2298 * @task: address of RPC task that manages state of connect request
2300 * TCP: If the remote end dropped the connection, delay reconnecting.
2302 * UDP socket connects are synchronous, but we use a work queue anyway
2303 * to guarantee that even unprivileged user processes can set up a
2304 * socket on a privileged port.
2306 * If a UDP socket connect fails, the delay behavior here prevents
2307 * retry floods (hard mounts).
2309 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2311 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2312 unsigned long delay = 0;
2314 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2316 if (transport->sock != NULL) {
2317 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2318 "seconds\n",
2319 xprt, xprt->reestablish_timeout / HZ);
2321 /* Start by resetting any existing state */
2322 xs_reset_transport(transport);
2324 delay = xprt_reconnect_delay(xprt);
2325 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2327 } else
2328 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2330 queue_delayed_work(xprtiod_workqueue,
2331 &transport->connect_worker,
2332 delay);
2335 static void xs_wake_disconnect(struct sock_xprt *transport)
2337 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2338 xs_tcp_force_close(&transport->xprt);
2341 static void xs_wake_write(struct sock_xprt *transport)
2343 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2344 xprt_write_space(&transport->xprt);
2347 static void xs_wake_error(struct sock_xprt *transport)
2349 int sockerr;
2351 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2352 return;
2353 mutex_lock(&transport->recv_mutex);
2354 if (transport->sock == NULL)
2355 goto out;
2356 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2357 goto out;
2358 sockerr = xchg(&transport->xprt_err, 0);
2359 if (sockerr < 0)
2360 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2361 out:
2362 mutex_unlock(&transport->recv_mutex);
2365 static void xs_wake_pending(struct sock_xprt *transport)
2367 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2368 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2371 static void xs_error_handle(struct work_struct *work)
2373 struct sock_xprt *transport = container_of(work,
2374 struct sock_xprt, error_worker);
2376 xs_wake_disconnect(transport);
2377 xs_wake_write(transport);
2378 xs_wake_error(transport);
2379 xs_wake_pending(transport);
2383 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2384 * @xprt: rpc_xprt struct containing statistics
2385 * @seq: output file
2388 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2390 long idle_time = 0;
2392 if (xprt_connected(xprt))
2393 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2395 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2396 "%llu %llu %lu %llu %llu\n",
2397 xprt->stat.bind_count,
2398 xprt->stat.connect_count,
2399 xprt->stat.connect_time / HZ,
2400 idle_time,
2401 xprt->stat.sends,
2402 xprt->stat.recvs,
2403 xprt->stat.bad_xids,
2404 xprt->stat.req_u,
2405 xprt->stat.bklog_u,
2406 xprt->stat.max_slots,
2407 xprt->stat.sending_u,
2408 xprt->stat.pending_u);
2412 * xs_udp_print_stats - display UDP socket-specifc stats
2413 * @xprt: rpc_xprt struct containing statistics
2414 * @seq: output file
2417 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2419 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2421 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2422 "%lu %llu %llu\n",
2423 transport->srcport,
2424 xprt->stat.bind_count,
2425 xprt->stat.sends,
2426 xprt->stat.recvs,
2427 xprt->stat.bad_xids,
2428 xprt->stat.req_u,
2429 xprt->stat.bklog_u,
2430 xprt->stat.max_slots,
2431 xprt->stat.sending_u,
2432 xprt->stat.pending_u);
2436 * xs_tcp_print_stats - display TCP socket-specifc stats
2437 * @xprt: rpc_xprt struct containing statistics
2438 * @seq: output file
2441 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2443 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2444 long idle_time = 0;
2446 if (xprt_connected(xprt))
2447 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2449 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2450 "%llu %llu %lu %llu %llu\n",
2451 transport->srcport,
2452 xprt->stat.bind_count,
2453 xprt->stat.connect_count,
2454 xprt->stat.connect_time / HZ,
2455 idle_time,
2456 xprt->stat.sends,
2457 xprt->stat.recvs,
2458 xprt->stat.bad_xids,
2459 xprt->stat.req_u,
2460 xprt->stat.bklog_u,
2461 xprt->stat.max_slots,
2462 xprt->stat.sending_u,
2463 xprt->stat.pending_u);
2467 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2468 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2469 * to use the server side send routines.
2471 static int bc_malloc(struct rpc_task *task)
2473 struct rpc_rqst *rqst = task->tk_rqstp;
2474 size_t size = rqst->rq_callsize;
2475 struct page *page;
2476 struct rpc_buffer *buf;
2478 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2479 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2480 size);
2481 return -EINVAL;
2484 page = alloc_page(GFP_KERNEL);
2485 if (!page)
2486 return -ENOMEM;
2488 buf = page_address(page);
2489 buf->len = PAGE_SIZE;
2491 rqst->rq_buffer = buf->data;
2492 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2493 return 0;
2497 * Free the space allocated in the bc_alloc routine
2499 static void bc_free(struct rpc_task *task)
2501 void *buffer = task->tk_rqstp->rq_buffer;
2502 struct rpc_buffer *buf;
2504 buf = container_of(buffer, struct rpc_buffer, data);
2505 free_page((unsigned long)buf);
2508 static int bc_sendto(struct rpc_rqst *req)
2510 struct xdr_buf *xdr = &req->rq_snd_buf;
2511 struct sock_xprt *transport =
2512 container_of(req->rq_xprt, struct sock_xprt, xprt);
2513 struct msghdr msg = {
2514 .msg_flags = 0,
2516 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2517 (u32)xdr->len);
2518 unsigned int sent = 0;
2519 int err;
2521 req->rq_xtime = ktime_get();
2522 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2523 xdr_free_bvec(xdr);
2524 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2525 return -EAGAIN;
2526 return sent;
2530 * bc_send_request - Send a backchannel Call on a TCP socket
2531 * @req: rpc_rqst containing Call message to be sent
2533 * xpt_mutex ensures @rqstp's whole message is written to the socket
2534 * without interruption.
2536 * Return values:
2537 * %0 if the message was sent successfully
2538 * %ENOTCONN if the message was not sent
2540 static int bc_send_request(struct rpc_rqst *req)
2542 struct svc_xprt *xprt;
2543 int len;
2546 * Get the server socket associated with this callback xprt
2548 xprt = req->rq_xprt->bc_xprt;
2551 * Grab the mutex to serialize data as the connection is shared
2552 * with the fore channel
2554 mutex_lock(&xprt->xpt_mutex);
2555 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2556 len = -ENOTCONN;
2557 else
2558 len = bc_sendto(req);
2559 mutex_unlock(&xprt->xpt_mutex);
2561 if (len > 0)
2562 len = 0;
2564 return len;
2568 * The close routine. Since this is client initiated, we do nothing
2571 static void bc_close(struct rpc_xprt *xprt)
2573 xprt_disconnect_done(xprt);
2577 * The xprt destroy routine. Again, because this connection is client
2578 * initiated, we do nothing
2581 static void bc_destroy(struct rpc_xprt *xprt)
2583 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2585 xs_xprt_free(xprt);
2586 module_put(THIS_MODULE);
2589 static const struct rpc_xprt_ops xs_local_ops = {
2590 .reserve_xprt = xprt_reserve_xprt,
2591 .release_xprt = xprt_release_xprt,
2592 .alloc_slot = xprt_alloc_slot,
2593 .free_slot = xprt_free_slot,
2594 .rpcbind = xs_local_rpcbind,
2595 .set_port = xs_local_set_port,
2596 .connect = xs_local_connect,
2597 .buf_alloc = rpc_malloc,
2598 .buf_free = rpc_free,
2599 .prepare_request = xs_stream_prepare_request,
2600 .send_request = xs_local_send_request,
2601 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2602 .close = xs_close,
2603 .destroy = xs_destroy,
2604 .print_stats = xs_local_print_stats,
2605 .enable_swap = xs_enable_swap,
2606 .disable_swap = xs_disable_swap,
2609 static const struct rpc_xprt_ops xs_udp_ops = {
2610 .set_buffer_size = xs_udp_set_buffer_size,
2611 .reserve_xprt = xprt_reserve_xprt_cong,
2612 .release_xprt = xprt_release_xprt_cong,
2613 .alloc_slot = xprt_alloc_slot,
2614 .free_slot = xprt_free_slot,
2615 .rpcbind = rpcb_getport_async,
2616 .set_port = xs_set_port,
2617 .connect = xs_connect,
2618 .buf_alloc = rpc_malloc,
2619 .buf_free = rpc_free,
2620 .send_request = xs_udp_send_request,
2621 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2622 .timer = xs_udp_timer,
2623 .release_request = xprt_release_rqst_cong,
2624 .close = xs_close,
2625 .destroy = xs_destroy,
2626 .print_stats = xs_udp_print_stats,
2627 .enable_swap = xs_enable_swap,
2628 .disable_swap = xs_disable_swap,
2629 .inject_disconnect = xs_inject_disconnect,
2632 static const struct rpc_xprt_ops xs_tcp_ops = {
2633 .reserve_xprt = xprt_reserve_xprt,
2634 .release_xprt = xprt_release_xprt,
2635 .alloc_slot = xprt_alloc_slot,
2636 .free_slot = xprt_free_slot,
2637 .rpcbind = rpcb_getport_async,
2638 .set_port = xs_set_port,
2639 .connect = xs_connect,
2640 .buf_alloc = rpc_malloc,
2641 .buf_free = rpc_free,
2642 .prepare_request = xs_stream_prepare_request,
2643 .send_request = xs_tcp_send_request,
2644 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2645 .close = xs_tcp_shutdown,
2646 .destroy = xs_destroy,
2647 .set_connect_timeout = xs_tcp_set_connect_timeout,
2648 .print_stats = xs_tcp_print_stats,
2649 .enable_swap = xs_enable_swap,
2650 .disable_swap = xs_disable_swap,
2651 .inject_disconnect = xs_inject_disconnect,
2652 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2653 .bc_setup = xprt_setup_bc,
2654 .bc_maxpayload = xs_tcp_bc_maxpayload,
2655 .bc_num_slots = xprt_bc_max_slots,
2656 .bc_free_rqst = xprt_free_bc_rqst,
2657 .bc_destroy = xprt_destroy_bc,
2658 #endif
2662 * The rpc_xprt_ops for the server backchannel
2665 static const struct rpc_xprt_ops bc_tcp_ops = {
2666 .reserve_xprt = xprt_reserve_xprt,
2667 .release_xprt = xprt_release_xprt,
2668 .alloc_slot = xprt_alloc_slot,
2669 .free_slot = xprt_free_slot,
2670 .buf_alloc = bc_malloc,
2671 .buf_free = bc_free,
2672 .send_request = bc_send_request,
2673 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2674 .close = bc_close,
2675 .destroy = bc_destroy,
2676 .print_stats = xs_tcp_print_stats,
2677 .enable_swap = xs_enable_swap,
2678 .disable_swap = xs_disable_swap,
2679 .inject_disconnect = xs_inject_disconnect,
2682 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2684 static const struct sockaddr_in sin = {
2685 .sin_family = AF_INET,
2686 .sin_addr.s_addr = htonl(INADDR_ANY),
2688 static const struct sockaddr_in6 sin6 = {
2689 .sin6_family = AF_INET6,
2690 .sin6_addr = IN6ADDR_ANY_INIT,
2693 switch (family) {
2694 case AF_LOCAL:
2695 break;
2696 case AF_INET:
2697 memcpy(sap, &sin, sizeof(sin));
2698 break;
2699 case AF_INET6:
2700 memcpy(sap, &sin6, sizeof(sin6));
2701 break;
2702 default:
2703 dprintk("RPC: %s: Bad address family\n", __func__);
2704 return -EAFNOSUPPORT;
2706 return 0;
2709 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2710 unsigned int slot_table_size,
2711 unsigned int max_slot_table_size)
2713 struct rpc_xprt *xprt;
2714 struct sock_xprt *new;
2716 if (args->addrlen > sizeof(xprt->addr)) {
2717 dprintk("RPC: xs_setup_xprt: address too large\n");
2718 return ERR_PTR(-EBADF);
2721 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2722 max_slot_table_size);
2723 if (xprt == NULL) {
2724 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2725 "rpc_xprt\n");
2726 return ERR_PTR(-ENOMEM);
2729 new = container_of(xprt, struct sock_xprt, xprt);
2730 mutex_init(&new->recv_mutex);
2731 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2732 xprt->addrlen = args->addrlen;
2733 if (args->srcaddr)
2734 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2735 else {
2736 int err;
2737 err = xs_init_anyaddr(args->dstaddr->sa_family,
2738 (struct sockaddr *)&new->srcaddr);
2739 if (err != 0) {
2740 xprt_free(xprt);
2741 return ERR_PTR(err);
2745 return xprt;
2748 static const struct rpc_timeout xs_local_default_timeout = {
2749 .to_initval = 10 * HZ,
2750 .to_maxval = 10 * HZ,
2751 .to_retries = 2,
2755 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2756 * @args: rpc transport creation arguments
2758 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2760 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2762 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2763 struct sock_xprt *transport;
2764 struct rpc_xprt *xprt;
2765 struct rpc_xprt *ret;
2767 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2768 xprt_max_tcp_slot_table_entries);
2769 if (IS_ERR(xprt))
2770 return xprt;
2771 transport = container_of(xprt, struct sock_xprt, xprt);
2773 xprt->prot = 0;
2774 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2776 xprt->bind_timeout = XS_BIND_TO;
2777 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2778 xprt->idle_timeout = XS_IDLE_DISC_TO;
2780 xprt->ops = &xs_local_ops;
2781 xprt->timeout = &xs_local_default_timeout;
2783 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2784 INIT_WORK(&transport->error_worker, xs_error_handle);
2785 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2787 switch (sun->sun_family) {
2788 case AF_LOCAL:
2789 if (sun->sun_path[0] != '/') {
2790 dprintk("RPC: bad AF_LOCAL address: %s\n",
2791 sun->sun_path);
2792 ret = ERR_PTR(-EINVAL);
2793 goto out_err;
2795 xprt_set_bound(xprt);
2796 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2797 ret = ERR_PTR(xs_local_setup_socket(transport));
2798 if (ret)
2799 goto out_err;
2800 break;
2801 default:
2802 ret = ERR_PTR(-EAFNOSUPPORT);
2803 goto out_err;
2806 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2807 xprt->address_strings[RPC_DISPLAY_ADDR]);
2809 if (try_module_get(THIS_MODULE))
2810 return xprt;
2811 ret = ERR_PTR(-EINVAL);
2812 out_err:
2813 xs_xprt_free(xprt);
2814 return ret;
2817 static const struct rpc_timeout xs_udp_default_timeout = {
2818 .to_initval = 5 * HZ,
2819 .to_maxval = 30 * HZ,
2820 .to_increment = 5 * HZ,
2821 .to_retries = 5,
2825 * xs_setup_udp - Set up transport to use a UDP socket
2826 * @args: rpc transport creation arguments
2829 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2831 struct sockaddr *addr = args->dstaddr;
2832 struct rpc_xprt *xprt;
2833 struct sock_xprt *transport;
2834 struct rpc_xprt *ret;
2836 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2837 xprt_udp_slot_table_entries);
2838 if (IS_ERR(xprt))
2839 return xprt;
2840 transport = container_of(xprt, struct sock_xprt, xprt);
2842 xprt->prot = IPPROTO_UDP;
2843 /* XXX: header size can vary due to auth type, IPv6, etc. */
2844 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2846 xprt->bind_timeout = XS_BIND_TO;
2847 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2848 xprt->idle_timeout = XS_IDLE_DISC_TO;
2850 xprt->ops = &xs_udp_ops;
2852 xprt->timeout = &xs_udp_default_timeout;
2854 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2855 INIT_WORK(&transport->error_worker, xs_error_handle);
2856 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2858 switch (addr->sa_family) {
2859 case AF_INET:
2860 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2861 xprt_set_bound(xprt);
2863 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2864 break;
2865 case AF_INET6:
2866 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2867 xprt_set_bound(xprt);
2869 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2870 break;
2871 default:
2872 ret = ERR_PTR(-EAFNOSUPPORT);
2873 goto out_err;
2876 if (xprt_bound(xprt))
2877 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2878 xprt->address_strings[RPC_DISPLAY_ADDR],
2879 xprt->address_strings[RPC_DISPLAY_PORT],
2880 xprt->address_strings[RPC_DISPLAY_PROTO]);
2881 else
2882 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2883 xprt->address_strings[RPC_DISPLAY_ADDR],
2884 xprt->address_strings[RPC_DISPLAY_PROTO]);
2886 if (try_module_get(THIS_MODULE))
2887 return xprt;
2888 ret = ERR_PTR(-EINVAL);
2889 out_err:
2890 xs_xprt_free(xprt);
2891 return ret;
2894 static const struct rpc_timeout xs_tcp_default_timeout = {
2895 .to_initval = 60 * HZ,
2896 .to_maxval = 60 * HZ,
2897 .to_retries = 2,
2901 * xs_setup_tcp - Set up transport to use a TCP socket
2902 * @args: rpc transport creation arguments
2905 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2907 struct sockaddr *addr = args->dstaddr;
2908 struct rpc_xprt *xprt;
2909 struct sock_xprt *transport;
2910 struct rpc_xprt *ret;
2911 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2913 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2914 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2916 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2917 max_slot_table_size);
2918 if (IS_ERR(xprt))
2919 return xprt;
2920 transport = container_of(xprt, struct sock_xprt, xprt);
2922 xprt->prot = IPPROTO_TCP;
2923 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2925 xprt->bind_timeout = XS_BIND_TO;
2926 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2927 xprt->idle_timeout = XS_IDLE_DISC_TO;
2929 xprt->ops = &xs_tcp_ops;
2930 xprt->timeout = &xs_tcp_default_timeout;
2932 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2933 xprt->connect_timeout = xprt->timeout->to_initval *
2934 (xprt->timeout->to_retries + 1);
2936 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2937 INIT_WORK(&transport->error_worker, xs_error_handle);
2938 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2940 switch (addr->sa_family) {
2941 case AF_INET:
2942 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2943 xprt_set_bound(xprt);
2945 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2946 break;
2947 case AF_INET6:
2948 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2949 xprt_set_bound(xprt);
2951 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2952 break;
2953 default:
2954 ret = ERR_PTR(-EAFNOSUPPORT);
2955 goto out_err;
2958 if (xprt_bound(xprt))
2959 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2960 xprt->address_strings[RPC_DISPLAY_ADDR],
2961 xprt->address_strings[RPC_DISPLAY_PORT],
2962 xprt->address_strings[RPC_DISPLAY_PROTO]);
2963 else
2964 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2965 xprt->address_strings[RPC_DISPLAY_ADDR],
2966 xprt->address_strings[RPC_DISPLAY_PROTO]);
2968 if (try_module_get(THIS_MODULE))
2969 return xprt;
2970 ret = ERR_PTR(-EINVAL);
2971 out_err:
2972 xs_xprt_free(xprt);
2973 return ret;
2977 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2978 * @args: rpc transport creation arguments
2981 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2983 struct sockaddr *addr = args->dstaddr;
2984 struct rpc_xprt *xprt;
2985 struct sock_xprt *transport;
2986 struct svc_sock *bc_sock;
2987 struct rpc_xprt *ret;
2989 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2990 xprt_tcp_slot_table_entries);
2991 if (IS_ERR(xprt))
2992 return xprt;
2993 transport = container_of(xprt, struct sock_xprt, xprt);
2995 xprt->prot = IPPROTO_TCP;
2996 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2997 xprt->timeout = &xs_tcp_default_timeout;
2999 /* backchannel */
3000 xprt_set_bound(xprt);
3001 xprt->bind_timeout = 0;
3002 xprt->reestablish_timeout = 0;
3003 xprt->idle_timeout = 0;
3005 xprt->ops = &bc_tcp_ops;
3007 switch (addr->sa_family) {
3008 case AF_INET:
3009 xs_format_peer_addresses(xprt, "tcp",
3010 RPCBIND_NETID_TCP);
3011 break;
3012 case AF_INET6:
3013 xs_format_peer_addresses(xprt, "tcp",
3014 RPCBIND_NETID_TCP6);
3015 break;
3016 default:
3017 ret = ERR_PTR(-EAFNOSUPPORT);
3018 goto out_err;
3021 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3022 xprt->address_strings[RPC_DISPLAY_ADDR],
3023 xprt->address_strings[RPC_DISPLAY_PORT],
3024 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027 * Once we've associated a backchannel xprt with a connection,
3028 * we want to keep it around as long as the connection lasts,
3029 * in case we need to start using it for a backchannel again;
3030 * this reference won't be dropped until bc_xprt is destroyed.
3032 xprt_get(xprt);
3033 args->bc_xprt->xpt_bc_xprt = xprt;
3034 xprt->bc_xprt = args->bc_xprt;
3035 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3036 transport->sock = bc_sock->sk_sock;
3037 transport->inet = bc_sock->sk_sk;
3040 * Since we don't want connections for the backchannel, we set
3041 * the xprt status to connected
3043 xprt_set_connected(xprt);
3045 if (try_module_get(THIS_MODULE))
3046 return xprt;
3048 args->bc_xprt->xpt_bc_xprt = NULL;
3049 args->bc_xprt->xpt_bc_xps = NULL;
3050 xprt_put(xprt);
3051 ret = ERR_PTR(-EINVAL);
3052 out_err:
3053 xs_xprt_free(xprt);
3054 return ret;
3057 static struct xprt_class xs_local_transport = {
3058 .list = LIST_HEAD_INIT(xs_local_transport.list),
3059 .name = "named UNIX socket",
3060 .owner = THIS_MODULE,
3061 .ident = XPRT_TRANSPORT_LOCAL,
3062 .setup = xs_setup_local,
3063 .netid = { "" },
3066 static struct xprt_class xs_udp_transport = {
3067 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3068 .name = "udp",
3069 .owner = THIS_MODULE,
3070 .ident = XPRT_TRANSPORT_UDP,
3071 .setup = xs_setup_udp,
3072 .netid = { "udp", "udp6", "" },
3075 static struct xprt_class xs_tcp_transport = {
3076 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3077 .name = "tcp",
3078 .owner = THIS_MODULE,
3079 .ident = XPRT_TRANSPORT_TCP,
3080 .setup = xs_setup_tcp,
3081 .netid = { "tcp", "tcp6", "" },
3084 static struct xprt_class xs_bc_tcp_transport = {
3085 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3086 .name = "tcp NFSv4.1 backchannel",
3087 .owner = THIS_MODULE,
3088 .ident = XPRT_TRANSPORT_BC_TCP,
3089 .setup = xs_setup_bc_tcp,
3090 .netid = { "" },
3094 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3097 int init_socket_xprt(void)
3099 if (!sunrpc_table_header)
3100 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3102 xprt_register_transport(&xs_local_transport);
3103 xprt_register_transport(&xs_udp_transport);
3104 xprt_register_transport(&xs_tcp_transport);
3105 xprt_register_transport(&xs_bc_tcp_transport);
3107 return 0;
3111 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3114 void cleanup_socket_xprt(void)
3116 if (sunrpc_table_header) {
3117 unregister_sysctl_table(sunrpc_table_header);
3118 sunrpc_table_header = NULL;
3121 xprt_unregister_transport(&xs_local_transport);
3122 xprt_unregister_transport(&xs_udp_transport);
3123 xprt_unregister_transport(&xs_tcp_transport);
3124 xprt_unregister_transport(&xs_bc_tcp_transport);
3127 static int param_set_uint_minmax(const char *val,
3128 const struct kernel_param *kp,
3129 unsigned int min, unsigned int max)
3131 unsigned int num;
3132 int ret;
3134 if (!val)
3135 return -EINVAL;
3136 ret = kstrtouint(val, 0, &num);
3137 if (ret)
3138 return ret;
3139 if (num < min || num > max)
3140 return -EINVAL;
3141 *((unsigned int *)kp->arg) = num;
3142 return 0;
3145 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3147 return param_set_uint_minmax(val, kp,
3148 RPC_MIN_RESVPORT,
3149 RPC_MAX_RESVPORT);
3152 static const struct kernel_param_ops param_ops_portnr = {
3153 .set = param_set_portnr,
3154 .get = param_get_uint,
3157 #define param_check_portnr(name, p) \
3158 __param_check(name, p, unsigned int);
3160 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3161 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3163 static int param_set_slot_table_size(const char *val,
3164 const struct kernel_param *kp)
3166 return param_set_uint_minmax(val, kp,
3167 RPC_MIN_SLOT_TABLE,
3168 RPC_MAX_SLOT_TABLE);
3171 static const struct kernel_param_ops param_ops_slot_table_size = {
3172 .set = param_set_slot_table_size,
3173 .get = param_get_uint,
3176 #define param_check_slot_table_size(name, p) \
3177 __param_check(name, p, unsigned int);
3179 static int param_set_max_slot_table_size(const char *val,
3180 const struct kernel_param *kp)
3182 return param_set_uint_minmax(val, kp,
3183 RPC_MIN_SLOT_TABLE,
3184 RPC_MAX_SLOT_TABLE_LIMIT);
3187 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3188 .set = param_set_max_slot_table_size,
3189 .get = param_get_uint,
3192 #define param_check_max_slot_table_size(name, p) \
3193 __param_check(name, p, unsigned int);
3195 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3196 slot_table_size, 0644);
3197 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3198 max_slot_table_size, 0644);
3199 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3200 slot_table_size, 0644);