1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
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>
11 *
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>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
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
45
46 #include <net/sock.h>
47 #include <net/checksum.h>
48 #include <net/udp.h>
49 #include <net/tcp.h>
50 #include <net/tls_prot.h>
51 #include <net/handshake.h>
52
53 #include <linux/bvec.h>
54 #include <linux/highmem.h>
55 #include <linux/uio.h>
56 #include <linux/sched/mm.h>
57
58 #include <trace/events/sock.h>
59 #include <trace/events/sunrpc.h>
60
61 #include "socklib.h"
62 #include "sunrpc.h"
63
64 static void xs_close(struct rpc_xprt *xprt);
65 static void xs_reset_srcport(struct sock_xprt *transport);
66 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
67 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
68 struct socket *sock);
69
70 /*
71 * xprtsock tunables
72 */
73 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
74 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
75 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
76
77 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
78 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
79
80 #define XS_TCP_LINGER_TO (15U * HZ)
81 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
82
83 /*
84 * We can register our own files under /proc/sys/sunrpc by
85 * calling register_sysctl() again. The files in that
86 * directory become the union of all files registered there.
87 *
88 * We simply need to make sure that we don't collide with
89 * someone else's file names!
90 */
91
92 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
93 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
94 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
95 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
96 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
97
98 static struct ctl_table_header *sunrpc_table_header;
99
100 static struct xprt_class xs_local_transport;
101 static struct xprt_class xs_udp_transport;
102 static struct xprt_class xs_tcp_transport;
103 static struct xprt_class xs_tcp_tls_transport;
104 static struct xprt_class xs_bc_tcp_transport;
105
106 /*
107 * FIXME: changing the UDP slot table size should also resize the UDP
108 * socket buffers for existing UDP transports
109 */
110 static struct ctl_table xs_tunables_table[] = {
111 {
112 .procname = "udp_slot_table_entries",
113 .data = &xprt_udp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_slot_table_size
119 },
120 {
121 .procname = "tcp_slot_table_entries",
122 .data = &xprt_tcp_slot_table_entries,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &min_slot_table_size,
127 .extra2 = &max_slot_table_size
128 },
129 {
130 .procname = "tcp_max_slot_table_entries",
131 .data = &xprt_max_tcp_slot_table_entries,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &min_slot_table_size,
136 .extra2 = &max_tcp_slot_table_limit
137 },
138 {
139 .procname = "min_resvport",
140 .data = &xprt_min_resvport,
141 .maxlen = sizeof(unsigned int),
142 .mode = 0644,
143 .proc_handler = proc_dointvec_minmax,
144 .extra1 = &xprt_min_resvport_limit,
145 .extra2 = &xprt_max_resvport_limit
146 },
147 {
148 .procname = "max_resvport",
149 .data = &xprt_max_resvport,
150 .maxlen = sizeof(unsigned int),
151 .mode = 0644,
152 .proc_handler = proc_dointvec_minmax,
153 .extra1 = &xprt_min_resvport_limit,
154 .extra2 = &xprt_max_resvport_limit
155 },
156 {
157 .procname = "tcp_fin_timeout",
158 .data = &xs_tcp_fin_timeout,
159 .maxlen = sizeof(xs_tcp_fin_timeout),
160 .mode = 0644,
161 .proc_handler = proc_dointvec_jiffies,
162 },
163 };
164
165 /*
166 * Wait duration for a reply from the RPC portmapper.
167 */
168 #define XS_BIND_TO (60U * HZ)
169
170 /*
171 * Delay if a UDP socket connect error occurs. This is most likely some
172 * kind of resource problem on the local host.
173 */
174 #define XS_UDP_REEST_TO (2U * HZ)
175
176 /*
177 * The reestablish timeout allows clients to delay for a bit before attempting
178 * to reconnect to a server that just dropped our connection.
179 *
180 * We implement an exponential backoff when trying to reestablish a TCP
181 * transport connection with the server. Some servers like to drop a TCP
182 * connection when they are overworked, so we start with a short timeout and
183 * increase over time if the server is down or not responding.
184 */
185 #define XS_TCP_INIT_REEST_TO (3U * HZ)
186
187 /*
188 * TCP idle timeout; client drops the transport socket if it is idle
189 * for this long. Note that we also timeout UDP sockets to prevent
190 * holding port numbers when there is no RPC traffic.
191 */
192 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
193
194 /*
195 * TLS handshake timeout.
196 */
197 #define XS_TLS_HANDSHAKE_TO (10U * HZ)
198
199 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
200 # undef RPC_DEBUG_DATA
201 # define RPCDBG_FACILITY RPCDBG_TRANS
202 #endif
203
204 #ifdef RPC_DEBUG_DATA
205 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
206 {
207 u8 *buf = (u8 *) packet;
208 int j;
209
210 dprintk("RPC: %s\n", msg);
211 for (j = 0; j < count && j < 128; j += 4) {
212 if (!(j & 31)) {
213 if (j)
214 dprintk("\n");
215 dprintk("0x%04x ", j);
216 }
217 dprintk("%02x%02x%02x%02x ",
218 buf[j], buf[j+1], buf[j+2], buf[j+3]);
219 }
220 dprintk("\n");
221 }
222 #else
223 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
224 {
225 /* NOP */
226 }
227 #endif
228
229 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
230 {
231 return (struct rpc_xprt *) sk->sk_user_data;
232 }
233
234 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
235 {
236 return (struct sockaddr *) &xprt->addr;
237 }
238
239 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
240 {
241 return (struct sockaddr_un *) &xprt->addr;
242 }
243
244 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
245 {
246 return (struct sockaddr_in *) &xprt->addr;
247 }
248
249 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
250 {
251 return (struct sockaddr_in6 *) &xprt->addr;
252 }
253
254 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
255 {
256 struct sockaddr *sap = xs_addr(xprt);
257 struct sockaddr_in6 *sin6;
258 struct sockaddr_in *sin;
259 struct sockaddr_un *sun;
260 char buf[128];
261
262 switch (sap->sa_family) {
263 case AF_LOCAL:
264 sun = xs_addr_un(xprt);
265 if (sun->sun_path[0]) {
266 strscpy(buf, sun->sun_path, sizeof(buf));
267 } else {
268 buf[0] = '@';
269 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
270 }
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 break;
274 case AF_INET:
275 (void)rpc_ntop(sap, buf, sizeof(buf));
276 xprt->address_strings[RPC_DISPLAY_ADDR] =
277 kstrdup(buf, GFP_KERNEL);
278 sin = xs_addr_in(xprt);
279 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
280 break;
281 case AF_INET6:
282 (void)rpc_ntop(sap, buf, sizeof(buf));
283 xprt->address_strings[RPC_DISPLAY_ADDR] =
284 kstrdup(buf, GFP_KERNEL);
285 sin6 = xs_addr_in6(xprt);
286 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
287 break;
288 default:
289 BUG();
290 }
291
292 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
293 }
294
295 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
296 {
297 struct sockaddr *sap = xs_addr(xprt);
298 char buf[128];
299
300 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
301 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
302
303 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
304 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
305 }
306
307 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
308 const char *protocol,
309 const char *netid)
310 {
311 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
312 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
313 xs_format_common_peer_addresses(xprt);
314 xs_format_common_peer_ports(xprt);
315 }
316
317 static void xs_update_peer_port(struct rpc_xprt *xprt)
318 {
319 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
320 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
321
322 xs_format_common_peer_ports(xprt);
323 }
324
325 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
326 {
327 unsigned int i;
328
329 for (i = 0; i < RPC_DISPLAY_MAX; i++)
330 switch (i) {
331 case RPC_DISPLAY_PROTO:
332 case RPC_DISPLAY_NETID:
333 continue;
334 default:
335 kfree(xprt->address_strings[i]);
336 }
337 }
338
339 static size_t
340 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
341 {
342 size_t i,n;
343
344 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
345 return want;
346 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
347 for (i = 0; i < n; i++) {
348 if (buf->pages[i])
349 continue;
350 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
351 if (!buf->pages[i]) {
352 i *= PAGE_SIZE;
353 return i > buf->page_base ? i - buf->page_base : 0;
354 }
355 }
356 return want;
357 }
358
359 static int
360 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
361 struct cmsghdr *cmsg, int ret)
362 {
363 u8 content_type = tls_get_record_type(sock->sk, cmsg);
364 u8 level, description;
365
366 switch (content_type) {
367 case 0:
368 break;
369 case TLS_RECORD_TYPE_DATA:
370 /* TLS sets EOR at the end of each application data
371 * record, even though there might be more frames
372 * waiting to be decrypted.
373 */
374 msg->msg_flags &= ~MSG_EOR;
375 break;
376 case TLS_RECORD_TYPE_ALERT:
377 tls_alert_recv(sock->sk, msg, &level, &description);
378 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
379 -EACCES : -EAGAIN;
380 break;
381 default:
382 /* discard this record type */
383 ret = -EAGAIN;
384 }
385 return ret;
386 }
387
388 static int
389 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
390 {
391 union {
392 struct cmsghdr cmsg;
393 u8 buf[CMSG_SPACE(sizeof(u8))];
394 } u;
395 int ret;
396
397 msg->msg_control = &u;
398 msg->msg_controllen = sizeof(u);
399 ret = sock_recvmsg(sock, msg, flags);
400 if (msg->msg_controllen != sizeof(u))
401 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
402 return ret;
403 }
404
405 static ssize_t
406 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
407 {
408 ssize_t ret;
409 if (seek != 0)
410 iov_iter_advance(&msg->msg_iter, seek);
411 ret = xs_sock_recv_cmsg(sock, msg, flags);
412 return ret > 0 ? ret + seek : ret;
413 }
414
415 static ssize_t
416 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
417 struct kvec *kvec, size_t count, size_t seek)
418 {
419 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
420 return xs_sock_recvmsg(sock, msg, flags, seek);
421 }
422
423 static ssize_t
424 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
425 struct bio_vec *bvec, unsigned long nr, size_t count,
426 size_t seek)
427 {
428 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
429 return xs_sock_recvmsg(sock, msg, flags, seek);
430 }
431
432 static ssize_t
433 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
434 size_t count)
435 {
436 iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
437 return xs_sock_recv_cmsg(sock, msg, flags);
438 }
439
440 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
441 static void
442 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
443 {
444 struct bvec_iter bi = {
445 .bi_size = count,
446 };
447 struct bio_vec bv;
448
449 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
450 for_each_bvec(bv, bvec, bi, bi)
451 flush_dcache_page(bv.bv_page);
452 }
453 #else
454 static inline void
455 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
456 {
457 }
458 #endif
459
460 static ssize_t
461 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
462 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
463 {
464 size_t want, seek_init = seek, offset = 0;
465 ssize_t ret;
466
467 want = min_t(size_t, count, buf->head[0].iov_len);
468 if (seek < want) {
469 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
470 if (ret <= 0)
471 goto sock_err;
472 offset += ret;
473 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
474 goto out;
475 if (ret != want)
476 goto out;
477 seek = 0;
478 } else {
479 seek -= want;
480 offset += want;
481 }
482
483 want = xs_alloc_sparse_pages(
484 buf, min_t(size_t, count - offset, buf->page_len),
485 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
486 if (seek < want) {
487 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
488 xdr_buf_pagecount(buf),
489 want + buf->page_base,
490 seek + buf->page_base);
491 if (ret <= 0)
492 goto sock_err;
493 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
494 ret -= buf->page_base;
495 offset += ret;
496 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
497 goto out;
498 if (ret != want)
499 goto out;
500 seek = 0;
501 } else {
502 seek -= want;
503 offset += want;
504 }
505
506 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
507 if (seek < want) {
508 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
509 if (ret <= 0)
510 goto sock_err;
511 offset += ret;
512 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
513 goto out;
514 if (ret != want)
515 goto out;
516 } else if (offset < seek_init)
517 offset = seek_init;
518 ret = -EMSGSIZE;
519 out:
520 *read = offset - seek_init;
521 return ret;
522 sock_err:
523 offset += seek;
524 goto out;
525 }
526
527 static void
528 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
529 {
530 if (!transport->recv.copied) {
531 if (buf->head[0].iov_len >= transport->recv.offset)
532 memcpy(buf->head[0].iov_base,
533 &transport->recv.xid,
534 transport->recv.offset);
535 transport->recv.copied = transport->recv.offset;
536 }
537 }
538
539 static bool
540 xs_read_stream_request_done(struct sock_xprt *transport)
541 {
542 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
543 }
544
545 static void
546 xs_read_stream_check_eor(struct sock_xprt *transport,
547 struct msghdr *msg)
548 {
549 if (xs_read_stream_request_done(transport))
550 msg->msg_flags |= MSG_EOR;
551 }
552
553 static ssize_t
554 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
555 int flags, struct rpc_rqst *req)
556 {
557 struct xdr_buf *buf = &req->rq_private_buf;
558 size_t want, read;
559 ssize_t ret;
560
561 xs_read_header(transport, buf);
562
563 want = transport->recv.len - transport->recv.offset;
564 if (want != 0) {
565 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
566 transport->recv.copied + want,
567 transport->recv.copied,
568 &read);
569 transport->recv.offset += read;
570 transport->recv.copied += read;
571 }
572
573 if (transport->recv.offset == transport->recv.len)
574 xs_read_stream_check_eor(transport, msg);
575
576 if (want == 0)
577 return 0;
578
579 switch (ret) {
580 default:
581 break;
582 case -EFAULT:
583 case -EMSGSIZE:
584 msg->msg_flags |= MSG_TRUNC;
585 return read;
586 case 0:
587 return -ESHUTDOWN;
588 }
589 return ret < 0 ? ret : read;
590 }
591
592 static size_t
593 xs_read_stream_headersize(bool isfrag)
594 {
595 if (isfrag)
596 return sizeof(__be32);
597 return 3 * sizeof(__be32);
598 }
599
600 static ssize_t
601 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
602 int flags, size_t want, size_t seek)
603 {
604 struct kvec kvec = {
605 .iov_base = &transport->recv.fraghdr,
606 .iov_len = want,
607 };
608 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
609 }
610
611 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
612 static ssize_t
613 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
614 {
615 struct rpc_xprt *xprt = &transport->xprt;
616 struct rpc_rqst *req;
617 ssize_t ret;
618
619 /* Is this transport associated with the backchannel? */
620 if (!xprt->bc_serv)
621 return -ESHUTDOWN;
622
623 /* Look up and lock the request corresponding to the given XID */
624 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
625 if (!req) {
626 printk(KERN_WARNING "Callback slot table overflowed\n");
627 return -ESHUTDOWN;
628 }
629 if (transport->recv.copied && !req->rq_private_buf.len)
630 return -ESHUTDOWN;
631
632 ret = xs_read_stream_request(transport, msg, flags, req);
633 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
634 xprt_complete_bc_request(req, transport->recv.copied);
635 else
636 req->rq_private_buf.len = transport->recv.copied;
637
638 return ret;
639 }
640 #else /* CONFIG_SUNRPC_BACKCHANNEL */
641 static ssize_t
642 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
643 {
644 return -ESHUTDOWN;
645 }
646 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
647
648 static ssize_t
649 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
650 {
651 struct rpc_xprt *xprt = &transport->xprt;
652 struct rpc_rqst *req;
653 ssize_t ret = 0;
654
655 /* Look up and lock the request corresponding to the given XID */
656 spin_lock(&xprt->queue_lock);
657 req = xprt_lookup_rqst(xprt, transport->recv.xid);
658 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
659 msg->msg_flags |= MSG_TRUNC;
660 goto out;
661 }
662 xprt_pin_rqst(req);
663 spin_unlock(&xprt->queue_lock);
664
665 ret = xs_read_stream_request(transport, msg, flags, req);
666
667 spin_lock(&xprt->queue_lock);
668 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
669 xprt_complete_rqst(req->rq_task, transport->recv.copied);
670 else
671 req->rq_private_buf.len = transport->recv.copied;
672 xprt_unpin_rqst(req);
673 out:
674 spin_unlock(&xprt->queue_lock);
675 return ret;
676 }
677
678 static ssize_t
679 xs_read_stream(struct sock_xprt *transport, int flags)
680 {
681 struct msghdr msg = { 0 };
682 size_t want, read = 0;
683 ssize_t ret = 0;
684
685 if (transport->recv.len == 0) {
686 want = xs_read_stream_headersize(transport->recv.copied != 0);
687 ret = xs_read_stream_header(transport, &msg, flags, want,
688 transport->recv.offset);
689 if (ret <= 0)
690 goto out_err;
691 transport->recv.offset = ret;
692 if (transport->recv.offset != want)
693 return transport->recv.offset;
694 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
695 RPC_FRAGMENT_SIZE_MASK;
696 transport->recv.offset -= sizeof(transport->recv.fraghdr);
697 read = ret;
698 }
699
700 switch (be32_to_cpu(transport->recv.calldir)) {
701 default:
702 msg.msg_flags |= MSG_TRUNC;
703 break;
704 case RPC_CALL:
705 ret = xs_read_stream_call(transport, &msg, flags);
706 break;
707 case RPC_REPLY:
708 ret = xs_read_stream_reply(transport, &msg, flags);
709 }
710 if (msg.msg_flags & MSG_TRUNC) {
711 transport->recv.calldir = cpu_to_be32(-1);
712 transport->recv.copied = -1;
713 }
714 if (ret < 0)
715 goto out_err;
716 read += ret;
717 if (transport->recv.offset < transport->recv.len) {
718 if (!(msg.msg_flags & MSG_TRUNC))
719 return read;
720 msg.msg_flags = 0;
721 ret = xs_read_discard(transport->sock, &msg, flags,
722 transport->recv.len - transport->recv.offset);
723 if (ret <= 0)
724 goto out_err;
725 transport->recv.offset += ret;
726 read += ret;
727 if (transport->recv.offset != transport->recv.len)
728 return read;
729 }
730 if (xs_read_stream_request_done(transport)) {
731 trace_xs_stream_read_request(transport);
732 transport->recv.copied = 0;
733 }
734 transport->recv.offset = 0;
735 transport->recv.len = 0;
736 return read;
737 out_err:
738 return ret != 0 ? ret : -ESHUTDOWN;
739 }
740
741 static __poll_t xs_poll_socket(struct sock_xprt *transport)
742 {
743 return transport->sock->ops->poll(transport->file, transport->sock,
744 NULL);
745 }
746
747 static bool xs_poll_socket_readable(struct sock_xprt *transport)
748 {
749 __poll_t events = xs_poll_socket(transport);
750
751 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
752 }
753
754 static void xs_poll_check_readable(struct sock_xprt *transport)
755 {
756
757 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
758 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
759 return;
760 if (!xs_poll_socket_readable(transport))
761 return;
762 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
763 queue_work(xprtiod_workqueue, &transport->recv_worker);
764 }
765
766 static void xs_stream_data_receive(struct sock_xprt *transport)
767 {
768 size_t read = 0;
769 ssize_t ret = 0;
770
771 mutex_lock(&transport->recv_mutex);
772 if (transport->sock == NULL)
773 goto out;
774 for (;;) {
775 ret = xs_read_stream(transport, MSG_DONTWAIT);
776 if (ret < 0)
777 break;
778 read += ret;
779 cond_resched();
780 }
781 if (ret == -ESHUTDOWN)
782 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
783 else if (ret == -EACCES)
784 xprt_wake_pending_tasks(&transport->xprt, -EACCES);
785 else
786 xs_poll_check_readable(transport);
787 out:
788 mutex_unlock(&transport->recv_mutex);
789 trace_xs_stream_read_data(&transport->xprt, ret, read);
790 }
791
792 static void xs_stream_data_receive_workfn(struct work_struct *work)
793 {
794 struct sock_xprt *transport =
795 container_of(work, struct sock_xprt, recv_worker);
796 unsigned int pflags = memalloc_nofs_save();
797
798 xs_stream_data_receive(transport);
799 memalloc_nofs_restore(pflags);
800 }
801
802 static void
803 xs_stream_reset_connect(struct sock_xprt *transport)
804 {
805 transport->recv.offset = 0;
806 transport->recv.len = 0;
807 transport->recv.copied = 0;
808 transport->xmit.offset = 0;
809 }
810
811 static void
812 xs_stream_start_connect(struct sock_xprt *transport)
813 {
814 transport->xprt.stat.connect_count++;
815 transport->xprt.stat.connect_start = jiffies;
816 }
817
818 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
819
820 /**
821 * xs_nospace - handle transmit was incomplete
822 * @req: pointer to RPC request
823 * @transport: pointer to struct sock_xprt
824 *
825 */
826 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
827 {
828 struct rpc_xprt *xprt = &transport->xprt;
829 struct sock *sk = transport->inet;
830 int ret = -EAGAIN;
831
832 trace_rpc_socket_nospace(req, transport);
833
834 /* Protect against races with write_space */
835 spin_lock(&xprt->transport_lock);
836
837 /* Don't race with disconnect */
838 if (xprt_connected(xprt)) {
839 /* wait for more buffer space */
840 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
841 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
842 sk->sk_write_pending++;
843 xprt_wait_for_buffer_space(xprt);
844 } else
845 ret = -ENOTCONN;
846
847 spin_unlock(&xprt->transport_lock);
848 return ret;
849 }
850
851 static int xs_sock_nospace(struct rpc_rqst *req)
852 {
853 struct sock_xprt *transport =
854 container_of(req->rq_xprt, struct sock_xprt, xprt);
855 struct sock *sk = transport->inet;
856 int ret = -EAGAIN;
857
858 lock_sock(sk);
859 if (!sock_writeable(sk))
860 ret = xs_nospace(req, transport);
861 release_sock(sk);
862 return ret;
863 }
864
865 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
866 {
867 struct sock_xprt *transport =
868 container_of(req->rq_xprt, struct sock_xprt, xprt);
869 struct sock *sk = transport->inet;
870 int ret = -EAGAIN;
871
872 if (vm_wait)
873 return -ENOBUFS;
874 lock_sock(sk);
875 if (!sk_stream_memory_free(sk))
876 ret = xs_nospace(req, transport);
877 release_sock(sk);
878 return ret;
879 }
880
881 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
882 {
883 return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
884 }
885
886 static void xs_stream_abort_send_request(struct rpc_rqst *req)
887 {
888 struct rpc_xprt *xprt = req->rq_xprt;
889 struct sock_xprt *transport =
890 container_of(xprt, struct sock_xprt, xprt);
891
892 if (transport->xmit.offset != 0 &&
893 !test_bit(XPRT_CLOSE_WAIT, &xprt->state))
894 xprt_force_disconnect(xprt);
895 }
896
897 /*
898 * Determine if the previous message in the stream was aborted before it
899 * could complete transmission.
900 */
901 static bool
902 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
903 {
904 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
905 }
906
907 /*
908 * Return the stream record marker field for a record of length < 2^31-1
909 */
910 static rpc_fraghdr
911 xs_stream_record_marker(struct xdr_buf *xdr)
912 {
913 if (!xdr->len)
914 return 0;
915 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
916 }
917
918 /**
919 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
920 * @req: pointer to RPC request
921 *
922 * Return values:
923 * 0: The request has been sent
924 * EAGAIN: The socket was blocked, please call again later to
925 * complete the request
926 * ENOTCONN: Caller needs to invoke connect logic then call again
927 * other: Some other error occurred, the request was not sent
928 */
929 static int xs_local_send_request(struct rpc_rqst *req)
930 {
931 struct rpc_xprt *xprt = req->rq_xprt;
932 struct sock_xprt *transport =
933 container_of(xprt, struct sock_xprt, xprt);
934 struct xdr_buf *xdr = &req->rq_snd_buf;
935 rpc_fraghdr rm = xs_stream_record_marker(xdr);
936 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
937 struct msghdr msg = {
938 .msg_flags = XS_SENDMSG_FLAGS,
939 };
940 bool vm_wait;
941 unsigned int sent;
942 int status;
943
944 /* Close the stream if the previous transmission was incomplete */
945 if (xs_send_request_was_aborted(transport, req)) {
946 xprt_force_disconnect(xprt);
947 return -ENOTCONN;
948 }
949
950 xs_pktdump("packet data:",
951 req->rq_svec->iov_base, req->rq_svec->iov_len);
952
953 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
954
955 req->rq_xtime = ktime_get();
956 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
957 transport->xmit.offset, rm, &sent);
958 dprintk("RPC: %s(%u) = %d\n",
959 __func__, xdr->len - transport->xmit.offset, status);
960
961 if (likely(sent > 0) || status == 0) {
962 transport->xmit.offset += sent;
963 req->rq_bytes_sent = transport->xmit.offset;
964 if (likely(req->rq_bytes_sent >= msglen)) {
965 req->rq_xmit_bytes_sent += transport->xmit.offset;
966 transport->xmit.offset = 0;
967 return 0;
968 }
969 status = -EAGAIN;
970 vm_wait = false;
971 }
972
973 switch (status) {
974 case -EAGAIN:
975 status = xs_stream_nospace(req, vm_wait);
976 break;
977 default:
978 dprintk("RPC: sendmsg returned unrecognized error %d\n",
979 -status);
980 fallthrough;
981 case -EPIPE:
982 xprt_force_disconnect(xprt);
983 status = -ENOTCONN;
984 }
985
986 return status;
987 }
988
989 /**
990 * xs_udp_send_request - write an RPC request to a UDP socket
991 * @req: pointer to RPC request
992 *
993 * Return values:
994 * 0: The request has been sent
995 * EAGAIN: The socket was blocked, please call again later to
996 * complete the request
997 * ENOTCONN: Caller needs to invoke connect logic then call again
998 * other: Some other error occurred, the request was not sent
999 */
1000 static int xs_udp_send_request(struct rpc_rqst *req)
1001 {
1002 struct rpc_xprt *xprt = req->rq_xprt;
1003 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1004 struct xdr_buf *xdr = &req->rq_snd_buf;
1005 struct msghdr msg = {
1006 .msg_name = xs_addr(xprt),
1007 .msg_namelen = xprt->addrlen,
1008 .msg_flags = XS_SENDMSG_FLAGS,
1009 };
1010 unsigned int sent;
1011 int status;
1012
1013 xs_pktdump("packet data:",
1014 req->rq_svec->iov_base,
1015 req->rq_svec->iov_len);
1016
1017 if (!xprt_bound(xprt))
1018 return -ENOTCONN;
1019
1020 if (!xprt_request_get_cong(xprt, req))
1021 return -EBADSLT;
1022
1023 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1024 if (status < 0)
1025 return status;
1026 req->rq_xtime = ktime_get();
1027 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1028
1029 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1030 xdr->len, status);
1031
1032 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1033 if (status == -EPERM)
1034 goto process_status;
1035
1036 if (status == -EAGAIN && sock_writeable(transport->inet))
1037 status = -ENOBUFS;
1038
1039 if (sent > 0 || status == 0) {
1040 req->rq_xmit_bytes_sent += sent;
1041 if (sent >= req->rq_slen)
1042 return 0;
1043 /* Still some bytes left; set up for a retry later. */
1044 status = -EAGAIN;
1045 }
1046
1047 process_status:
1048 switch (status) {
1049 case -ENOTSOCK:
1050 status = -ENOTCONN;
1051 /* Should we call xs_close() here? */
1052 break;
1053 case -EAGAIN:
1054 status = xs_sock_nospace(req);
1055 break;
1056 case -ENETUNREACH:
1057 case -ENOBUFS:
1058 case -EPIPE:
1059 case -ECONNREFUSED:
1060 case -EPERM:
1061 /* When the server has died, an ICMP port unreachable message
1062 * prompts ECONNREFUSED. */
1063 break;
1064 default:
1065 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1066 -status);
1067 }
1068
1069 return status;
1070 }
1071
1072 /**
1073 * xs_tcp_send_request - write an RPC request to a TCP socket
1074 * @req: pointer to RPC request
1075 *
1076 * Return values:
1077 * 0: The request has been sent
1078 * EAGAIN: The socket was blocked, please call again later to
1079 * complete the request
1080 * ENOTCONN: Caller needs to invoke connect logic then call again
1081 * other: Some other error occurred, the request was not sent
1082 *
1083 * XXX: In the case of soft timeouts, should we eventually give up
1084 * if sendmsg is not able to make progress?
1085 */
1086 static int xs_tcp_send_request(struct rpc_rqst *req)
1087 {
1088 struct rpc_xprt *xprt = req->rq_xprt;
1089 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1090 struct xdr_buf *xdr = &req->rq_snd_buf;
1091 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1092 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1093 struct msghdr msg = {
1094 .msg_flags = XS_SENDMSG_FLAGS,
1095 };
1096 bool vm_wait;
1097 unsigned int sent;
1098 int status;
1099
1100 /* Close the stream if the previous transmission was incomplete */
1101 if (xs_send_request_was_aborted(transport, req)) {
1102 if (transport->sock != NULL)
1103 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1104 return -ENOTCONN;
1105 }
1106 if (!transport->inet)
1107 return -ENOTCONN;
1108
1109 xs_pktdump("packet data:",
1110 req->rq_svec->iov_base,
1111 req->rq_svec->iov_len);
1112
1113 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1114 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1115
1116 xs_set_srcport(transport, transport->sock);
1117
1118 /* Continue transmitting the packet/record. We must be careful
1119 * to cope with writespace callbacks arriving _after_ we have
1120 * called sendmsg(). */
1121 req->rq_xtime = ktime_get();
1122 tcp_sock_set_cork(transport->inet, true);
1123
1124 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1125
1126 do {
1127 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1128 transport->xmit.offset, rm, &sent);
1129
1130 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1131 xdr->len - transport->xmit.offset, status);
1132
1133 /* If we've sent the entire packet, immediately
1134 * reset the count of bytes sent. */
1135 transport->xmit.offset += sent;
1136 req->rq_bytes_sent = transport->xmit.offset;
1137 if (likely(req->rq_bytes_sent >= msglen)) {
1138 req->rq_xmit_bytes_sent += transport->xmit.offset;
1139 transport->xmit.offset = 0;
1140 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1141 tcp_sock_set_cork(transport->inet, false);
1142 return 0;
1143 }
1144
1145 WARN_ON_ONCE(sent == 0 && status == 0);
1146
1147 if (sent > 0)
1148 vm_wait = false;
1149
1150 } while (status == 0);
1151
1152 switch (status) {
1153 case -ENOTSOCK:
1154 status = -ENOTCONN;
1155 /* Should we call xs_close() here? */
1156 break;
1157 case -EAGAIN:
1158 status = xs_stream_nospace(req, vm_wait);
1159 break;
1160 case -ECONNRESET:
1161 case -ECONNREFUSED:
1162 case -ENOTCONN:
1163 case -EADDRINUSE:
1164 case -ENOBUFS:
1165 case -EPIPE:
1166 break;
1167 default:
1168 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1169 -status);
1170 }
1171
1172 return status;
1173 }
1174
1175 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1176 {
1177 transport->old_data_ready = sk->sk_data_ready;
1178 transport->old_state_change = sk->sk_state_change;
1179 transport->old_write_space = sk->sk_write_space;
1180 transport->old_error_report = sk->sk_error_report;
1181 }
1182
1183 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1184 {
1185 sk->sk_data_ready = transport->old_data_ready;
1186 sk->sk_state_change = transport->old_state_change;
1187 sk->sk_write_space = transport->old_write_space;
1188 sk->sk_error_report = transport->old_error_report;
1189 }
1190
1191 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1192 {
1193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1194
1195 transport->xprt_err = 0;
1196 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1197 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1198 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1199 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1200 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1201 }
1202
1203 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1204 {
1205 set_bit(nr, &transport->sock_state);
1206 queue_work(xprtiod_workqueue, &transport->error_worker);
1207 }
1208
1209 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1210 {
1211 xprt->connect_cookie++;
1212 smp_mb__before_atomic();
1213 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1214 clear_bit(XPRT_CLOSING, &xprt->state);
1215 xs_sock_reset_state_flags(xprt);
1216 smp_mb__after_atomic();
1217 }
1218
1219 /**
1220 * xs_error_report - callback to handle TCP socket state errors
1221 * @sk: socket
1222 *
1223 * Note: we don't call sock_error() since there may be a rpc_task
1224 * using the socket, and so we don't want to clear sk->sk_err.
1225 */
1226 static void xs_error_report(struct sock *sk)
1227 {
1228 struct sock_xprt *transport;
1229 struct rpc_xprt *xprt;
1230
1231 if (!(xprt = xprt_from_sock(sk)))
1232 return;
1233
1234 transport = container_of(xprt, struct sock_xprt, xprt);
1235 transport->xprt_err = -sk->sk_err;
1236 if (transport->xprt_err == 0)
1237 return;
1238 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1239 xprt, -transport->xprt_err);
1240 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1241
1242 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1243 smp_mb__before_atomic();
1244 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1245 }
1246
1247 static void xs_reset_transport(struct sock_xprt *transport)
1248 {
1249 struct socket *sock = transport->sock;
1250 struct sock *sk = transport->inet;
1251 struct rpc_xprt *xprt = &transport->xprt;
1252 struct file *filp = transport->file;
1253
1254 if (sk == NULL)
1255 return;
1256 /*
1257 * Make sure we're calling this in a context from which it is safe
1258 * to call __fput_sync(). In practice that means rpciod and the
1259 * system workqueue.
1260 */
1261 if (!(current->flags & PF_WQ_WORKER)) {
1262 WARN_ON_ONCE(1);
1263 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1264 return;
1265 }
1266
1267 if (atomic_read(&transport->xprt.swapper))
1268 sk_clear_memalloc(sk);
1269
1270 tls_handshake_cancel(sk);
1271
1272 kernel_sock_shutdown(sock, SHUT_RDWR);
1273
1274 mutex_lock(&transport->recv_mutex);
1275 lock_sock(sk);
1276 transport->inet = NULL;
1277 transport->sock = NULL;
1278 transport->file = NULL;
1279
1280 sk->sk_user_data = NULL;
1281
1282 xs_restore_old_callbacks(transport, sk);
1283 xprt_clear_connected(xprt);
1284 xs_sock_reset_connection_flags(xprt);
1285 /* Reset stream record info */
1286 xs_stream_reset_connect(transport);
1287 release_sock(sk);
1288 mutex_unlock(&transport->recv_mutex);
1289
1290 trace_rpc_socket_close(xprt, sock);
1291 __fput_sync(filp);
1292
1293 xprt_disconnect_done(xprt);
1294 }
1295
1296 /**
1297 * xs_close - close a socket
1298 * @xprt: transport
1299 *
1300 * This is used when all requests are complete; ie, no DRC state remains
1301 * on the server we want to save.
1302 *
1303 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1304 * xs_reset_transport() zeroing the socket from underneath a writer.
1305 */
1306 static void xs_close(struct rpc_xprt *xprt)
1307 {
1308 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1309
1310 dprintk("RPC: xs_close xprt %p\n", xprt);
1311
1312 if (transport->sock)
1313 tls_handshake_close(transport->sock);
1314 xs_reset_transport(transport);
1315 xprt->reestablish_timeout = 0;
1316 }
1317
1318 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1319 {
1320 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1321 xprt);
1322 xprt_disconnect_done(xprt);
1323 }
1324
1325 static void xs_xprt_free(struct rpc_xprt *xprt)
1326 {
1327 xs_free_peer_addresses(xprt);
1328 xprt_free(xprt);
1329 }
1330
1331 /**
1332 * xs_destroy - prepare to shutdown a transport
1333 * @xprt: doomed transport
1334 *
1335 */
1336 static void xs_destroy(struct rpc_xprt *xprt)
1337 {
1338 struct sock_xprt *transport = container_of(xprt,
1339 struct sock_xprt, xprt);
1340 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1341
1342 cancel_delayed_work_sync(&transport->connect_worker);
1343 xs_close(xprt);
1344 cancel_work_sync(&transport->recv_worker);
1345 cancel_work_sync(&transport->error_worker);
1346 xs_xprt_free(xprt);
1347 module_put(THIS_MODULE);
1348 }
1349
1350 /**
1351 * xs_udp_data_read_skb - receive callback for UDP sockets
1352 * @xprt: transport
1353 * @sk: socket
1354 * @skb: skbuff
1355 *
1356 */
1357 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1358 struct sock *sk,
1359 struct sk_buff *skb)
1360 {
1361 struct rpc_task *task;
1362 struct rpc_rqst *rovr;
1363 int repsize, copied;
1364 u32 _xid;
1365 __be32 *xp;
1366
1367 repsize = skb->len;
1368 if (repsize < 4) {
1369 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1370 return;
1371 }
1372
1373 /* Copy the XID from the skb... */
1374 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1375 if (xp == NULL)
1376 return;
1377
1378 /* Look up and lock the request corresponding to the given XID */
1379 spin_lock(&xprt->queue_lock);
1380 rovr = xprt_lookup_rqst(xprt, *xp);
1381 if (!rovr)
1382 goto out_unlock;
1383 xprt_pin_rqst(rovr);
1384 xprt_update_rtt(rovr->rq_task);
1385 spin_unlock(&xprt->queue_lock);
1386 task = rovr->rq_task;
1387
1388 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1389 copied = repsize;
1390
1391 /* Suck it into the iovec, verify checksum if not done by hw. */
1392 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1393 spin_lock(&xprt->queue_lock);
1394 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1395 goto out_unpin;
1396 }
1397
1398
1399 spin_lock(&xprt->transport_lock);
1400 xprt_adjust_cwnd(xprt, task, copied);
1401 spin_unlock(&xprt->transport_lock);
1402 spin_lock(&xprt->queue_lock);
1403 xprt_complete_rqst(task, copied);
1404 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1405 out_unpin:
1406 xprt_unpin_rqst(rovr);
1407 out_unlock:
1408 spin_unlock(&xprt->queue_lock);
1409 }
1410
1411 static void xs_udp_data_receive(struct sock_xprt *transport)
1412 {
1413 struct sk_buff *skb;
1414 struct sock *sk;
1415 int err;
1416
1417 mutex_lock(&transport->recv_mutex);
1418 sk = transport->inet;
1419 if (sk == NULL)
1420 goto out;
1421 for (;;) {
1422 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1423 if (skb == NULL)
1424 break;
1425 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1426 consume_skb(skb);
1427 cond_resched();
1428 }
1429 xs_poll_check_readable(transport);
1430 out:
1431 mutex_unlock(&transport->recv_mutex);
1432 }
1433
1434 static void xs_udp_data_receive_workfn(struct work_struct *work)
1435 {
1436 struct sock_xprt *transport =
1437 container_of(work, struct sock_xprt, recv_worker);
1438 unsigned int pflags = memalloc_nofs_save();
1439
1440 xs_udp_data_receive(transport);
1441 memalloc_nofs_restore(pflags);
1442 }
1443
1444 /**
1445 * xs_data_ready - "data ready" callback for sockets
1446 * @sk: socket with data to read
1447 *
1448 */
1449 static void xs_data_ready(struct sock *sk)
1450 {
1451 struct rpc_xprt *xprt;
1452
1453 trace_sk_data_ready(sk);
1454
1455 xprt = xprt_from_sock(sk);
1456 if (xprt != NULL) {
1457 struct sock_xprt *transport = container_of(xprt,
1458 struct sock_xprt, xprt);
1459
1460 trace_xs_data_ready(xprt);
1461
1462 transport->old_data_ready(sk);
1463
1464 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1465 return;
1466
1467 /* Any data means we had a useful conversation, so
1468 * then we don't need to delay the next reconnect
1469 */
1470 if (xprt->reestablish_timeout)
1471 xprt->reestablish_timeout = 0;
1472 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1473 queue_work(xprtiod_workqueue, &transport->recv_worker);
1474 }
1475 }
1476
1477 /*
1478 * Helper function to force a TCP close if the server is sending
1479 * junk and/or it has put us in CLOSE_WAIT
1480 */
1481 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1482 {
1483 xprt_force_disconnect(xprt);
1484 }
1485
1486 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1487 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1488 {
1489 return PAGE_SIZE;
1490 }
1491 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1492
1493 /**
1494 * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1495 * @sk: socket whose state has changed
1496 *
1497 */
1498 static void xs_local_state_change(struct sock *sk)
1499 {
1500 struct rpc_xprt *xprt;
1501 struct sock_xprt *transport;
1502
1503 if (!(xprt = xprt_from_sock(sk)))
1504 return;
1505 transport = container_of(xprt, struct sock_xprt, xprt);
1506 if (sk->sk_shutdown & SHUTDOWN_MASK) {
1507 clear_bit(XPRT_CONNECTED, &xprt->state);
1508 /* Trigger the socket release */
1509 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1510 }
1511 }
1512
1513 /**
1514 * xs_tcp_state_change - callback to handle TCP socket state changes
1515 * @sk: socket whose state has changed
1516 *
1517 */
1518 static void xs_tcp_state_change(struct sock *sk)
1519 {
1520 struct rpc_xprt *xprt;
1521 struct sock_xprt *transport;
1522
1523 if (!(xprt = xprt_from_sock(sk)))
1524 return;
1525 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1526 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1527 sk->sk_state, xprt_connected(xprt),
1528 sock_flag(sk, SOCK_DEAD),
1529 sock_flag(sk, SOCK_ZAPPED),
1530 sk->sk_shutdown);
1531
1532 transport = container_of(xprt, struct sock_xprt, xprt);
1533 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1534 switch (sk->sk_state) {
1535 case TCP_ESTABLISHED:
1536 if (!xprt_test_and_set_connected(xprt)) {
1537 xprt->connect_cookie++;
1538 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1539 xprt_clear_connecting(xprt);
1540
1541 xprt->stat.connect_count++;
1542 xprt->stat.connect_time += (long)jiffies -
1543 xprt->stat.connect_start;
1544 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1545 }
1546 break;
1547 case TCP_FIN_WAIT1:
1548 /* The client initiated a shutdown of the socket */
1549 xprt->connect_cookie++;
1550 xprt->reestablish_timeout = 0;
1551 set_bit(XPRT_CLOSING, &xprt->state);
1552 smp_mb__before_atomic();
1553 clear_bit(XPRT_CONNECTED, &xprt->state);
1554 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1555 smp_mb__after_atomic();
1556 break;
1557 case TCP_CLOSE_WAIT:
1558 /* The server initiated a shutdown of the socket */
1559 xprt->connect_cookie++;
1560 clear_bit(XPRT_CONNECTED, &xprt->state);
1561 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1562 fallthrough;
1563 case TCP_CLOSING:
1564 /*
1565 * If the server closed down the connection, make sure that
1566 * we back off before reconnecting
1567 */
1568 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1569 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1570 break;
1571 case TCP_LAST_ACK:
1572 set_bit(XPRT_CLOSING, &xprt->state);
1573 smp_mb__before_atomic();
1574 clear_bit(XPRT_CONNECTED, &xprt->state);
1575 smp_mb__after_atomic();
1576 break;
1577 case TCP_CLOSE:
1578 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1579 &transport->sock_state)) {
1580 xs_reset_srcport(transport);
1581 xprt_clear_connecting(xprt);
1582 }
1583 clear_bit(XPRT_CLOSING, &xprt->state);
1584 /* Trigger the socket release */
1585 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1586 }
1587 }
1588
1589 static void xs_write_space(struct sock *sk)
1590 {
1591 struct sock_xprt *transport;
1592 struct rpc_xprt *xprt;
1593
1594 if (!sk->sk_socket)
1595 return;
1596 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1597
1598 if (unlikely(!(xprt = xprt_from_sock(sk))))
1599 return;
1600 transport = container_of(xprt, struct sock_xprt, xprt);
1601 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1602 return;
1603 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1604 sk->sk_write_pending--;
1605 }
1606
1607 /**
1608 * xs_udp_write_space - callback invoked when socket buffer space
1609 * becomes available
1610 * @sk: socket whose state has changed
1611 *
1612 * Called when more output buffer space is available for this socket.
1613 * We try not to wake our writers until they can make "significant"
1614 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1615 * with a bunch of small requests.
1616 */
1617 static void xs_udp_write_space(struct sock *sk)
1618 {
1619 /* from net/core/sock.c:sock_def_write_space */
1620 if (sock_writeable(sk))
1621 xs_write_space(sk);
1622 }
1623
1624 /**
1625 * xs_tcp_write_space - callback invoked when socket buffer space
1626 * becomes available
1627 * @sk: socket whose state has changed
1628 *
1629 * Called when more output buffer space is available for this socket.
1630 * We try not to wake our writers until they can make "significant"
1631 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1632 * with a bunch of small requests.
1633 */
1634 static void xs_tcp_write_space(struct sock *sk)
1635 {
1636 /* from net/core/stream.c:sk_stream_write_space */
1637 if (sk_stream_is_writeable(sk))
1638 xs_write_space(sk);
1639 }
1640
1641 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1642 {
1643 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1644 struct sock *sk = transport->inet;
1645
1646 if (transport->rcvsize) {
1647 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1648 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1649 }
1650 if (transport->sndsize) {
1651 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1652 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1653 sk->sk_write_space(sk);
1654 }
1655 }
1656
1657 /**
1658 * xs_udp_set_buffer_size - set send and receive limits
1659 * @xprt: generic transport
1660 * @sndsize: requested size of send buffer, in bytes
1661 * @rcvsize: requested size of receive buffer, in bytes
1662 *
1663 * Set socket send and receive buffer size limits.
1664 */
1665 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1666 {
1667 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1668
1669 transport->sndsize = 0;
1670 if (sndsize)
1671 transport->sndsize = sndsize + 1024;
1672 transport->rcvsize = 0;
1673 if (rcvsize)
1674 transport->rcvsize = rcvsize + 1024;
1675
1676 xs_udp_do_set_buffer_size(xprt);
1677 }
1678
1679 /**
1680 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1681 * @xprt: controlling transport
1682 * @task: task that timed out
1683 *
1684 * Adjust the congestion window after a retransmit timeout has occurred.
1685 */
1686 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1687 {
1688 spin_lock(&xprt->transport_lock);
1689 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1690 spin_unlock(&xprt->transport_lock);
1691 }
1692
1693 static int xs_get_random_port(void)
1694 {
1695 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1696 unsigned short range;
1697 unsigned short rand;
1698
1699 if (max < min)
1700 return -EADDRINUSE;
1701 range = max - min + 1;
1702 rand = get_random_u32_below(range);
1703 return rand + min;
1704 }
1705
1706 static unsigned short xs_sock_getport(struct socket *sock)
1707 {
1708 struct sockaddr_storage buf;
1709 unsigned short port = 0;
1710
1711 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1712 goto out;
1713 switch (buf.ss_family) {
1714 case AF_INET6:
1715 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1716 break;
1717 case AF_INET:
1718 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1719 }
1720 out:
1721 return port;
1722 }
1723
1724 /**
1725 * xs_set_port - reset the port number in the remote endpoint address
1726 * @xprt: generic transport
1727 * @port: new port number
1728 *
1729 */
1730 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1731 {
1732 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1733
1734 rpc_set_port(xs_addr(xprt), port);
1735 xs_update_peer_port(xprt);
1736 }
1737
1738 static void xs_reset_srcport(struct sock_xprt *transport)
1739 {
1740 transport->srcport = 0;
1741 }
1742
1743 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1744 {
1745 if (transport->srcport == 0 && transport->xprt.reuseport)
1746 transport->srcport = xs_sock_getport(sock);
1747 }
1748
1749 static int xs_get_srcport(struct sock_xprt *transport)
1750 {
1751 int port = transport->srcport;
1752
1753 if (port == 0 && transport->xprt.resvport)
1754 port = xs_get_random_port();
1755 return port;
1756 }
1757
1758 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1759 {
1760 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1761 unsigned short ret = 0;
1762 mutex_lock(&sock->recv_mutex);
1763 if (sock->sock)
1764 ret = xs_sock_getport(sock->sock);
1765 mutex_unlock(&sock->recv_mutex);
1766 return ret;
1767 }
1768
1769 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1770 {
1771 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1772 union {
1773 struct sockaddr sa;
1774 struct sockaddr_storage st;
1775 } saddr;
1776 int ret = -ENOTCONN;
1777
1778 mutex_lock(&sock->recv_mutex);
1779 if (sock->sock) {
1780 ret = kernel_getsockname(sock->sock, &saddr.sa);
1781 if (ret >= 0)
1782 ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1783 }
1784 mutex_unlock(&sock->recv_mutex);
1785 return ret;
1786 }
1787
1788 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1789 {
1790 if (transport->srcport != 0)
1791 transport->srcport = 0;
1792 if (!transport->xprt.resvport)
1793 return 0;
1794 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1795 return xprt_max_resvport;
1796 return --port;
1797 }
1798 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1799 {
1800 struct sockaddr_storage myaddr;
1801 int err, nloop = 0;
1802 int port = xs_get_srcport(transport);
1803 unsigned short last;
1804
1805 /*
1806 * If we are asking for any ephemeral port (i.e. port == 0 &&
1807 * transport->xprt.resvport == 0), don't bind. Let the local
1808 * port selection happen implicitly when the socket is used
1809 * (for example at connect time).
1810 *
1811 * This ensures that we can continue to establish TCP
1812 * connections even when all local ephemeral ports are already
1813 * a part of some TCP connection. This makes no difference
1814 * for UDP sockets, but also doesn't harm them.
1815 *
1816 * If we're asking for any reserved port (i.e. port == 0 &&
1817 * transport->xprt.resvport == 1) xs_get_srcport above will
1818 * ensure that port is non-zero and we will bind as needed.
1819 */
1820 if (port <= 0)
1821 return port;
1822
1823 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1824 do {
1825 rpc_set_port((struct sockaddr *)&myaddr, port);
1826 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1827 transport->xprt.addrlen);
1828 if (err == 0) {
1829 if (transport->xprt.reuseport)
1830 transport->srcport = port;
1831 break;
1832 }
1833 last = port;
1834 port = xs_next_srcport(transport, port);
1835 if (port > last)
1836 nloop++;
1837 } while (err == -EADDRINUSE && nloop != 2);
1838
1839 if (myaddr.ss_family == AF_INET)
1840 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1841 &((struct sockaddr_in *)&myaddr)->sin_addr,
1842 port, err ? "failed" : "ok", err);
1843 else
1844 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1845 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1846 port, err ? "failed" : "ok", err);
1847 return err;
1848 }
1849
1850 /*
1851 * We don't support autobind on AF_LOCAL sockets
1852 */
1853 static void xs_local_rpcbind(struct rpc_task *task)
1854 {
1855 xprt_set_bound(task->tk_xprt);
1856 }
1857
1858 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1859 {
1860 }
1861
1862 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1863 static struct lock_class_key xs_key[3];
1864 static struct lock_class_key xs_slock_key[3];
1865
1866 static inline void xs_reclassify_socketu(struct socket *sock)
1867 {
1868 struct sock *sk = sock->sk;
1869
1870 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1871 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1872 }
1873
1874 static inline void xs_reclassify_socket4(struct socket *sock)
1875 {
1876 struct sock *sk = sock->sk;
1877
1878 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1879 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1880 }
1881
1882 static inline void xs_reclassify_socket6(struct socket *sock)
1883 {
1884 struct sock *sk = sock->sk;
1885
1886 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1887 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1888 }
1889
1890 static inline void xs_reclassify_socket(int family, struct socket *sock)
1891 {
1892 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1893 return;
1894
1895 switch (family) {
1896 case AF_LOCAL:
1897 xs_reclassify_socketu(sock);
1898 break;
1899 case AF_INET:
1900 xs_reclassify_socket4(sock);
1901 break;
1902 case AF_INET6:
1903 xs_reclassify_socket6(sock);
1904 break;
1905 }
1906 }
1907 #else
1908 static inline void xs_reclassify_socket(int family, struct socket *sock)
1909 {
1910 }
1911 #endif
1912
1913 static void xs_dummy_setup_socket(struct work_struct *work)
1914 {
1915 }
1916
1917 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1918 struct sock_xprt *transport, int family, int type,
1919 int protocol, bool reuseport)
1920 {
1921 struct file *filp;
1922 struct socket *sock;
1923 int err;
1924
1925 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1926 if (err < 0) {
1927 dprintk("RPC: can't create %d transport socket (%d).\n",
1928 protocol, -err);
1929 goto out;
1930 }
1931 xs_reclassify_socket(family, sock);
1932
1933 if (reuseport)
1934 sock_set_reuseport(sock->sk);
1935
1936 err = xs_bind(transport, sock);
1937 if (err) {
1938 sock_release(sock);
1939 goto out;
1940 }
1941
1942 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1943 if (IS_ERR(filp))
1944 return ERR_CAST(filp);
1945 transport->file = filp;
1946
1947 return sock;
1948 out:
1949 return ERR_PTR(err);
1950 }
1951
1952 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1953 struct socket *sock)
1954 {
1955 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1956 xprt);
1957
1958 if (!transport->inet) {
1959 struct sock *sk = sock->sk;
1960
1961 lock_sock(sk);
1962
1963 xs_save_old_callbacks(transport, sk);
1964
1965 sk->sk_user_data = xprt;
1966 sk->sk_data_ready = xs_data_ready;
1967 sk->sk_write_space = xs_udp_write_space;
1968 sk->sk_state_change = xs_local_state_change;
1969 sk->sk_error_report = xs_error_report;
1970 sk->sk_use_task_frag = false;
1971
1972 xprt_clear_connected(xprt);
1973
1974 /* Reset to new socket */
1975 transport->sock = sock;
1976 transport->inet = sk;
1977
1978 release_sock(sk);
1979 }
1980
1981 xs_stream_start_connect(transport);
1982
1983 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1984 }
1985
1986 /**
1987 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1988 * @transport: socket transport to connect
1989 */
1990 static int xs_local_setup_socket(struct sock_xprt *transport)
1991 {
1992 struct rpc_xprt *xprt = &transport->xprt;
1993 struct file *filp;
1994 struct socket *sock;
1995 int status;
1996
1997 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1998 SOCK_STREAM, 0, &sock, 1);
1999 if (status < 0) {
2000 dprintk("RPC: can't create AF_LOCAL "
2001 "transport socket (%d).\n", -status);
2002 goto out;
2003 }
2004 xs_reclassify_socket(AF_LOCAL, sock);
2005
2006 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
2007 if (IS_ERR(filp)) {
2008 status = PTR_ERR(filp);
2009 goto out;
2010 }
2011 transport->file = filp;
2012
2013 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2014 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2015
2016 status = xs_local_finish_connecting(xprt, sock);
2017 trace_rpc_socket_connect(xprt, sock, status);
2018 switch (status) {
2019 case 0:
2020 dprintk("RPC: xprt %p connected to %s\n",
2021 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2022 xprt->stat.connect_count++;
2023 xprt->stat.connect_time += (long)jiffies -
2024 xprt->stat.connect_start;
2025 xprt_set_connected(xprt);
2026 break;
2027 case -ENOBUFS:
2028 break;
2029 case -ENOENT:
2030 dprintk("RPC: xprt %p: socket %s does not exist\n",
2031 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2032 break;
2033 case -ECONNREFUSED:
2034 dprintk("RPC: xprt %p: connection refused for %s\n",
2035 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036 break;
2037 default:
2038 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2039 __func__, -status,
2040 xprt->address_strings[RPC_DISPLAY_ADDR]);
2041 }
2042
2043 out:
2044 xprt_clear_connecting(xprt);
2045 xprt_wake_pending_tasks(xprt, status);
2046 return status;
2047 }
2048
2049 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2050 {
2051 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2052 int ret;
2053
2054 if (transport->file)
2055 goto force_disconnect;
2056
2057 if (RPC_IS_ASYNC(task)) {
2058 /*
2059 * We want the AF_LOCAL connect to be resolved in the
2060 * filesystem namespace of the process making the rpc
2061 * call. Thus we connect synchronously.
2062 *
2063 * If we want to support asynchronous AF_LOCAL calls,
2064 * we'll need to figure out how to pass a namespace to
2065 * connect.
2066 */
2067 rpc_task_set_rpc_status(task, -ENOTCONN);
2068 goto out_wake;
2069 }
2070 ret = xs_local_setup_socket(transport);
2071 if (ret && !RPC_IS_SOFTCONN(task))
2072 msleep_interruptible(15000);
2073 return;
2074 force_disconnect:
2075 xprt_force_disconnect(xprt);
2076 out_wake:
2077 xprt_clear_connecting(xprt);
2078 xprt_wake_pending_tasks(xprt, -ENOTCONN);
2079 }
2080
2081 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2082 /*
2083 * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2084 * held, or when we otherwise know that we have exclusive access to the
2085 * socket, to guard against races with xs_reset_transport.
2086 */
2087 static void xs_set_memalloc(struct rpc_xprt *xprt)
2088 {
2089 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2090 xprt);
2091
2092 /*
2093 * If there's no sock, then we have nothing to set. The
2094 * reconnecting process will get it for us.
2095 */
2096 if (!transport->inet)
2097 return;
2098 if (atomic_read(&xprt->swapper))
2099 sk_set_memalloc(transport->inet);
2100 }
2101
2102 /**
2103 * xs_enable_swap - Tag this transport as being used for swap.
2104 * @xprt: transport to tag
2105 *
2106 * Take a reference to this transport on behalf of the rpc_clnt, and
2107 * optionally mark it for swapping if it wasn't already.
2108 */
2109 static int
2110 xs_enable_swap(struct rpc_xprt *xprt)
2111 {
2112 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2113
2114 mutex_lock(&xs->recv_mutex);
2115 if (atomic_inc_return(&xprt->swapper) == 1 &&
2116 xs->inet)
2117 sk_set_memalloc(xs->inet);
2118 mutex_unlock(&xs->recv_mutex);
2119 return 0;
2120 }
2121
2122 /**
2123 * xs_disable_swap - Untag this transport as being used for swap.
2124 * @xprt: transport to tag
2125 *
2126 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2127 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2128 */
2129 static void
2130 xs_disable_swap(struct rpc_xprt *xprt)
2131 {
2132 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2133
2134 mutex_lock(&xs->recv_mutex);
2135 if (atomic_dec_and_test(&xprt->swapper) &&
2136 xs->inet)
2137 sk_clear_memalloc(xs->inet);
2138 mutex_unlock(&xs->recv_mutex);
2139 }
2140 #else
2141 static void xs_set_memalloc(struct rpc_xprt *xprt)
2142 {
2143 }
2144
2145 static int
2146 xs_enable_swap(struct rpc_xprt *xprt)
2147 {
2148 return -EINVAL;
2149 }
2150
2151 static void
2152 xs_disable_swap(struct rpc_xprt *xprt)
2153 {
2154 }
2155 #endif
2156
2157 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2158 {
2159 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2160
2161 if (!transport->inet) {
2162 struct sock *sk = sock->sk;
2163
2164 lock_sock(sk);
2165
2166 xs_save_old_callbacks(transport, sk);
2167
2168 sk->sk_user_data = xprt;
2169 sk->sk_data_ready = xs_data_ready;
2170 sk->sk_write_space = xs_udp_write_space;
2171 sk->sk_use_task_frag = false;
2172
2173 xprt_set_connected(xprt);
2174
2175 /* Reset to new socket */
2176 transport->sock = sock;
2177 transport->inet = sk;
2178
2179 xs_set_memalloc(xprt);
2180
2181 release_sock(sk);
2182 }
2183 xs_udp_do_set_buffer_size(xprt);
2184
2185 xprt->stat.connect_start = jiffies;
2186 }
2187
2188 static void xs_udp_setup_socket(struct work_struct *work)
2189 {
2190 struct sock_xprt *transport =
2191 container_of(work, struct sock_xprt, connect_worker.work);
2192 struct rpc_xprt *xprt = &transport->xprt;
2193 struct socket *sock;
2194 int status = -EIO;
2195 unsigned int pflags = current->flags;
2196
2197 if (atomic_read(&xprt->swapper))
2198 current->flags |= PF_MEMALLOC;
2199 sock = xs_create_sock(xprt, transport,
2200 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2201 IPPROTO_UDP, false);
2202 if (IS_ERR(sock))
2203 goto out;
2204
2205 dprintk("RPC: worker connecting xprt %p via %s to "
2206 "%s (port %s)\n", xprt,
2207 xprt->address_strings[RPC_DISPLAY_PROTO],
2208 xprt->address_strings[RPC_DISPLAY_ADDR],
2209 xprt->address_strings[RPC_DISPLAY_PORT]);
2210
2211 xs_udp_finish_connecting(xprt, sock);
2212 trace_rpc_socket_connect(xprt, sock, 0);
2213 status = 0;
2214 out:
2215 xprt_clear_connecting(xprt);
2216 xprt_unlock_connect(xprt, transport);
2217 xprt_wake_pending_tasks(xprt, status);
2218 current_restore_flags(pflags, PF_MEMALLOC);
2219 }
2220
2221 /**
2222 * xs_tcp_shutdown - gracefully shut down a TCP socket
2223 * @xprt: transport
2224 *
2225 * Initiates a graceful shutdown of the TCP socket by calling the
2226 * equivalent of shutdown(SHUT_RDWR);
2227 */
2228 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2229 {
2230 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2231 struct socket *sock = transport->sock;
2232 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2233
2234 if (sock == NULL)
2235 return;
2236 if (!xprt->reuseport) {
2237 xs_close(xprt);
2238 return;
2239 }
2240 switch (skst) {
2241 case TCP_FIN_WAIT1:
2242 case TCP_FIN_WAIT2:
2243 case TCP_LAST_ACK:
2244 break;
2245 case TCP_ESTABLISHED:
2246 case TCP_CLOSE_WAIT:
2247 kernel_sock_shutdown(sock, SHUT_RDWR);
2248 trace_rpc_socket_shutdown(xprt, sock);
2249 break;
2250 default:
2251 xs_reset_transport(transport);
2252 }
2253 }
2254
2255 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2256 struct socket *sock)
2257 {
2258 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2259 struct net *net = sock_net(sock->sk);
2260 unsigned long connect_timeout;
2261 unsigned long syn_retries;
2262 unsigned int keepidle;
2263 unsigned int keepcnt;
2264 unsigned int timeo;
2265 unsigned long t;
2266
2267 spin_lock(&xprt->transport_lock);
2268 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2269 keepcnt = xprt->timeout->to_retries + 1;
2270 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2271 (xprt->timeout->to_retries + 1);
2272 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2273 spin_unlock(&xprt->transport_lock);
2274
2275 /* TCP Keepalive options */
2276 sock_set_keepalive(sock->sk);
2277 tcp_sock_set_keepidle(sock->sk, keepidle);
2278 tcp_sock_set_keepintvl(sock->sk, keepidle);
2279 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2280
2281 /* TCP user timeout (see RFC5482) */
2282 tcp_sock_set_user_timeout(sock->sk, timeo);
2283
2284 /* Connect timeout */
2285 connect_timeout = max_t(unsigned long,
2286 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2287 syn_retries = max_t(unsigned long,
2288 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2289 for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2290 ;
2291 if (t <= syn_retries)
2292 tcp_sock_set_syncnt(sock->sk, t - 1);
2293 }
2294
2295 static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2296 unsigned long connect_timeout)
2297 {
2298 struct sock_xprt *transport =
2299 container_of(xprt, struct sock_xprt, xprt);
2300 struct rpc_timeout to;
2301 unsigned long initval;
2302
2303 memcpy(&to, xprt->timeout, sizeof(to));
2304 /* Arbitrary lower limit */
2305 initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2306 to.to_initval = initval;
2307 to.to_maxval = initval;
2308 to.to_retries = 0;
2309 memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2310 xprt->timeout = &transport->tcp_timeout;
2311 xprt->connect_timeout = connect_timeout;
2312 }
2313
2314 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2315 unsigned long connect_timeout,
2316 unsigned long reconnect_timeout)
2317 {
2318 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2319
2320 spin_lock(&xprt->transport_lock);
2321 if (reconnect_timeout < xprt->max_reconnect_timeout)
2322 xprt->max_reconnect_timeout = reconnect_timeout;
2323 if (connect_timeout < xprt->connect_timeout)
2324 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2325 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2326 spin_unlock(&xprt->transport_lock);
2327 }
2328
2329 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2330 {
2331 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2332
2333 if (!transport->inet) {
2334 struct sock *sk = sock->sk;
2335
2336 /* Avoid temporary address, they are bad for long-lived
2337 * connections such as NFS mounts.
2338 * RFC4941, section 3.6 suggests that:
2339 * Individual applications, which have specific
2340 * knowledge about the normal duration of connections,
2341 * MAY override this as appropriate.
2342 */
2343 if (xs_addr(xprt)->sa_family == PF_INET6) {
2344 ip6_sock_set_addr_preferences(sk,
2345 IPV6_PREFER_SRC_PUBLIC);
2346 }
2347
2348 xs_tcp_set_socket_timeouts(xprt, sock);
2349 tcp_sock_set_nodelay(sk);
2350
2351 lock_sock(sk);
2352
2353 xs_save_old_callbacks(transport, sk);
2354
2355 sk->sk_user_data = xprt;
2356 sk->sk_data_ready = xs_data_ready;
2357 sk->sk_state_change = xs_tcp_state_change;
2358 sk->sk_write_space = xs_tcp_write_space;
2359 sk->sk_error_report = xs_error_report;
2360 sk->sk_use_task_frag = false;
2361
2362 /* socket options */
2363 sock_reset_flag(sk, SOCK_LINGER);
2364
2365 xprt_clear_connected(xprt);
2366
2367 /* Reset to new socket */
2368 transport->sock = sock;
2369 transport->inet = sk;
2370
2371 release_sock(sk);
2372 }
2373
2374 if (!xprt_bound(xprt))
2375 return -ENOTCONN;
2376
2377 xs_set_memalloc(xprt);
2378
2379 xs_stream_start_connect(transport);
2380
2381 /* Tell the socket layer to start connecting... */
2382 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2383 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2384 }
2385
2386 /**
2387 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2388 * @work: queued work item
2389 *
2390 * Invoked by a work queue tasklet.
2391 */
2392 static void xs_tcp_setup_socket(struct work_struct *work)
2393 {
2394 struct sock_xprt *transport =
2395 container_of(work, struct sock_xprt, connect_worker.work);
2396 struct socket *sock = transport->sock;
2397 struct rpc_xprt *xprt = &transport->xprt;
2398 int status;
2399 unsigned int pflags = current->flags;
2400
2401 if (atomic_read(&xprt->swapper))
2402 current->flags |= PF_MEMALLOC;
2403
2404 if (xprt_connected(xprt))
2405 goto out;
2406 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2407 &transport->sock_state) ||
2408 !sock) {
2409 xs_reset_transport(transport);
2410 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2411 SOCK_STREAM, IPPROTO_TCP, true);
2412 if (IS_ERR(sock)) {
2413 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2414 goto out;
2415 }
2416 }
2417
2418 dprintk("RPC: worker connecting xprt %p via %s to "
2419 "%s (port %s)\n", xprt,
2420 xprt->address_strings[RPC_DISPLAY_PROTO],
2421 xprt->address_strings[RPC_DISPLAY_ADDR],
2422 xprt->address_strings[RPC_DISPLAY_PORT]);
2423
2424 status = xs_tcp_finish_connecting(xprt, sock);
2425 trace_rpc_socket_connect(xprt, sock, status);
2426 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2427 xprt, -status, xprt_connected(xprt),
2428 sock->sk->sk_state);
2429 switch (status) {
2430 case 0:
2431 case -EINPROGRESS:
2432 /* SYN_SENT! */
2433 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2434 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2435 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2436 fallthrough;
2437 case -EALREADY:
2438 goto out_unlock;
2439 case -EADDRNOTAVAIL:
2440 /* Source port number is unavailable. Try a new one! */
2441 transport->srcport = 0;
2442 status = -EAGAIN;
2443 break;
2444 case -EPERM:
2445 /* Happens, for instance, if a BPF program is preventing
2446 * the connect. Remap the error so upper layers can better
2447 * deal with it.
2448 */
2449 status = -ECONNREFUSED;
2450 fallthrough;
2451 case -EINVAL:
2452 /* Happens, for instance, if the user specified a link
2453 * local IPv6 address without a scope-id.
2454 */
2455 case -ECONNREFUSED:
2456 case -ECONNRESET:
2457 case -ENETDOWN:
2458 case -ENETUNREACH:
2459 case -EHOSTUNREACH:
2460 case -EADDRINUSE:
2461 case -ENOBUFS:
2462 break;
2463 default:
2464 printk("%s: connect returned unhandled error %d\n",
2465 __func__, status);
2466 status = -EAGAIN;
2467 }
2468
2469 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2470 * We need to wake them first to ensure the correct error code.
2471 */
2472 xprt_wake_pending_tasks(xprt, status);
2473 xs_tcp_force_close(xprt);
2474 out:
2475 xprt_clear_connecting(xprt);
2476 out_unlock:
2477 xprt_unlock_connect(xprt, transport);
2478 current_restore_flags(pflags, PF_MEMALLOC);
2479 }
2480
2481 /*
2482 * Transfer the connected socket to @upper_transport, then mark that
2483 * xprt CONNECTED.
2484 */
2485 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2486 struct sock_xprt *upper_transport)
2487 {
2488 struct sock_xprt *lower_transport =
2489 container_of(lower_xprt, struct sock_xprt, xprt);
2490 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2491
2492 if (!upper_transport->inet) {
2493 struct socket *sock = lower_transport->sock;
2494 struct sock *sk = sock->sk;
2495
2496 /* Avoid temporary address, they are bad for long-lived
2497 * connections such as NFS mounts.
2498 * RFC4941, section 3.6 suggests that:
2499 * Individual applications, which have specific
2500 * knowledge about the normal duration of connections,
2501 * MAY override this as appropriate.
2502 */
2503 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2504 ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2505
2506 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2507 tcp_sock_set_nodelay(sk);
2508
2509 lock_sock(sk);
2510
2511 /* @sk is already connected, so it now has the RPC callbacks.
2512 * Reach into @lower_transport to save the original ones.
2513 */
2514 upper_transport->old_data_ready = lower_transport->old_data_ready;
2515 upper_transport->old_state_change = lower_transport->old_state_change;
2516 upper_transport->old_write_space = lower_transport->old_write_space;
2517 upper_transport->old_error_report = lower_transport->old_error_report;
2518 sk->sk_user_data = upper_xprt;
2519
2520 /* socket options */
2521 sock_reset_flag(sk, SOCK_LINGER);
2522
2523 xprt_clear_connected(upper_xprt);
2524
2525 upper_transport->sock = sock;
2526 upper_transport->inet = sk;
2527 upper_transport->file = lower_transport->file;
2528
2529 release_sock(sk);
2530
2531 /* Reset lower_transport before shutting down its clnt */
2532 mutex_lock(&lower_transport->recv_mutex);
2533 lower_transport->inet = NULL;
2534 lower_transport->sock = NULL;
2535 lower_transport->file = NULL;
2536
2537 xprt_clear_connected(lower_xprt);
2538 xs_sock_reset_connection_flags(lower_xprt);
2539 xs_stream_reset_connect(lower_transport);
2540 mutex_unlock(&lower_transport->recv_mutex);
2541 }
2542
2543 if (!xprt_bound(upper_xprt))
2544 return -ENOTCONN;
2545
2546 xs_set_memalloc(upper_xprt);
2547
2548 if (!xprt_test_and_set_connected(upper_xprt)) {
2549 upper_xprt->connect_cookie++;
2550 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2551 xprt_clear_connecting(upper_xprt);
2552
2553 upper_xprt->stat.connect_count++;
2554 upper_xprt->stat.connect_time += (long)jiffies -
2555 upper_xprt->stat.connect_start;
2556 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2557 }
2558 return 0;
2559 }
2560
2561 /**
2562 * xs_tls_handshake_done - TLS handshake completion handler
2563 * @data: address of xprt to wake
2564 * @status: status of handshake
2565 * @peerid: serial number of key containing the remote's identity
2566 *
2567 */
2568 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2569 {
2570 struct rpc_xprt *lower_xprt = data;
2571 struct sock_xprt *lower_transport =
2572 container_of(lower_xprt, struct sock_xprt, xprt);
2573
2574 lower_transport->xprt_err = status ? -EACCES : 0;
2575 complete(&lower_transport->handshake_done);
2576 xprt_put(lower_xprt);
2577 }
2578
2579 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2580 {
2581 struct sock_xprt *lower_transport =
2582 container_of(lower_xprt, struct sock_xprt, xprt);
2583 struct tls_handshake_args args = {
2584 .ta_sock = lower_transport->sock,
2585 .ta_done = xs_tls_handshake_done,
2586 .ta_data = xprt_get(lower_xprt),
2587 .ta_peername = lower_xprt->servername,
2588 };
2589 struct sock *sk = lower_transport->inet;
2590 int rc;
2591
2592 init_completion(&lower_transport->handshake_done);
2593 set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2594 lower_transport->xprt_err = -ETIMEDOUT;
2595 switch (xprtsec->policy) {
2596 case RPC_XPRTSEC_TLS_ANON:
2597 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2598 if (rc)
2599 goto out_put_xprt;
2600 break;
2601 case RPC_XPRTSEC_TLS_X509:
2602 args.ta_my_cert = xprtsec->cert_serial;
2603 args.ta_my_privkey = xprtsec->privkey_serial;
2604 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2605 if (rc)
2606 goto out_put_xprt;
2607 break;
2608 default:
2609 rc = -EACCES;
2610 goto out_put_xprt;
2611 }
2612
2613 rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2614 XS_TLS_HANDSHAKE_TO);
2615 if (rc <= 0) {
2616 if (!tls_handshake_cancel(sk)) {
2617 if (rc == 0)
2618 rc = -ETIMEDOUT;
2619 goto out_put_xprt;
2620 }
2621 }
2622
2623 rc = lower_transport->xprt_err;
2624
2625 out:
2626 xs_stream_reset_connect(lower_transport);
2627 clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2628 return rc;
2629
2630 out_put_xprt:
2631 xprt_put(lower_xprt);
2632 goto out;
2633 }
2634
2635 /**
2636 * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2637 * @work: queued work item
2638 *
2639 * Invoked by a work queue tasklet.
2640 *
2641 * For RPC-with-TLS, there is a two-stage connection process.
2642 *
2643 * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2644 * been marked connected, the consumer knows that a TCP connection and
2645 * a TLS session have been established.
2646 *
2647 * A "lower-layer xprt", created in this function, handles the mechanics
2648 * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2649 * then driving the TLS handshake. Once all that is complete, the upper
2650 * layer xprt is marked connected.
2651 */
2652 static void xs_tcp_tls_setup_socket(struct work_struct *work)
2653 {
2654 struct sock_xprt *upper_transport =
2655 container_of(work, struct sock_xprt, connect_worker.work);
2656 struct rpc_clnt *upper_clnt = upper_transport->clnt;
2657 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2658 struct rpc_create_args args = {
2659 .net = upper_xprt->xprt_net,
2660 .protocol = upper_xprt->prot,
2661 .address = (struct sockaddr *)&upper_xprt->addr,
2662 .addrsize = upper_xprt->addrlen,
2663 .timeout = upper_clnt->cl_timeout,
2664 .servername = upper_xprt->servername,
2665 .program = upper_clnt->cl_program,
2666 .prognumber = upper_clnt->cl_prog,
2667 .version = upper_clnt->cl_vers,
2668 .authflavor = RPC_AUTH_TLS,
2669 .cred = upper_clnt->cl_cred,
2670 .xprtsec = {
2671 .policy = RPC_XPRTSEC_NONE,
2672 },
2673 .stats = upper_clnt->cl_stats,
2674 };
2675 unsigned int pflags = current->flags;
2676 struct rpc_clnt *lower_clnt;
2677 struct rpc_xprt *lower_xprt;
2678 int status;
2679
2680 if (atomic_read(&upper_xprt->swapper))
2681 current->flags |= PF_MEMALLOC;
2682
2683 xs_stream_start_connect(upper_transport);
2684
2685 /* This implicitly sends an RPC_AUTH_TLS probe */
2686 lower_clnt = rpc_create(&args);
2687 if (IS_ERR(lower_clnt)) {
2688 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2689 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2690 xprt_clear_connecting(upper_xprt);
2691 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2692 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2693 goto out_unlock;
2694 }
2695
2696 /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2697 * the lower xprt.
2698 */
2699 rcu_read_lock();
2700 lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2701 rcu_read_unlock();
2702
2703 if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2704 goto out_unlock;
2705
2706 status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2707 if (status) {
2708 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2709 goto out_close;
2710 }
2711
2712 status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2713 if (status)
2714 goto out_close;
2715 xprt_release_write(lower_xprt, NULL);
2716
2717 trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2718 if (!xprt_test_and_set_connected(upper_xprt)) {
2719 upper_xprt->connect_cookie++;
2720 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2721 xprt_clear_connecting(upper_xprt);
2722
2723 upper_xprt->stat.connect_count++;
2724 upper_xprt->stat.connect_time += (long)jiffies -
2725 upper_xprt->stat.connect_start;
2726 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2727 }
2728 rpc_shutdown_client(lower_clnt);
2729
2730 out_unlock:
2731 current_restore_flags(pflags, PF_MEMALLOC);
2732 upper_transport->clnt = NULL;
2733 xprt_unlock_connect(upper_xprt, upper_transport);
2734 return;
2735
2736 out_close:
2737 xprt_release_write(lower_xprt, NULL);
2738 rpc_shutdown_client(lower_clnt);
2739
2740 /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2741 * Wake them first here to ensure they get our tk_status code.
2742 */
2743 xprt_wake_pending_tasks(upper_xprt, status);
2744 xs_tcp_force_close(upper_xprt);
2745 xprt_clear_connecting(upper_xprt);
2746 goto out_unlock;
2747 }
2748
2749 /**
2750 * xs_connect - connect a socket to a remote endpoint
2751 * @xprt: pointer to transport structure
2752 * @task: address of RPC task that manages state of connect request
2753 *
2754 * TCP: If the remote end dropped the connection, delay reconnecting.
2755 *
2756 * UDP socket connects are synchronous, but we use a work queue anyway
2757 * to guarantee that even unprivileged user processes can set up a
2758 * socket on a privileged port.
2759 *
2760 * If a UDP socket connect fails, the delay behavior here prevents
2761 * retry floods (hard mounts).
2762 */
2763 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2764 {
2765 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2766 unsigned long delay = 0;
2767
2768 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2769
2770 if (transport->sock != NULL) {
2771 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2772 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2773
2774 delay = xprt_reconnect_delay(xprt);
2775 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2776
2777 } else
2778 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2779
2780 transport->clnt = task->tk_client;
2781 queue_delayed_work(xprtiod_workqueue,
2782 &transport->connect_worker,
2783 delay);
2784 }
2785
2786 static void xs_wake_disconnect(struct sock_xprt *transport)
2787 {
2788 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2789 xs_tcp_force_close(&transport->xprt);
2790 }
2791
2792 static void xs_wake_write(struct sock_xprt *transport)
2793 {
2794 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2795 xprt_write_space(&transport->xprt);
2796 }
2797
2798 static void xs_wake_error(struct sock_xprt *transport)
2799 {
2800 int sockerr;
2801
2802 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2803 return;
2804 sockerr = xchg(&transport->xprt_err, 0);
2805 if (sockerr < 0) {
2806 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2807 xs_tcp_force_close(&transport->xprt);
2808 }
2809 }
2810
2811 static void xs_wake_pending(struct sock_xprt *transport)
2812 {
2813 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2814 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2815 }
2816
2817 static void xs_error_handle(struct work_struct *work)
2818 {
2819 struct sock_xprt *transport = container_of(work,
2820 struct sock_xprt, error_worker);
2821
2822 xs_wake_disconnect(transport);
2823 xs_wake_write(transport);
2824 xs_wake_error(transport);
2825 xs_wake_pending(transport);
2826 }
2827
2828 /**
2829 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2830 * @xprt: rpc_xprt struct containing statistics
2831 * @seq: output file
2832 *
2833 */
2834 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2835 {
2836 long idle_time = 0;
2837
2838 if (xprt_connected(xprt))
2839 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2840
2841 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2842 "%llu %llu %lu %llu %llu\n",
2843 xprt->stat.bind_count,
2844 xprt->stat.connect_count,
2845 xprt->stat.connect_time / HZ,
2846 idle_time,
2847 xprt->stat.sends,
2848 xprt->stat.recvs,
2849 xprt->stat.bad_xids,
2850 xprt->stat.req_u,
2851 xprt->stat.bklog_u,
2852 xprt->stat.max_slots,
2853 xprt->stat.sending_u,
2854 xprt->stat.pending_u);
2855 }
2856
2857 /**
2858 * xs_udp_print_stats - display UDP socket-specific stats
2859 * @xprt: rpc_xprt struct containing statistics
2860 * @seq: output file
2861 *
2862 */
2863 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2864 {
2865 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2866
2867 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2868 "%lu %llu %llu\n",
2869 transport->srcport,
2870 xprt->stat.bind_count,
2871 xprt->stat.sends,
2872 xprt->stat.recvs,
2873 xprt->stat.bad_xids,
2874 xprt->stat.req_u,
2875 xprt->stat.bklog_u,
2876 xprt->stat.max_slots,
2877 xprt->stat.sending_u,
2878 xprt->stat.pending_u);
2879 }
2880
2881 /**
2882 * xs_tcp_print_stats - display TCP socket-specific stats
2883 * @xprt: rpc_xprt struct containing statistics
2884 * @seq: output file
2885 *
2886 */
2887 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2888 {
2889 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2890 long idle_time = 0;
2891
2892 if (xprt_connected(xprt))
2893 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2894
2895 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2896 "%llu %llu %lu %llu %llu\n",
2897 transport->srcport,
2898 xprt->stat.bind_count,
2899 xprt->stat.connect_count,
2900 xprt->stat.connect_time / HZ,
2901 idle_time,
2902 xprt->stat.sends,
2903 xprt->stat.recvs,
2904 xprt->stat.bad_xids,
2905 xprt->stat.req_u,
2906 xprt->stat.bklog_u,
2907 xprt->stat.max_slots,
2908 xprt->stat.sending_u,
2909 xprt->stat.pending_u);
2910 }
2911
2912 /*
2913 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2914 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2915 * to use the server side send routines.
2916 */
2917 static int bc_malloc(struct rpc_task *task)
2918 {
2919 struct rpc_rqst *rqst = task->tk_rqstp;
2920 size_t size = rqst->rq_callsize;
2921 struct page *page;
2922 struct rpc_buffer *buf;
2923
2924 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2925 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2926 size);
2927 return -EINVAL;
2928 }
2929
2930 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2931 if (!page)
2932 return -ENOMEM;
2933
2934 buf = page_address(page);
2935 buf->len = PAGE_SIZE;
2936
2937 rqst->rq_buffer = buf->data;
2938 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2939 return 0;
2940 }
2941
2942 /*
2943 * Free the space allocated in the bc_alloc routine
2944 */
2945 static void bc_free(struct rpc_task *task)
2946 {
2947 void *buffer = task->tk_rqstp->rq_buffer;
2948 struct rpc_buffer *buf;
2949
2950 buf = container_of(buffer, struct rpc_buffer, data);
2951 free_page((unsigned long)buf);
2952 }
2953
2954 static int bc_sendto(struct rpc_rqst *req)
2955 {
2956 struct xdr_buf *xdr = &req->rq_snd_buf;
2957 struct sock_xprt *transport =
2958 container_of(req->rq_xprt, struct sock_xprt, xprt);
2959 struct msghdr msg = {
2960 .msg_flags = 0,
2961 };
2962 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2963 (u32)xdr->len);
2964 unsigned int sent = 0;
2965 int err;
2966
2967 req->rq_xtime = ktime_get();
2968 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2969 if (err < 0)
2970 return err;
2971 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2972 xdr_free_bvec(xdr);
2973 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2974 return -EAGAIN;
2975 return sent;
2976 }
2977
2978 /**
2979 * bc_send_request - Send a backchannel Call on a TCP socket
2980 * @req: rpc_rqst containing Call message to be sent
2981 *
2982 * xpt_mutex ensures @rqstp's whole message is written to the socket
2983 * without interruption.
2984 *
2985 * Return values:
2986 * %0 if the message was sent successfully
2987 * %ENOTCONN if the message was not sent
2988 */
2989 static int bc_send_request(struct rpc_rqst *req)
2990 {
2991 struct svc_xprt *xprt;
2992 int len;
2993
2994 /*
2995 * Get the server socket associated with this callback xprt
2996 */
2997 xprt = req->rq_xprt->bc_xprt;
2998
2999 /*
3000 * Grab the mutex to serialize data as the connection is shared
3001 * with the fore channel
3002 */
3003 mutex_lock(&xprt->xpt_mutex);
3004 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
3005 len = -ENOTCONN;
3006 else
3007 len = bc_sendto(req);
3008 mutex_unlock(&xprt->xpt_mutex);
3009
3010 if (len > 0)
3011 len = 0;
3012
3013 return len;
3014 }
3015
3016 static void bc_close(struct rpc_xprt *xprt)
3017 {
3018 xprt_disconnect_done(xprt);
3019 }
3020
3021 static void bc_destroy(struct rpc_xprt *xprt)
3022 {
3023 dprintk("RPC: bc_destroy xprt %p\n", xprt);
3024
3025 xs_xprt_free(xprt);
3026 module_put(THIS_MODULE);
3027 }
3028
3029 static const struct rpc_xprt_ops xs_local_ops = {
3030 .reserve_xprt = xprt_reserve_xprt,
3031 .release_xprt = xprt_release_xprt,
3032 .alloc_slot = xprt_alloc_slot,
3033 .free_slot = xprt_free_slot,
3034 .rpcbind = xs_local_rpcbind,
3035 .set_port = xs_local_set_port,
3036 .connect = xs_local_connect,
3037 .buf_alloc = rpc_malloc,
3038 .buf_free = rpc_free,
3039 .prepare_request = xs_stream_prepare_request,
3040 .send_request = xs_local_send_request,
3041 .abort_send_request = xs_stream_abort_send_request,
3042 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3043 .close = xs_close,
3044 .destroy = xs_destroy,
3045 .print_stats = xs_local_print_stats,
3046 .enable_swap = xs_enable_swap,
3047 .disable_swap = xs_disable_swap,
3048 };
3049
3050 static const struct rpc_xprt_ops xs_udp_ops = {
3051 .set_buffer_size = xs_udp_set_buffer_size,
3052 .reserve_xprt = xprt_reserve_xprt_cong,
3053 .release_xprt = xprt_release_xprt_cong,
3054 .alloc_slot = xprt_alloc_slot,
3055 .free_slot = xprt_free_slot,
3056 .rpcbind = rpcb_getport_async,
3057 .set_port = xs_set_port,
3058 .connect = xs_connect,
3059 .get_srcaddr = xs_sock_srcaddr,
3060 .get_srcport = xs_sock_srcport,
3061 .buf_alloc = rpc_malloc,
3062 .buf_free = rpc_free,
3063 .send_request = xs_udp_send_request,
3064 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3065 .timer = xs_udp_timer,
3066 .release_request = xprt_release_rqst_cong,
3067 .close = xs_close,
3068 .destroy = xs_destroy,
3069 .print_stats = xs_udp_print_stats,
3070 .enable_swap = xs_enable_swap,
3071 .disable_swap = xs_disable_swap,
3072 .inject_disconnect = xs_inject_disconnect,
3073 };
3074
3075 static const struct rpc_xprt_ops xs_tcp_ops = {
3076 .reserve_xprt = xprt_reserve_xprt,
3077 .release_xprt = xprt_release_xprt,
3078 .alloc_slot = xprt_alloc_slot,
3079 .free_slot = xprt_free_slot,
3080 .rpcbind = rpcb_getport_async,
3081 .set_port = xs_set_port,
3082 .connect = xs_connect,
3083 .get_srcaddr = xs_sock_srcaddr,
3084 .get_srcport = xs_sock_srcport,
3085 .buf_alloc = rpc_malloc,
3086 .buf_free = rpc_free,
3087 .prepare_request = xs_stream_prepare_request,
3088 .send_request = xs_tcp_send_request,
3089 .abort_send_request = xs_stream_abort_send_request,
3090 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3091 .close = xs_tcp_shutdown,
3092 .destroy = xs_destroy,
3093 .set_connect_timeout = xs_tcp_set_connect_timeout,
3094 .print_stats = xs_tcp_print_stats,
3095 .enable_swap = xs_enable_swap,
3096 .disable_swap = xs_disable_swap,
3097 .inject_disconnect = xs_inject_disconnect,
3098 #ifdef CONFIG_SUNRPC_BACKCHANNEL
3099 .bc_setup = xprt_setup_bc,
3100 .bc_maxpayload = xs_tcp_bc_maxpayload,
3101 .bc_num_slots = xprt_bc_max_slots,
3102 .bc_free_rqst = xprt_free_bc_rqst,
3103 .bc_destroy = xprt_destroy_bc,
3104 #endif
3105 };
3106
3107 /*
3108 * The rpc_xprt_ops for the server backchannel
3109 */
3110
3111 static const struct rpc_xprt_ops bc_tcp_ops = {
3112 .reserve_xprt = xprt_reserve_xprt,
3113 .release_xprt = xprt_release_xprt,
3114 .alloc_slot = xprt_alloc_slot,
3115 .free_slot = xprt_free_slot,
3116 .buf_alloc = bc_malloc,
3117 .buf_free = bc_free,
3118 .send_request = bc_send_request,
3119 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3120 .close = bc_close,
3121 .destroy = bc_destroy,
3122 .print_stats = xs_tcp_print_stats,
3123 .enable_swap = xs_enable_swap,
3124 .disable_swap = xs_disable_swap,
3125 .inject_disconnect = xs_inject_disconnect,
3126 };
3127
3128 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3129 {
3130 static const struct sockaddr_in sin = {
3131 .sin_family = AF_INET,
3132 .sin_addr.s_addr = htonl(INADDR_ANY),
3133 };
3134 static const struct sockaddr_in6 sin6 = {
3135 .sin6_family = AF_INET6,
3136 .sin6_addr = IN6ADDR_ANY_INIT,
3137 };
3138
3139 switch (family) {
3140 case AF_LOCAL:
3141 break;
3142 case AF_INET:
3143 memcpy(sap, &sin, sizeof(sin));
3144 break;
3145 case AF_INET6:
3146 memcpy(sap, &sin6, sizeof(sin6));
3147 break;
3148 default:
3149 dprintk("RPC: %s: Bad address family\n", __func__);
3150 return -EAFNOSUPPORT;
3151 }
3152 return 0;
3153 }
3154
3155 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3156 unsigned int slot_table_size,
3157 unsigned int max_slot_table_size)
3158 {
3159 struct rpc_xprt *xprt;
3160 struct sock_xprt *new;
3161
3162 if (args->addrlen > sizeof(xprt->addr)) {
3163 dprintk("RPC: xs_setup_xprt: address too large\n");
3164 return ERR_PTR(-EBADF);
3165 }
3166
3167 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3168 max_slot_table_size);
3169 if (xprt == NULL) {
3170 dprintk("RPC: xs_setup_xprt: couldn't allocate "
3171 "rpc_xprt\n");
3172 return ERR_PTR(-ENOMEM);
3173 }
3174
3175 new = container_of(xprt, struct sock_xprt, xprt);
3176 mutex_init(&new->recv_mutex);
3177 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3178 xprt->addrlen = args->addrlen;
3179 if (args->srcaddr)
3180 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3181 else {
3182 int err;
3183 err = xs_init_anyaddr(args->dstaddr->sa_family,
3184 (struct sockaddr *)&new->srcaddr);
3185 if (err != 0) {
3186 xprt_free(xprt);
3187 return ERR_PTR(err);
3188 }
3189 }
3190
3191 return xprt;
3192 }
3193
3194 static const struct rpc_timeout xs_local_default_timeout = {
3195 .to_initval = 10 * HZ,
3196 .to_maxval = 10 * HZ,
3197 .to_retries = 2,
3198 };
3199
3200 /**
3201 * xs_setup_local - Set up transport to use an AF_LOCAL socket
3202 * @args: rpc transport creation arguments
3203 *
3204 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3205 */
3206 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3207 {
3208 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3209 struct sock_xprt *transport;
3210 struct rpc_xprt *xprt;
3211 struct rpc_xprt *ret;
3212
3213 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3214 xprt_max_tcp_slot_table_entries);
3215 if (IS_ERR(xprt))
3216 return xprt;
3217 transport = container_of(xprt, struct sock_xprt, xprt);
3218
3219 xprt->prot = 0;
3220 xprt->xprt_class = &xs_local_transport;
3221 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3222
3223 xprt->bind_timeout = XS_BIND_TO;
3224 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3225 xprt->idle_timeout = XS_IDLE_DISC_TO;
3226
3227 xprt->ops = &xs_local_ops;
3228 xprt->timeout = &xs_local_default_timeout;
3229
3230 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3231 INIT_WORK(&transport->error_worker, xs_error_handle);
3232 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3233
3234 switch (sun->sun_family) {
3235 case AF_LOCAL:
3236 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3237 dprintk("RPC: bad AF_LOCAL address: %s\n",
3238 sun->sun_path);
3239 ret = ERR_PTR(-EINVAL);
3240 goto out_err;
3241 }
3242 xprt_set_bound(xprt);
3243 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3244 break;
3245 default:
3246 ret = ERR_PTR(-EAFNOSUPPORT);
3247 goto out_err;
3248 }
3249
3250 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
3251 xprt->address_strings[RPC_DISPLAY_ADDR]);
3252
3253 if (try_module_get(THIS_MODULE))
3254 return xprt;
3255 ret = ERR_PTR(-EINVAL);
3256 out_err:
3257 xs_xprt_free(xprt);
3258 return ret;
3259 }
3260
3261 static const struct rpc_timeout xs_udp_default_timeout = {
3262 .to_initval = 5 * HZ,
3263 .to_maxval = 30 * HZ,
3264 .to_increment = 5 * HZ,
3265 .to_retries = 5,
3266 };
3267
3268 /**
3269 * xs_setup_udp - Set up transport to use a UDP socket
3270 * @args: rpc transport creation arguments
3271 *
3272 */
3273 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3274 {
3275 struct sockaddr *addr = args->dstaddr;
3276 struct rpc_xprt *xprt;
3277 struct sock_xprt *transport;
3278 struct rpc_xprt *ret;
3279
3280 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3281 xprt_udp_slot_table_entries);
3282 if (IS_ERR(xprt))
3283 return xprt;
3284 transport = container_of(xprt, struct sock_xprt, xprt);
3285
3286 xprt->prot = IPPROTO_UDP;
3287 xprt->xprt_class = &xs_udp_transport;
3288 /* XXX: header size can vary due to auth type, IPv6, etc. */
3289 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3290
3291 xprt->bind_timeout = XS_BIND_TO;
3292 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3293 xprt->idle_timeout = XS_IDLE_DISC_TO;
3294
3295 xprt->ops = &xs_udp_ops;
3296
3297 xprt->timeout = &xs_udp_default_timeout;
3298
3299 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3300 INIT_WORK(&transport->error_worker, xs_error_handle);
3301 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3302
3303 switch (addr->sa_family) {
3304 case AF_INET:
3305 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3306 xprt_set_bound(xprt);
3307
3308 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3309 break;
3310 case AF_INET6:
3311 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3312 xprt_set_bound(xprt);
3313
3314 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3315 break;
3316 default:
3317 ret = ERR_PTR(-EAFNOSUPPORT);
3318 goto out_err;
3319 }
3320
3321 if (xprt_bound(xprt))
3322 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3323 xprt->address_strings[RPC_DISPLAY_ADDR],
3324 xprt->address_strings[RPC_DISPLAY_PORT],
3325 xprt->address_strings[RPC_DISPLAY_PROTO]);
3326 else
3327 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3328 xprt->address_strings[RPC_DISPLAY_ADDR],
3329 xprt->address_strings[RPC_DISPLAY_PROTO]);
3330
3331 if (try_module_get(THIS_MODULE))
3332 return xprt;
3333 ret = ERR_PTR(-EINVAL);
3334 out_err:
3335 xs_xprt_free(xprt);
3336 return ret;
3337 }
3338
3339 static const struct rpc_timeout xs_tcp_default_timeout = {
3340 .to_initval = 60 * HZ,
3341 .to_maxval = 60 * HZ,
3342 .to_retries = 2,
3343 };
3344
3345 /**
3346 * xs_setup_tcp - Set up transport to use a TCP socket
3347 * @args: rpc transport creation arguments
3348 *
3349 */
3350 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3351 {
3352 struct sockaddr *addr = args->dstaddr;
3353 struct rpc_xprt *xprt;
3354 struct sock_xprt *transport;
3355 struct rpc_xprt *ret;
3356 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3357
3358 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3359 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3360
3361 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3362 max_slot_table_size);
3363 if (IS_ERR(xprt))
3364 return xprt;
3365 transport = container_of(xprt, struct sock_xprt, xprt);
3366
3367 xprt->prot = IPPROTO_TCP;
3368 xprt->xprt_class = &xs_tcp_transport;
3369 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3370
3371 xprt->bind_timeout = XS_BIND_TO;
3372 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3373 xprt->idle_timeout = XS_IDLE_DISC_TO;
3374
3375 xprt->ops = &xs_tcp_ops;
3376 xprt->timeout = &xs_tcp_default_timeout;
3377
3378 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3379 if (args->reconnect_timeout)
3380 xprt->max_reconnect_timeout = args->reconnect_timeout;
3381
3382 xprt->connect_timeout = xprt->timeout->to_initval *
3383 (xprt->timeout->to_retries + 1);
3384 if (args->connect_timeout)
3385 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3386
3387 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3388 INIT_WORK(&transport->error_worker, xs_error_handle);
3389 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3390
3391 switch (addr->sa_family) {
3392 case AF_INET:
3393 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3394 xprt_set_bound(xprt);
3395
3396 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3397 break;
3398 case AF_INET6:
3399 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3400 xprt_set_bound(xprt);
3401
3402 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3403 break;
3404 default:
3405 ret = ERR_PTR(-EAFNOSUPPORT);
3406 goto out_err;
3407 }
3408
3409 if (xprt_bound(xprt))
3410 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3411 xprt->address_strings[RPC_DISPLAY_ADDR],
3412 xprt->address_strings[RPC_DISPLAY_PORT],
3413 xprt->address_strings[RPC_DISPLAY_PROTO]);
3414 else
3415 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3416 xprt->address_strings[RPC_DISPLAY_ADDR],
3417 xprt->address_strings[RPC_DISPLAY_PROTO]);
3418
3419 if (try_module_get(THIS_MODULE))
3420 return xprt;
3421 ret = ERR_PTR(-EINVAL);
3422 out_err:
3423 xs_xprt_free(xprt);
3424 return ret;
3425 }
3426
3427 /**
3428 * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3429 * @args: rpc transport creation arguments
3430 *
3431 */
3432 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3433 {
3434 struct sockaddr *addr = args->dstaddr;
3435 struct rpc_xprt *xprt;
3436 struct sock_xprt *transport;
3437 struct rpc_xprt *ret;
3438 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3439
3440 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3441 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3442
3443 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3444 max_slot_table_size);
3445 if (IS_ERR(xprt))
3446 return xprt;
3447 transport = container_of(xprt, struct sock_xprt, xprt);
3448
3449 xprt->prot = IPPROTO_TCP;
3450 xprt->xprt_class = &xs_tcp_transport;
3451 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3452
3453 xprt->bind_timeout = XS_BIND_TO;
3454 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3455 xprt->idle_timeout = XS_IDLE_DISC_TO;
3456
3457 xprt->ops = &xs_tcp_ops;
3458 xprt->timeout = &xs_tcp_default_timeout;
3459
3460 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3461 xprt->connect_timeout = xprt->timeout->to_initval *
3462 (xprt->timeout->to_retries + 1);
3463
3464 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3465 INIT_WORK(&transport->error_worker, xs_error_handle);
3466
3467 switch (args->xprtsec.policy) {
3468 case RPC_XPRTSEC_TLS_ANON:
3469 case RPC_XPRTSEC_TLS_X509:
3470 xprt->xprtsec = args->xprtsec;
3471 INIT_DELAYED_WORK(&transport->connect_worker,
3472 xs_tcp_tls_setup_socket);
3473 break;
3474 default:
3475 ret = ERR_PTR(-EACCES);
3476 goto out_err;
3477 }
3478
3479 switch (addr->sa_family) {
3480 case AF_INET:
3481 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3482 xprt_set_bound(xprt);
3483
3484 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3485 break;
3486 case AF_INET6:
3487 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3488 xprt_set_bound(xprt);
3489
3490 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3491 break;
3492 default:
3493 ret = ERR_PTR(-EAFNOSUPPORT);
3494 goto out_err;
3495 }
3496
3497 if (xprt_bound(xprt))
3498 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3499 xprt->address_strings[RPC_DISPLAY_ADDR],
3500 xprt->address_strings[RPC_DISPLAY_PORT],
3501 xprt->address_strings[RPC_DISPLAY_PROTO]);
3502 else
3503 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3504 xprt->address_strings[RPC_DISPLAY_ADDR],
3505 xprt->address_strings[RPC_DISPLAY_PROTO]);
3506
3507 if (try_module_get(THIS_MODULE))
3508 return xprt;
3509 ret = ERR_PTR(-EINVAL);
3510 out_err:
3511 xs_xprt_free(xprt);
3512 return ret;
3513 }
3514
3515 /**
3516 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3517 * @args: rpc transport creation arguments
3518 *
3519 */
3520 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3521 {
3522 struct sockaddr *addr = args->dstaddr;
3523 struct rpc_xprt *xprt;
3524 struct sock_xprt *transport;
3525 struct svc_sock *bc_sock;
3526 struct rpc_xprt *ret;
3527
3528 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3529 xprt_tcp_slot_table_entries);
3530 if (IS_ERR(xprt))
3531 return xprt;
3532 transport = container_of(xprt, struct sock_xprt, xprt);
3533
3534 xprt->prot = IPPROTO_TCP;
3535 xprt->xprt_class = &xs_bc_tcp_transport;
3536 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3537 xprt->timeout = &xs_tcp_default_timeout;
3538
3539 /* backchannel */
3540 xprt_set_bound(xprt);
3541 xprt->bind_timeout = 0;
3542 xprt->reestablish_timeout = 0;
3543 xprt->idle_timeout = 0;
3544
3545 xprt->ops = &bc_tcp_ops;
3546
3547 switch (addr->sa_family) {
3548 case AF_INET:
3549 xs_format_peer_addresses(xprt, "tcp",
3550 RPCBIND_NETID_TCP);
3551 break;
3552 case AF_INET6:
3553 xs_format_peer_addresses(xprt, "tcp",
3554 RPCBIND_NETID_TCP6);
3555 break;
3556 default:
3557 ret = ERR_PTR(-EAFNOSUPPORT);
3558 goto out_err;
3559 }
3560
3561 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3562 xprt->address_strings[RPC_DISPLAY_ADDR],
3563 xprt->address_strings[RPC_DISPLAY_PORT],
3564 xprt->address_strings[RPC_DISPLAY_PROTO]);
3565
3566 /*
3567 * Once we've associated a backchannel xprt with a connection,
3568 * we want to keep it around as long as the connection lasts,
3569 * in case we need to start using it for a backchannel again;
3570 * this reference won't be dropped until bc_xprt is destroyed.
3571 */
3572 xprt_get(xprt);
3573 args->bc_xprt->xpt_bc_xprt = xprt;
3574 xprt->bc_xprt = args->bc_xprt;
3575 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3576 transport->sock = bc_sock->sk_sock;
3577 transport->inet = bc_sock->sk_sk;
3578
3579 /*
3580 * Since we don't want connections for the backchannel, we set
3581 * the xprt status to connected
3582 */
3583 xprt_set_connected(xprt);
3584
3585 if (try_module_get(THIS_MODULE))
3586 return xprt;
3587
3588 args->bc_xprt->xpt_bc_xprt = NULL;
3589 args->bc_xprt->xpt_bc_xps = NULL;
3590 xprt_put(xprt);
3591 ret = ERR_PTR(-EINVAL);
3592 out_err:
3593 xs_xprt_free(xprt);
3594 return ret;
3595 }
3596
3597 static struct xprt_class xs_local_transport = {
3598 .list = LIST_HEAD_INIT(xs_local_transport.list),
3599 .name = "named UNIX socket",
3600 .owner = THIS_MODULE,
3601 .ident = XPRT_TRANSPORT_LOCAL,
3602 .setup = xs_setup_local,
3603 .netid = { "" },
3604 };
3605
3606 static struct xprt_class xs_udp_transport = {
3607 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3608 .name = "udp",
3609 .owner = THIS_MODULE,
3610 .ident = XPRT_TRANSPORT_UDP,
3611 .setup = xs_setup_udp,
3612 .netid = { "udp", "udp6", "" },
3613 };
3614
3615 static struct xprt_class xs_tcp_transport = {
3616 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3617 .name = "tcp",
3618 .owner = THIS_MODULE,
3619 .ident = XPRT_TRANSPORT_TCP,
3620 .setup = xs_setup_tcp,
3621 .netid = { "tcp", "tcp6", "" },
3622 };
3623
3624 static struct xprt_class xs_tcp_tls_transport = {
3625 .list = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3626 .name = "tcp-with-tls",
3627 .owner = THIS_MODULE,
3628 .ident = XPRT_TRANSPORT_TCP_TLS,
3629 .setup = xs_setup_tcp_tls,
3630 .netid = { "tcp", "tcp6", "" },
3631 };
3632
3633 static struct xprt_class xs_bc_tcp_transport = {
3634 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3635 .name = "tcp NFSv4.1 backchannel",
3636 .owner = THIS_MODULE,
3637 .ident = XPRT_TRANSPORT_BC_TCP,
3638 .setup = xs_setup_bc_tcp,
3639 .netid = { "" },
3640 };
3641
3642 /**
3643 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3644 *
3645 */
3646 int init_socket_xprt(void)
3647 {
3648 if (!sunrpc_table_header)
3649 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3650
3651 xprt_register_transport(&xs_local_transport);
3652 xprt_register_transport(&xs_udp_transport);
3653 xprt_register_transport(&xs_tcp_transport);
3654 xprt_register_transport(&xs_tcp_tls_transport);
3655 xprt_register_transport(&xs_bc_tcp_transport);
3656
3657 return 0;
3658 }
3659
3660 /**
3661 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3662 *
3663 */
3664 void cleanup_socket_xprt(void)
3665 {
3666 if (sunrpc_table_header) {
3667 unregister_sysctl_table(sunrpc_table_header);
3668 sunrpc_table_header = NULL;
3669 }
3670
3671 xprt_unregister_transport(&xs_local_transport);
3672 xprt_unregister_transport(&xs_udp_transport);
3673 xprt_unregister_transport(&xs_tcp_transport);
3674 xprt_unregister_transport(&xs_tcp_tls_transport);
3675 xprt_unregister_transport(&xs_bc_tcp_transport);
3676 }
3677
3678 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3679 {
3680 return param_set_uint_minmax(val, kp,
3681 RPC_MIN_RESVPORT,
3682 RPC_MAX_RESVPORT);
3683 }
3684
3685 static const struct kernel_param_ops param_ops_portnr = {
3686 .set = param_set_portnr,
3687 .get = param_get_uint,
3688 };
3689
3690 #define param_check_portnr(name, p) \
3691 __param_check(name, p, unsigned int);
3692
3693 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3694 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3695
3696 static int param_set_slot_table_size(const char *val,
3697 const struct kernel_param *kp)
3698 {
3699 return param_set_uint_minmax(val, kp,
3700 RPC_MIN_SLOT_TABLE,
3701 RPC_MAX_SLOT_TABLE);
3702 }
3703
3704 static const struct kernel_param_ops param_ops_slot_table_size = {
3705 .set = param_set_slot_table_size,
3706 .get = param_get_uint,
3707 };
3708
3709 #define param_check_slot_table_size(name, p) \
3710 __param_check(name, p, unsigned int);
3711
3712 static int param_set_max_slot_table_size(const char *val,
3713 const struct kernel_param *kp)
3714 {
3715 return param_set_uint_minmax(val, kp,
3716 RPC_MIN_SLOT_TABLE,
3717 RPC_MAX_SLOT_TABLE_LIMIT);
3718 }
3719
3720 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3721 .set = param_set_max_slot_table_size,
3722 .get = param_get_uint,
3723 };
3724
3725 #define param_check_max_slot_table_size(name, p) \
3726 __param_check(name, p, unsigned int);
3727
3728 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3729 slot_table_size, 0644);
3730 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3731 max_slot_table_size, 0644);
3732 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3733 slot_table_size, 0644);
3734
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.