Loading...
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
51#include <trace/events/sunrpc.h>
52
53#include "sunrpc.h"
54
55#define RPC_TCP_READ_CHUNK_SZ (3*512*1024)
56
57static void xs_close(struct rpc_xprt *xprt);
58static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
59 struct socket *sock);
60
61/*
62 * xprtsock tunables
63 */
64static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
67
68static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
70
71#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
72
73#define XS_TCP_LINGER_TO (15U * HZ)
74static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
75
76/*
77 * We can register our own files under /proc/sys/sunrpc by
78 * calling register_sysctl_table() again. The files in that
79 * directory become the union of all files registered there.
80 *
81 * We simply need to make sure that we don't collide with
82 * someone else's file names!
83 */
84
85static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
90
91static struct ctl_table_header *sunrpc_table_header;
92
93/*
94 * FIXME: changing the UDP slot table size should also resize the UDP
95 * socket buffers for existing UDP transports
96 */
97static struct ctl_table xs_tunables_table[] = {
98 {
99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
102 .mode = 0644,
103 .proc_handler = proc_dointvec_minmax,
104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
106 },
107 {
108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
111 .mode = 0644,
112 .proc_handler = proc_dointvec_minmax,
113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
115 },
116 {
117 .procname = "tcp_max_slot_table_entries",
118 .data = &xprt_max_tcp_slot_table_entries,
119 .maxlen = sizeof(unsigned int),
120 .mode = 0644,
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_slot_table_size,
123 .extra2 = &max_tcp_slot_table_limit
124 },
125 {
126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
129 .mode = 0644,
130 .proc_handler = proc_dointvec_minmax,
131 .extra1 = &xprt_min_resvport_limit,
132 .extra2 = &xprt_max_resvport
133 },
134 {
135 .procname = "max_resvport",
136 .data = &xprt_max_resvport,
137 .maxlen = sizeof(unsigned int),
138 .mode = 0644,
139 .proc_handler = proc_dointvec_minmax,
140 .extra1 = &xprt_min_resvport,
141 .extra2 = &xprt_max_resvport_limit
142 },
143 {
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
147 .mode = 0644,
148 .proc_handler = proc_dointvec_jiffies,
149 },
150 { },
151};
152
153static struct ctl_table sunrpc_table[] = {
154 {
155 .procname = "sunrpc",
156 .mode = 0555,
157 .child = xs_tunables_table
158 },
159 { },
160};
161
162#endif
163
164/*
165 * Wait duration for a reply from the RPC portmapper.
166 */
167#define XS_BIND_TO (60U * HZ)
168
169/*
170 * Delay if a UDP socket connect error occurs. This is most likely some
171 * kind of resource problem on the local host.
172 */
173#define XS_UDP_REEST_TO (2U * HZ)
174
175/*
176 * The reestablish timeout allows clients to delay for a bit before attempting
177 * to reconnect to a server that just dropped our connection.
178 *
179 * We implement an exponential backoff when trying to reestablish a TCP
180 * transport connection with the server. Some servers like to drop a TCP
181 * connection when they are overworked, so we start with a short timeout and
182 * increase over time if the server is down or not responding.
183 */
184#define XS_TCP_INIT_REEST_TO (3U * HZ)
185
186/*
187 * TCP idle timeout; client drops the transport socket if it is idle
188 * for this long. Note that we also timeout UDP sockets to prevent
189 * holding port numbers when there is no RPC traffic.
190 */
191#define XS_IDLE_DISC_TO (5U * 60 * HZ)
192
193#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
194# undef RPC_DEBUG_DATA
195# define RPCDBG_FACILITY RPCDBG_TRANS
196#endif
197
198#ifdef RPC_DEBUG_DATA
199static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
200{
201 u8 *buf = (u8 *) packet;
202 int j;
203
204 dprintk("RPC: %s\n", msg);
205 for (j = 0; j < count && j < 128; j += 4) {
206 if (!(j & 31)) {
207 if (j)
208 dprintk("\n");
209 dprintk("0x%04x ", j);
210 }
211 dprintk("%02x%02x%02x%02x ",
212 buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 }
214 dprintk("\n");
215}
216#else
217static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
218{
219 /* NOP */
220}
221#endif
222
223static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
224{
225 return (struct rpc_xprt *) sk->sk_user_data;
226}
227
228static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
229{
230 return (struct sockaddr *) &xprt->addr;
231}
232
233static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
234{
235 return (struct sockaddr_un *) &xprt->addr;
236}
237
238static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
239{
240 return (struct sockaddr_in *) &xprt->addr;
241}
242
243static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
244{
245 return (struct sockaddr_in6 *) &xprt->addr;
246}
247
248static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
249{
250 struct sockaddr *sap = xs_addr(xprt);
251 struct sockaddr_in6 *sin6;
252 struct sockaddr_in *sin;
253 struct sockaddr_un *sun;
254 char buf[128];
255
256 switch (sap->sa_family) {
257 case AF_LOCAL:
258 sun = xs_addr_un(xprt);
259 strlcpy(buf, sun->sun_path, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
262 break;
263 case AF_INET:
264 (void)rpc_ntop(sap, buf, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
267 sin = xs_addr_in(xprt);
268 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
269 break;
270 case AF_INET6:
271 (void)rpc_ntop(sap, buf, sizeof(buf));
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
274 sin6 = xs_addr_in6(xprt);
275 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
276 break;
277 default:
278 BUG();
279 }
280
281 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
282}
283
284static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
285{
286 struct sockaddr *sap = xs_addr(xprt);
287 char buf[128];
288
289 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
290 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
291
292 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
293 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
294}
295
296static void xs_format_peer_addresses(struct rpc_xprt *xprt,
297 const char *protocol,
298 const char *netid)
299{
300 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
301 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
302 xs_format_common_peer_addresses(xprt);
303 xs_format_common_peer_ports(xprt);
304}
305
306static void xs_update_peer_port(struct rpc_xprt *xprt)
307{
308 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
309 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
310
311 xs_format_common_peer_ports(xprt);
312}
313
314static void xs_free_peer_addresses(struct rpc_xprt *xprt)
315{
316 unsigned int i;
317
318 for (i = 0; i < RPC_DISPLAY_MAX; i++)
319 switch (i) {
320 case RPC_DISPLAY_PROTO:
321 case RPC_DISPLAY_NETID:
322 continue;
323 default:
324 kfree(xprt->address_strings[i]);
325 }
326}
327
328#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
329
330static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
331{
332 struct msghdr msg = {
333 .msg_name = addr,
334 .msg_namelen = addrlen,
335 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
336 };
337 struct kvec iov = {
338 .iov_base = vec->iov_base + base,
339 .iov_len = vec->iov_len - base,
340 };
341
342 if (iov.iov_len != 0)
343 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
344 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
345}
346
347static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
348{
349 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
350 int offset, size_t size, int flags);
351 struct page **ppage;
352 unsigned int remainder;
353 int err;
354
355 remainder = xdr->page_len - base;
356 base += xdr->page_base;
357 ppage = xdr->pages + (base >> PAGE_SHIFT);
358 base &= ~PAGE_MASK;
359 do_sendpage = sock->ops->sendpage;
360 if (!zerocopy)
361 do_sendpage = sock_no_sendpage;
362 for(;;) {
363 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
364 int flags = XS_SENDMSG_FLAGS;
365
366 remainder -= len;
367 if (more)
368 flags |= MSG_MORE;
369 if (remainder != 0)
370 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
371 err = do_sendpage(sock, *ppage, base, len, flags);
372 if (remainder == 0 || err != len)
373 break;
374 *sent_p += err;
375 ppage++;
376 base = 0;
377 }
378 if (err > 0) {
379 *sent_p += err;
380 err = 0;
381 }
382 return err;
383}
384
385/**
386 * xs_sendpages - write pages directly to a socket
387 * @sock: socket to send on
388 * @addr: UDP only -- address of destination
389 * @addrlen: UDP only -- length of destination address
390 * @xdr: buffer containing this request
391 * @base: starting position in the buffer
392 * @zerocopy: true if it is safe to use sendpage()
393 * @sent_p: return the total number of bytes successfully queued for sending
394 *
395 */
396static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
397{
398 unsigned int remainder = xdr->len - base;
399 int err = 0;
400 int sent = 0;
401
402 if (unlikely(!sock))
403 return -ENOTSOCK;
404
405 if (base != 0) {
406 addr = NULL;
407 addrlen = 0;
408 }
409
410 if (base < xdr->head[0].iov_len || addr != NULL) {
411 unsigned int len = xdr->head[0].iov_len - base;
412 remainder -= len;
413 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
414 if (remainder == 0 || err != len)
415 goto out;
416 *sent_p += err;
417 base = 0;
418 } else
419 base -= xdr->head[0].iov_len;
420
421 if (base < xdr->page_len) {
422 unsigned int len = xdr->page_len - base;
423 remainder -= len;
424 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
425 *sent_p += sent;
426 if (remainder == 0 || sent != len)
427 goto out;
428 base = 0;
429 } else
430 base -= xdr->page_len;
431
432 if (base >= xdr->tail[0].iov_len)
433 return 0;
434 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
435out:
436 if (err > 0) {
437 *sent_p += err;
438 err = 0;
439 }
440 return err;
441}
442
443static void xs_nospace_callback(struct rpc_task *task)
444{
445 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
446
447 transport->inet->sk_write_pending--;
448}
449
450/**
451 * xs_nospace - place task on wait queue if transmit was incomplete
452 * @task: task to put to sleep
453 *
454 */
455static int xs_nospace(struct rpc_task *task)
456{
457 struct rpc_rqst *req = task->tk_rqstp;
458 struct rpc_xprt *xprt = req->rq_xprt;
459 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
460 struct sock *sk = transport->inet;
461 int ret = -EAGAIN;
462
463 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
464 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
465 req->rq_slen);
466
467 /* Protect against races with write_space */
468 spin_lock_bh(&xprt->transport_lock);
469
470 /* Don't race with disconnect */
471 if (xprt_connected(xprt)) {
472 /* wait for more buffer space */
473 sk->sk_write_pending++;
474 xprt_wait_for_buffer_space(task, xs_nospace_callback);
475 } else
476 ret = -ENOTCONN;
477
478 spin_unlock_bh(&xprt->transport_lock);
479
480 /* Race breaker in case memory is freed before above code is called */
481 if (ret == -EAGAIN) {
482 struct socket_wq *wq;
483
484 rcu_read_lock();
485 wq = rcu_dereference(sk->sk_wq);
486 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
487 rcu_read_unlock();
488
489 sk->sk_write_space(sk);
490 }
491 return ret;
492}
493
494/*
495 * Construct a stream transport record marker in @buf.
496 */
497static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
498{
499 u32 reclen = buf->len - sizeof(rpc_fraghdr);
500 rpc_fraghdr *base = buf->head[0].iov_base;
501 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
502}
503
504/**
505 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
506 * @task: RPC task that manages the state of an RPC request
507 *
508 * Return values:
509 * 0: The request has been sent
510 * EAGAIN: The socket was blocked, please call again later to
511 * complete the request
512 * ENOTCONN: Caller needs to invoke connect logic then call again
513 * other: Some other error occured, the request was not sent
514 */
515static int xs_local_send_request(struct rpc_task *task)
516{
517 struct rpc_rqst *req = task->tk_rqstp;
518 struct rpc_xprt *xprt = req->rq_xprt;
519 struct sock_xprt *transport =
520 container_of(xprt, struct sock_xprt, xprt);
521 struct xdr_buf *xdr = &req->rq_snd_buf;
522 int status;
523 int sent = 0;
524
525 xs_encode_stream_record_marker(&req->rq_snd_buf);
526
527 xs_pktdump("packet data:",
528 req->rq_svec->iov_base, req->rq_svec->iov_len);
529
530 req->rq_xtime = ktime_get();
531 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
532 true, &sent);
533 dprintk("RPC: %s(%u) = %d\n",
534 __func__, xdr->len - req->rq_bytes_sent, status);
535
536 if (status == -EAGAIN && sock_writeable(transport->inet))
537 status = -ENOBUFS;
538
539 if (likely(sent > 0) || status == 0) {
540 req->rq_bytes_sent += sent;
541 req->rq_xmit_bytes_sent += sent;
542 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
543 req->rq_bytes_sent = 0;
544 return 0;
545 }
546 status = -EAGAIN;
547 }
548
549 switch (status) {
550 case -ENOBUFS:
551 break;
552 case -EAGAIN:
553 status = xs_nospace(task);
554 break;
555 default:
556 dprintk("RPC: sendmsg returned unrecognized error %d\n",
557 -status);
558 /* fall through */
559 case -EPIPE:
560 xs_close(xprt);
561 status = -ENOTCONN;
562 }
563
564 return status;
565}
566
567/**
568 * xs_udp_send_request - write an RPC request to a UDP socket
569 * @task: address of RPC task that manages the state of an RPC request
570 *
571 * Return values:
572 * 0: The request has been sent
573 * EAGAIN: The socket was blocked, please call again later to
574 * complete the request
575 * ENOTCONN: Caller needs to invoke connect logic then call again
576 * other: Some other error occurred, the request was not sent
577 */
578static int xs_udp_send_request(struct rpc_task *task)
579{
580 struct rpc_rqst *req = task->tk_rqstp;
581 struct rpc_xprt *xprt = req->rq_xprt;
582 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
583 struct xdr_buf *xdr = &req->rq_snd_buf;
584 int sent = 0;
585 int status;
586
587 xs_pktdump("packet data:",
588 req->rq_svec->iov_base,
589 req->rq_svec->iov_len);
590
591 if (!xprt_bound(xprt))
592 return -ENOTCONN;
593 req->rq_xtime = ktime_get();
594 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
595 xdr, req->rq_bytes_sent, true, &sent);
596
597 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
598 xdr->len - req->rq_bytes_sent, status);
599
600 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
601 if (status == -EPERM)
602 goto process_status;
603
604 if (status == -EAGAIN && sock_writeable(transport->inet))
605 status = -ENOBUFS;
606
607 if (sent > 0 || status == 0) {
608 req->rq_xmit_bytes_sent += sent;
609 if (sent >= req->rq_slen)
610 return 0;
611 /* Still some bytes left; set up for a retry later. */
612 status = -EAGAIN;
613 }
614
615process_status:
616 switch (status) {
617 case -ENOTSOCK:
618 status = -ENOTCONN;
619 /* Should we call xs_close() here? */
620 break;
621 case -EAGAIN:
622 status = xs_nospace(task);
623 break;
624 case -ENETUNREACH:
625 case -ENOBUFS:
626 case -EPIPE:
627 case -ECONNREFUSED:
628 case -EPERM:
629 /* When the server has died, an ICMP port unreachable message
630 * prompts ECONNREFUSED. */
631 break;
632 default:
633 dprintk("RPC: sendmsg returned unrecognized error %d\n",
634 -status);
635 }
636
637 return status;
638}
639
640/**
641 * xs_tcp_send_request - write an RPC request to a TCP socket
642 * @task: address of RPC task that manages the state of an RPC request
643 *
644 * Return values:
645 * 0: The request has been sent
646 * EAGAIN: The socket was blocked, please call again later to
647 * complete the request
648 * ENOTCONN: Caller needs to invoke connect logic then call again
649 * other: Some other error occurred, the request was not sent
650 *
651 * XXX: In the case of soft timeouts, should we eventually give up
652 * if sendmsg is not able to make progress?
653 */
654static int xs_tcp_send_request(struct rpc_task *task)
655{
656 struct rpc_rqst *req = task->tk_rqstp;
657 struct rpc_xprt *xprt = req->rq_xprt;
658 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
659 struct xdr_buf *xdr = &req->rq_snd_buf;
660 bool zerocopy = true;
661 bool vm_wait = false;
662 int status;
663 int sent;
664
665 xs_encode_stream_record_marker(&req->rq_snd_buf);
666
667 xs_pktdump("packet data:",
668 req->rq_svec->iov_base,
669 req->rq_svec->iov_len);
670 /* Don't use zero copy if this is a resend. If the RPC call
671 * completes while the socket holds a reference to the pages,
672 * then we may end up resending corrupted data.
673 */
674 if (task->tk_flags & RPC_TASK_SENT)
675 zerocopy = false;
676
677 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
678 xs_tcp_set_socket_timeouts(xprt, transport->sock);
679
680 /* Continue transmitting the packet/record. We must be careful
681 * to cope with writespace callbacks arriving _after_ we have
682 * called sendmsg(). */
683 req->rq_xtime = ktime_get();
684 while (1) {
685 sent = 0;
686 status = xs_sendpages(transport->sock, NULL, 0, xdr,
687 req->rq_bytes_sent, zerocopy, &sent);
688
689 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
690 xdr->len - req->rq_bytes_sent, status);
691
692 /* If we've sent the entire packet, immediately
693 * reset the count of bytes sent. */
694 req->rq_bytes_sent += sent;
695 req->rq_xmit_bytes_sent += sent;
696 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
697 req->rq_bytes_sent = 0;
698 return 0;
699 }
700
701 WARN_ON_ONCE(sent == 0 && status == 0);
702
703 if (status == -EAGAIN ) {
704 /*
705 * Return EAGAIN if we're sure we're hitting the
706 * socket send buffer limits.
707 */
708 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
709 break;
710 /*
711 * Did we hit a memory allocation failure?
712 */
713 if (sent == 0) {
714 status = -ENOBUFS;
715 if (vm_wait)
716 break;
717 /* Retry, knowing now that we're below the
718 * socket send buffer limit
719 */
720 vm_wait = true;
721 }
722 continue;
723 }
724 if (status < 0)
725 break;
726 vm_wait = false;
727 }
728
729 switch (status) {
730 case -ENOTSOCK:
731 status = -ENOTCONN;
732 /* Should we call xs_close() here? */
733 break;
734 case -EAGAIN:
735 status = xs_nospace(task);
736 break;
737 case -ECONNRESET:
738 case -ECONNREFUSED:
739 case -ENOTCONN:
740 case -EADDRINUSE:
741 case -ENOBUFS:
742 case -EPIPE:
743 break;
744 default:
745 dprintk("RPC: sendmsg returned unrecognized error %d\n",
746 -status);
747 }
748
749 return status;
750}
751
752/**
753 * xs_tcp_release_xprt - clean up after a tcp transmission
754 * @xprt: transport
755 * @task: rpc task
756 *
757 * This cleans up if an error causes us to abort the transmission of a request.
758 * In this case, the socket may need to be reset in order to avoid confusing
759 * the server.
760 */
761static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
762{
763 struct rpc_rqst *req;
764
765 if (task != xprt->snd_task)
766 return;
767 if (task == NULL)
768 goto out_release;
769 req = task->tk_rqstp;
770 if (req == NULL)
771 goto out_release;
772 if (req->rq_bytes_sent == 0)
773 goto out_release;
774 if (req->rq_bytes_sent == req->rq_snd_buf.len)
775 goto out_release;
776 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
777out_release:
778 xprt_release_xprt(xprt, task);
779}
780
781static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
782{
783 transport->old_data_ready = sk->sk_data_ready;
784 transport->old_state_change = sk->sk_state_change;
785 transport->old_write_space = sk->sk_write_space;
786 transport->old_error_report = sk->sk_error_report;
787}
788
789static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
790{
791 sk->sk_data_ready = transport->old_data_ready;
792 sk->sk_state_change = transport->old_state_change;
793 sk->sk_write_space = transport->old_write_space;
794 sk->sk_error_report = transport->old_error_report;
795}
796
797static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
798{
799 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
800
801 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
802}
803
804static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
805{
806 smp_mb__before_atomic();
807 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
808 clear_bit(XPRT_CLOSING, &xprt->state);
809 xs_sock_reset_state_flags(xprt);
810 smp_mb__after_atomic();
811}
812
813/**
814 * xs_error_report - callback to handle TCP socket state errors
815 * @sk: socket
816 *
817 * Note: we don't call sock_error() since there may be a rpc_task
818 * using the socket, and so we don't want to clear sk->sk_err.
819 */
820static void xs_error_report(struct sock *sk)
821{
822 struct rpc_xprt *xprt;
823 int err;
824
825 read_lock_bh(&sk->sk_callback_lock);
826 if (!(xprt = xprt_from_sock(sk)))
827 goto out;
828
829 err = -sk->sk_err;
830 if (err == 0)
831 goto out;
832 dprintk("RPC: xs_error_report client %p, error=%d...\n",
833 xprt, -err);
834 trace_rpc_socket_error(xprt, sk->sk_socket, err);
835 xprt_wake_pending_tasks(xprt, err);
836 out:
837 read_unlock_bh(&sk->sk_callback_lock);
838}
839
840static void xs_reset_transport(struct sock_xprt *transport)
841{
842 struct socket *sock = transport->sock;
843 struct sock *sk = transport->inet;
844 struct rpc_xprt *xprt = &transport->xprt;
845
846 if (sk == NULL)
847 return;
848
849 if (atomic_read(&transport->xprt.swapper))
850 sk_clear_memalloc(sk);
851
852 kernel_sock_shutdown(sock, SHUT_RDWR);
853
854 mutex_lock(&transport->recv_mutex);
855 write_lock_bh(&sk->sk_callback_lock);
856 transport->inet = NULL;
857 transport->sock = NULL;
858
859 sk->sk_user_data = NULL;
860
861 xs_restore_old_callbacks(transport, sk);
862 xprt_clear_connected(xprt);
863 write_unlock_bh(&sk->sk_callback_lock);
864 xs_sock_reset_connection_flags(xprt);
865 mutex_unlock(&transport->recv_mutex);
866
867 trace_rpc_socket_close(xprt, sock);
868 sock_release(sock);
869}
870
871/**
872 * xs_close - close a socket
873 * @xprt: transport
874 *
875 * This is used when all requests are complete; ie, no DRC state remains
876 * on the server we want to save.
877 *
878 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
879 * xs_reset_transport() zeroing the socket from underneath a writer.
880 */
881static void xs_close(struct rpc_xprt *xprt)
882{
883 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
884
885 dprintk("RPC: xs_close xprt %p\n", xprt);
886
887 xs_reset_transport(transport);
888 xprt->reestablish_timeout = 0;
889
890 xprt_disconnect_done(xprt);
891}
892
893static void xs_inject_disconnect(struct rpc_xprt *xprt)
894{
895 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
896 xprt);
897 xprt_disconnect_done(xprt);
898}
899
900static void xs_xprt_free(struct rpc_xprt *xprt)
901{
902 xs_free_peer_addresses(xprt);
903 xprt_free(xprt);
904}
905
906/**
907 * xs_destroy - prepare to shutdown a transport
908 * @xprt: doomed transport
909 *
910 */
911static void xs_destroy(struct rpc_xprt *xprt)
912{
913 struct sock_xprt *transport = container_of(xprt,
914 struct sock_xprt, xprt);
915 dprintk("RPC: xs_destroy xprt %p\n", xprt);
916
917 cancel_delayed_work_sync(&transport->connect_worker);
918 xs_close(xprt);
919 cancel_work_sync(&transport->recv_worker);
920 xs_xprt_free(xprt);
921 module_put(THIS_MODULE);
922}
923
924static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
925{
926 struct xdr_skb_reader desc = {
927 .skb = skb,
928 .offset = sizeof(rpc_fraghdr),
929 .count = skb->len - sizeof(rpc_fraghdr),
930 };
931
932 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
933 return -1;
934 if (desc.count)
935 return -1;
936 return 0;
937}
938
939/**
940 * xs_local_data_read_skb
941 * @xprt: transport
942 * @sk: socket
943 * @skb: skbuff
944 *
945 * Currently this assumes we can read the whole reply in a single gulp.
946 */
947static void xs_local_data_read_skb(struct rpc_xprt *xprt,
948 struct sock *sk,
949 struct sk_buff *skb)
950{
951 struct rpc_task *task;
952 struct rpc_rqst *rovr;
953 int repsize, copied;
954 u32 _xid;
955 __be32 *xp;
956
957 repsize = skb->len - sizeof(rpc_fraghdr);
958 if (repsize < 4) {
959 dprintk("RPC: impossible RPC reply size %d\n", repsize);
960 return;
961 }
962
963 /* Copy the XID from the skb... */
964 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
965 if (xp == NULL)
966 return;
967
968 /* Look up and lock the request corresponding to the given XID */
969 spin_lock(&xprt->recv_lock);
970 rovr = xprt_lookup_rqst(xprt, *xp);
971 if (!rovr)
972 goto out_unlock;
973 xprt_pin_rqst(rovr);
974 spin_unlock(&xprt->recv_lock);
975 task = rovr->rq_task;
976
977 copied = rovr->rq_private_buf.buflen;
978 if (copied > repsize)
979 copied = repsize;
980
981 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
982 dprintk("RPC: sk_buff copy failed\n");
983 spin_lock(&xprt->recv_lock);
984 goto out_unpin;
985 }
986
987 spin_lock(&xprt->recv_lock);
988 xprt_complete_rqst(task, copied);
989out_unpin:
990 xprt_unpin_rqst(rovr);
991 out_unlock:
992 spin_unlock(&xprt->recv_lock);
993}
994
995static void xs_local_data_receive(struct sock_xprt *transport)
996{
997 struct sk_buff *skb;
998 struct sock *sk;
999 int err;
1000
1001restart:
1002 mutex_lock(&transport->recv_mutex);
1003 sk = transport->inet;
1004 if (sk == NULL)
1005 goto out;
1006 for (;;) {
1007 skb = skb_recv_datagram(sk, 0, 1, &err);
1008 if (skb != NULL) {
1009 xs_local_data_read_skb(&transport->xprt, sk, skb);
1010 skb_free_datagram(sk, skb);
1011 continue;
1012 }
1013 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1014 break;
1015 if (need_resched()) {
1016 mutex_unlock(&transport->recv_mutex);
1017 cond_resched();
1018 goto restart;
1019 }
1020 }
1021out:
1022 mutex_unlock(&transport->recv_mutex);
1023}
1024
1025static void xs_local_data_receive_workfn(struct work_struct *work)
1026{
1027 struct sock_xprt *transport =
1028 container_of(work, struct sock_xprt, recv_worker);
1029 xs_local_data_receive(transport);
1030}
1031
1032/**
1033 * xs_udp_data_read_skb - receive callback for UDP sockets
1034 * @xprt: transport
1035 * @sk: socket
1036 * @skb: skbuff
1037 *
1038 */
1039static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1040 struct sock *sk,
1041 struct sk_buff *skb)
1042{
1043 struct rpc_task *task;
1044 struct rpc_rqst *rovr;
1045 int repsize, copied;
1046 u32 _xid;
1047 __be32 *xp;
1048
1049 repsize = skb->len;
1050 if (repsize < 4) {
1051 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1052 return;
1053 }
1054
1055 /* Copy the XID from the skb... */
1056 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1057 if (xp == NULL)
1058 return;
1059
1060 /* Look up and lock the request corresponding to the given XID */
1061 spin_lock(&xprt->recv_lock);
1062 rovr = xprt_lookup_rqst(xprt, *xp);
1063 if (!rovr)
1064 goto out_unlock;
1065 xprt_pin_rqst(rovr);
1066 xprt_update_rtt(rovr->rq_task);
1067 spin_unlock(&xprt->recv_lock);
1068 task = rovr->rq_task;
1069
1070 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1071 copied = repsize;
1072
1073 /* Suck it into the iovec, verify checksum if not done by hw. */
1074 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1075 spin_lock(&xprt->recv_lock);
1076 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1077 goto out_unpin;
1078 }
1079
1080
1081 spin_lock_bh(&xprt->transport_lock);
1082 xprt_adjust_cwnd(xprt, task, copied);
1083 spin_unlock_bh(&xprt->transport_lock);
1084 spin_lock(&xprt->recv_lock);
1085 xprt_complete_rqst(task, copied);
1086 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1087out_unpin:
1088 xprt_unpin_rqst(rovr);
1089 out_unlock:
1090 spin_unlock(&xprt->recv_lock);
1091}
1092
1093static void xs_udp_data_receive(struct sock_xprt *transport)
1094{
1095 struct sk_buff *skb;
1096 struct sock *sk;
1097 int err;
1098
1099restart:
1100 mutex_lock(&transport->recv_mutex);
1101 sk = transport->inet;
1102 if (sk == NULL)
1103 goto out;
1104 for (;;) {
1105 skb = skb_recv_udp(sk, 0, 1, &err);
1106 if (skb != NULL) {
1107 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1108 consume_skb(skb);
1109 continue;
1110 }
1111 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1112 break;
1113 if (need_resched()) {
1114 mutex_unlock(&transport->recv_mutex);
1115 cond_resched();
1116 goto restart;
1117 }
1118 }
1119out:
1120 mutex_unlock(&transport->recv_mutex);
1121}
1122
1123static void xs_udp_data_receive_workfn(struct work_struct *work)
1124{
1125 struct sock_xprt *transport =
1126 container_of(work, struct sock_xprt, recv_worker);
1127 xs_udp_data_receive(transport);
1128}
1129
1130/**
1131 * xs_data_ready - "data ready" callback for UDP sockets
1132 * @sk: socket with data to read
1133 *
1134 */
1135static void xs_data_ready(struct sock *sk)
1136{
1137 struct rpc_xprt *xprt;
1138
1139 read_lock_bh(&sk->sk_callback_lock);
1140 dprintk("RPC: xs_data_ready...\n");
1141 xprt = xprt_from_sock(sk);
1142 if (xprt != NULL) {
1143 struct sock_xprt *transport = container_of(xprt,
1144 struct sock_xprt, xprt);
1145 transport->old_data_ready(sk);
1146 /* Any data means we had a useful conversation, so
1147 * then we don't need to delay the next reconnect
1148 */
1149 if (xprt->reestablish_timeout)
1150 xprt->reestablish_timeout = 0;
1151 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1152 queue_work(xprtiod_workqueue, &transport->recv_worker);
1153 }
1154 read_unlock_bh(&sk->sk_callback_lock);
1155}
1156
1157/*
1158 * Helper function to force a TCP close if the server is sending
1159 * junk and/or it has put us in CLOSE_WAIT
1160 */
1161static void xs_tcp_force_close(struct rpc_xprt *xprt)
1162{
1163 xprt_force_disconnect(xprt);
1164}
1165
1166static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1167{
1168 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1169 size_t len, used;
1170 char *p;
1171
1172 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1173 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1174 used = xdr_skb_read_bits(desc, p, len);
1175 transport->tcp_offset += used;
1176 if (used != len)
1177 return;
1178
1179 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1180 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1181 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1182 else
1183 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1184 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1185
1186 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1187 transport->tcp_offset = 0;
1188
1189 /* Sanity check of the record length */
1190 if (unlikely(transport->tcp_reclen < 8)) {
1191 dprintk("RPC: invalid TCP record fragment length\n");
1192 xs_tcp_force_close(xprt);
1193 return;
1194 }
1195 dprintk("RPC: reading TCP record fragment of length %d\n",
1196 transport->tcp_reclen);
1197}
1198
1199static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1200{
1201 if (transport->tcp_offset == transport->tcp_reclen) {
1202 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1203 transport->tcp_offset = 0;
1204 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1205 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1206 transport->tcp_flags |= TCP_RCV_COPY_XID;
1207 transport->tcp_copied = 0;
1208 }
1209 }
1210}
1211
1212static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1213{
1214 size_t len, used;
1215 char *p;
1216
1217 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1218 dprintk("RPC: reading XID (%zu bytes)\n", len);
1219 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1220 used = xdr_skb_read_bits(desc, p, len);
1221 transport->tcp_offset += used;
1222 if (used != len)
1223 return;
1224 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1225 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1226 transport->tcp_copied = 4;
1227 dprintk("RPC: reading %s XID %08x\n",
1228 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1229 : "request with",
1230 ntohl(transport->tcp_xid));
1231 xs_tcp_check_fraghdr(transport);
1232}
1233
1234static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1235 struct xdr_skb_reader *desc)
1236{
1237 size_t len, used;
1238 u32 offset;
1239 char *p;
1240
1241 /*
1242 * We want transport->tcp_offset to be 8 at the end of this routine
1243 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1244 * When this function is called for the first time,
1245 * transport->tcp_offset is 4 (after having already read the xid).
1246 */
1247 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1248 len = sizeof(transport->tcp_calldir) - offset;
1249 dprintk("RPC: reading CALL/REPLY flag (%zu bytes)\n", len);
1250 p = ((char *) &transport->tcp_calldir) + offset;
1251 used = xdr_skb_read_bits(desc, p, len);
1252 transport->tcp_offset += used;
1253 if (used != len)
1254 return;
1255 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1256 /*
1257 * We don't yet have the XDR buffer, so we will write the calldir
1258 * out after we get the buffer from the 'struct rpc_rqst'
1259 */
1260 switch (ntohl(transport->tcp_calldir)) {
1261 case RPC_REPLY:
1262 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1263 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1264 transport->tcp_flags |= TCP_RPC_REPLY;
1265 break;
1266 case RPC_CALL:
1267 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1268 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1269 transport->tcp_flags &= ~TCP_RPC_REPLY;
1270 break;
1271 default:
1272 dprintk("RPC: invalid request message type\n");
1273 xs_tcp_force_close(&transport->xprt);
1274 }
1275 xs_tcp_check_fraghdr(transport);
1276}
1277
1278static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1279 struct xdr_skb_reader *desc,
1280 struct rpc_rqst *req)
1281{
1282 struct sock_xprt *transport =
1283 container_of(xprt, struct sock_xprt, xprt);
1284 struct xdr_buf *rcvbuf;
1285 size_t len;
1286 ssize_t r;
1287
1288 rcvbuf = &req->rq_private_buf;
1289
1290 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1291 /*
1292 * Save the RPC direction in the XDR buffer
1293 */
1294 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1295 &transport->tcp_calldir,
1296 sizeof(transport->tcp_calldir));
1297 transport->tcp_copied += sizeof(transport->tcp_calldir);
1298 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1299 }
1300
1301 len = desc->count;
1302 if (len > transport->tcp_reclen - transport->tcp_offset)
1303 desc->count = transport->tcp_reclen - transport->tcp_offset;
1304 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1305 desc, xdr_skb_read_bits);
1306
1307 if (desc->count) {
1308 /* Error when copying to the receive buffer,
1309 * usually because we weren't able to allocate
1310 * additional buffer pages. All we can do now
1311 * is turn off TCP_RCV_COPY_DATA, so the request
1312 * will not receive any additional updates,
1313 * and time out.
1314 * Any remaining data from this record will
1315 * be discarded.
1316 */
1317 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1318 dprintk("RPC: XID %08x truncated request\n",
1319 ntohl(transport->tcp_xid));
1320 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1321 "tcp_offset = %u, tcp_reclen = %u\n",
1322 xprt, transport->tcp_copied,
1323 transport->tcp_offset, transport->tcp_reclen);
1324 return;
1325 }
1326
1327 transport->tcp_copied += r;
1328 transport->tcp_offset += r;
1329 desc->count = len - r;
1330
1331 dprintk("RPC: XID %08x read %zd bytes\n",
1332 ntohl(transport->tcp_xid), r);
1333 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1334 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1335 transport->tcp_offset, transport->tcp_reclen);
1336
1337 if (transport->tcp_copied == req->rq_private_buf.buflen)
1338 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1339 else if (transport->tcp_offset == transport->tcp_reclen) {
1340 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1341 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1342 }
1343}
1344
1345/*
1346 * Finds the request corresponding to the RPC xid and invokes the common
1347 * tcp read code to read the data.
1348 */
1349static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1350 struct xdr_skb_reader *desc)
1351{
1352 struct sock_xprt *transport =
1353 container_of(xprt, struct sock_xprt, xprt);
1354 struct rpc_rqst *req;
1355
1356 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1357
1358 /* Find and lock the request corresponding to this xid */
1359 spin_lock(&xprt->recv_lock);
1360 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1361 if (!req) {
1362 dprintk("RPC: XID %08x request not found!\n",
1363 ntohl(transport->tcp_xid));
1364 spin_unlock(&xprt->recv_lock);
1365 return -1;
1366 }
1367 xprt_pin_rqst(req);
1368 spin_unlock(&xprt->recv_lock);
1369
1370 xs_tcp_read_common(xprt, desc, req);
1371
1372 spin_lock(&xprt->recv_lock);
1373 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1374 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1375 xprt_unpin_rqst(req);
1376 spin_unlock(&xprt->recv_lock);
1377 return 0;
1378}
1379
1380#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1381/*
1382 * Obtains an rpc_rqst previously allocated and invokes the common
1383 * tcp read code to read the data. The result is placed in the callback
1384 * queue.
1385 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1386 * connection and return -1.
1387 */
1388static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1389 struct xdr_skb_reader *desc)
1390{
1391 struct sock_xprt *transport =
1392 container_of(xprt, struct sock_xprt, xprt);
1393 struct rpc_rqst *req;
1394
1395 /* Look up the request corresponding to the given XID */
1396 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1397 if (req == NULL) {
1398 printk(KERN_WARNING "Callback slot table overflowed\n");
1399 xprt_force_disconnect(xprt);
1400 return -1;
1401 }
1402
1403 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1404 xs_tcp_read_common(xprt, desc, req);
1405
1406 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1407 xprt_complete_bc_request(req, transport->tcp_copied);
1408
1409 return 0;
1410}
1411
1412static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1413 struct xdr_skb_reader *desc)
1414{
1415 struct sock_xprt *transport =
1416 container_of(xprt, struct sock_xprt, xprt);
1417
1418 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1419 xs_tcp_read_reply(xprt, desc) :
1420 xs_tcp_read_callback(xprt, desc);
1421}
1422
1423static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1424{
1425 int ret;
1426
1427 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1428 SVC_SOCK_ANONYMOUS);
1429 if (ret < 0)
1430 return ret;
1431 return 0;
1432}
1433
1434static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1435{
1436 return PAGE_SIZE;
1437}
1438#else
1439static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1440 struct xdr_skb_reader *desc)
1441{
1442 return xs_tcp_read_reply(xprt, desc);
1443}
1444#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1445
1446/*
1447 * Read data off the transport. This can be either an RPC_CALL or an
1448 * RPC_REPLY. Relay the processing to helper functions.
1449 */
1450static void xs_tcp_read_data(struct rpc_xprt *xprt,
1451 struct xdr_skb_reader *desc)
1452{
1453 struct sock_xprt *transport =
1454 container_of(xprt, struct sock_xprt, xprt);
1455
1456 if (_xs_tcp_read_data(xprt, desc) == 0)
1457 xs_tcp_check_fraghdr(transport);
1458 else {
1459 /*
1460 * The transport_lock protects the request handling.
1461 * There's no need to hold it to update the tcp_flags.
1462 */
1463 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1464 }
1465}
1466
1467static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1468{
1469 size_t len;
1470
1471 len = transport->tcp_reclen - transport->tcp_offset;
1472 if (len > desc->count)
1473 len = desc->count;
1474 desc->count -= len;
1475 desc->offset += len;
1476 transport->tcp_offset += len;
1477 dprintk("RPC: discarded %zu bytes\n", len);
1478 xs_tcp_check_fraghdr(transport);
1479}
1480
1481static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1482{
1483 struct rpc_xprt *xprt = rd_desc->arg.data;
1484 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1485 struct xdr_skb_reader desc = {
1486 .skb = skb,
1487 .offset = offset,
1488 .count = len,
1489 };
1490 size_t ret;
1491
1492 dprintk("RPC: xs_tcp_data_recv started\n");
1493 do {
1494 trace_xs_tcp_data_recv(transport);
1495 /* Read in a new fragment marker if necessary */
1496 /* Can we ever really expect to get completely empty fragments? */
1497 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1498 xs_tcp_read_fraghdr(xprt, &desc);
1499 continue;
1500 }
1501 /* Read in the xid if necessary */
1502 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1503 xs_tcp_read_xid(transport, &desc);
1504 continue;
1505 }
1506 /* Read in the call/reply flag */
1507 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1508 xs_tcp_read_calldir(transport, &desc);
1509 continue;
1510 }
1511 /* Read in the request data */
1512 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1513 xs_tcp_read_data(xprt, &desc);
1514 continue;
1515 }
1516 /* Skip over any trailing bytes on short reads */
1517 xs_tcp_read_discard(transport, &desc);
1518 } while (desc.count);
1519 ret = len - desc.count;
1520 if (ret < rd_desc->count)
1521 rd_desc->count -= ret;
1522 else
1523 rd_desc->count = 0;
1524 trace_xs_tcp_data_recv(transport);
1525 dprintk("RPC: xs_tcp_data_recv done\n");
1526 return ret;
1527}
1528
1529static void xs_tcp_data_receive(struct sock_xprt *transport)
1530{
1531 struct rpc_xprt *xprt = &transport->xprt;
1532 struct sock *sk;
1533 read_descriptor_t rd_desc = {
1534 .arg.data = xprt,
1535 };
1536 unsigned long total = 0;
1537 int read = 0;
1538
1539restart:
1540 mutex_lock(&transport->recv_mutex);
1541 sk = transport->inet;
1542 if (sk == NULL)
1543 goto out;
1544
1545 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1546 for (;;) {
1547 rd_desc.count = RPC_TCP_READ_CHUNK_SZ;
1548 lock_sock(sk);
1549 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1550 if (rd_desc.count != 0 || read < 0) {
1551 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1552 release_sock(sk);
1553 break;
1554 }
1555 release_sock(sk);
1556 total += read;
1557 if (need_resched()) {
1558 mutex_unlock(&transport->recv_mutex);
1559 cond_resched();
1560 goto restart;
1561 }
1562 }
1563 if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1564 queue_work(xprtiod_workqueue, &transport->recv_worker);
1565out:
1566 mutex_unlock(&transport->recv_mutex);
1567 trace_xs_tcp_data_ready(xprt, read, total);
1568}
1569
1570static void xs_tcp_data_receive_workfn(struct work_struct *work)
1571{
1572 struct sock_xprt *transport =
1573 container_of(work, struct sock_xprt, recv_worker);
1574 xs_tcp_data_receive(transport);
1575}
1576
1577/**
1578 * xs_tcp_state_change - callback to handle TCP socket state changes
1579 * @sk: socket whose state has changed
1580 *
1581 */
1582static void xs_tcp_state_change(struct sock *sk)
1583{
1584 struct rpc_xprt *xprt;
1585 struct sock_xprt *transport;
1586
1587 read_lock_bh(&sk->sk_callback_lock);
1588 if (!(xprt = xprt_from_sock(sk)))
1589 goto out;
1590 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1591 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1592 sk->sk_state, xprt_connected(xprt),
1593 sock_flag(sk, SOCK_DEAD),
1594 sock_flag(sk, SOCK_ZAPPED),
1595 sk->sk_shutdown);
1596
1597 transport = container_of(xprt, struct sock_xprt, xprt);
1598 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1599 switch (sk->sk_state) {
1600 case TCP_ESTABLISHED:
1601 spin_lock(&xprt->transport_lock);
1602 if (!xprt_test_and_set_connected(xprt)) {
1603
1604 /* Reset TCP record info */
1605 transport->tcp_offset = 0;
1606 transport->tcp_reclen = 0;
1607 transport->tcp_copied = 0;
1608 transport->tcp_flags =
1609 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1610 xprt->connect_cookie++;
1611 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1612 xprt_clear_connecting(xprt);
1613
1614 xprt_wake_pending_tasks(xprt, -EAGAIN);
1615 }
1616 spin_unlock(&xprt->transport_lock);
1617 break;
1618 case TCP_FIN_WAIT1:
1619 /* The client initiated a shutdown of the socket */
1620 xprt->connect_cookie++;
1621 xprt->reestablish_timeout = 0;
1622 set_bit(XPRT_CLOSING, &xprt->state);
1623 smp_mb__before_atomic();
1624 clear_bit(XPRT_CONNECTED, &xprt->state);
1625 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1626 smp_mb__after_atomic();
1627 break;
1628 case TCP_CLOSE_WAIT:
1629 /* The server initiated a shutdown of the socket */
1630 xprt->connect_cookie++;
1631 clear_bit(XPRT_CONNECTED, &xprt->state);
1632 xs_tcp_force_close(xprt);
1633 /* fall through */
1634 case TCP_CLOSING:
1635 /*
1636 * If the server closed down the connection, make sure that
1637 * we back off before reconnecting
1638 */
1639 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1640 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1641 break;
1642 case TCP_LAST_ACK:
1643 set_bit(XPRT_CLOSING, &xprt->state);
1644 smp_mb__before_atomic();
1645 clear_bit(XPRT_CONNECTED, &xprt->state);
1646 smp_mb__after_atomic();
1647 break;
1648 case TCP_CLOSE:
1649 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1650 &transport->sock_state))
1651 xprt_clear_connecting(xprt);
1652 clear_bit(XPRT_CLOSING, &xprt->state);
1653 if (sk->sk_err)
1654 xprt_wake_pending_tasks(xprt, -sk->sk_err);
1655 /* Trigger the socket release */
1656 xs_tcp_force_close(xprt);
1657 }
1658 out:
1659 read_unlock_bh(&sk->sk_callback_lock);
1660}
1661
1662static void xs_write_space(struct sock *sk)
1663{
1664 struct socket_wq *wq;
1665 struct rpc_xprt *xprt;
1666
1667 if (!sk->sk_socket)
1668 return;
1669 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1670
1671 if (unlikely(!(xprt = xprt_from_sock(sk))))
1672 return;
1673 rcu_read_lock();
1674 wq = rcu_dereference(sk->sk_wq);
1675 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1676 goto out;
1677
1678 xprt_write_space(xprt);
1679out:
1680 rcu_read_unlock();
1681}
1682
1683/**
1684 * xs_udp_write_space - callback invoked when socket buffer space
1685 * becomes available
1686 * @sk: socket whose state has changed
1687 *
1688 * Called when more output buffer space is available for this socket.
1689 * We try not to wake our writers until they can make "significant"
1690 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1691 * with a bunch of small requests.
1692 */
1693static void xs_udp_write_space(struct sock *sk)
1694{
1695 read_lock_bh(&sk->sk_callback_lock);
1696
1697 /* from net/core/sock.c:sock_def_write_space */
1698 if (sock_writeable(sk))
1699 xs_write_space(sk);
1700
1701 read_unlock_bh(&sk->sk_callback_lock);
1702}
1703
1704/**
1705 * xs_tcp_write_space - callback invoked when socket buffer space
1706 * becomes available
1707 * @sk: socket whose state has changed
1708 *
1709 * Called when more output buffer space is available for this socket.
1710 * We try not to wake our writers until they can make "significant"
1711 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1712 * with a bunch of small requests.
1713 */
1714static void xs_tcp_write_space(struct sock *sk)
1715{
1716 read_lock_bh(&sk->sk_callback_lock);
1717
1718 /* from net/core/stream.c:sk_stream_write_space */
1719 if (sk_stream_is_writeable(sk))
1720 xs_write_space(sk);
1721
1722 read_unlock_bh(&sk->sk_callback_lock);
1723}
1724
1725static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1726{
1727 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1728 struct sock *sk = transport->inet;
1729
1730 if (transport->rcvsize) {
1731 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1732 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1733 }
1734 if (transport->sndsize) {
1735 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1736 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1737 sk->sk_write_space(sk);
1738 }
1739}
1740
1741/**
1742 * xs_udp_set_buffer_size - set send and receive limits
1743 * @xprt: generic transport
1744 * @sndsize: requested size of send buffer, in bytes
1745 * @rcvsize: requested size of receive buffer, in bytes
1746 *
1747 * Set socket send and receive buffer size limits.
1748 */
1749static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1750{
1751 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1752
1753 transport->sndsize = 0;
1754 if (sndsize)
1755 transport->sndsize = sndsize + 1024;
1756 transport->rcvsize = 0;
1757 if (rcvsize)
1758 transport->rcvsize = rcvsize + 1024;
1759
1760 xs_udp_do_set_buffer_size(xprt);
1761}
1762
1763/**
1764 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1765 * @task: task that timed out
1766 *
1767 * Adjust the congestion window after a retransmit timeout has occurred.
1768 */
1769static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1770{
1771 spin_lock_bh(&xprt->transport_lock);
1772 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1773 spin_unlock_bh(&xprt->transport_lock);
1774}
1775
1776static unsigned short xs_get_random_port(void)
1777{
1778 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1779 unsigned short rand = (unsigned short) prandom_u32() % range;
1780 return rand + xprt_min_resvport;
1781}
1782
1783/**
1784 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1785 * @sock: socket
1786 *
1787 * Note that this function has to be called on all sockets that share the
1788 * same port, and it must be called before binding.
1789 */
1790static void xs_sock_set_reuseport(struct socket *sock)
1791{
1792 int opt = 1;
1793
1794 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1795 (char *)&opt, sizeof(opt));
1796}
1797
1798static unsigned short xs_sock_getport(struct socket *sock)
1799{
1800 struct sockaddr_storage buf;
1801 unsigned short port = 0;
1802
1803 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1804 goto out;
1805 switch (buf.ss_family) {
1806 case AF_INET6:
1807 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1808 break;
1809 case AF_INET:
1810 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1811 }
1812out:
1813 return port;
1814}
1815
1816/**
1817 * xs_set_port - reset the port number in the remote endpoint address
1818 * @xprt: generic transport
1819 * @port: new port number
1820 *
1821 */
1822static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1823{
1824 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1825
1826 rpc_set_port(xs_addr(xprt), port);
1827 xs_update_peer_port(xprt);
1828}
1829
1830static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1831{
1832 if (transport->srcport == 0)
1833 transport->srcport = xs_sock_getport(sock);
1834}
1835
1836static unsigned short xs_get_srcport(struct sock_xprt *transport)
1837{
1838 unsigned short port = transport->srcport;
1839
1840 if (port == 0 && transport->xprt.resvport)
1841 port = xs_get_random_port();
1842 return port;
1843}
1844
1845static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1846{
1847 if (transport->srcport != 0)
1848 transport->srcport = 0;
1849 if (!transport->xprt.resvport)
1850 return 0;
1851 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1852 return xprt_max_resvport;
1853 return --port;
1854}
1855static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1856{
1857 struct sockaddr_storage myaddr;
1858 int err, nloop = 0;
1859 unsigned short port = xs_get_srcport(transport);
1860 unsigned short last;
1861
1862 /*
1863 * If we are asking for any ephemeral port (i.e. port == 0 &&
1864 * transport->xprt.resvport == 0), don't bind. Let the local
1865 * port selection happen implicitly when the socket is used
1866 * (for example at connect time).
1867 *
1868 * This ensures that we can continue to establish TCP
1869 * connections even when all local ephemeral ports are already
1870 * a part of some TCP connection. This makes no difference
1871 * for UDP sockets, but also doens't harm them.
1872 *
1873 * If we're asking for any reserved port (i.e. port == 0 &&
1874 * transport->xprt.resvport == 1) xs_get_srcport above will
1875 * ensure that port is non-zero and we will bind as needed.
1876 */
1877 if (port == 0)
1878 return 0;
1879
1880 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1881 do {
1882 rpc_set_port((struct sockaddr *)&myaddr, port);
1883 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1884 transport->xprt.addrlen);
1885 if (err == 0) {
1886 transport->srcport = port;
1887 break;
1888 }
1889 last = port;
1890 port = xs_next_srcport(transport, port);
1891 if (port > last)
1892 nloop++;
1893 } while (err == -EADDRINUSE && nloop != 2);
1894
1895 if (myaddr.ss_family == AF_INET)
1896 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1897 &((struct sockaddr_in *)&myaddr)->sin_addr,
1898 port, err ? "failed" : "ok", err);
1899 else
1900 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1901 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1902 port, err ? "failed" : "ok", err);
1903 return err;
1904}
1905
1906/*
1907 * We don't support autobind on AF_LOCAL sockets
1908 */
1909static void xs_local_rpcbind(struct rpc_task *task)
1910{
1911 xprt_set_bound(task->tk_xprt);
1912}
1913
1914static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1915{
1916}
1917
1918#ifdef CONFIG_DEBUG_LOCK_ALLOC
1919static struct lock_class_key xs_key[2];
1920static struct lock_class_key xs_slock_key[2];
1921
1922static inline void xs_reclassify_socketu(struct socket *sock)
1923{
1924 struct sock *sk = sock->sk;
1925
1926 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1927 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1928}
1929
1930static inline void xs_reclassify_socket4(struct socket *sock)
1931{
1932 struct sock *sk = sock->sk;
1933
1934 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1935 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1936}
1937
1938static inline void xs_reclassify_socket6(struct socket *sock)
1939{
1940 struct sock *sk = sock->sk;
1941
1942 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1943 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1944}
1945
1946static inline void xs_reclassify_socket(int family, struct socket *sock)
1947{
1948 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1949 return;
1950
1951 switch (family) {
1952 case AF_LOCAL:
1953 xs_reclassify_socketu(sock);
1954 break;
1955 case AF_INET:
1956 xs_reclassify_socket4(sock);
1957 break;
1958 case AF_INET6:
1959 xs_reclassify_socket6(sock);
1960 break;
1961 }
1962}
1963#else
1964static inline void xs_reclassify_socket(int family, struct socket *sock)
1965{
1966}
1967#endif
1968
1969static void xs_dummy_setup_socket(struct work_struct *work)
1970{
1971}
1972
1973static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1974 struct sock_xprt *transport, int family, int type,
1975 int protocol, bool reuseport)
1976{
1977 struct socket *sock;
1978 int err;
1979
1980 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1981 if (err < 0) {
1982 dprintk("RPC: can't create %d transport socket (%d).\n",
1983 protocol, -err);
1984 goto out;
1985 }
1986 xs_reclassify_socket(family, sock);
1987
1988 if (reuseport)
1989 xs_sock_set_reuseport(sock);
1990
1991 err = xs_bind(transport, sock);
1992 if (err) {
1993 sock_release(sock);
1994 goto out;
1995 }
1996
1997 return sock;
1998out:
1999 return ERR_PTR(err);
2000}
2001
2002static int xs_local_finish_connecting(struct rpc_xprt *xprt,
2003 struct socket *sock)
2004{
2005 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2006 xprt);
2007
2008 if (!transport->inet) {
2009 struct sock *sk = sock->sk;
2010
2011 write_lock_bh(&sk->sk_callback_lock);
2012
2013 xs_save_old_callbacks(transport, sk);
2014
2015 sk->sk_user_data = xprt;
2016 sk->sk_data_ready = xs_data_ready;
2017 sk->sk_write_space = xs_udp_write_space;
2018 sock_set_flag(sk, SOCK_FASYNC);
2019 sk->sk_error_report = xs_error_report;
2020 sk->sk_allocation = GFP_NOIO;
2021
2022 xprt_clear_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 write_unlock_bh(&sk->sk_callback_lock);
2029 }
2030
2031 /* Tell the socket layer to start connecting... */
2032 xprt->stat.connect_count++;
2033 xprt->stat.connect_start = jiffies;
2034 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2035}
2036
2037/**
2038 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2039 * @transport: socket transport to connect
2040 */
2041static int xs_local_setup_socket(struct sock_xprt *transport)
2042{
2043 struct rpc_xprt *xprt = &transport->xprt;
2044 struct socket *sock;
2045 int status = -EIO;
2046
2047 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2048 SOCK_STREAM, 0, &sock, 1);
2049 if (status < 0) {
2050 dprintk("RPC: can't create AF_LOCAL "
2051 "transport socket (%d).\n", -status);
2052 goto out;
2053 }
2054 xs_reclassify_socket(AF_LOCAL, sock);
2055
2056 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2057 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2058
2059 status = xs_local_finish_connecting(xprt, sock);
2060 trace_rpc_socket_connect(xprt, sock, status);
2061 switch (status) {
2062 case 0:
2063 dprintk("RPC: xprt %p connected to %s\n",
2064 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2065 xprt_set_connected(xprt);
2066 case -ENOBUFS:
2067 break;
2068 case -ENOENT:
2069 dprintk("RPC: xprt %p: socket %s does not exist\n",
2070 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2071 break;
2072 case -ECONNREFUSED:
2073 dprintk("RPC: xprt %p: connection refused for %s\n",
2074 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2075 break;
2076 default:
2077 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2078 __func__, -status,
2079 xprt->address_strings[RPC_DISPLAY_ADDR]);
2080 }
2081
2082out:
2083 xprt_clear_connecting(xprt);
2084 xprt_wake_pending_tasks(xprt, status);
2085 return status;
2086}
2087
2088static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2089{
2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2091 int ret;
2092
2093 if (RPC_IS_ASYNC(task)) {
2094 /*
2095 * We want the AF_LOCAL connect to be resolved in the
2096 * filesystem namespace of the process making the rpc
2097 * call. Thus we connect synchronously.
2098 *
2099 * If we want to support asynchronous AF_LOCAL calls,
2100 * we'll need to figure out how to pass a namespace to
2101 * connect.
2102 */
2103 rpc_exit(task, -ENOTCONN);
2104 return;
2105 }
2106 ret = xs_local_setup_socket(transport);
2107 if (ret && !RPC_IS_SOFTCONN(task))
2108 msleep_interruptible(15000);
2109}
2110
2111#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2112/*
2113 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2114 * know that we have exclusive access to the socket), to guard against
2115 * races with xs_reset_transport.
2116 */
2117static void xs_set_memalloc(struct rpc_xprt *xprt)
2118{
2119 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2120 xprt);
2121
2122 /*
2123 * If there's no sock, then we have nothing to set. The
2124 * reconnecting process will get it for us.
2125 */
2126 if (!transport->inet)
2127 return;
2128 if (atomic_read(&xprt->swapper))
2129 sk_set_memalloc(transport->inet);
2130}
2131
2132/**
2133 * xs_enable_swap - Tag this transport as being used for swap.
2134 * @xprt: transport to tag
2135 *
2136 * Take a reference to this transport on behalf of the rpc_clnt, and
2137 * optionally mark it for swapping if it wasn't already.
2138 */
2139static int
2140xs_enable_swap(struct rpc_xprt *xprt)
2141{
2142 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2143
2144 if (atomic_inc_return(&xprt->swapper) != 1)
2145 return 0;
2146 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2147 return -ERESTARTSYS;
2148 if (xs->inet)
2149 sk_set_memalloc(xs->inet);
2150 xprt_release_xprt(xprt, NULL);
2151 return 0;
2152}
2153
2154/**
2155 * xs_disable_swap - Untag this transport as being used for swap.
2156 * @xprt: transport to tag
2157 *
2158 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2159 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2160 */
2161static void
2162xs_disable_swap(struct rpc_xprt *xprt)
2163{
2164 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2165
2166 if (!atomic_dec_and_test(&xprt->swapper))
2167 return;
2168 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2169 return;
2170 if (xs->inet)
2171 sk_clear_memalloc(xs->inet);
2172 xprt_release_xprt(xprt, NULL);
2173}
2174#else
2175static void xs_set_memalloc(struct rpc_xprt *xprt)
2176{
2177}
2178
2179static int
2180xs_enable_swap(struct rpc_xprt *xprt)
2181{
2182 return -EINVAL;
2183}
2184
2185static void
2186xs_disable_swap(struct rpc_xprt *xprt)
2187{
2188}
2189#endif
2190
2191static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2192{
2193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2194
2195 if (!transport->inet) {
2196 struct sock *sk = sock->sk;
2197
2198 write_lock_bh(&sk->sk_callback_lock);
2199
2200 xs_save_old_callbacks(transport, sk);
2201
2202 sk->sk_user_data = xprt;
2203 sk->sk_data_ready = xs_data_ready;
2204 sk->sk_write_space = xs_udp_write_space;
2205 sock_set_flag(sk, SOCK_FASYNC);
2206 sk->sk_allocation = GFP_NOIO;
2207
2208 xprt_set_connected(xprt);
2209
2210 /* Reset to new socket */
2211 transport->sock = sock;
2212 transport->inet = sk;
2213
2214 xs_set_memalloc(xprt);
2215
2216 write_unlock_bh(&sk->sk_callback_lock);
2217 }
2218 xs_udp_do_set_buffer_size(xprt);
2219
2220 xprt->stat.connect_start = jiffies;
2221}
2222
2223static void xs_udp_setup_socket(struct work_struct *work)
2224{
2225 struct sock_xprt *transport =
2226 container_of(work, struct sock_xprt, connect_worker.work);
2227 struct rpc_xprt *xprt = &transport->xprt;
2228 struct socket *sock;
2229 int status = -EIO;
2230
2231 sock = xs_create_sock(xprt, transport,
2232 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2233 IPPROTO_UDP, false);
2234 if (IS_ERR(sock))
2235 goto out;
2236
2237 dprintk("RPC: worker connecting xprt %p via %s to "
2238 "%s (port %s)\n", xprt,
2239 xprt->address_strings[RPC_DISPLAY_PROTO],
2240 xprt->address_strings[RPC_DISPLAY_ADDR],
2241 xprt->address_strings[RPC_DISPLAY_PORT]);
2242
2243 xs_udp_finish_connecting(xprt, sock);
2244 trace_rpc_socket_connect(xprt, sock, 0);
2245 status = 0;
2246out:
2247 xprt_unlock_connect(xprt, transport);
2248 xprt_clear_connecting(xprt);
2249 xprt_wake_pending_tasks(xprt, status);
2250}
2251
2252/**
2253 * xs_tcp_shutdown - gracefully shut down a TCP socket
2254 * @xprt: transport
2255 *
2256 * Initiates a graceful shutdown of the TCP socket by calling the
2257 * equivalent of shutdown(SHUT_RDWR);
2258 */
2259static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2260{
2261 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2262 struct socket *sock = transport->sock;
2263 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2264
2265 if (sock == NULL)
2266 return;
2267 switch (skst) {
2268 default:
2269 kernel_sock_shutdown(sock, SHUT_RDWR);
2270 trace_rpc_socket_shutdown(xprt, sock);
2271 break;
2272 case TCP_CLOSE:
2273 case TCP_TIME_WAIT:
2274 xs_reset_transport(transport);
2275 }
2276}
2277
2278static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2279 struct socket *sock)
2280{
2281 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2282 unsigned int keepidle;
2283 unsigned int keepcnt;
2284 unsigned int opt_on = 1;
2285 unsigned int timeo;
2286
2287 spin_lock_bh(&xprt->transport_lock);
2288 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2289 keepcnt = xprt->timeout->to_retries + 1;
2290 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2291 (xprt->timeout->to_retries + 1);
2292 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2293 spin_unlock_bh(&xprt->transport_lock);
2294
2295 /* TCP Keepalive options */
2296 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2297 (char *)&opt_on, sizeof(opt_on));
2298 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2299 (char *)&keepidle, sizeof(keepidle));
2300 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2301 (char *)&keepidle, sizeof(keepidle));
2302 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2303 (char *)&keepcnt, sizeof(keepcnt));
2304
2305 /* TCP user timeout (see RFC5482) */
2306 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2307 (char *)&timeo, sizeof(timeo));
2308}
2309
2310static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2311 unsigned long connect_timeout,
2312 unsigned long reconnect_timeout)
2313{
2314 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2315 struct rpc_timeout to;
2316 unsigned long initval;
2317
2318 spin_lock_bh(&xprt->transport_lock);
2319 if (reconnect_timeout < xprt->max_reconnect_timeout)
2320 xprt->max_reconnect_timeout = reconnect_timeout;
2321 if (connect_timeout < xprt->connect_timeout) {
2322 memcpy(&to, xprt->timeout, sizeof(to));
2323 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2324 /* Arbitrary lower limit */
2325 if (initval < XS_TCP_INIT_REEST_TO << 1)
2326 initval = XS_TCP_INIT_REEST_TO << 1;
2327 to.to_initval = initval;
2328 to.to_maxval = initval;
2329 memcpy(&transport->tcp_timeout, &to,
2330 sizeof(transport->tcp_timeout));
2331 xprt->timeout = &transport->tcp_timeout;
2332 xprt->connect_timeout = connect_timeout;
2333 }
2334 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2335 spin_unlock_bh(&xprt->transport_lock);
2336}
2337
2338static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2339{
2340 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2341 int ret = -ENOTCONN;
2342
2343 if (!transport->inet) {
2344 struct sock *sk = sock->sk;
2345 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2346
2347 /* Avoid temporary address, they are bad for long-lived
2348 * connections such as NFS mounts.
2349 * RFC4941, section 3.6 suggests that:
2350 * Individual applications, which have specific
2351 * knowledge about the normal duration of connections,
2352 * MAY override this as appropriate.
2353 */
2354 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2355 (char *)&addr_pref, sizeof(addr_pref));
2356
2357 xs_tcp_set_socket_timeouts(xprt, sock);
2358
2359 write_lock_bh(&sk->sk_callback_lock);
2360
2361 xs_save_old_callbacks(transport, sk);
2362
2363 sk->sk_user_data = xprt;
2364 sk->sk_data_ready = xs_data_ready;
2365 sk->sk_state_change = xs_tcp_state_change;
2366 sk->sk_write_space = xs_tcp_write_space;
2367 sock_set_flag(sk, SOCK_FASYNC);
2368 sk->sk_error_report = xs_error_report;
2369 sk->sk_allocation = GFP_NOIO;
2370
2371 /* socket options */
2372 sock_reset_flag(sk, SOCK_LINGER);
2373 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2374
2375 xprt_clear_connected(xprt);
2376
2377 /* Reset to new socket */
2378 transport->sock = sock;
2379 transport->inet = sk;
2380
2381 write_unlock_bh(&sk->sk_callback_lock);
2382 }
2383
2384 if (!xprt_bound(xprt))
2385 goto out;
2386
2387 xs_set_memalloc(xprt);
2388
2389 /* Tell the socket layer to start connecting... */
2390 xprt->stat.connect_count++;
2391 xprt->stat.connect_start = jiffies;
2392 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2393 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2394 switch (ret) {
2395 case 0:
2396 xs_set_srcport(transport, sock);
2397 /* fall through */
2398 case -EINPROGRESS:
2399 /* SYN_SENT! */
2400 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2401 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2402 break;
2403 case -EADDRNOTAVAIL:
2404 /* Source port number is unavailable. Try a new one! */
2405 transport->srcport = 0;
2406 }
2407out:
2408 return ret;
2409}
2410
2411/**
2412 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2413 *
2414 * Invoked by a work queue tasklet.
2415 */
2416static void xs_tcp_setup_socket(struct work_struct *work)
2417{
2418 struct sock_xprt *transport =
2419 container_of(work, struct sock_xprt, connect_worker.work);
2420 struct socket *sock = transport->sock;
2421 struct rpc_xprt *xprt = &transport->xprt;
2422 int status = -EIO;
2423
2424 if (!sock) {
2425 sock = xs_create_sock(xprt, transport,
2426 xs_addr(xprt)->sa_family, SOCK_STREAM,
2427 IPPROTO_TCP, true);
2428 if (IS_ERR(sock)) {
2429 status = PTR_ERR(sock);
2430 goto out;
2431 }
2432 }
2433
2434 dprintk("RPC: worker connecting xprt %p via %s to "
2435 "%s (port %s)\n", xprt,
2436 xprt->address_strings[RPC_DISPLAY_PROTO],
2437 xprt->address_strings[RPC_DISPLAY_ADDR],
2438 xprt->address_strings[RPC_DISPLAY_PORT]);
2439
2440 status = xs_tcp_finish_connecting(xprt, sock);
2441 trace_rpc_socket_connect(xprt, sock, status);
2442 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2443 xprt, -status, xprt_connected(xprt),
2444 sock->sk->sk_state);
2445 switch (status) {
2446 default:
2447 printk("%s: connect returned unhandled error %d\n",
2448 __func__, status);
2449 /* fall through */
2450 case -EADDRNOTAVAIL:
2451 /* We're probably in TIME_WAIT. Get rid of existing socket,
2452 * and retry
2453 */
2454 xs_tcp_force_close(xprt);
2455 break;
2456 case 0:
2457 case -EINPROGRESS:
2458 case -EALREADY:
2459 xprt_unlock_connect(xprt, transport);
2460 return;
2461 case -EINVAL:
2462 /* Happens, for instance, if the user specified a link
2463 * local IPv6 address without a scope-id.
2464 */
2465 case -ECONNREFUSED:
2466 case -ECONNRESET:
2467 case -ENETDOWN:
2468 case -ENETUNREACH:
2469 case -EHOSTUNREACH:
2470 case -EADDRINUSE:
2471 case -ENOBUFS:
2472 /*
2473 * xs_tcp_force_close() wakes tasks with -EIO.
2474 * We need to wake them first to ensure the
2475 * correct error code.
2476 */
2477 xprt_wake_pending_tasks(xprt, status);
2478 xs_tcp_force_close(xprt);
2479 goto out;
2480 }
2481 status = -EAGAIN;
2482out:
2483 xprt_unlock_connect(xprt, transport);
2484 xprt_clear_connecting(xprt);
2485 xprt_wake_pending_tasks(xprt, status);
2486}
2487
2488static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2489{
2490 unsigned long start, now = jiffies;
2491
2492 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2493 if (time_after(start, now))
2494 return start - now;
2495 return 0;
2496}
2497
2498static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2499{
2500 xprt->reestablish_timeout <<= 1;
2501 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2502 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2503 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2504 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2505}
2506
2507/**
2508 * xs_connect - connect a socket to a remote endpoint
2509 * @xprt: pointer to transport structure
2510 * @task: address of RPC task that manages state of connect request
2511 *
2512 * TCP: If the remote end dropped the connection, delay reconnecting.
2513 *
2514 * UDP socket connects are synchronous, but we use a work queue anyway
2515 * to guarantee that even unprivileged user processes can set up a
2516 * socket on a privileged port.
2517 *
2518 * If a UDP socket connect fails, the delay behavior here prevents
2519 * retry floods (hard mounts).
2520 */
2521static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2522{
2523 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2524 unsigned long delay = 0;
2525
2526 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2527
2528 if (transport->sock != NULL) {
2529 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2530 "seconds\n",
2531 xprt, xprt->reestablish_timeout / HZ);
2532
2533 /* Start by resetting any existing state */
2534 xs_reset_transport(transport);
2535
2536 delay = xs_reconnect_delay(xprt);
2537 xs_reconnect_backoff(xprt);
2538
2539 } else
2540 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2541
2542 queue_delayed_work(xprtiod_workqueue,
2543 &transport->connect_worker,
2544 delay);
2545}
2546
2547/**
2548 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2549 * @xprt: rpc_xprt struct containing statistics
2550 * @seq: output file
2551 *
2552 */
2553static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2554{
2555 long idle_time = 0;
2556
2557 if (xprt_connected(xprt))
2558 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2559
2560 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2561 "%llu %llu %lu %llu %llu\n",
2562 xprt->stat.bind_count,
2563 xprt->stat.connect_count,
2564 xprt->stat.connect_time,
2565 idle_time,
2566 xprt->stat.sends,
2567 xprt->stat.recvs,
2568 xprt->stat.bad_xids,
2569 xprt->stat.req_u,
2570 xprt->stat.bklog_u,
2571 xprt->stat.max_slots,
2572 xprt->stat.sending_u,
2573 xprt->stat.pending_u);
2574}
2575
2576/**
2577 * xs_udp_print_stats - display UDP socket-specifc stats
2578 * @xprt: rpc_xprt struct containing statistics
2579 * @seq: output file
2580 *
2581 */
2582static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2583{
2584 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2585
2586 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2587 "%lu %llu %llu\n",
2588 transport->srcport,
2589 xprt->stat.bind_count,
2590 xprt->stat.sends,
2591 xprt->stat.recvs,
2592 xprt->stat.bad_xids,
2593 xprt->stat.req_u,
2594 xprt->stat.bklog_u,
2595 xprt->stat.max_slots,
2596 xprt->stat.sending_u,
2597 xprt->stat.pending_u);
2598}
2599
2600/**
2601 * xs_tcp_print_stats - display TCP socket-specifc stats
2602 * @xprt: rpc_xprt struct containing statistics
2603 * @seq: output file
2604 *
2605 */
2606static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2607{
2608 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2609 long idle_time = 0;
2610
2611 if (xprt_connected(xprt))
2612 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2613
2614 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2615 "%llu %llu %lu %llu %llu\n",
2616 transport->srcport,
2617 xprt->stat.bind_count,
2618 xprt->stat.connect_count,
2619 xprt->stat.connect_time,
2620 idle_time,
2621 xprt->stat.sends,
2622 xprt->stat.recvs,
2623 xprt->stat.bad_xids,
2624 xprt->stat.req_u,
2625 xprt->stat.bklog_u,
2626 xprt->stat.max_slots,
2627 xprt->stat.sending_u,
2628 xprt->stat.pending_u);
2629}
2630
2631/*
2632 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2633 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2634 * to use the server side send routines.
2635 */
2636static int bc_malloc(struct rpc_task *task)
2637{
2638 struct rpc_rqst *rqst = task->tk_rqstp;
2639 size_t size = rqst->rq_callsize;
2640 struct page *page;
2641 struct rpc_buffer *buf;
2642
2643 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2644 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2645 size);
2646 return -EINVAL;
2647 }
2648
2649 page = alloc_page(GFP_KERNEL);
2650 if (!page)
2651 return -ENOMEM;
2652
2653 buf = page_address(page);
2654 buf->len = PAGE_SIZE;
2655
2656 rqst->rq_buffer = buf->data;
2657 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2658 return 0;
2659}
2660
2661/*
2662 * Free the space allocated in the bc_alloc routine
2663 */
2664static void bc_free(struct rpc_task *task)
2665{
2666 void *buffer = task->tk_rqstp->rq_buffer;
2667 struct rpc_buffer *buf;
2668
2669 buf = container_of(buffer, struct rpc_buffer, data);
2670 free_page((unsigned long)buf);
2671}
2672
2673/*
2674 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2675 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2676 */
2677static int bc_sendto(struct rpc_rqst *req)
2678{
2679 int len;
2680 struct xdr_buf *xbufp = &req->rq_snd_buf;
2681 struct rpc_xprt *xprt = req->rq_xprt;
2682 struct sock_xprt *transport =
2683 container_of(xprt, struct sock_xprt, xprt);
2684 struct socket *sock = transport->sock;
2685 unsigned long headoff;
2686 unsigned long tailoff;
2687
2688 xs_encode_stream_record_marker(xbufp);
2689
2690 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2691 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2692 len = svc_send_common(sock, xbufp,
2693 virt_to_page(xbufp->head[0].iov_base), headoff,
2694 xbufp->tail[0].iov_base, tailoff);
2695
2696 if (len != xbufp->len) {
2697 printk(KERN_NOTICE "Error sending entire callback!\n");
2698 len = -EAGAIN;
2699 }
2700
2701 return len;
2702}
2703
2704/*
2705 * The send routine. Borrows from svc_send
2706 */
2707static int bc_send_request(struct rpc_task *task)
2708{
2709 struct rpc_rqst *req = task->tk_rqstp;
2710 struct svc_xprt *xprt;
2711 int len;
2712
2713 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2714 /*
2715 * Get the server socket associated with this callback xprt
2716 */
2717 xprt = req->rq_xprt->bc_xprt;
2718
2719 /*
2720 * Grab the mutex to serialize data as the connection is shared
2721 * with the fore channel
2722 */
2723 if (!mutex_trylock(&xprt->xpt_mutex)) {
2724 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2725 if (!mutex_trylock(&xprt->xpt_mutex))
2726 return -EAGAIN;
2727 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2728 }
2729 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2730 len = -ENOTCONN;
2731 else
2732 len = bc_sendto(req);
2733 mutex_unlock(&xprt->xpt_mutex);
2734
2735 if (len > 0)
2736 len = 0;
2737
2738 return len;
2739}
2740
2741/*
2742 * The close routine. Since this is client initiated, we do nothing
2743 */
2744
2745static void bc_close(struct rpc_xprt *xprt)
2746{
2747}
2748
2749/*
2750 * The xprt destroy routine. Again, because this connection is client
2751 * initiated, we do nothing
2752 */
2753
2754static void bc_destroy(struct rpc_xprt *xprt)
2755{
2756 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2757
2758 xs_xprt_free(xprt);
2759 module_put(THIS_MODULE);
2760}
2761
2762static const struct rpc_xprt_ops xs_local_ops = {
2763 .reserve_xprt = xprt_reserve_xprt,
2764 .release_xprt = xs_tcp_release_xprt,
2765 .alloc_slot = xprt_alloc_slot,
2766 .rpcbind = xs_local_rpcbind,
2767 .set_port = xs_local_set_port,
2768 .connect = xs_local_connect,
2769 .buf_alloc = rpc_malloc,
2770 .buf_free = rpc_free,
2771 .send_request = xs_local_send_request,
2772 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2773 .close = xs_close,
2774 .destroy = xs_destroy,
2775 .print_stats = xs_local_print_stats,
2776 .enable_swap = xs_enable_swap,
2777 .disable_swap = xs_disable_swap,
2778};
2779
2780static const struct rpc_xprt_ops xs_udp_ops = {
2781 .set_buffer_size = xs_udp_set_buffer_size,
2782 .reserve_xprt = xprt_reserve_xprt_cong,
2783 .release_xprt = xprt_release_xprt_cong,
2784 .alloc_slot = xprt_alloc_slot,
2785 .rpcbind = rpcb_getport_async,
2786 .set_port = xs_set_port,
2787 .connect = xs_connect,
2788 .buf_alloc = rpc_malloc,
2789 .buf_free = rpc_free,
2790 .send_request = xs_udp_send_request,
2791 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2792 .timer = xs_udp_timer,
2793 .release_request = xprt_release_rqst_cong,
2794 .close = xs_close,
2795 .destroy = xs_destroy,
2796 .print_stats = xs_udp_print_stats,
2797 .enable_swap = xs_enable_swap,
2798 .disable_swap = xs_disable_swap,
2799 .inject_disconnect = xs_inject_disconnect,
2800};
2801
2802static const struct rpc_xprt_ops xs_tcp_ops = {
2803 .reserve_xprt = xprt_reserve_xprt,
2804 .release_xprt = xs_tcp_release_xprt,
2805 .alloc_slot = xprt_lock_and_alloc_slot,
2806 .rpcbind = rpcb_getport_async,
2807 .set_port = xs_set_port,
2808 .connect = xs_connect,
2809 .buf_alloc = rpc_malloc,
2810 .buf_free = rpc_free,
2811 .send_request = xs_tcp_send_request,
2812 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2813 .close = xs_tcp_shutdown,
2814 .destroy = xs_destroy,
2815 .set_connect_timeout = xs_tcp_set_connect_timeout,
2816 .print_stats = xs_tcp_print_stats,
2817 .enable_swap = xs_enable_swap,
2818 .disable_swap = xs_disable_swap,
2819 .inject_disconnect = xs_inject_disconnect,
2820#ifdef CONFIG_SUNRPC_BACKCHANNEL
2821 .bc_setup = xprt_setup_bc,
2822 .bc_up = xs_tcp_bc_up,
2823 .bc_maxpayload = xs_tcp_bc_maxpayload,
2824 .bc_free_rqst = xprt_free_bc_rqst,
2825 .bc_destroy = xprt_destroy_bc,
2826#endif
2827};
2828
2829/*
2830 * The rpc_xprt_ops for the server backchannel
2831 */
2832
2833static const struct rpc_xprt_ops bc_tcp_ops = {
2834 .reserve_xprt = xprt_reserve_xprt,
2835 .release_xprt = xprt_release_xprt,
2836 .alloc_slot = xprt_alloc_slot,
2837 .buf_alloc = bc_malloc,
2838 .buf_free = bc_free,
2839 .send_request = bc_send_request,
2840 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2841 .close = bc_close,
2842 .destroy = bc_destroy,
2843 .print_stats = xs_tcp_print_stats,
2844 .enable_swap = xs_enable_swap,
2845 .disable_swap = xs_disable_swap,
2846 .inject_disconnect = xs_inject_disconnect,
2847};
2848
2849static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2850{
2851 static const struct sockaddr_in sin = {
2852 .sin_family = AF_INET,
2853 .sin_addr.s_addr = htonl(INADDR_ANY),
2854 };
2855 static const struct sockaddr_in6 sin6 = {
2856 .sin6_family = AF_INET6,
2857 .sin6_addr = IN6ADDR_ANY_INIT,
2858 };
2859
2860 switch (family) {
2861 case AF_LOCAL:
2862 break;
2863 case AF_INET:
2864 memcpy(sap, &sin, sizeof(sin));
2865 break;
2866 case AF_INET6:
2867 memcpy(sap, &sin6, sizeof(sin6));
2868 break;
2869 default:
2870 dprintk("RPC: %s: Bad address family\n", __func__);
2871 return -EAFNOSUPPORT;
2872 }
2873 return 0;
2874}
2875
2876static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2877 unsigned int slot_table_size,
2878 unsigned int max_slot_table_size)
2879{
2880 struct rpc_xprt *xprt;
2881 struct sock_xprt *new;
2882
2883 if (args->addrlen > sizeof(xprt->addr)) {
2884 dprintk("RPC: xs_setup_xprt: address too large\n");
2885 return ERR_PTR(-EBADF);
2886 }
2887
2888 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2889 max_slot_table_size);
2890 if (xprt == NULL) {
2891 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2892 "rpc_xprt\n");
2893 return ERR_PTR(-ENOMEM);
2894 }
2895
2896 new = container_of(xprt, struct sock_xprt, xprt);
2897 mutex_init(&new->recv_mutex);
2898 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2899 xprt->addrlen = args->addrlen;
2900 if (args->srcaddr)
2901 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2902 else {
2903 int err;
2904 err = xs_init_anyaddr(args->dstaddr->sa_family,
2905 (struct sockaddr *)&new->srcaddr);
2906 if (err != 0) {
2907 xprt_free(xprt);
2908 return ERR_PTR(err);
2909 }
2910 }
2911
2912 return xprt;
2913}
2914
2915static const struct rpc_timeout xs_local_default_timeout = {
2916 .to_initval = 10 * HZ,
2917 .to_maxval = 10 * HZ,
2918 .to_retries = 2,
2919};
2920
2921/**
2922 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2923 * @args: rpc transport creation arguments
2924 *
2925 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2926 */
2927static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2928{
2929 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2930 struct sock_xprt *transport;
2931 struct rpc_xprt *xprt;
2932 struct rpc_xprt *ret;
2933
2934 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2935 xprt_max_tcp_slot_table_entries);
2936 if (IS_ERR(xprt))
2937 return xprt;
2938 transport = container_of(xprt, struct sock_xprt, xprt);
2939
2940 xprt->prot = 0;
2941 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2942 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2943
2944 xprt->bind_timeout = XS_BIND_TO;
2945 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2946 xprt->idle_timeout = XS_IDLE_DISC_TO;
2947
2948 xprt->ops = &xs_local_ops;
2949 xprt->timeout = &xs_local_default_timeout;
2950
2951 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2952 INIT_DELAYED_WORK(&transport->connect_worker,
2953 xs_dummy_setup_socket);
2954
2955 switch (sun->sun_family) {
2956 case AF_LOCAL:
2957 if (sun->sun_path[0] != '/') {
2958 dprintk("RPC: bad AF_LOCAL address: %s\n",
2959 sun->sun_path);
2960 ret = ERR_PTR(-EINVAL);
2961 goto out_err;
2962 }
2963 xprt_set_bound(xprt);
2964 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2965 ret = ERR_PTR(xs_local_setup_socket(transport));
2966 if (ret)
2967 goto out_err;
2968 break;
2969 default:
2970 ret = ERR_PTR(-EAFNOSUPPORT);
2971 goto out_err;
2972 }
2973
2974 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2975 xprt->address_strings[RPC_DISPLAY_ADDR]);
2976
2977 if (try_module_get(THIS_MODULE))
2978 return xprt;
2979 ret = ERR_PTR(-EINVAL);
2980out_err:
2981 xs_xprt_free(xprt);
2982 return ret;
2983}
2984
2985static const struct rpc_timeout xs_udp_default_timeout = {
2986 .to_initval = 5 * HZ,
2987 .to_maxval = 30 * HZ,
2988 .to_increment = 5 * HZ,
2989 .to_retries = 5,
2990};
2991
2992/**
2993 * xs_setup_udp - Set up transport to use a UDP socket
2994 * @args: rpc transport creation arguments
2995 *
2996 */
2997static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2998{
2999 struct sockaddr *addr = args->dstaddr;
3000 struct rpc_xprt *xprt;
3001 struct sock_xprt *transport;
3002 struct rpc_xprt *ret;
3003
3004 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3005 xprt_udp_slot_table_entries);
3006 if (IS_ERR(xprt))
3007 return xprt;
3008 transport = container_of(xprt, struct sock_xprt, xprt);
3009
3010 xprt->prot = IPPROTO_UDP;
3011 xprt->tsh_size = 0;
3012 /* XXX: header size can vary due to auth type, IPv6, etc. */
3013 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3014
3015 xprt->bind_timeout = XS_BIND_TO;
3016 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3017 xprt->idle_timeout = XS_IDLE_DISC_TO;
3018
3019 xprt->ops = &xs_udp_ops;
3020
3021 xprt->timeout = &xs_udp_default_timeout;
3022
3023 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3024 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3025
3026 switch (addr->sa_family) {
3027 case AF_INET:
3028 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3029 xprt_set_bound(xprt);
3030
3031 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3032 break;
3033 case AF_INET6:
3034 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3035 xprt_set_bound(xprt);
3036
3037 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3038 break;
3039 default:
3040 ret = ERR_PTR(-EAFNOSUPPORT);
3041 goto out_err;
3042 }
3043
3044 if (xprt_bound(xprt))
3045 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3046 xprt->address_strings[RPC_DISPLAY_ADDR],
3047 xprt->address_strings[RPC_DISPLAY_PORT],
3048 xprt->address_strings[RPC_DISPLAY_PROTO]);
3049 else
3050 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3051 xprt->address_strings[RPC_DISPLAY_ADDR],
3052 xprt->address_strings[RPC_DISPLAY_PROTO]);
3053
3054 if (try_module_get(THIS_MODULE))
3055 return xprt;
3056 ret = ERR_PTR(-EINVAL);
3057out_err:
3058 xs_xprt_free(xprt);
3059 return ret;
3060}
3061
3062static const struct rpc_timeout xs_tcp_default_timeout = {
3063 .to_initval = 60 * HZ,
3064 .to_maxval = 60 * HZ,
3065 .to_retries = 2,
3066};
3067
3068/**
3069 * xs_setup_tcp - Set up transport to use a TCP socket
3070 * @args: rpc transport creation arguments
3071 *
3072 */
3073static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3074{
3075 struct sockaddr *addr = args->dstaddr;
3076 struct rpc_xprt *xprt;
3077 struct sock_xprt *transport;
3078 struct rpc_xprt *ret;
3079 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3080
3081 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3082 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3083
3084 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3085 max_slot_table_size);
3086 if (IS_ERR(xprt))
3087 return xprt;
3088 transport = container_of(xprt, struct sock_xprt, xprt);
3089
3090 xprt->prot = IPPROTO_TCP;
3091 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3092 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3093
3094 xprt->bind_timeout = XS_BIND_TO;
3095 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3096 xprt->idle_timeout = XS_IDLE_DISC_TO;
3097
3098 xprt->ops = &xs_tcp_ops;
3099 xprt->timeout = &xs_tcp_default_timeout;
3100
3101 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3102 xprt->connect_timeout = xprt->timeout->to_initval *
3103 (xprt->timeout->to_retries + 1);
3104
3105 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3106 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3107
3108 switch (addr->sa_family) {
3109 case AF_INET:
3110 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3111 xprt_set_bound(xprt);
3112
3113 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3114 break;
3115 case AF_INET6:
3116 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3117 xprt_set_bound(xprt);
3118
3119 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3120 break;
3121 default:
3122 ret = ERR_PTR(-EAFNOSUPPORT);
3123 goto out_err;
3124 }
3125
3126 if (xprt_bound(xprt))
3127 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3128 xprt->address_strings[RPC_DISPLAY_ADDR],
3129 xprt->address_strings[RPC_DISPLAY_PORT],
3130 xprt->address_strings[RPC_DISPLAY_PROTO]);
3131 else
3132 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3133 xprt->address_strings[RPC_DISPLAY_ADDR],
3134 xprt->address_strings[RPC_DISPLAY_PROTO]);
3135
3136 if (try_module_get(THIS_MODULE))
3137 return xprt;
3138 ret = ERR_PTR(-EINVAL);
3139out_err:
3140 xs_xprt_free(xprt);
3141 return ret;
3142}
3143
3144/**
3145 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3146 * @args: rpc transport creation arguments
3147 *
3148 */
3149static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3150{
3151 struct sockaddr *addr = args->dstaddr;
3152 struct rpc_xprt *xprt;
3153 struct sock_xprt *transport;
3154 struct svc_sock *bc_sock;
3155 struct rpc_xprt *ret;
3156
3157 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3158 xprt_tcp_slot_table_entries);
3159 if (IS_ERR(xprt))
3160 return xprt;
3161 transport = container_of(xprt, struct sock_xprt, xprt);
3162
3163 xprt->prot = IPPROTO_TCP;
3164 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3165 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3166 xprt->timeout = &xs_tcp_default_timeout;
3167
3168 /* backchannel */
3169 xprt_set_bound(xprt);
3170 xprt->bind_timeout = 0;
3171 xprt->reestablish_timeout = 0;
3172 xprt->idle_timeout = 0;
3173
3174 xprt->ops = &bc_tcp_ops;
3175
3176 switch (addr->sa_family) {
3177 case AF_INET:
3178 xs_format_peer_addresses(xprt, "tcp",
3179 RPCBIND_NETID_TCP);
3180 break;
3181 case AF_INET6:
3182 xs_format_peer_addresses(xprt, "tcp",
3183 RPCBIND_NETID_TCP6);
3184 break;
3185 default:
3186 ret = ERR_PTR(-EAFNOSUPPORT);
3187 goto out_err;
3188 }
3189
3190 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3191 xprt->address_strings[RPC_DISPLAY_ADDR],
3192 xprt->address_strings[RPC_DISPLAY_PORT],
3193 xprt->address_strings[RPC_DISPLAY_PROTO]);
3194
3195 /*
3196 * Once we've associated a backchannel xprt with a connection,
3197 * we want to keep it around as long as the connection lasts,
3198 * in case we need to start using it for a backchannel again;
3199 * this reference won't be dropped until bc_xprt is destroyed.
3200 */
3201 xprt_get(xprt);
3202 args->bc_xprt->xpt_bc_xprt = xprt;
3203 xprt->bc_xprt = args->bc_xprt;
3204 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3205 transport->sock = bc_sock->sk_sock;
3206 transport->inet = bc_sock->sk_sk;
3207
3208 /*
3209 * Since we don't want connections for the backchannel, we set
3210 * the xprt status to connected
3211 */
3212 xprt_set_connected(xprt);
3213
3214 if (try_module_get(THIS_MODULE))
3215 return xprt;
3216
3217 args->bc_xprt->xpt_bc_xprt = NULL;
3218 args->bc_xprt->xpt_bc_xps = NULL;
3219 xprt_put(xprt);
3220 ret = ERR_PTR(-EINVAL);
3221out_err:
3222 xs_xprt_free(xprt);
3223 return ret;
3224}
3225
3226static struct xprt_class xs_local_transport = {
3227 .list = LIST_HEAD_INIT(xs_local_transport.list),
3228 .name = "named UNIX socket",
3229 .owner = THIS_MODULE,
3230 .ident = XPRT_TRANSPORT_LOCAL,
3231 .setup = xs_setup_local,
3232};
3233
3234static struct xprt_class xs_udp_transport = {
3235 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3236 .name = "udp",
3237 .owner = THIS_MODULE,
3238 .ident = XPRT_TRANSPORT_UDP,
3239 .setup = xs_setup_udp,
3240};
3241
3242static struct xprt_class xs_tcp_transport = {
3243 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3244 .name = "tcp",
3245 .owner = THIS_MODULE,
3246 .ident = XPRT_TRANSPORT_TCP,
3247 .setup = xs_setup_tcp,
3248};
3249
3250static struct xprt_class xs_bc_tcp_transport = {
3251 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3252 .name = "tcp NFSv4.1 backchannel",
3253 .owner = THIS_MODULE,
3254 .ident = XPRT_TRANSPORT_BC_TCP,
3255 .setup = xs_setup_bc_tcp,
3256};
3257
3258/**
3259 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3260 *
3261 */
3262int init_socket_xprt(void)
3263{
3264#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3265 if (!sunrpc_table_header)
3266 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3267#endif
3268
3269 xprt_register_transport(&xs_local_transport);
3270 xprt_register_transport(&xs_udp_transport);
3271 xprt_register_transport(&xs_tcp_transport);
3272 xprt_register_transport(&xs_bc_tcp_transport);
3273
3274 return 0;
3275}
3276
3277/**
3278 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3279 *
3280 */
3281void cleanup_socket_xprt(void)
3282{
3283#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3284 if (sunrpc_table_header) {
3285 unregister_sysctl_table(sunrpc_table_header);
3286 sunrpc_table_header = NULL;
3287 }
3288#endif
3289
3290 xprt_unregister_transport(&xs_local_transport);
3291 xprt_unregister_transport(&xs_udp_transport);
3292 xprt_unregister_transport(&xs_tcp_transport);
3293 xprt_unregister_transport(&xs_bc_tcp_transport);
3294}
3295
3296static int param_set_uint_minmax(const char *val,
3297 const struct kernel_param *kp,
3298 unsigned int min, unsigned int max)
3299{
3300 unsigned int num;
3301 int ret;
3302
3303 if (!val)
3304 return -EINVAL;
3305 ret = kstrtouint(val, 0, &num);
3306 if (ret)
3307 return ret;
3308 if (num < min || num > max)
3309 return -EINVAL;
3310 *((unsigned int *)kp->arg) = num;
3311 return 0;
3312}
3313
3314static int param_set_portnr(const char *val, const struct kernel_param *kp)
3315{
3316 if (kp->arg == &xprt_min_resvport)
3317 return param_set_uint_minmax(val, kp,
3318 RPC_MIN_RESVPORT,
3319 xprt_max_resvport);
3320 return param_set_uint_minmax(val, kp,
3321 xprt_min_resvport,
3322 RPC_MAX_RESVPORT);
3323}
3324
3325static const struct kernel_param_ops param_ops_portnr = {
3326 .set = param_set_portnr,
3327 .get = param_get_uint,
3328};
3329
3330#define param_check_portnr(name, p) \
3331 __param_check(name, p, unsigned int);
3332
3333module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3334module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3335
3336static int param_set_slot_table_size(const char *val,
3337 const struct kernel_param *kp)
3338{
3339 return param_set_uint_minmax(val, kp,
3340 RPC_MIN_SLOT_TABLE,
3341 RPC_MAX_SLOT_TABLE);
3342}
3343
3344static const struct kernel_param_ops param_ops_slot_table_size = {
3345 .set = param_set_slot_table_size,
3346 .get = param_get_uint,
3347};
3348
3349#define param_check_slot_table_size(name, p) \
3350 __param_check(name, p, unsigned int);
3351
3352static int param_set_max_slot_table_size(const char *val,
3353 const struct kernel_param *kp)
3354{
3355 return param_set_uint_minmax(val, kp,
3356 RPC_MIN_SLOT_TABLE,
3357 RPC_MAX_SLOT_TABLE_LIMIT);
3358}
3359
3360static const struct kernel_param_ops param_ops_max_slot_table_size = {
3361 .set = param_set_max_slot_table_size,
3362 .get = param_get_uint,
3363};
3364
3365#define param_check_max_slot_table_size(name, p) \
3366 __param_check(name, p, unsigned int);
3367
3368module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3369 slot_table_size, 0644);
3370module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3371 max_slot_table_size, 0644);
3372module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3373 slot_table_size, 0644);
3374
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 <linux/bvec.h>
51#include <linux/highmem.h>
52#include <linux/uio.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "socklib.h"
58#include "sunrpc.h"
59
60static void xs_close(struct rpc_xprt *xprt);
61static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
62 struct socket *sock);
63
64/*
65 * xprtsock tunables
66 */
67static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
70
71static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
73
74#define XS_TCP_LINGER_TO (15U * HZ)
75static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
76
77/*
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
81 *
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
84 */
85
86static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91
92static struct ctl_table_header *sunrpc_table_header;
93
94/*
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
97 */
98static struct ctl_table xs_tunables_table[] = {
99 {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
108 {
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
116 },
117 {
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
125 },
126 {
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
135 {
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
143 },
144 {
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_jiffies,
150 },
151 { },
152};
153
154static struct ctl_table sunrpc_table[] = {
155 {
156 .procname = "sunrpc",
157 .mode = 0555,
158 .child = xs_tunables_table
159 },
160 { },
161};
162
163/*
164 * Wait duration for a reply from the RPC portmapper.
165 */
166#define XS_BIND_TO (60U * HZ)
167
168/*
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
171 */
172#define XS_UDP_REEST_TO (2U * HZ)
173
174/*
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
177 *
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
182 */
183#define XS_TCP_INIT_REEST_TO (3U * HZ)
184
185/*
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
189 */
190#define XS_IDLE_DISC_TO (5U * 60 * HZ)
191
192#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193# undef RPC_DEBUG_DATA
194# define RPCDBG_FACILITY RPCDBG_TRANS
195#endif
196
197#ifdef RPC_DEBUG_DATA
198static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
199{
200 u8 *buf = (u8 *) packet;
201 int j;
202
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
205 if (!(j & 31)) {
206 if (j)
207 dprintk("\n");
208 dprintk("0x%04x ", j);
209 }
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 }
213 dprintk("\n");
214}
215#else
216static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217{
218 /* NOP */
219}
220#endif
221
222static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
223{
224 return (struct rpc_xprt *) sk->sk_user_data;
225}
226
227static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
228{
229 return (struct sockaddr *) &xprt->addr;
230}
231
232static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
233{
234 return (struct sockaddr_un *) &xprt->addr;
235}
236
237static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
238{
239 return (struct sockaddr_in *) &xprt->addr;
240}
241
242static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
243{
244 return (struct sockaddr_in6 *) &xprt->addr;
245}
246
247static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
248{
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
253 char buf[128];
254
255 switch (sap->sa_family) {
256 case AF_LOCAL:
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
261 break;
262 case AF_INET:
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
268 break;
269 case AF_INET6:
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
275 break;
276 default:
277 BUG();
278 }
279
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
281}
282
283static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
284{
285 struct sockaddr *sap = xs_addr(xprt);
286 char buf[128];
287
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
290
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
293}
294
295static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
297 const char *netid)
298{
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
303}
304
305static void xs_update_peer_port(struct rpc_xprt *xprt)
306{
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
309
310 xs_format_common_peer_ports(xprt);
311}
312
313static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314{
315 unsigned int i;
316
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 switch (i) {
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
321 continue;
322 default:
323 kfree(xprt->address_strings[i]);
324 }
325}
326
327static size_t
328xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
329{
330 size_t i,n;
331
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 return want;
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
336 if (buf->pages[i])
337 continue;
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
340 i *= PAGE_SIZE;
341 return i > buf->page_base ? i - buf->page_base : 0;
342 }
343 }
344 return want;
345}
346
347static ssize_t
348xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
349{
350 ssize_t ret;
351 if (seek != 0)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
355}
356
357static ssize_t
358xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
360{
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
363}
364
365static ssize_t
366xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
368 size_t seek)
369{
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
372}
373
374static ssize_t
375xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
376 size_t count)
377{
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
380}
381
382#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383static void
384xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
385{
386 struct bvec_iter bi = {
387 .bi_size = count,
388 };
389 struct bio_vec bv;
390
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
394}
395#else
396static inline void
397xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
398{
399}
400#endif
401
402static ssize_t
403xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
405{
406 size_t want, seek_init = seek, offset = 0;
407 ssize_t ret;
408
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 if (seek < want) {
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
412 if (ret <= 0)
413 goto sock_err;
414 offset += ret;
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
416 goto out;
417 if (ret != want)
418 goto out;
419 seek = 0;
420 } else {
421 seek -= want;
422 offset += want;
423 }
424
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
427 GFP_KERNEL);
428 if (seek < want) {
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
433 if (ret <= 0)
434 goto sock_err;
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 offset += ret - buf->page_base;
437 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
438 goto out;
439 if (ret != want)
440 goto out;
441 seek = 0;
442 } else {
443 seek -= want;
444 offset += want;
445 }
446
447 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
448 if (seek < want) {
449 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
450 if (ret <= 0)
451 goto sock_err;
452 offset += ret;
453 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
454 goto out;
455 if (ret != want)
456 goto out;
457 } else if (offset < seek_init)
458 offset = seek_init;
459 ret = -EMSGSIZE;
460out:
461 *read = offset - seek_init;
462 return ret;
463sock_err:
464 offset += seek;
465 goto out;
466}
467
468static void
469xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
470{
471 if (!transport->recv.copied) {
472 if (buf->head[0].iov_len >= transport->recv.offset)
473 memcpy(buf->head[0].iov_base,
474 &transport->recv.xid,
475 transport->recv.offset);
476 transport->recv.copied = transport->recv.offset;
477 }
478}
479
480static bool
481xs_read_stream_request_done(struct sock_xprt *transport)
482{
483 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
484}
485
486static void
487xs_read_stream_check_eor(struct sock_xprt *transport,
488 struct msghdr *msg)
489{
490 if (xs_read_stream_request_done(transport))
491 msg->msg_flags |= MSG_EOR;
492}
493
494static ssize_t
495xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
496 int flags, struct rpc_rqst *req)
497{
498 struct xdr_buf *buf = &req->rq_private_buf;
499 size_t want, read;
500 ssize_t ret;
501
502 xs_read_header(transport, buf);
503
504 want = transport->recv.len - transport->recv.offset;
505 if (want != 0) {
506 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
507 transport->recv.copied + want,
508 transport->recv.copied,
509 &read);
510 transport->recv.offset += read;
511 transport->recv.copied += read;
512 }
513
514 if (transport->recv.offset == transport->recv.len)
515 xs_read_stream_check_eor(transport, msg);
516
517 if (want == 0)
518 return 0;
519
520 switch (ret) {
521 default:
522 break;
523 case -EFAULT:
524 case -EMSGSIZE:
525 msg->msg_flags |= MSG_TRUNC;
526 return read;
527 case 0:
528 return -ESHUTDOWN;
529 }
530 return ret < 0 ? ret : read;
531}
532
533static size_t
534xs_read_stream_headersize(bool isfrag)
535{
536 if (isfrag)
537 return sizeof(__be32);
538 return 3 * sizeof(__be32);
539}
540
541static ssize_t
542xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
543 int flags, size_t want, size_t seek)
544{
545 struct kvec kvec = {
546 .iov_base = &transport->recv.fraghdr,
547 .iov_len = want,
548 };
549 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
550}
551
552#if defined(CONFIG_SUNRPC_BACKCHANNEL)
553static ssize_t
554xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
555{
556 struct rpc_xprt *xprt = &transport->xprt;
557 struct rpc_rqst *req;
558 ssize_t ret;
559
560 /* Look up and lock the request corresponding to the given XID */
561 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
562 if (!req) {
563 printk(KERN_WARNING "Callback slot table overflowed\n");
564 return -ESHUTDOWN;
565 }
566 if (transport->recv.copied && !req->rq_private_buf.len)
567 return -ESHUTDOWN;
568
569 ret = xs_read_stream_request(transport, msg, flags, req);
570 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
571 xprt_complete_bc_request(req, transport->recv.copied);
572 else
573 req->rq_private_buf.len = transport->recv.copied;
574
575 return ret;
576}
577#else /* CONFIG_SUNRPC_BACKCHANNEL */
578static ssize_t
579xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
580{
581 return -ESHUTDOWN;
582}
583#endif /* CONFIG_SUNRPC_BACKCHANNEL */
584
585static ssize_t
586xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
587{
588 struct rpc_xprt *xprt = &transport->xprt;
589 struct rpc_rqst *req;
590 ssize_t ret = 0;
591
592 /* Look up and lock the request corresponding to the given XID */
593 spin_lock(&xprt->queue_lock);
594 req = xprt_lookup_rqst(xprt, transport->recv.xid);
595 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
596 msg->msg_flags |= MSG_TRUNC;
597 goto out;
598 }
599 xprt_pin_rqst(req);
600 spin_unlock(&xprt->queue_lock);
601
602 ret = xs_read_stream_request(transport, msg, flags, req);
603
604 spin_lock(&xprt->queue_lock);
605 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
606 xprt_complete_rqst(req->rq_task, transport->recv.copied);
607 else
608 req->rq_private_buf.len = transport->recv.copied;
609 xprt_unpin_rqst(req);
610out:
611 spin_unlock(&xprt->queue_lock);
612 return ret;
613}
614
615static ssize_t
616xs_read_stream(struct sock_xprt *transport, int flags)
617{
618 struct msghdr msg = { 0 };
619 size_t want, read = 0;
620 ssize_t ret = 0;
621
622 if (transport->recv.len == 0) {
623 want = xs_read_stream_headersize(transport->recv.copied != 0);
624 ret = xs_read_stream_header(transport, &msg, flags, want,
625 transport->recv.offset);
626 if (ret <= 0)
627 goto out_err;
628 transport->recv.offset = ret;
629 if (transport->recv.offset != want)
630 return transport->recv.offset;
631 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
632 RPC_FRAGMENT_SIZE_MASK;
633 transport->recv.offset -= sizeof(transport->recv.fraghdr);
634 read = ret;
635 }
636
637 switch (be32_to_cpu(transport->recv.calldir)) {
638 default:
639 msg.msg_flags |= MSG_TRUNC;
640 break;
641 case RPC_CALL:
642 ret = xs_read_stream_call(transport, &msg, flags);
643 break;
644 case RPC_REPLY:
645 ret = xs_read_stream_reply(transport, &msg, flags);
646 }
647 if (msg.msg_flags & MSG_TRUNC) {
648 transport->recv.calldir = cpu_to_be32(-1);
649 transport->recv.copied = -1;
650 }
651 if (ret < 0)
652 goto out_err;
653 read += ret;
654 if (transport->recv.offset < transport->recv.len) {
655 if (!(msg.msg_flags & MSG_TRUNC))
656 return read;
657 msg.msg_flags = 0;
658 ret = xs_read_discard(transport->sock, &msg, flags,
659 transport->recv.len - transport->recv.offset);
660 if (ret <= 0)
661 goto out_err;
662 transport->recv.offset += ret;
663 read += ret;
664 if (transport->recv.offset != transport->recv.len)
665 return read;
666 }
667 if (xs_read_stream_request_done(transport)) {
668 trace_xs_stream_read_request(transport);
669 transport->recv.copied = 0;
670 }
671 transport->recv.offset = 0;
672 transport->recv.len = 0;
673 return read;
674out_err:
675 return ret != 0 ? ret : -ESHUTDOWN;
676}
677
678static __poll_t xs_poll_socket(struct sock_xprt *transport)
679{
680 return transport->sock->ops->poll(transport->file, transport->sock,
681 NULL);
682}
683
684static bool xs_poll_socket_readable(struct sock_xprt *transport)
685{
686 __poll_t events = xs_poll_socket(transport);
687
688 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
689}
690
691static void xs_poll_check_readable(struct sock_xprt *transport)
692{
693
694 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
695 if (!xs_poll_socket_readable(transport))
696 return;
697 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
698 queue_work(xprtiod_workqueue, &transport->recv_worker);
699}
700
701static void xs_stream_data_receive(struct sock_xprt *transport)
702{
703 size_t read = 0;
704 ssize_t ret = 0;
705
706 mutex_lock(&transport->recv_mutex);
707 if (transport->sock == NULL)
708 goto out;
709 for (;;) {
710 ret = xs_read_stream(transport, MSG_DONTWAIT);
711 if (ret < 0)
712 break;
713 read += ret;
714 cond_resched();
715 }
716 if (ret == -ESHUTDOWN)
717 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
718 else
719 xs_poll_check_readable(transport);
720out:
721 mutex_unlock(&transport->recv_mutex);
722 trace_xs_stream_read_data(&transport->xprt, ret, read);
723}
724
725static void xs_stream_data_receive_workfn(struct work_struct *work)
726{
727 struct sock_xprt *transport =
728 container_of(work, struct sock_xprt, recv_worker);
729 unsigned int pflags = memalloc_nofs_save();
730
731 xs_stream_data_receive(transport);
732 memalloc_nofs_restore(pflags);
733}
734
735static void
736xs_stream_reset_connect(struct sock_xprt *transport)
737{
738 transport->recv.offset = 0;
739 transport->recv.len = 0;
740 transport->recv.copied = 0;
741 transport->xmit.offset = 0;
742}
743
744static void
745xs_stream_start_connect(struct sock_xprt *transport)
746{
747 transport->xprt.stat.connect_count++;
748 transport->xprt.stat.connect_start = jiffies;
749}
750
751#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
752
753/**
754 * xs_nospace - handle transmit was incomplete
755 * @req: pointer to RPC request
756 *
757 */
758static int xs_nospace(struct rpc_rqst *req)
759{
760 struct rpc_xprt *xprt = req->rq_xprt;
761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
762 struct sock *sk = transport->inet;
763 int ret = -EAGAIN;
764
765 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
766 req->rq_task->tk_pid,
767 req->rq_slen - transport->xmit.offset,
768 req->rq_slen);
769
770 /* Protect against races with write_space */
771 spin_lock(&xprt->transport_lock);
772
773 /* Don't race with disconnect */
774 if (xprt_connected(xprt)) {
775 /* wait for more buffer space */
776 sk->sk_write_pending++;
777 xprt_wait_for_buffer_space(xprt);
778 } else
779 ret = -ENOTCONN;
780
781 spin_unlock(&xprt->transport_lock);
782
783 /* Race breaker in case memory is freed before above code is called */
784 if (ret == -EAGAIN) {
785 struct socket_wq *wq;
786
787 rcu_read_lock();
788 wq = rcu_dereference(sk->sk_wq);
789 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
790 rcu_read_unlock();
791
792 sk->sk_write_space(sk);
793 }
794 return ret;
795}
796
797static void
798xs_stream_prepare_request(struct rpc_rqst *req)
799{
800 xdr_free_bvec(&req->rq_rcv_buf);
801 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
802}
803
804/*
805 * Determine if the previous message in the stream was aborted before it
806 * could complete transmission.
807 */
808static bool
809xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
810{
811 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
812}
813
814/*
815 * Return the stream record marker field for a record of length < 2^31-1
816 */
817static rpc_fraghdr
818xs_stream_record_marker(struct xdr_buf *xdr)
819{
820 if (!xdr->len)
821 return 0;
822 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
823}
824
825/**
826 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
827 * @req: pointer to RPC request
828 *
829 * Return values:
830 * 0: The request has been sent
831 * EAGAIN: The socket was blocked, please call again later to
832 * complete the request
833 * ENOTCONN: Caller needs to invoke connect logic then call again
834 * other: Some other error occured, the request was not sent
835 */
836static int xs_local_send_request(struct rpc_rqst *req)
837{
838 struct rpc_xprt *xprt = req->rq_xprt;
839 struct sock_xprt *transport =
840 container_of(xprt, struct sock_xprt, xprt);
841 struct xdr_buf *xdr = &req->rq_snd_buf;
842 rpc_fraghdr rm = xs_stream_record_marker(xdr);
843 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
844 struct msghdr msg = {
845 .msg_flags = XS_SENDMSG_FLAGS,
846 };
847 unsigned int sent;
848 int status;
849
850 /* Close the stream if the previous transmission was incomplete */
851 if (xs_send_request_was_aborted(transport, req)) {
852 xs_close(xprt);
853 return -ENOTCONN;
854 }
855
856 xs_pktdump("packet data:",
857 req->rq_svec->iov_base, req->rq_svec->iov_len);
858
859 req->rq_xtime = ktime_get();
860 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
861 transport->xmit.offset, rm, &sent);
862 dprintk("RPC: %s(%u) = %d\n",
863 __func__, xdr->len - transport->xmit.offset, status);
864
865 if (status == -EAGAIN && sock_writeable(transport->inet))
866 status = -ENOBUFS;
867
868 if (likely(sent > 0) || status == 0) {
869 transport->xmit.offset += sent;
870 req->rq_bytes_sent = transport->xmit.offset;
871 if (likely(req->rq_bytes_sent >= msglen)) {
872 req->rq_xmit_bytes_sent += transport->xmit.offset;
873 transport->xmit.offset = 0;
874 return 0;
875 }
876 status = -EAGAIN;
877 }
878
879 switch (status) {
880 case -ENOBUFS:
881 break;
882 case -EAGAIN:
883 status = xs_nospace(req);
884 break;
885 default:
886 dprintk("RPC: sendmsg returned unrecognized error %d\n",
887 -status);
888 fallthrough;
889 case -EPIPE:
890 xs_close(xprt);
891 status = -ENOTCONN;
892 }
893
894 return status;
895}
896
897/**
898 * xs_udp_send_request - write an RPC request to a UDP socket
899 * @req: pointer to RPC request
900 *
901 * Return values:
902 * 0: The request has been sent
903 * EAGAIN: The socket was blocked, please call again later to
904 * complete the request
905 * ENOTCONN: Caller needs to invoke connect logic then call again
906 * other: Some other error occurred, the request was not sent
907 */
908static int xs_udp_send_request(struct rpc_rqst *req)
909{
910 struct rpc_xprt *xprt = req->rq_xprt;
911 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
912 struct xdr_buf *xdr = &req->rq_snd_buf;
913 struct msghdr msg = {
914 .msg_name = xs_addr(xprt),
915 .msg_namelen = xprt->addrlen,
916 .msg_flags = XS_SENDMSG_FLAGS,
917 };
918 unsigned int sent;
919 int status;
920
921 xs_pktdump("packet data:",
922 req->rq_svec->iov_base,
923 req->rq_svec->iov_len);
924
925 if (!xprt_bound(xprt))
926 return -ENOTCONN;
927
928 if (!xprt_request_get_cong(xprt, req))
929 return -EBADSLT;
930
931 req->rq_xtime = ktime_get();
932 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
933
934 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
935 xdr->len, status);
936
937 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
938 if (status == -EPERM)
939 goto process_status;
940
941 if (status == -EAGAIN && sock_writeable(transport->inet))
942 status = -ENOBUFS;
943
944 if (sent > 0 || status == 0) {
945 req->rq_xmit_bytes_sent += sent;
946 if (sent >= req->rq_slen)
947 return 0;
948 /* Still some bytes left; set up for a retry later. */
949 status = -EAGAIN;
950 }
951
952process_status:
953 switch (status) {
954 case -ENOTSOCK:
955 status = -ENOTCONN;
956 /* Should we call xs_close() here? */
957 break;
958 case -EAGAIN:
959 status = xs_nospace(req);
960 break;
961 case -ENETUNREACH:
962 case -ENOBUFS:
963 case -EPIPE:
964 case -ECONNREFUSED:
965 case -EPERM:
966 /* When the server has died, an ICMP port unreachable message
967 * prompts ECONNREFUSED. */
968 break;
969 default:
970 dprintk("RPC: sendmsg returned unrecognized error %d\n",
971 -status);
972 }
973
974 return status;
975}
976
977/**
978 * xs_tcp_send_request - write an RPC request to a TCP socket
979 * @req: pointer to RPC request
980 *
981 * Return values:
982 * 0: The request has been sent
983 * EAGAIN: The socket was blocked, please call again later to
984 * complete the request
985 * ENOTCONN: Caller needs to invoke connect logic then call again
986 * other: Some other error occurred, the request was not sent
987 *
988 * XXX: In the case of soft timeouts, should we eventually give up
989 * if sendmsg is not able to make progress?
990 */
991static int xs_tcp_send_request(struct rpc_rqst *req)
992{
993 struct rpc_xprt *xprt = req->rq_xprt;
994 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
995 struct xdr_buf *xdr = &req->rq_snd_buf;
996 rpc_fraghdr rm = xs_stream_record_marker(xdr);
997 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
998 struct msghdr msg = {
999 .msg_flags = XS_SENDMSG_FLAGS,
1000 };
1001 bool vm_wait = false;
1002 unsigned int sent;
1003 int status;
1004
1005 /* Close the stream if the previous transmission was incomplete */
1006 if (xs_send_request_was_aborted(transport, req)) {
1007 if (transport->sock != NULL)
1008 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1009 return -ENOTCONN;
1010 }
1011
1012 xs_pktdump("packet data:",
1013 req->rq_svec->iov_base,
1014 req->rq_svec->iov_len);
1015
1016 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1017 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1018
1019 /* Continue transmitting the packet/record. We must be careful
1020 * to cope with writespace callbacks arriving _after_ we have
1021 * called sendmsg(). */
1022 req->rq_xtime = ktime_get();
1023 while (1) {
1024 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1025 transport->xmit.offset, rm, &sent);
1026
1027 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1028 xdr->len - transport->xmit.offset, status);
1029
1030 /* If we've sent the entire packet, immediately
1031 * reset the count of bytes sent. */
1032 transport->xmit.offset += sent;
1033 req->rq_bytes_sent = transport->xmit.offset;
1034 if (likely(req->rq_bytes_sent >= msglen)) {
1035 req->rq_xmit_bytes_sent += transport->xmit.offset;
1036 transport->xmit.offset = 0;
1037 return 0;
1038 }
1039
1040 WARN_ON_ONCE(sent == 0 && status == 0);
1041
1042 if (status == -EAGAIN ) {
1043 /*
1044 * Return EAGAIN if we're sure we're hitting the
1045 * socket send buffer limits.
1046 */
1047 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1048 break;
1049 /*
1050 * Did we hit a memory allocation failure?
1051 */
1052 if (sent == 0) {
1053 status = -ENOBUFS;
1054 if (vm_wait)
1055 break;
1056 /* Retry, knowing now that we're below the
1057 * socket send buffer limit
1058 */
1059 vm_wait = true;
1060 }
1061 continue;
1062 }
1063 if (status < 0)
1064 break;
1065 vm_wait = false;
1066 }
1067
1068 switch (status) {
1069 case -ENOTSOCK:
1070 status = -ENOTCONN;
1071 /* Should we call xs_close() here? */
1072 break;
1073 case -EAGAIN:
1074 status = xs_nospace(req);
1075 break;
1076 case -ECONNRESET:
1077 case -ECONNREFUSED:
1078 case -ENOTCONN:
1079 case -EADDRINUSE:
1080 case -ENOBUFS:
1081 case -EPIPE:
1082 break;
1083 default:
1084 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1085 -status);
1086 }
1087
1088 return status;
1089}
1090
1091static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1092{
1093 transport->old_data_ready = sk->sk_data_ready;
1094 transport->old_state_change = sk->sk_state_change;
1095 transport->old_write_space = sk->sk_write_space;
1096 transport->old_error_report = sk->sk_error_report;
1097}
1098
1099static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1100{
1101 sk->sk_data_ready = transport->old_data_ready;
1102 sk->sk_state_change = transport->old_state_change;
1103 sk->sk_write_space = transport->old_write_space;
1104 sk->sk_error_report = transport->old_error_report;
1105}
1106
1107static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1108{
1109 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1110
1111 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1112 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1113 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1114 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1115}
1116
1117static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1118{
1119 set_bit(nr, &transport->sock_state);
1120 queue_work(xprtiod_workqueue, &transport->error_worker);
1121}
1122
1123static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1124{
1125 smp_mb__before_atomic();
1126 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1127 clear_bit(XPRT_CLOSING, &xprt->state);
1128 xs_sock_reset_state_flags(xprt);
1129 smp_mb__after_atomic();
1130}
1131
1132/**
1133 * xs_error_report - callback to handle TCP socket state errors
1134 * @sk: socket
1135 *
1136 * Note: we don't call sock_error() since there may be a rpc_task
1137 * using the socket, and so we don't want to clear sk->sk_err.
1138 */
1139static void xs_error_report(struct sock *sk)
1140{
1141 struct sock_xprt *transport;
1142 struct rpc_xprt *xprt;
1143
1144 read_lock_bh(&sk->sk_callback_lock);
1145 if (!(xprt = xprt_from_sock(sk)))
1146 goto out;
1147
1148 transport = container_of(xprt, struct sock_xprt, xprt);
1149 transport->xprt_err = -sk->sk_err;
1150 if (transport->xprt_err == 0)
1151 goto out;
1152 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1153 xprt, -transport->xprt_err);
1154 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1155
1156 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1157 smp_mb__before_atomic();
1158 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1159 out:
1160 read_unlock_bh(&sk->sk_callback_lock);
1161}
1162
1163static void xs_reset_transport(struct sock_xprt *transport)
1164{
1165 struct socket *sock = transport->sock;
1166 struct sock *sk = transport->inet;
1167 struct rpc_xprt *xprt = &transport->xprt;
1168 struct file *filp = transport->file;
1169
1170 if (sk == NULL)
1171 return;
1172
1173 if (atomic_read(&transport->xprt.swapper))
1174 sk_clear_memalloc(sk);
1175
1176 kernel_sock_shutdown(sock, SHUT_RDWR);
1177
1178 mutex_lock(&transport->recv_mutex);
1179 write_lock_bh(&sk->sk_callback_lock);
1180 transport->inet = NULL;
1181 transport->sock = NULL;
1182 transport->file = NULL;
1183
1184 sk->sk_user_data = NULL;
1185
1186 xs_restore_old_callbacks(transport, sk);
1187 xprt_clear_connected(xprt);
1188 write_unlock_bh(&sk->sk_callback_lock);
1189 xs_sock_reset_connection_flags(xprt);
1190 /* Reset stream record info */
1191 xs_stream_reset_connect(transport);
1192 mutex_unlock(&transport->recv_mutex);
1193
1194 trace_rpc_socket_close(xprt, sock);
1195 fput(filp);
1196
1197 xprt_disconnect_done(xprt);
1198}
1199
1200/**
1201 * xs_close - close a socket
1202 * @xprt: transport
1203 *
1204 * This is used when all requests are complete; ie, no DRC state remains
1205 * on the server we want to save.
1206 *
1207 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1208 * xs_reset_transport() zeroing the socket from underneath a writer.
1209 */
1210static void xs_close(struct rpc_xprt *xprt)
1211{
1212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1213
1214 dprintk("RPC: xs_close xprt %p\n", xprt);
1215
1216 xs_reset_transport(transport);
1217 xprt->reestablish_timeout = 0;
1218}
1219
1220static void xs_inject_disconnect(struct rpc_xprt *xprt)
1221{
1222 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1223 xprt);
1224 xprt_disconnect_done(xprt);
1225}
1226
1227static void xs_xprt_free(struct rpc_xprt *xprt)
1228{
1229 xs_free_peer_addresses(xprt);
1230 xprt_free(xprt);
1231}
1232
1233/**
1234 * xs_destroy - prepare to shutdown a transport
1235 * @xprt: doomed transport
1236 *
1237 */
1238static void xs_destroy(struct rpc_xprt *xprt)
1239{
1240 struct sock_xprt *transport = container_of(xprt,
1241 struct sock_xprt, xprt);
1242 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1243
1244 cancel_delayed_work_sync(&transport->connect_worker);
1245 xs_close(xprt);
1246 cancel_work_sync(&transport->recv_worker);
1247 cancel_work_sync(&transport->error_worker);
1248 xs_xprt_free(xprt);
1249 module_put(THIS_MODULE);
1250}
1251
1252/**
1253 * xs_udp_data_read_skb - receive callback for UDP sockets
1254 * @xprt: transport
1255 * @sk: socket
1256 * @skb: skbuff
1257 *
1258 */
1259static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1260 struct sock *sk,
1261 struct sk_buff *skb)
1262{
1263 struct rpc_task *task;
1264 struct rpc_rqst *rovr;
1265 int repsize, copied;
1266 u32 _xid;
1267 __be32 *xp;
1268
1269 repsize = skb->len;
1270 if (repsize < 4) {
1271 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1272 return;
1273 }
1274
1275 /* Copy the XID from the skb... */
1276 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1277 if (xp == NULL)
1278 return;
1279
1280 /* Look up and lock the request corresponding to the given XID */
1281 spin_lock(&xprt->queue_lock);
1282 rovr = xprt_lookup_rqst(xprt, *xp);
1283 if (!rovr)
1284 goto out_unlock;
1285 xprt_pin_rqst(rovr);
1286 xprt_update_rtt(rovr->rq_task);
1287 spin_unlock(&xprt->queue_lock);
1288 task = rovr->rq_task;
1289
1290 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1291 copied = repsize;
1292
1293 /* Suck it into the iovec, verify checksum if not done by hw. */
1294 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1295 spin_lock(&xprt->queue_lock);
1296 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1297 goto out_unpin;
1298 }
1299
1300
1301 spin_lock(&xprt->transport_lock);
1302 xprt_adjust_cwnd(xprt, task, copied);
1303 spin_unlock(&xprt->transport_lock);
1304 spin_lock(&xprt->queue_lock);
1305 xprt_complete_rqst(task, copied);
1306 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1307out_unpin:
1308 xprt_unpin_rqst(rovr);
1309 out_unlock:
1310 spin_unlock(&xprt->queue_lock);
1311}
1312
1313static void xs_udp_data_receive(struct sock_xprt *transport)
1314{
1315 struct sk_buff *skb;
1316 struct sock *sk;
1317 int err;
1318
1319 mutex_lock(&transport->recv_mutex);
1320 sk = transport->inet;
1321 if (sk == NULL)
1322 goto out;
1323 for (;;) {
1324 skb = skb_recv_udp(sk, 0, 1, &err);
1325 if (skb == NULL)
1326 break;
1327 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1328 consume_skb(skb);
1329 cond_resched();
1330 }
1331 xs_poll_check_readable(transport);
1332out:
1333 mutex_unlock(&transport->recv_mutex);
1334}
1335
1336static void xs_udp_data_receive_workfn(struct work_struct *work)
1337{
1338 struct sock_xprt *transport =
1339 container_of(work, struct sock_xprt, recv_worker);
1340 unsigned int pflags = memalloc_nofs_save();
1341
1342 xs_udp_data_receive(transport);
1343 memalloc_nofs_restore(pflags);
1344}
1345
1346/**
1347 * xs_data_ready - "data ready" callback for UDP sockets
1348 * @sk: socket with data to read
1349 *
1350 */
1351static void xs_data_ready(struct sock *sk)
1352{
1353 struct rpc_xprt *xprt;
1354
1355 read_lock_bh(&sk->sk_callback_lock);
1356 dprintk("RPC: xs_data_ready...\n");
1357 xprt = xprt_from_sock(sk);
1358 if (xprt != NULL) {
1359 struct sock_xprt *transport = container_of(xprt,
1360 struct sock_xprt, xprt);
1361 transport->old_data_ready(sk);
1362 /* Any data means we had a useful conversation, so
1363 * then we don't need to delay the next reconnect
1364 */
1365 if (xprt->reestablish_timeout)
1366 xprt->reestablish_timeout = 0;
1367 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1368 queue_work(xprtiod_workqueue, &transport->recv_worker);
1369 }
1370 read_unlock_bh(&sk->sk_callback_lock);
1371}
1372
1373/*
1374 * Helper function to force a TCP close if the server is sending
1375 * junk and/or it has put us in CLOSE_WAIT
1376 */
1377static void xs_tcp_force_close(struct rpc_xprt *xprt)
1378{
1379 xprt_force_disconnect(xprt);
1380}
1381
1382#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1383static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1384{
1385 return PAGE_SIZE;
1386}
1387#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1388
1389/**
1390 * xs_tcp_state_change - callback to handle TCP socket state changes
1391 * @sk: socket whose state has changed
1392 *
1393 */
1394static void xs_tcp_state_change(struct sock *sk)
1395{
1396 struct rpc_xprt *xprt;
1397 struct sock_xprt *transport;
1398
1399 read_lock_bh(&sk->sk_callback_lock);
1400 if (!(xprt = xprt_from_sock(sk)))
1401 goto out;
1402 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1403 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1404 sk->sk_state, xprt_connected(xprt),
1405 sock_flag(sk, SOCK_DEAD),
1406 sock_flag(sk, SOCK_ZAPPED),
1407 sk->sk_shutdown);
1408
1409 transport = container_of(xprt, struct sock_xprt, xprt);
1410 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1411 switch (sk->sk_state) {
1412 case TCP_ESTABLISHED:
1413 if (!xprt_test_and_set_connected(xprt)) {
1414 xprt->connect_cookie++;
1415 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1416 xprt_clear_connecting(xprt);
1417
1418 xprt->stat.connect_count++;
1419 xprt->stat.connect_time += (long)jiffies -
1420 xprt->stat.connect_start;
1421 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1422 }
1423 break;
1424 case TCP_FIN_WAIT1:
1425 /* The client initiated a shutdown of the socket */
1426 xprt->connect_cookie++;
1427 xprt->reestablish_timeout = 0;
1428 set_bit(XPRT_CLOSING, &xprt->state);
1429 smp_mb__before_atomic();
1430 clear_bit(XPRT_CONNECTED, &xprt->state);
1431 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1432 smp_mb__after_atomic();
1433 break;
1434 case TCP_CLOSE_WAIT:
1435 /* The server initiated a shutdown of the socket */
1436 xprt->connect_cookie++;
1437 clear_bit(XPRT_CONNECTED, &xprt->state);
1438 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1439 fallthrough;
1440 case TCP_CLOSING:
1441 /*
1442 * If the server closed down the connection, make sure that
1443 * we back off before reconnecting
1444 */
1445 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1446 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1447 break;
1448 case TCP_LAST_ACK:
1449 set_bit(XPRT_CLOSING, &xprt->state);
1450 smp_mb__before_atomic();
1451 clear_bit(XPRT_CONNECTED, &xprt->state);
1452 smp_mb__after_atomic();
1453 break;
1454 case TCP_CLOSE:
1455 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1456 &transport->sock_state))
1457 xprt_clear_connecting(xprt);
1458 clear_bit(XPRT_CLOSING, &xprt->state);
1459 /* Trigger the socket release */
1460 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1461 }
1462 out:
1463 read_unlock_bh(&sk->sk_callback_lock);
1464}
1465
1466static void xs_write_space(struct sock *sk)
1467{
1468 struct socket_wq *wq;
1469 struct sock_xprt *transport;
1470 struct rpc_xprt *xprt;
1471
1472 if (!sk->sk_socket)
1473 return;
1474 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1475
1476 if (unlikely(!(xprt = xprt_from_sock(sk))))
1477 return;
1478 transport = container_of(xprt, struct sock_xprt, xprt);
1479 rcu_read_lock();
1480 wq = rcu_dereference(sk->sk_wq);
1481 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1482 goto out;
1483
1484 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1485 sk->sk_write_pending--;
1486out:
1487 rcu_read_unlock();
1488}
1489
1490/**
1491 * xs_udp_write_space - callback invoked when socket buffer space
1492 * becomes available
1493 * @sk: socket whose state has changed
1494 *
1495 * Called when more output buffer space is available for this socket.
1496 * We try not to wake our writers until they can make "significant"
1497 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1498 * with a bunch of small requests.
1499 */
1500static void xs_udp_write_space(struct sock *sk)
1501{
1502 read_lock_bh(&sk->sk_callback_lock);
1503
1504 /* from net/core/sock.c:sock_def_write_space */
1505 if (sock_writeable(sk))
1506 xs_write_space(sk);
1507
1508 read_unlock_bh(&sk->sk_callback_lock);
1509}
1510
1511/**
1512 * xs_tcp_write_space - callback invoked when socket buffer space
1513 * becomes available
1514 * @sk: socket whose state has changed
1515 *
1516 * Called when more output buffer space is available for this socket.
1517 * We try not to wake our writers until they can make "significant"
1518 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1519 * with a bunch of small requests.
1520 */
1521static void xs_tcp_write_space(struct sock *sk)
1522{
1523 read_lock_bh(&sk->sk_callback_lock);
1524
1525 /* from net/core/stream.c:sk_stream_write_space */
1526 if (sk_stream_is_writeable(sk))
1527 xs_write_space(sk);
1528
1529 read_unlock_bh(&sk->sk_callback_lock);
1530}
1531
1532static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1533{
1534 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1535 struct sock *sk = transport->inet;
1536
1537 if (transport->rcvsize) {
1538 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1539 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1540 }
1541 if (transport->sndsize) {
1542 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1543 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1544 sk->sk_write_space(sk);
1545 }
1546}
1547
1548/**
1549 * xs_udp_set_buffer_size - set send and receive limits
1550 * @xprt: generic transport
1551 * @sndsize: requested size of send buffer, in bytes
1552 * @rcvsize: requested size of receive buffer, in bytes
1553 *
1554 * Set socket send and receive buffer size limits.
1555 */
1556static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1557{
1558 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1559
1560 transport->sndsize = 0;
1561 if (sndsize)
1562 transport->sndsize = sndsize + 1024;
1563 transport->rcvsize = 0;
1564 if (rcvsize)
1565 transport->rcvsize = rcvsize + 1024;
1566
1567 xs_udp_do_set_buffer_size(xprt);
1568}
1569
1570/**
1571 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1572 * @xprt: controlling transport
1573 * @task: task that timed out
1574 *
1575 * Adjust the congestion window after a retransmit timeout has occurred.
1576 */
1577static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1578{
1579 spin_lock(&xprt->transport_lock);
1580 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1581 spin_unlock(&xprt->transport_lock);
1582}
1583
1584static int xs_get_random_port(void)
1585{
1586 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1587 unsigned short range;
1588 unsigned short rand;
1589
1590 if (max < min)
1591 return -EADDRINUSE;
1592 range = max - min + 1;
1593 rand = (unsigned short) prandom_u32() % range;
1594 return rand + min;
1595}
1596
1597static unsigned short xs_sock_getport(struct socket *sock)
1598{
1599 struct sockaddr_storage buf;
1600 unsigned short port = 0;
1601
1602 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1603 goto out;
1604 switch (buf.ss_family) {
1605 case AF_INET6:
1606 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1607 break;
1608 case AF_INET:
1609 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1610 }
1611out:
1612 return port;
1613}
1614
1615/**
1616 * xs_set_port - reset the port number in the remote endpoint address
1617 * @xprt: generic transport
1618 * @port: new port number
1619 *
1620 */
1621static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1622{
1623 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1624
1625 rpc_set_port(xs_addr(xprt), port);
1626 xs_update_peer_port(xprt);
1627}
1628
1629static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1630{
1631 if (transport->srcport == 0 && transport->xprt.reuseport)
1632 transport->srcport = xs_sock_getport(sock);
1633}
1634
1635static int xs_get_srcport(struct sock_xprt *transport)
1636{
1637 int port = transport->srcport;
1638
1639 if (port == 0 && transport->xprt.resvport)
1640 port = xs_get_random_port();
1641 return port;
1642}
1643
1644static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1645{
1646 if (transport->srcport != 0)
1647 transport->srcport = 0;
1648 if (!transport->xprt.resvport)
1649 return 0;
1650 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1651 return xprt_max_resvport;
1652 return --port;
1653}
1654static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1655{
1656 struct sockaddr_storage myaddr;
1657 int err, nloop = 0;
1658 int port = xs_get_srcport(transport);
1659 unsigned short last;
1660
1661 /*
1662 * If we are asking for any ephemeral port (i.e. port == 0 &&
1663 * transport->xprt.resvport == 0), don't bind. Let the local
1664 * port selection happen implicitly when the socket is used
1665 * (for example at connect time).
1666 *
1667 * This ensures that we can continue to establish TCP
1668 * connections even when all local ephemeral ports are already
1669 * a part of some TCP connection. This makes no difference
1670 * for UDP sockets, but also doens't harm them.
1671 *
1672 * If we're asking for any reserved port (i.e. port == 0 &&
1673 * transport->xprt.resvport == 1) xs_get_srcport above will
1674 * ensure that port is non-zero and we will bind as needed.
1675 */
1676 if (port <= 0)
1677 return port;
1678
1679 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1680 do {
1681 rpc_set_port((struct sockaddr *)&myaddr, port);
1682 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1683 transport->xprt.addrlen);
1684 if (err == 0) {
1685 transport->srcport = port;
1686 break;
1687 }
1688 last = port;
1689 port = xs_next_srcport(transport, port);
1690 if (port > last)
1691 nloop++;
1692 } while (err == -EADDRINUSE && nloop != 2);
1693
1694 if (myaddr.ss_family == AF_INET)
1695 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1696 &((struct sockaddr_in *)&myaddr)->sin_addr,
1697 port, err ? "failed" : "ok", err);
1698 else
1699 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1700 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1701 port, err ? "failed" : "ok", err);
1702 return err;
1703}
1704
1705/*
1706 * We don't support autobind on AF_LOCAL sockets
1707 */
1708static void xs_local_rpcbind(struct rpc_task *task)
1709{
1710 xprt_set_bound(task->tk_xprt);
1711}
1712
1713static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1714{
1715}
1716
1717#ifdef CONFIG_DEBUG_LOCK_ALLOC
1718static struct lock_class_key xs_key[2];
1719static struct lock_class_key xs_slock_key[2];
1720
1721static inline void xs_reclassify_socketu(struct socket *sock)
1722{
1723 struct sock *sk = sock->sk;
1724
1725 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1726 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1727}
1728
1729static inline void xs_reclassify_socket4(struct socket *sock)
1730{
1731 struct sock *sk = sock->sk;
1732
1733 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1734 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1735}
1736
1737static inline void xs_reclassify_socket6(struct socket *sock)
1738{
1739 struct sock *sk = sock->sk;
1740
1741 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1742 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1743}
1744
1745static inline void xs_reclassify_socket(int family, struct socket *sock)
1746{
1747 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1748 return;
1749
1750 switch (family) {
1751 case AF_LOCAL:
1752 xs_reclassify_socketu(sock);
1753 break;
1754 case AF_INET:
1755 xs_reclassify_socket4(sock);
1756 break;
1757 case AF_INET6:
1758 xs_reclassify_socket6(sock);
1759 break;
1760 }
1761}
1762#else
1763static inline void xs_reclassify_socket(int family, struct socket *sock)
1764{
1765}
1766#endif
1767
1768static void xs_dummy_setup_socket(struct work_struct *work)
1769{
1770}
1771
1772static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1773 struct sock_xprt *transport, int family, int type,
1774 int protocol, bool reuseport)
1775{
1776 struct file *filp;
1777 struct socket *sock;
1778 int err;
1779
1780 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1781 if (err < 0) {
1782 dprintk("RPC: can't create %d transport socket (%d).\n",
1783 protocol, -err);
1784 goto out;
1785 }
1786 xs_reclassify_socket(family, sock);
1787
1788 if (reuseport)
1789 sock_set_reuseport(sock->sk);
1790
1791 err = xs_bind(transport, sock);
1792 if (err) {
1793 sock_release(sock);
1794 goto out;
1795 }
1796
1797 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1798 if (IS_ERR(filp))
1799 return ERR_CAST(filp);
1800 transport->file = filp;
1801
1802 return sock;
1803out:
1804 return ERR_PTR(err);
1805}
1806
1807static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1808 struct socket *sock)
1809{
1810 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1811 xprt);
1812
1813 if (!transport->inet) {
1814 struct sock *sk = sock->sk;
1815
1816 write_lock_bh(&sk->sk_callback_lock);
1817
1818 xs_save_old_callbacks(transport, sk);
1819
1820 sk->sk_user_data = xprt;
1821 sk->sk_data_ready = xs_data_ready;
1822 sk->sk_write_space = xs_udp_write_space;
1823 sock_set_flag(sk, SOCK_FASYNC);
1824 sk->sk_error_report = xs_error_report;
1825
1826 xprt_clear_connected(xprt);
1827
1828 /* Reset to new socket */
1829 transport->sock = sock;
1830 transport->inet = sk;
1831
1832 write_unlock_bh(&sk->sk_callback_lock);
1833 }
1834
1835 xs_stream_start_connect(transport);
1836
1837 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1838}
1839
1840/**
1841 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1842 * @transport: socket transport to connect
1843 */
1844static int xs_local_setup_socket(struct sock_xprt *transport)
1845{
1846 struct rpc_xprt *xprt = &transport->xprt;
1847 struct file *filp;
1848 struct socket *sock;
1849 int status;
1850
1851 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1852 SOCK_STREAM, 0, &sock, 1);
1853 if (status < 0) {
1854 dprintk("RPC: can't create AF_LOCAL "
1855 "transport socket (%d).\n", -status);
1856 goto out;
1857 }
1858 xs_reclassify_socket(AF_LOCAL, sock);
1859
1860 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1861 if (IS_ERR(filp)) {
1862 status = PTR_ERR(filp);
1863 goto out;
1864 }
1865 transport->file = filp;
1866
1867 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1868 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1869
1870 status = xs_local_finish_connecting(xprt, sock);
1871 trace_rpc_socket_connect(xprt, sock, status);
1872 switch (status) {
1873 case 0:
1874 dprintk("RPC: xprt %p connected to %s\n",
1875 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1876 xprt->stat.connect_count++;
1877 xprt->stat.connect_time += (long)jiffies -
1878 xprt->stat.connect_start;
1879 xprt_set_connected(xprt);
1880 case -ENOBUFS:
1881 break;
1882 case -ENOENT:
1883 dprintk("RPC: xprt %p: socket %s does not exist\n",
1884 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1885 break;
1886 case -ECONNREFUSED:
1887 dprintk("RPC: xprt %p: connection refused for %s\n",
1888 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1889 break;
1890 default:
1891 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1892 __func__, -status,
1893 xprt->address_strings[RPC_DISPLAY_ADDR]);
1894 }
1895
1896out:
1897 xprt_clear_connecting(xprt);
1898 xprt_wake_pending_tasks(xprt, status);
1899 return status;
1900}
1901
1902static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1903{
1904 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1905 int ret;
1906
1907 if (RPC_IS_ASYNC(task)) {
1908 /*
1909 * We want the AF_LOCAL connect to be resolved in the
1910 * filesystem namespace of the process making the rpc
1911 * call. Thus we connect synchronously.
1912 *
1913 * If we want to support asynchronous AF_LOCAL calls,
1914 * we'll need to figure out how to pass a namespace to
1915 * connect.
1916 */
1917 task->tk_rpc_status = -ENOTCONN;
1918 rpc_exit(task, -ENOTCONN);
1919 return;
1920 }
1921 ret = xs_local_setup_socket(transport);
1922 if (ret && !RPC_IS_SOFTCONN(task))
1923 msleep_interruptible(15000);
1924}
1925
1926#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1927/*
1928 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1929 * know that we have exclusive access to the socket), to guard against
1930 * races with xs_reset_transport.
1931 */
1932static void xs_set_memalloc(struct rpc_xprt *xprt)
1933{
1934 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 xprt);
1936
1937 /*
1938 * If there's no sock, then we have nothing to set. The
1939 * reconnecting process will get it for us.
1940 */
1941 if (!transport->inet)
1942 return;
1943 if (atomic_read(&xprt->swapper))
1944 sk_set_memalloc(transport->inet);
1945}
1946
1947/**
1948 * xs_enable_swap - Tag this transport as being used for swap.
1949 * @xprt: transport to tag
1950 *
1951 * Take a reference to this transport on behalf of the rpc_clnt, and
1952 * optionally mark it for swapping if it wasn't already.
1953 */
1954static int
1955xs_enable_swap(struct rpc_xprt *xprt)
1956{
1957 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1958
1959 if (atomic_inc_return(&xprt->swapper) != 1)
1960 return 0;
1961 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1962 return -ERESTARTSYS;
1963 if (xs->inet)
1964 sk_set_memalloc(xs->inet);
1965 xprt_release_xprt(xprt, NULL);
1966 return 0;
1967}
1968
1969/**
1970 * xs_disable_swap - Untag this transport as being used for swap.
1971 * @xprt: transport to tag
1972 *
1973 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1974 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1975 */
1976static void
1977xs_disable_swap(struct rpc_xprt *xprt)
1978{
1979 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1980
1981 if (!atomic_dec_and_test(&xprt->swapper))
1982 return;
1983 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1984 return;
1985 if (xs->inet)
1986 sk_clear_memalloc(xs->inet);
1987 xprt_release_xprt(xprt, NULL);
1988}
1989#else
1990static void xs_set_memalloc(struct rpc_xprt *xprt)
1991{
1992}
1993
1994static int
1995xs_enable_swap(struct rpc_xprt *xprt)
1996{
1997 return -EINVAL;
1998}
1999
2000static void
2001xs_disable_swap(struct rpc_xprt *xprt)
2002{
2003}
2004#endif
2005
2006static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2007{
2008 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2009
2010 if (!transport->inet) {
2011 struct sock *sk = sock->sk;
2012
2013 write_lock_bh(&sk->sk_callback_lock);
2014
2015 xs_save_old_callbacks(transport, sk);
2016
2017 sk->sk_user_data = xprt;
2018 sk->sk_data_ready = xs_data_ready;
2019 sk->sk_write_space = xs_udp_write_space;
2020 sock_set_flag(sk, SOCK_FASYNC);
2021
2022 xprt_set_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 xs_set_memalloc(xprt);
2029
2030 write_unlock_bh(&sk->sk_callback_lock);
2031 }
2032 xs_udp_do_set_buffer_size(xprt);
2033
2034 xprt->stat.connect_start = jiffies;
2035}
2036
2037static void xs_udp_setup_socket(struct work_struct *work)
2038{
2039 struct sock_xprt *transport =
2040 container_of(work, struct sock_xprt, connect_worker.work);
2041 struct rpc_xprt *xprt = &transport->xprt;
2042 struct socket *sock;
2043 int status = -EIO;
2044
2045 sock = xs_create_sock(xprt, transport,
2046 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2047 IPPROTO_UDP, false);
2048 if (IS_ERR(sock))
2049 goto out;
2050
2051 dprintk("RPC: worker connecting xprt %p via %s to "
2052 "%s (port %s)\n", xprt,
2053 xprt->address_strings[RPC_DISPLAY_PROTO],
2054 xprt->address_strings[RPC_DISPLAY_ADDR],
2055 xprt->address_strings[RPC_DISPLAY_PORT]);
2056
2057 xs_udp_finish_connecting(xprt, sock);
2058 trace_rpc_socket_connect(xprt, sock, 0);
2059 status = 0;
2060out:
2061 xprt_clear_connecting(xprt);
2062 xprt_unlock_connect(xprt, transport);
2063 xprt_wake_pending_tasks(xprt, status);
2064}
2065
2066/**
2067 * xs_tcp_shutdown - gracefully shut down a TCP socket
2068 * @xprt: transport
2069 *
2070 * Initiates a graceful shutdown of the TCP socket by calling the
2071 * equivalent of shutdown(SHUT_RDWR);
2072 */
2073static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2074{
2075 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2076 struct socket *sock = transport->sock;
2077 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2078
2079 if (sock == NULL)
2080 return;
2081 switch (skst) {
2082 default:
2083 kernel_sock_shutdown(sock, SHUT_RDWR);
2084 trace_rpc_socket_shutdown(xprt, sock);
2085 break;
2086 case TCP_CLOSE:
2087 case TCP_TIME_WAIT:
2088 xs_reset_transport(transport);
2089 }
2090}
2091
2092static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2093 struct socket *sock)
2094{
2095 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2096 unsigned int keepidle;
2097 unsigned int keepcnt;
2098 unsigned int timeo;
2099
2100 spin_lock(&xprt->transport_lock);
2101 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2102 keepcnt = xprt->timeout->to_retries + 1;
2103 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2104 (xprt->timeout->to_retries + 1);
2105 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2106 spin_unlock(&xprt->transport_lock);
2107
2108 /* TCP Keepalive options */
2109 sock_set_keepalive(sock->sk);
2110 tcp_sock_set_keepidle(sock->sk, keepidle);
2111 tcp_sock_set_keepintvl(sock->sk, keepidle);
2112 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2113
2114 /* TCP user timeout (see RFC5482) */
2115 tcp_sock_set_user_timeout(sock->sk, timeo);
2116}
2117
2118static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2119 unsigned long connect_timeout,
2120 unsigned long reconnect_timeout)
2121{
2122 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2123 struct rpc_timeout to;
2124 unsigned long initval;
2125
2126 spin_lock(&xprt->transport_lock);
2127 if (reconnect_timeout < xprt->max_reconnect_timeout)
2128 xprt->max_reconnect_timeout = reconnect_timeout;
2129 if (connect_timeout < xprt->connect_timeout) {
2130 memcpy(&to, xprt->timeout, sizeof(to));
2131 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2132 /* Arbitrary lower limit */
2133 if (initval < XS_TCP_INIT_REEST_TO << 1)
2134 initval = XS_TCP_INIT_REEST_TO << 1;
2135 to.to_initval = initval;
2136 to.to_maxval = initval;
2137 memcpy(&transport->tcp_timeout, &to,
2138 sizeof(transport->tcp_timeout));
2139 xprt->timeout = &transport->tcp_timeout;
2140 xprt->connect_timeout = connect_timeout;
2141 }
2142 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2143 spin_unlock(&xprt->transport_lock);
2144}
2145
2146static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2147{
2148 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2149 int ret = -ENOTCONN;
2150
2151 if (!transport->inet) {
2152 struct sock *sk = sock->sk;
2153
2154 /* Avoid temporary address, they are bad for long-lived
2155 * connections such as NFS mounts.
2156 * RFC4941, section 3.6 suggests that:
2157 * Individual applications, which have specific
2158 * knowledge about the normal duration of connections,
2159 * MAY override this as appropriate.
2160 */
2161 if (xs_addr(xprt)->sa_family == PF_INET6) {
2162 ip6_sock_set_addr_preferences(sk,
2163 IPV6_PREFER_SRC_PUBLIC);
2164 }
2165
2166 xs_tcp_set_socket_timeouts(xprt, sock);
2167
2168 write_lock_bh(&sk->sk_callback_lock);
2169
2170 xs_save_old_callbacks(transport, sk);
2171
2172 sk->sk_user_data = xprt;
2173 sk->sk_data_ready = xs_data_ready;
2174 sk->sk_state_change = xs_tcp_state_change;
2175 sk->sk_write_space = xs_tcp_write_space;
2176 sock_set_flag(sk, SOCK_FASYNC);
2177 sk->sk_error_report = xs_error_report;
2178
2179 /* socket options */
2180 sock_reset_flag(sk, SOCK_LINGER);
2181 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2182
2183 xprt_clear_connected(xprt);
2184
2185 /* Reset to new socket */
2186 transport->sock = sock;
2187 transport->inet = sk;
2188
2189 write_unlock_bh(&sk->sk_callback_lock);
2190 }
2191
2192 if (!xprt_bound(xprt))
2193 goto out;
2194
2195 xs_set_memalloc(xprt);
2196
2197 xs_stream_start_connect(transport);
2198
2199 /* Tell the socket layer to start connecting... */
2200 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2201 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2202 switch (ret) {
2203 case 0:
2204 xs_set_srcport(transport, sock);
2205 fallthrough;
2206 case -EINPROGRESS:
2207 /* SYN_SENT! */
2208 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2209 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2210 break;
2211 case -EADDRNOTAVAIL:
2212 /* Source port number is unavailable. Try a new one! */
2213 transport->srcport = 0;
2214 }
2215out:
2216 return ret;
2217}
2218
2219/**
2220 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2221 * @work: queued work item
2222 *
2223 * Invoked by a work queue tasklet.
2224 */
2225static void xs_tcp_setup_socket(struct work_struct *work)
2226{
2227 struct sock_xprt *transport =
2228 container_of(work, struct sock_xprt, connect_worker.work);
2229 struct socket *sock = transport->sock;
2230 struct rpc_xprt *xprt = &transport->xprt;
2231 int status = -EIO;
2232
2233 if (!sock) {
2234 sock = xs_create_sock(xprt, transport,
2235 xs_addr(xprt)->sa_family, SOCK_STREAM,
2236 IPPROTO_TCP, true);
2237 if (IS_ERR(sock)) {
2238 status = PTR_ERR(sock);
2239 goto out;
2240 }
2241 }
2242
2243 dprintk("RPC: worker connecting xprt %p via %s to "
2244 "%s (port %s)\n", xprt,
2245 xprt->address_strings[RPC_DISPLAY_PROTO],
2246 xprt->address_strings[RPC_DISPLAY_ADDR],
2247 xprt->address_strings[RPC_DISPLAY_PORT]);
2248
2249 status = xs_tcp_finish_connecting(xprt, sock);
2250 trace_rpc_socket_connect(xprt, sock, status);
2251 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2252 xprt, -status, xprt_connected(xprt),
2253 sock->sk->sk_state);
2254 switch (status) {
2255 default:
2256 printk("%s: connect returned unhandled error %d\n",
2257 __func__, status);
2258 fallthrough;
2259 case -EADDRNOTAVAIL:
2260 /* We're probably in TIME_WAIT. Get rid of existing socket,
2261 * and retry
2262 */
2263 xs_tcp_force_close(xprt);
2264 break;
2265 case 0:
2266 case -EINPROGRESS:
2267 case -EALREADY:
2268 xprt_unlock_connect(xprt, transport);
2269 return;
2270 case -EINVAL:
2271 /* Happens, for instance, if the user specified a link
2272 * local IPv6 address without a scope-id.
2273 */
2274 case -ECONNREFUSED:
2275 case -ECONNRESET:
2276 case -ENETDOWN:
2277 case -ENETUNREACH:
2278 case -EHOSTUNREACH:
2279 case -EADDRINUSE:
2280 case -ENOBUFS:
2281 /*
2282 * xs_tcp_force_close() wakes tasks with -EIO.
2283 * We need to wake them first to ensure the
2284 * correct error code.
2285 */
2286 xprt_wake_pending_tasks(xprt, status);
2287 xs_tcp_force_close(xprt);
2288 goto out;
2289 }
2290 status = -EAGAIN;
2291out:
2292 xprt_clear_connecting(xprt);
2293 xprt_unlock_connect(xprt, transport);
2294 xprt_wake_pending_tasks(xprt, status);
2295}
2296
2297/**
2298 * xs_connect - connect a socket to a remote endpoint
2299 * @xprt: pointer to transport structure
2300 * @task: address of RPC task that manages state of connect request
2301 *
2302 * TCP: If the remote end dropped the connection, delay reconnecting.
2303 *
2304 * UDP socket connects are synchronous, but we use a work queue anyway
2305 * to guarantee that even unprivileged user processes can set up a
2306 * socket on a privileged port.
2307 *
2308 * If a UDP socket connect fails, the delay behavior here prevents
2309 * retry floods (hard mounts).
2310 */
2311static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2312{
2313 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314 unsigned long delay = 0;
2315
2316 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2317
2318 if (transport->sock != NULL) {
2319 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2320 "seconds\n",
2321 xprt, xprt->reestablish_timeout / HZ);
2322
2323 /* Start by resetting any existing state */
2324 xs_reset_transport(transport);
2325
2326 delay = xprt_reconnect_delay(xprt);
2327 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2328
2329 } else
2330 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2331
2332 queue_delayed_work(xprtiod_workqueue,
2333 &transport->connect_worker,
2334 delay);
2335}
2336
2337static void xs_wake_disconnect(struct sock_xprt *transport)
2338{
2339 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2340 xs_tcp_force_close(&transport->xprt);
2341}
2342
2343static void xs_wake_write(struct sock_xprt *transport)
2344{
2345 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2346 xprt_write_space(&transport->xprt);
2347}
2348
2349static void xs_wake_error(struct sock_xprt *transport)
2350{
2351 int sockerr;
2352
2353 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2354 return;
2355 mutex_lock(&transport->recv_mutex);
2356 if (transport->sock == NULL)
2357 goto out;
2358 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2359 goto out;
2360 sockerr = xchg(&transport->xprt_err, 0);
2361 if (sockerr < 0)
2362 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2363out:
2364 mutex_unlock(&transport->recv_mutex);
2365}
2366
2367static void xs_wake_pending(struct sock_xprt *transport)
2368{
2369 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2370 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2371}
2372
2373static void xs_error_handle(struct work_struct *work)
2374{
2375 struct sock_xprt *transport = container_of(work,
2376 struct sock_xprt, error_worker);
2377
2378 xs_wake_disconnect(transport);
2379 xs_wake_write(transport);
2380 xs_wake_error(transport);
2381 xs_wake_pending(transport);
2382}
2383
2384/**
2385 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2386 * @xprt: rpc_xprt struct containing statistics
2387 * @seq: output file
2388 *
2389 */
2390static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2391{
2392 long idle_time = 0;
2393
2394 if (xprt_connected(xprt))
2395 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2396
2397 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2398 "%llu %llu %lu %llu %llu\n",
2399 xprt->stat.bind_count,
2400 xprt->stat.connect_count,
2401 xprt->stat.connect_time / HZ,
2402 idle_time,
2403 xprt->stat.sends,
2404 xprt->stat.recvs,
2405 xprt->stat.bad_xids,
2406 xprt->stat.req_u,
2407 xprt->stat.bklog_u,
2408 xprt->stat.max_slots,
2409 xprt->stat.sending_u,
2410 xprt->stat.pending_u);
2411}
2412
2413/**
2414 * xs_udp_print_stats - display UDP socket-specifc stats
2415 * @xprt: rpc_xprt struct containing statistics
2416 * @seq: output file
2417 *
2418 */
2419static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2420{
2421 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2422
2423 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2424 "%lu %llu %llu\n",
2425 transport->srcport,
2426 xprt->stat.bind_count,
2427 xprt->stat.sends,
2428 xprt->stat.recvs,
2429 xprt->stat.bad_xids,
2430 xprt->stat.req_u,
2431 xprt->stat.bklog_u,
2432 xprt->stat.max_slots,
2433 xprt->stat.sending_u,
2434 xprt->stat.pending_u);
2435}
2436
2437/**
2438 * xs_tcp_print_stats - display TCP socket-specifc stats
2439 * @xprt: rpc_xprt struct containing statistics
2440 * @seq: output file
2441 *
2442 */
2443static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2444{
2445 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2446 long idle_time = 0;
2447
2448 if (xprt_connected(xprt))
2449 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2450
2451 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2452 "%llu %llu %lu %llu %llu\n",
2453 transport->srcport,
2454 xprt->stat.bind_count,
2455 xprt->stat.connect_count,
2456 xprt->stat.connect_time / HZ,
2457 idle_time,
2458 xprt->stat.sends,
2459 xprt->stat.recvs,
2460 xprt->stat.bad_xids,
2461 xprt->stat.req_u,
2462 xprt->stat.bklog_u,
2463 xprt->stat.max_slots,
2464 xprt->stat.sending_u,
2465 xprt->stat.pending_u);
2466}
2467
2468/*
2469 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2470 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2471 * to use the server side send routines.
2472 */
2473static int bc_malloc(struct rpc_task *task)
2474{
2475 struct rpc_rqst *rqst = task->tk_rqstp;
2476 size_t size = rqst->rq_callsize;
2477 struct page *page;
2478 struct rpc_buffer *buf;
2479
2480 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2481 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2482 size);
2483 return -EINVAL;
2484 }
2485
2486 page = alloc_page(GFP_KERNEL);
2487 if (!page)
2488 return -ENOMEM;
2489
2490 buf = page_address(page);
2491 buf->len = PAGE_SIZE;
2492
2493 rqst->rq_buffer = buf->data;
2494 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2495 return 0;
2496}
2497
2498/*
2499 * Free the space allocated in the bc_alloc routine
2500 */
2501static void bc_free(struct rpc_task *task)
2502{
2503 void *buffer = task->tk_rqstp->rq_buffer;
2504 struct rpc_buffer *buf;
2505
2506 buf = container_of(buffer, struct rpc_buffer, data);
2507 free_page((unsigned long)buf);
2508}
2509
2510static int bc_sendto(struct rpc_rqst *req)
2511{
2512 struct xdr_buf *xdr = &req->rq_snd_buf;
2513 struct sock_xprt *transport =
2514 container_of(req->rq_xprt, struct sock_xprt, xprt);
2515 struct msghdr msg = {
2516 .msg_flags = 0,
2517 };
2518 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2519 (u32)xdr->len);
2520 unsigned int sent = 0;
2521 int err;
2522
2523 req->rq_xtime = ktime_get();
2524 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2525 xdr_free_bvec(xdr);
2526 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2527 return -EAGAIN;
2528 return sent;
2529}
2530
2531/**
2532 * bc_send_request - Send a backchannel Call on a TCP socket
2533 * @req: rpc_rqst containing Call message to be sent
2534 *
2535 * xpt_mutex ensures @rqstp's whole message is written to the socket
2536 * without interruption.
2537 *
2538 * Return values:
2539 * %0 if the message was sent successfully
2540 * %ENOTCONN if the message was not sent
2541 */
2542static int bc_send_request(struct rpc_rqst *req)
2543{
2544 struct svc_xprt *xprt;
2545 int len;
2546
2547 /*
2548 * Get the server socket associated with this callback xprt
2549 */
2550 xprt = req->rq_xprt->bc_xprt;
2551
2552 /*
2553 * Grab the mutex to serialize data as the connection is shared
2554 * with the fore channel
2555 */
2556 mutex_lock(&xprt->xpt_mutex);
2557 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2558 len = -ENOTCONN;
2559 else
2560 len = bc_sendto(req);
2561 mutex_unlock(&xprt->xpt_mutex);
2562
2563 if (len > 0)
2564 len = 0;
2565
2566 return len;
2567}
2568
2569/*
2570 * The close routine. Since this is client initiated, we do nothing
2571 */
2572
2573static void bc_close(struct rpc_xprt *xprt)
2574{
2575 xprt_disconnect_done(xprt);
2576}
2577
2578/*
2579 * The xprt destroy routine. Again, because this connection is client
2580 * initiated, we do nothing
2581 */
2582
2583static void bc_destroy(struct rpc_xprt *xprt)
2584{
2585 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2586
2587 xs_xprt_free(xprt);
2588 module_put(THIS_MODULE);
2589}
2590
2591static const struct rpc_xprt_ops xs_local_ops = {
2592 .reserve_xprt = xprt_reserve_xprt,
2593 .release_xprt = xprt_release_xprt,
2594 .alloc_slot = xprt_alloc_slot,
2595 .free_slot = xprt_free_slot,
2596 .rpcbind = xs_local_rpcbind,
2597 .set_port = xs_local_set_port,
2598 .connect = xs_local_connect,
2599 .buf_alloc = rpc_malloc,
2600 .buf_free = rpc_free,
2601 .prepare_request = xs_stream_prepare_request,
2602 .send_request = xs_local_send_request,
2603 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2604 .close = xs_close,
2605 .destroy = xs_destroy,
2606 .print_stats = xs_local_print_stats,
2607 .enable_swap = xs_enable_swap,
2608 .disable_swap = xs_disable_swap,
2609};
2610
2611static const struct rpc_xprt_ops xs_udp_ops = {
2612 .set_buffer_size = xs_udp_set_buffer_size,
2613 .reserve_xprt = xprt_reserve_xprt_cong,
2614 .release_xprt = xprt_release_xprt_cong,
2615 .alloc_slot = xprt_alloc_slot,
2616 .free_slot = xprt_free_slot,
2617 .rpcbind = rpcb_getport_async,
2618 .set_port = xs_set_port,
2619 .connect = xs_connect,
2620 .buf_alloc = rpc_malloc,
2621 .buf_free = rpc_free,
2622 .send_request = xs_udp_send_request,
2623 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2624 .timer = xs_udp_timer,
2625 .release_request = xprt_release_rqst_cong,
2626 .close = xs_close,
2627 .destroy = xs_destroy,
2628 .print_stats = xs_udp_print_stats,
2629 .enable_swap = xs_enable_swap,
2630 .disable_swap = xs_disable_swap,
2631 .inject_disconnect = xs_inject_disconnect,
2632};
2633
2634static const struct rpc_xprt_ops xs_tcp_ops = {
2635 .reserve_xprt = xprt_reserve_xprt,
2636 .release_xprt = xprt_release_xprt,
2637 .alloc_slot = xprt_alloc_slot,
2638 .free_slot = xprt_free_slot,
2639 .rpcbind = rpcb_getport_async,
2640 .set_port = xs_set_port,
2641 .connect = xs_connect,
2642 .buf_alloc = rpc_malloc,
2643 .buf_free = rpc_free,
2644 .prepare_request = xs_stream_prepare_request,
2645 .send_request = xs_tcp_send_request,
2646 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2647 .close = xs_tcp_shutdown,
2648 .destroy = xs_destroy,
2649 .set_connect_timeout = xs_tcp_set_connect_timeout,
2650 .print_stats = xs_tcp_print_stats,
2651 .enable_swap = xs_enable_swap,
2652 .disable_swap = xs_disable_swap,
2653 .inject_disconnect = xs_inject_disconnect,
2654#ifdef CONFIG_SUNRPC_BACKCHANNEL
2655 .bc_setup = xprt_setup_bc,
2656 .bc_maxpayload = xs_tcp_bc_maxpayload,
2657 .bc_num_slots = xprt_bc_max_slots,
2658 .bc_free_rqst = xprt_free_bc_rqst,
2659 .bc_destroy = xprt_destroy_bc,
2660#endif
2661};
2662
2663/*
2664 * The rpc_xprt_ops for the server backchannel
2665 */
2666
2667static const struct rpc_xprt_ops bc_tcp_ops = {
2668 .reserve_xprt = xprt_reserve_xprt,
2669 .release_xprt = xprt_release_xprt,
2670 .alloc_slot = xprt_alloc_slot,
2671 .free_slot = xprt_free_slot,
2672 .buf_alloc = bc_malloc,
2673 .buf_free = bc_free,
2674 .send_request = bc_send_request,
2675 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2676 .close = bc_close,
2677 .destroy = bc_destroy,
2678 .print_stats = xs_tcp_print_stats,
2679 .enable_swap = xs_enable_swap,
2680 .disable_swap = xs_disable_swap,
2681 .inject_disconnect = xs_inject_disconnect,
2682};
2683
2684static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2685{
2686 static const struct sockaddr_in sin = {
2687 .sin_family = AF_INET,
2688 .sin_addr.s_addr = htonl(INADDR_ANY),
2689 };
2690 static const struct sockaddr_in6 sin6 = {
2691 .sin6_family = AF_INET6,
2692 .sin6_addr = IN6ADDR_ANY_INIT,
2693 };
2694
2695 switch (family) {
2696 case AF_LOCAL:
2697 break;
2698 case AF_INET:
2699 memcpy(sap, &sin, sizeof(sin));
2700 break;
2701 case AF_INET6:
2702 memcpy(sap, &sin6, sizeof(sin6));
2703 break;
2704 default:
2705 dprintk("RPC: %s: Bad address family\n", __func__);
2706 return -EAFNOSUPPORT;
2707 }
2708 return 0;
2709}
2710
2711static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2712 unsigned int slot_table_size,
2713 unsigned int max_slot_table_size)
2714{
2715 struct rpc_xprt *xprt;
2716 struct sock_xprt *new;
2717
2718 if (args->addrlen > sizeof(xprt->addr)) {
2719 dprintk("RPC: xs_setup_xprt: address too large\n");
2720 return ERR_PTR(-EBADF);
2721 }
2722
2723 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2724 max_slot_table_size);
2725 if (xprt == NULL) {
2726 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2727 "rpc_xprt\n");
2728 return ERR_PTR(-ENOMEM);
2729 }
2730
2731 new = container_of(xprt, struct sock_xprt, xprt);
2732 mutex_init(&new->recv_mutex);
2733 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2734 xprt->addrlen = args->addrlen;
2735 if (args->srcaddr)
2736 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2737 else {
2738 int err;
2739 err = xs_init_anyaddr(args->dstaddr->sa_family,
2740 (struct sockaddr *)&new->srcaddr);
2741 if (err != 0) {
2742 xprt_free(xprt);
2743 return ERR_PTR(err);
2744 }
2745 }
2746
2747 return xprt;
2748}
2749
2750static const struct rpc_timeout xs_local_default_timeout = {
2751 .to_initval = 10 * HZ,
2752 .to_maxval = 10 * HZ,
2753 .to_retries = 2,
2754};
2755
2756/**
2757 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2758 * @args: rpc transport creation arguments
2759 *
2760 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2761 */
2762static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2763{
2764 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2765 struct sock_xprt *transport;
2766 struct rpc_xprt *xprt;
2767 struct rpc_xprt *ret;
2768
2769 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2770 xprt_max_tcp_slot_table_entries);
2771 if (IS_ERR(xprt))
2772 return xprt;
2773 transport = container_of(xprt, struct sock_xprt, xprt);
2774
2775 xprt->prot = 0;
2776 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2777
2778 xprt->bind_timeout = XS_BIND_TO;
2779 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2780 xprt->idle_timeout = XS_IDLE_DISC_TO;
2781
2782 xprt->ops = &xs_local_ops;
2783 xprt->timeout = &xs_local_default_timeout;
2784
2785 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2786 INIT_WORK(&transport->error_worker, xs_error_handle);
2787 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2788
2789 switch (sun->sun_family) {
2790 case AF_LOCAL:
2791 if (sun->sun_path[0] != '/') {
2792 dprintk("RPC: bad AF_LOCAL address: %s\n",
2793 sun->sun_path);
2794 ret = ERR_PTR(-EINVAL);
2795 goto out_err;
2796 }
2797 xprt_set_bound(xprt);
2798 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2799 ret = ERR_PTR(xs_local_setup_socket(transport));
2800 if (ret)
2801 goto out_err;
2802 break;
2803 default:
2804 ret = ERR_PTR(-EAFNOSUPPORT);
2805 goto out_err;
2806 }
2807
2808 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2809 xprt->address_strings[RPC_DISPLAY_ADDR]);
2810
2811 if (try_module_get(THIS_MODULE))
2812 return xprt;
2813 ret = ERR_PTR(-EINVAL);
2814out_err:
2815 xs_xprt_free(xprt);
2816 return ret;
2817}
2818
2819static const struct rpc_timeout xs_udp_default_timeout = {
2820 .to_initval = 5 * HZ,
2821 .to_maxval = 30 * HZ,
2822 .to_increment = 5 * HZ,
2823 .to_retries = 5,
2824};
2825
2826/**
2827 * xs_setup_udp - Set up transport to use a UDP socket
2828 * @args: rpc transport creation arguments
2829 *
2830 */
2831static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2832{
2833 struct sockaddr *addr = args->dstaddr;
2834 struct rpc_xprt *xprt;
2835 struct sock_xprt *transport;
2836 struct rpc_xprt *ret;
2837
2838 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2839 xprt_udp_slot_table_entries);
2840 if (IS_ERR(xprt))
2841 return xprt;
2842 transport = container_of(xprt, struct sock_xprt, xprt);
2843
2844 xprt->prot = IPPROTO_UDP;
2845 /* XXX: header size can vary due to auth type, IPv6, etc. */
2846 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2847
2848 xprt->bind_timeout = XS_BIND_TO;
2849 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2850 xprt->idle_timeout = XS_IDLE_DISC_TO;
2851
2852 xprt->ops = &xs_udp_ops;
2853
2854 xprt->timeout = &xs_udp_default_timeout;
2855
2856 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2857 INIT_WORK(&transport->error_worker, xs_error_handle);
2858 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2859
2860 switch (addr->sa_family) {
2861 case AF_INET:
2862 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2863 xprt_set_bound(xprt);
2864
2865 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2866 break;
2867 case AF_INET6:
2868 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2869 xprt_set_bound(xprt);
2870
2871 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2872 break;
2873 default:
2874 ret = ERR_PTR(-EAFNOSUPPORT);
2875 goto out_err;
2876 }
2877
2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2883 else
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2887
2888 if (try_module_get(THIS_MODULE))
2889 return xprt;
2890 ret = ERR_PTR(-EINVAL);
2891out_err:
2892 xs_xprt_free(xprt);
2893 return ret;
2894}
2895
2896static const struct rpc_timeout xs_tcp_default_timeout = {
2897 .to_initval = 60 * HZ,
2898 .to_maxval = 60 * HZ,
2899 .to_retries = 2,
2900};
2901
2902/**
2903 * xs_setup_tcp - Set up transport to use a TCP socket
2904 * @args: rpc transport creation arguments
2905 *
2906 */
2907static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2908{
2909 struct sockaddr *addr = args->dstaddr;
2910 struct rpc_xprt *xprt;
2911 struct sock_xprt *transport;
2912 struct rpc_xprt *ret;
2913 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2914
2915 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2916 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2917
2918 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2919 max_slot_table_size);
2920 if (IS_ERR(xprt))
2921 return xprt;
2922 transport = container_of(xprt, struct sock_xprt, xprt);
2923
2924 xprt->prot = IPPROTO_TCP;
2925 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2926
2927 xprt->bind_timeout = XS_BIND_TO;
2928 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2929 xprt->idle_timeout = XS_IDLE_DISC_TO;
2930
2931 xprt->ops = &xs_tcp_ops;
2932 xprt->timeout = &xs_tcp_default_timeout;
2933
2934 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2935 xprt->connect_timeout = xprt->timeout->to_initval *
2936 (xprt->timeout->to_retries + 1);
2937
2938 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2939 INIT_WORK(&transport->error_worker, xs_error_handle);
2940 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2941
2942 switch (addr->sa_family) {
2943 case AF_INET:
2944 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2945 xprt_set_bound(xprt);
2946
2947 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2948 break;
2949 case AF_INET6:
2950 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2951 xprt_set_bound(xprt);
2952
2953 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2954 break;
2955 default:
2956 ret = ERR_PTR(-EAFNOSUPPORT);
2957 goto out_err;
2958 }
2959
2960 if (xprt_bound(xprt))
2961 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2962 xprt->address_strings[RPC_DISPLAY_ADDR],
2963 xprt->address_strings[RPC_DISPLAY_PORT],
2964 xprt->address_strings[RPC_DISPLAY_PROTO]);
2965 else
2966 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2967 xprt->address_strings[RPC_DISPLAY_ADDR],
2968 xprt->address_strings[RPC_DISPLAY_PROTO]);
2969
2970 if (try_module_get(THIS_MODULE))
2971 return xprt;
2972 ret = ERR_PTR(-EINVAL);
2973out_err:
2974 xs_xprt_free(xprt);
2975 return ret;
2976}
2977
2978/**
2979 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2980 * @args: rpc transport creation arguments
2981 *
2982 */
2983static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2984{
2985 struct sockaddr *addr = args->dstaddr;
2986 struct rpc_xprt *xprt;
2987 struct sock_xprt *transport;
2988 struct svc_sock *bc_sock;
2989 struct rpc_xprt *ret;
2990
2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2992 xprt_tcp_slot_table_entries);
2993 if (IS_ERR(xprt))
2994 return xprt;
2995 transport = container_of(xprt, struct sock_xprt, xprt);
2996
2997 xprt->prot = IPPROTO_TCP;
2998 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2999 xprt->timeout = &xs_tcp_default_timeout;
3000
3001 /* backchannel */
3002 xprt_set_bound(xprt);
3003 xprt->bind_timeout = 0;
3004 xprt->reestablish_timeout = 0;
3005 xprt->idle_timeout = 0;
3006
3007 xprt->ops = &bc_tcp_ops;
3008
3009 switch (addr->sa_family) {
3010 case AF_INET:
3011 xs_format_peer_addresses(xprt, "tcp",
3012 RPCBIND_NETID_TCP);
3013 break;
3014 case AF_INET6:
3015 xs_format_peer_addresses(xprt, "tcp",
3016 RPCBIND_NETID_TCP6);
3017 break;
3018 default:
3019 ret = ERR_PTR(-EAFNOSUPPORT);
3020 goto out_err;
3021 }
3022
3023 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3024 xprt->address_strings[RPC_DISPLAY_ADDR],
3025 xprt->address_strings[RPC_DISPLAY_PORT],
3026 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028 /*
3029 * Once we've associated a backchannel xprt with a connection,
3030 * we want to keep it around as long as the connection lasts,
3031 * in case we need to start using it for a backchannel again;
3032 * this reference won't be dropped until bc_xprt is destroyed.
3033 */
3034 xprt_get(xprt);
3035 args->bc_xprt->xpt_bc_xprt = xprt;
3036 xprt->bc_xprt = args->bc_xprt;
3037 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3038 transport->sock = bc_sock->sk_sock;
3039 transport->inet = bc_sock->sk_sk;
3040
3041 /*
3042 * Since we don't want connections for the backchannel, we set
3043 * the xprt status to connected
3044 */
3045 xprt_set_connected(xprt);
3046
3047 if (try_module_get(THIS_MODULE))
3048 return xprt;
3049
3050 args->bc_xprt->xpt_bc_xprt = NULL;
3051 args->bc_xprt->xpt_bc_xps = NULL;
3052 xprt_put(xprt);
3053 ret = ERR_PTR(-EINVAL);
3054out_err:
3055 xs_xprt_free(xprt);
3056 return ret;
3057}
3058
3059static struct xprt_class xs_local_transport = {
3060 .list = LIST_HEAD_INIT(xs_local_transport.list),
3061 .name = "named UNIX socket",
3062 .owner = THIS_MODULE,
3063 .ident = XPRT_TRANSPORT_LOCAL,
3064 .setup = xs_setup_local,
3065};
3066
3067static struct xprt_class xs_udp_transport = {
3068 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3069 .name = "udp",
3070 .owner = THIS_MODULE,
3071 .ident = XPRT_TRANSPORT_UDP,
3072 .setup = xs_setup_udp,
3073};
3074
3075static struct xprt_class xs_tcp_transport = {
3076 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3077 .name = "tcp",
3078 .owner = THIS_MODULE,
3079 .ident = XPRT_TRANSPORT_TCP,
3080 .setup = xs_setup_tcp,
3081};
3082
3083static struct xprt_class xs_bc_tcp_transport = {
3084 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3085 .name = "tcp NFSv4.1 backchannel",
3086 .owner = THIS_MODULE,
3087 .ident = XPRT_TRANSPORT_BC_TCP,
3088 .setup = xs_setup_bc_tcp,
3089};
3090
3091/**
3092 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3093 *
3094 */
3095int init_socket_xprt(void)
3096{
3097 if (!sunrpc_table_header)
3098 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3099
3100 xprt_register_transport(&xs_local_transport);
3101 xprt_register_transport(&xs_udp_transport);
3102 xprt_register_transport(&xs_tcp_transport);
3103 xprt_register_transport(&xs_bc_tcp_transport);
3104
3105 return 0;
3106}
3107
3108/**
3109 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3110 *
3111 */
3112void cleanup_socket_xprt(void)
3113{
3114 if (sunrpc_table_header) {
3115 unregister_sysctl_table(sunrpc_table_header);
3116 sunrpc_table_header = NULL;
3117 }
3118
3119 xprt_unregister_transport(&xs_local_transport);
3120 xprt_unregister_transport(&xs_udp_transport);
3121 xprt_unregister_transport(&xs_tcp_transport);
3122 xprt_unregister_transport(&xs_bc_tcp_transport);
3123}
3124
3125static int param_set_uint_minmax(const char *val,
3126 const struct kernel_param *kp,
3127 unsigned int min, unsigned int max)
3128{
3129 unsigned int num;
3130 int ret;
3131
3132 if (!val)
3133 return -EINVAL;
3134 ret = kstrtouint(val, 0, &num);
3135 if (ret)
3136 return ret;
3137 if (num < min || num > max)
3138 return -EINVAL;
3139 *((unsigned int *)kp->arg) = num;
3140 return 0;
3141}
3142
3143static int param_set_portnr(const char *val, const struct kernel_param *kp)
3144{
3145 return param_set_uint_minmax(val, kp,
3146 RPC_MIN_RESVPORT,
3147 RPC_MAX_RESVPORT);
3148}
3149
3150static const struct kernel_param_ops param_ops_portnr = {
3151 .set = param_set_portnr,
3152 .get = param_get_uint,
3153};
3154
3155#define param_check_portnr(name, p) \
3156 __param_check(name, p, unsigned int);
3157
3158module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3159module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3160
3161static int param_set_slot_table_size(const char *val,
3162 const struct kernel_param *kp)
3163{
3164 return param_set_uint_minmax(val, kp,
3165 RPC_MIN_SLOT_TABLE,
3166 RPC_MAX_SLOT_TABLE);
3167}
3168
3169static const struct kernel_param_ops param_ops_slot_table_size = {
3170 .set = param_set_slot_table_size,
3171 .get = param_get_uint,
3172};
3173
3174#define param_check_slot_table_size(name, p) \
3175 __param_check(name, p, unsigned int);
3176
3177static int param_set_max_slot_table_size(const char *val,
3178 const struct kernel_param *kp)
3179{
3180 return param_set_uint_minmax(val, kp,
3181 RPC_MIN_SLOT_TABLE,
3182 RPC_MAX_SLOT_TABLE_LIMIT);
3183}
3184
3185static const struct kernel_param_ops param_ops_max_slot_table_size = {
3186 .set = param_set_max_slot_table_size,
3187 .get = param_get_uint,
3188};
3189
3190#define param_check_max_slot_table_size(name, p) \
3191 __param_check(name, p, unsigned int);
3192
3193module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3194 slot_table_size, 0644);
3195module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3196 max_slot_table_size, 0644);
3197module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3198 slot_table_size, 0644);