?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
PNP: fix "work around Dell 1536/1546 BIOS MMCONFIG bug that breaks USB"
[linux/fpc-iii.git] / net / sunrpc / xprtsock.c
blobd37f07c8179fa9da4e466327fe836c2e168ebd25
1 /*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
19 */
20
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
28 #include <linux/in.h>
29 #include <linux/net.h>
30 #include <linux/mm.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/svcsock.h>
36 #include <linux/sunrpc/xprtsock.h>
37 #include <linux/file.h>
38 #ifdef CONFIG_NFS_V4_1
39 #include <linux/sunrpc/bc_xprt.h>
40 #endif
41
42 #include <net/sock.h>
43 #include <net/checksum.h>
44 #include <net/udp.h>
45 #include <net/tcp.h>
46
47 #include "sunrpc.h"
48 /*
49 * xprtsock tunables
50 */
51 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
52 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
53
54 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
55 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
56
57 #define XS_TCP_LINGER_TO (15U * HZ)
58 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
59
60 /*
61 * We can register our own files under /proc/sys/sunrpc by
62 * calling register_sysctl_table() again. The files in that
63 * directory become the union of all files registered there.
64 *
65 * We simply need to make sure that we don't collide with
66 * someone else's file names!
67 */
68
69 #ifdef RPC_DEBUG
70
71 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
72 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
73 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
74 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
75
76 static struct ctl_table_header *sunrpc_table_header;
77
78 /*
79 * FIXME: changing the UDP slot table size should also resize the UDP
80 * socket buffers for existing UDP transports
81 */
82 static ctl_table xs_tunables_table[] = {
83 {
84 .ctl_name = CTL_SLOTTABLE_UDP,
85 .procname = "udp_slot_table_entries",
86 .data = &xprt_udp_slot_table_entries,
87 .maxlen = sizeof(unsigned int),
88 .mode = 0644,
89 .proc_handler = &proc_dointvec_minmax,
90 .strategy = &sysctl_intvec,
91 .extra1 = &min_slot_table_size,
92 .extra2 = &max_slot_table_size
93 },
94 {
95 .ctl_name = CTL_SLOTTABLE_TCP,
96 .procname = "tcp_slot_table_entries",
97 .data = &xprt_tcp_slot_table_entries,
98 .maxlen = sizeof(unsigned int),
99 .mode = 0644,
100 .proc_handler = &proc_dointvec_minmax,
101 .strategy = &sysctl_intvec,
102 .extra1 = &min_slot_table_size,
103 .extra2 = &max_slot_table_size
104 },
105 {
106 .ctl_name = CTL_MIN_RESVPORT,
107 .procname = "min_resvport",
108 .data = &xprt_min_resvport,
109 .maxlen = sizeof(unsigned int),
110 .mode = 0644,
111 .proc_handler = &proc_dointvec_minmax,
112 .strategy = &sysctl_intvec,
113 .extra1 = &xprt_min_resvport_limit,
114 .extra2 = &xprt_max_resvport_limit
115 },
116 {
117 .ctl_name = CTL_MAX_RESVPORT,
118 .procname = "max_resvport",
119 .data = &xprt_max_resvport,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = &proc_dointvec_minmax,
123 .strategy = &sysctl_intvec,
124 .extra1 = &xprt_min_resvport_limit,
125 .extra2 = &xprt_max_resvport_limit
126 },
127 {
128 .procname = "tcp_fin_timeout",
129 .data = &xs_tcp_fin_timeout,
130 .maxlen = sizeof(xs_tcp_fin_timeout),
131 .mode = 0644,
132 .proc_handler = &proc_dointvec_jiffies,
133 .strategy = sysctl_jiffies
134 },
135 {
136 .ctl_name = 0,
137 },
138 };
139
140 static ctl_table sunrpc_table[] = {
141 {
142 .ctl_name = CTL_SUNRPC,
143 .procname = "sunrpc",
144 .mode = 0555,
145 .child = xs_tunables_table
146 },
147 {
148 .ctl_name = 0,
149 },
150 };
151
152 #endif
153
154 /*
155 * Time out for an RPC UDP socket connect. UDP socket connects are
156 * synchronous, but we set a timeout anyway in case of resource
157 * exhaustion on the local host.
158 */
159 #define XS_UDP_CONN_TO (5U * HZ)
160
161 /*
162 * Wait duration for an RPC TCP connection to be established. Solaris
163 * NFS over TCP uses 60 seconds, for example, which is in line with how
164 * long a server takes to reboot.
165 */
166 #define XS_TCP_CONN_TO (60U * HZ)
167
168 /*
169 * Wait duration for a reply from the RPC portmapper.
170 */
171 #define XS_BIND_TO (60U * HZ)
172
173 /*
174 * Delay if a UDP socket connect error occurs. This is most likely some
175 * kind of resource problem on the local host.
176 */
177 #define XS_UDP_REEST_TO (2U * HZ)
178
179 /*
180 * The reestablish timeout allows clients to delay for a bit before attempting
181 * to reconnect to a server that just dropped our connection.
182 *
183 * We implement an exponential backoff when trying to reestablish a TCP
184 * transport connection with the server. Some servers like to drop a TCP
185 * connection when they are overworked, so we start with a short timeout and
186 * increase over time if the server is down or not responding.
187 */
188 #define XS_TCP_INIT_REEST_TO (3U * HZ)
189 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
190
191 /*
192 * TCP idle timeout; client drops the transport socket if it is idle
193 * for this long. Note that we also timeout UDP sockets to prevent
194 * holding port numbers when there is no RPC traffic.
195 */
196 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
197
198 #ifdef RPC_DEBUG
199 # undef RPC_DEBUG_DATA
200 # define RPCDBG_FACILITY RPCDBG_TRANS
201 #endif
202
203 #ifdef RPC_DEBUG_DATA
204 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
205 {
206 u8 *buf = (u8 *) packet;
207 int j;
208
209 dprintk("RPC: %s\n", msg);
210 for (j = 0; j < count && j < 128; j += 4) {
211 if (!(j & 31)) {
212 if (j)
213 dprintk("\n");
214 dprintk("0x%04x ", j);
215 }
216 dprintk("%02x%02x%02x%02x ",
217 buf[j], buf[j+1], buf[j+2], buf[j+3]);
218 }
219 dprintk("\n");
220 }
221 #else
222 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
223 {
224 /* NOP */
225 }
226 #endif
227
228 struct sock_xprt {
229 struct rpc_xprt xprt;
230
231 /*
232 * Network layer
233 */
234 struct socket * sock;
235 struct sock * inet;
236
237 /*
238 * State of TCP reply receive
239 */
240 __be32 tcp_fraghdr,
241 tcp_xid,
242 tcp_calldir;
243
244 u32 tcp_offset,
245 tcp_reclen;
246
247 unsigned long tcp_copied,
248 tcp_flags;
249
250 /*
251 * Connection of transports
252 */
253 struct delayed_work connect_worker;
254 struct sockaddr_storage srcaddr;
255 unsigned short srcport;
256
257 /*
258 * UDP socket buffer size parameters
259 */
260 size_t rcvsize,
261 sndsize;
262
263 /*
264 * Saved socket callback addresses
265 */
266 void (*old_data_ready)(struct sock *, int);
267 void (*old_state_change)(struct sock *);
268 void (*old_write_space)(struct sock *);
269 void (*old_error_report)(struct sock *);
270 };
271
272 /*
273 * TCP receive state flags
274 */
275 #define TCP_RCV_LAST_FRAG (1UL << 0)
276 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
277 #define TCP_RCV_COPY_XID (1UL << 2)
278 #define TCP_RCV_COPY_DATA (1UL << 3)
279 #define TCP_RCV_READ_CALLDIR (1UL << 4)
280 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
281
282 /*
283 * TCP RPC flags
284 */
285 #define TCP_RPC_REPLY (1UL << 6)
286
287 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
288 {
289 return (struct sockaddr *) &xprt->addr;
290 }
291
292 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
293 {
294 return (struct sockaddr_in *) &xprt->addr;
295 }
296
297 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
298 {
299 return (struct sockaddr_in6 *) &xprt->addr;
300 }
301
302 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
303 {
304 struct sockaddr *sap = xs_addr(xprt);
305 struct sockaddr_in6 *sin6;
306 struct sockaddr_in *sin;
307 char buf[128];
308
309 (void)rpc_ntop(sap, buf, sizeof(buf));
310 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
311
312 switch (sap->sa_family) {
313 case AF_INET:
314 sin = xs_addr_in(xprt);
315 (void)snprintf(buf, sizeof(buf), "%02x%02x%02x%02x",
316 NIPQUAD(sin->sin_addr.s_addr));
317 break;
318 case AF_INET6:
319 sin6 = xs_addr_in6(xprt);
320 (void)snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
321 break;
322 default:
323 BUG();
324 }
325 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
326 }
327
328 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
329 {
330 struct sockaddr *sap = xs_addr(xprt);
331 char buf[128];
332
333 (void)snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
334 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
335
336 (void)snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
337 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
338 }
339
340 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
341 const char *protocol,
342 const char *netid)
343 {
344 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
345 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
346 xs_format_common_peer_addresses(xprt);
347 xs_format_common_peer_ports(xprt);
348 }
349
350 static void xs_update_peer_port(struct rpc_xprt *xprt)
351 {
352 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
353 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
354
355 xs_format_common_peer_ports(xprt);
356 }
357
358 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
359 {
360 unsigned int i;
361
362 for (i = 0; i < RPC_DISPLAY_MAX; i++)
363 switch (i) {
364 case RPC_DISPLAY_PROTO:
365 case RPC_DISPLAY_NETID:
366 continue;
367 default:
368 kfree(xprt->address_strings[i]);
369 }
370 }
371
372 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
373
374 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
375 {
376 struct msghdr msg = {
377 .msg_name = addr,
378 .msg_namelen = addrlen,
379 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
380 };
381 struct kvec iov = {
382 .iov_base = vec->iov_base + base,
383 .iov_len = vec->iov_len - base,
384 };
385
386 if (iov.iov_len != 0)
387 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
388 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
389 }
390
391 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
392 {
393 struct page **ppage;
394 unsigned int remainder;
395 int err, sent = 0;
396
397 remainder = xdr->page_len - base;
398 base += xdr->page_base;
399 ppage = xdr->pages + (base >> PAGE_SHIFT);
400 base &= ~PAGE_MASK;
401 for(;;) {
402 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
403 int flags = XS_SENDMSG_FLAGS;
404
405 remainder -= len;
406 if (remainder != 0 || more)
407 flags |= MSG_MORE;
408 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
409 if (remainder == 0 || err != len)
410 break;
411 sent += err;
412 ppage++;
413 base = 0;
414 }
415 if (sent == 0)
416 return err;
417 if (err > 0)
418 sent += err;
419 return sent;
420 }
421
422 /**
423 * xs_sendpages - write pages directly to a socket
424 * @sock: socket to send on
425 * @addr: UDP only -- address of destination
426 * @addrlen: UDP only -- length of destination address
427 * @xdr: buffer containing this request
428 * @base: starting position in the buffer
429 *
430 */
431 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
432 {
433 unsigned int remainder = xdr->len - base;
434 int err, sent = 0;
435
436 if (unlikely(!sock))
437 return -ENOTSOCK;
438
439 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
440 if (base != 0) {
441 addr = NULL;
442 addrlen = 0;
443 }
444
445 if (base < xdr->head[0].iov_len || addr != NULL) {
446 unsigned int len = xdr->head[0].iov_len - base;
447 remainder -= len;
448 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
449 if (remainder == 0 || err != len)
450 goto out;
451 sent += err;
452 base = 0;
453 } else
454 base -= xdr->head[0].iov_len;
455
456 if (base < xdr->page_len) {
457 unsigned int len = xdr->page_len - base;
458 remainder -= len;
459 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
460 if (remainder == 0 || err != len)
461 goto out;
462 sent += err;
463 base = 0;
464 } else
465 base -= xdr->page_len;
466
467 if (base >= xdr->tail[0].iov_len)
468 return sent;
469 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
470 out:
471 if (sent == 0)
472 return err;
473 if (err > 0)
474 sent += err;
475 return sent;
476 }
477
478 static void xs_nospace_callback(struct rpc_task *task)
479 {
480 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
481
482 transport->inet->sk_write_pending--;
483 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
484 }
485
486 /**
487 * xs_nospace - place task on wait queue if transmit was incomplete
488 * @task: task to put to sleep
489 *
490 */
491 static int xs_nospace(struct rpc_task *task)
492 {
493 struct rpc_rqst *req = task->tk_rqstp;
494 struct rpc_xprt *xprt = req->rq_xprt;
495 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
496 int ret = -EAGAIN;
497
498 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
499 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
500 req->rq_slen);
501
502 /* Protect against races with write_space */
503 spin_lock_bh(&xprt->transport_lock);
504
505 /* Don't race with disconnect */
506 if (xprt_connected(xprt)) {
507 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
508 /*
509 * Notify TCP that we're limited by the application
510 * window size
511 */
512 set_bit(SOCK_NOSPACE, &transport->sock->flags);
513 transport->inet->sk_write_pending++;
514 /* ...and wait for more buffer space */
515 xprt_wait_for_buffer_space(task, xs_nospace_callback);
516 }
517 } else {
518 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
519 ret = -ENOTCONN;
520 }
521
522 spin_unlock_bh(&xprt->transport_lock);
523 return ret;
524 }
525
526 /**
527 * xs_udp_send_request - write an RPC request to a UDP socket
528 * @task: address of RPC task that manages the state of an RPC request
529 *
530 * Return values:
531 * 0: The request has been sent
532 * EAGAIN: The socket was blocked, please call again later to
533 * complete the request
534 * ENOTCONN: Caller needs to invoke connect logic then call again
535 * other: Some other error occured, the request was not sent
536 */
537 static int xs_udp_send_request(struct rpc_task *task)
538 {
539 struct rpc_rqst *req = task->tk_rqstp;
540 struct rpc_xprt *xprt = req->rq_xprt;
541 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
542 struct xdr_buf *xdr = &req->rq_snd_buf;
543 int status;
544
545 xs_pktdump("packet data:",
546 req->rq_svec->iov_base,
547 req->rq_svec->iov_len);
548
549 if (!xprt_bound(xprt))
550 return -ENOTCONN;
551 status = xs_sendpages(transport->sock,
552 xs_addr(xprt),
553 xprt->addrlen, xdr,
554 req->rq_bytes_sent);
555
556 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
557 xdr->len - req->rq_bytes_sent, status);
558
559 if (status >= 0) {
560 task->tk_bytes_sent += status;
561 if (status >= req->rq_slen)
562 return 0;
563 /* Still some bytes left; set up for a retry later. */
564 status = -EAGAIN;
565 }
566
567 switch (status) {
568 case -ENOTSOCK:
569 status = -ENOTCONN;
570 /* Should we call xs_close() here? */
571 break;
572 case -EAGAIN:
573 status = xs_nospace(task);
574 break;
575 default:
576 dprintk("RPC: sendmsg returned unrecognized error %d\n",
577 -status);
578 case -ENETUNREACH:
579 case -EPIPE:
580 case -ECONNREFUSED:
581 /* When the server has died, an ICMP port unreachable message
582 * prompts ECONNREFUSED. */
583 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
584 }
585
586 return status;
587 }
588
589 /**
590 * xs_tcp_shutdown - gracefully shut down a TCP socket
591 * @xprt: transport
592 *
593 * Initiates a graceful shutdown of the TCP socket by calling the
594 * equivalent of shutdown(SHUT_WR);
595 */
596 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
597 {
598 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
599 struct socket *sock = transport->sock;
600
601 if (sock != NULL)
602 kernel_sock_shutdown(sock, SHUT_WR);
603 }
604
605 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
606 {
607 u32 reclen = buf->len - sizeof(rpc_fraghdr);
608 rpc_fraghdr *base = buf->head[0].iov_base;
609 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
610 }
611
612 /**
613 * xs_tcp_send_request - write an RPC request to a TCP socket
614 * @task: address of RPC task that manages the state of an RPC request
615 *
616 * Return values:
617 * 0: The request has been sent
618 * EAGAIN: The socket was blocked, please call again later to
619 * complete the request
620 * ENOTCONN: Caller needs to invoke connect logic then call again
621 * other: Some other error occured, the request was not sent
622 *
623 * XXX: In the case of soft timeouts, should we eventually give up
624 * if sendmsg is not able to make progress?
625 */
626 static int xs_tcp_send_request(struct rpc_task *task)
627 {
628 struct rpc_rqst *req = task->tk_rqstp;
629 struct rpc_xprt *xprt = req->rq_xprt;
630 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
631 struct xdr_buf *xdr = &req->rq_snd_buf;
632 int status;
633
634 xs_encode_tcp_record_marker(&req->rq_snd_buf);
635
636 xs_pktdump("packet data:",
637 req->rq_svec->iov_base,
638 req->rq_svec->iov_len);
639
640 /* Continue transmitting the packet/record. We must be careful
641 * to cope with writespace callbacks arriving _after_ we have
642 * called sendmsg(). */
643 while (1) {
644 status = xs_sendpages(transport->sock,
645 NULL, 0, xdr, req->rq_bytes_sent);
646
647 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
648 xdr->len - req->rq_bytes_sent, status);
649
650 if (unlikely(status < 0))
651 break;
652
653 /* If we've sent the entire packet, immediately
654 * reset the count of bytes sent. */
655 req->rq_bytes_sent += status;
656 task->tk_bytes_sent += status;
657 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
658 req->rq_bytes_sent = 0;
659 return 0;
660 }
661
662 if (status != 0)
663 continue;
664 status = -EAGAIN;
665 break;
666 }
667
668 switch (status) {
669 case -ENOTSOCK:
670 status = -ENOTCONN;
671 /* Should we call xs_close() here? */
672 break;
673 case -EAGAIN:
674 status = xs_nospace(task);
675 break;
676 default:
677 dprintk("RPC: sendmsg returned unrecognized error %d\n",
678 -status);
679 case -ECONNRESET:
680 case -EPIPE:
681 xs_tcp_shutdown(xprt);
682 case -ECONNREFUSED:
683 case -ENOTCONN:
684 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
685 }
686
687 return status;
688 }
689
690 /**
691 * xs_tcp_release_xprt - clean up after a tcp transmission
692 * @xprt: transport
693 * @task: rpc task
694 *
695 * This cleans up if an error causes us to abort the transmission of a request.
696 * In this case, the socket may need to be reset in order to avoid confusing
697 * the server.
698 */
699 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
700 {
701 struct rpc_rqst *req;
702
703 if (task != xprt->snd_task)
704 return;
705 if (task == NULL)
706 goto out_release;
707 req = task->tk_rqstp;
708 if (req->rq_bytes_sent == 0)
709 goto out_release;
710 if (req->rq_bytes_sent == req->rq_snd_buf.len)
711 goto out_release;
712 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
713 out_release:
714 xprt_release_xprt(xprt, task);
715 }
716
717 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
718 {
719 transport->old_data_ready = sk->sk_data_ready;
720 transport->old_state_change = sk->sk_state_change;
721 transport->old_write_space = sk->sk_write_space;
722 transport->old_error_report = sk->sk_error_report;
723 }
724
725 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
726 {
727 sk->sk_data_ready = transport->old_data_ready;
728 sk->sk_state_change = transport->old_state_change;
729 sk->sk_write_space = transport->old_write_space;
730 sk->sk_error_report = transport->old_error_report;
731 }
732
733 static void xs_reset_transport(struct sock_xprt *transport)
734 {
735 struct socket *sock = transport->sock;
736 struct sock *sk = transport->inet;
737
738 if (sk == NULL)
739 return;
740
741 transport->srcport = 0;
742
743 write_lock_bh(&sk->sk_callback_lock);
744 transport->inet = NULL;
745 transport->sock = NULL;
746
747 sk->sk_user_data = NULL;
748
749 xs_restore_old_callbacks(transport, sk);
750 write_unlock_bh(&sk->sk_callback_lock);
751
752 sk->sk_no_check = 0;
753
754 sock_release(sock);
755 }
756
757 /**
758 * xs_close - close a socket
759 * @xprt: transport
760 *
761 * This is used when all requests are complete; ie, no DRC state remains
762 * on the server we want to save.
763 *
764 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
765 * xs_reset_transport() zeroing the socket from underneath a writer.
766 */
767 static void xs_close(struct rpc_xprt *xprt)
768 {
769 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
770
771 dprintk("RPC: xs_close xprt %p\n", xprt);
772
773 xs_reset_transport(transport);
774 xprt->reestablish_timeout = 0;
775
776 smp_mb__before_clear_bit();
777 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
778 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
779 clear_bit(XPRT_CLOSING, &xprt->state);
780 smp_mb__after_clear_bit();
781 xprt_disconnect_done(xprt);
782 }
783
784 static void xs_tcp_close(struct rpc_xprt *xprt)
785 {
786 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
787 xs_close(xprt);
788 else
789 xs_tcp_shutdown(xprt);
790 }
791
792 /**
793 * xs_destroy - prepare to shutdown a transport
794 * @xprt: doomed transport
795 *
796 */
797 static void xs_destroy(struct rpc_xprt *xprt)
798 {
799 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
800
801 dprintk("RPC: xs_destroy xprt %p\n", xprt);
802
803 cancel_rearming_delayed_work(&transport->connect_worker);
804
805 xs_close(xprt);
806 xs_free_peer_addresses(xprt);
807 kfree(xprt->slot);
808 kfree(xprt);
809 module_put(THIS_MODULE);
810 }
811
812 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
813 {
814 return (struct rpc_xprt *) sk->sk_user_data;
815 }
816
817 /**
818 * xs_udp_data_ready - "data ready" callback for UDP sockets
819 * @sk: socket with data to read
820 * @len: how much data to read
821 *
822 */
823 static void xs_udp_data_ready(struct sock *sk, int len)
824 {
825 struct rpc_task *task;
826 struct rpc_xprt *xprt;
827 struct rpc_rqst *rovr;
828 struct sk_buff *skb;
829 int err, repsize, copied;
830 u32 _xid;
831 __be32 *xp;
832
833 read_lock(&sk->sk_callback_lock);
834 dprintk("RPC: xs_udp_data_ready...\n");
835 if (!(xprt = xprt_from_sock(sk)))
836 goto out;
837
838 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
839 goto out;
840
841 if (xprt->shutdown)
842 goto dropit;
843
844 repsize = skb->len - sizeof(struct udphdr);
845 if (repsize < 4) {
846 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
847 goto dropit;
848 }
849
850 /* Copy the XID from the skb... */
851 xp = skb_header_pointer(skb, sizeof(struct udphdr),
852 sizeof(_xid), &_xid);
853 if (xp == NULL)
854 goto dropit;
855
856 /* Look up and lock the request corresponding to the given XID */
857 spin_lock(&xprt->transport_lock);
858 rovr = xprt_lookup_rqst(xprt, *xp);
859 if (!rovr)
860 goto out_unlock;
861 task = rovr->rq_task;
862
863 if ((copied = rovr->rq_private_buf.buflen) > repsize)
864 copied = repsize;
865
866 /* Suck it into the iovec, verify checksum if not done by hw. */
867 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
868 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
869 goto out_unlock;
870 }
871
872 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
873
874 /* Something worked... */
875 dst_confirm(skb_dst(skb));
876
877 xprt_adjust_cwnd(task, copied);
878 xprt_update_rtt(task);
879 xprt_complete_rqst(task, copied);
880
881 out_unlock:
882 spin_unlock(&xprt->transport_lock);
883 dropit:
884 skb_free_datagram(sk, skb);
885 out:
886 read_unlock(&sk->sk_callback_lock);
887 }
888
889 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
890 {
891 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
892 size_t len, used;
893 char *p;
894
895 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
896 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
897 used = xdr_skb_read_bits(desc, p, len);
898 transport->tcp_offset += used;
899 if (used != len)
900 return;
901
902 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
903 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
904 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
905 else
906 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
907 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
908
909 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
910 transport->tcp_offset = 0;
911
912 /* Sanity check of the record length */
913 if (unlikely(transport->tcp_reclen < 8)) {
914 dprintk("RPC: invalid TCP record fragment length\n");
915 xprt_force_disconnect(xprt);
916 return;
917 }
918 dprintk("RPC: reading TCP record fragment of length %d\n",
919 transport->tcp_reclen);
920 }
921
922 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
923 {
924 if (transport->tcp_offset == transport->tcp_reclen) {
925 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
926 transport->tcp_offset = 0;
927 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
928 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
929 transport->tcp_flags |= TCP_RCV_COPY_XID;
930 transport->tcp_copied = 0;
931 }
932 }
933 }
934
935 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
936 {
937 size_t len, used;
938 char *p;
939
940 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
941 dprintk("RPC: reading XID (%Zu bytes)\n", len);
942 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
943 used = xdr_skb_read_bits(desc, p, len);
944 transport->tcp_offset += used;
945 if (used != len)
946 return;
947 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
948 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
949 transport->tcp_copied = 4;
950 dprintk("RPC: reading %s XID %08x\n",
951 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
952 : "request with",
953 ntohl(transport->tcp_xid));
954 xs_tcp_check_fraghdr(transport);
955 }
956
957 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
958 struct xdr_skb_reader *desc)
959 {
960 size_t len, used;
961 u32 offset;
962 char *p;
963
964 /*
965 * We want transport->tcp_offset to be 8 at the end of this routine
966 * (4 bytes for the xid and 4 bytes for the call/reply flag).
967 * When this function is called for the first time,
968 * transport->tcp_offset is 4 (after having already read the xid).
969 */
970 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
971 len = sizeof(transport->tcp_calldir) - offset;
972 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
973 p = ((char *) &transport->tcp_calldir) + offset;
974 used = xdr_skb_read_bits(desc, p, len);
975 transport->tcp_offset += used;
976 if (used != len)
977 return;
978 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
979 /*
980 * We don't yet have the XDR buffer, so we will write the calldir
981 * out after we get the buffer from the 'struct rpc_rqst'
982 */
983 switch (ntohl(transport->tcp_calldir)) {
984 case RPC_REPLY:
985 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
986 transport->tcp_flags |= TCP_RCV_COPY_DATA;
987 transport->tcp_flags |= TCP_RPC_REPLY;
988 break;
989 case RPC_CALL:
990 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
991 transport->tcp_flags |= TCP_RCV_COPY_DATA;
992 transport->tcp_flags &= ~TCP_RPC_REPLY;
993 break;
994 default:
995 dprintk("RPC: invalid request message type\n");
996 xprt_force_disconnect(&transport->xprt);
997 }
998 xs_tcp_check_fraghdr(transport);
999 }
1000
1001 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1002 struct xdr_skb_reader *desc,
1003 struct rpc_rqst *req)
1004 {
1005 struct sock_xprt *transport =
1006 container_of(xprt, struct sock_xprt, xprt);
1007 struct xdr_buf *rcvbuf;
1008 size_t len;
1009 ssize_t r;
1010
1011 rcvbuf = &req->rq_private_buf;
1012
1013 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1014 /*
1015 * Save the RPC direction in the XDR buffer
1016 */
1017 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1018 &transport->tcp_calldir,
1019 sizeof(transport->tcp_calldir));
1020 transport->tcp_copied += sizeof(transport->tcp_calldir);
1021 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1022 }
1023
1024 len = desc->count;
1025 if (len > transport->tcp_reclen - transport->tcp_offset) {
1026 struct xdr_skb_reader my_desc;
1027
1028 len = transport->tcp_reclen - transport->tcp_offset;
1029 memcpy(&my_desc, desc, sizeof(my_desc));
1030 my_desc.count = len;
1031 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1032 &my_desc, xdr_skb_read_bits);
1033 desc->count -= r;
1034 desc->offset += r;
1035 } else
1036 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1037 desc, xdr_skb_read_bits);
1038
1039 if (r > 0) {
1040 transport->tcp_copied += r;
1041 transport->tcp_offset += r;
1042 }
1043 if (r != len) {
1044 /* Error when copying to the receive buffer,
1045 * usually because we weren't able to allocate
1046 * additional buffer pages. All we can do now
1047 * is turn off TCP_RCV_COPY_DATA, so the request
1048 * will not receive any additional updates,
1049 * and time out.
1050 * Any remaining data from this record will
1051 * be discarded.
1052 */
1053 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1054 dprintk("RPC: XID %08x truncated request\n",
1055 ntohl(transport->tcp_xid));
1056 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1057 "tcp_offset = %u, tcp_reclen = %u\n",
1058 xprt, transport->tcp_copied,
1059 transport->tcp_offset, transport->tcp_reclen);
1060 return;
1061 }
1062
1063 dprintk("RPC: XID %08x read %Zd bytes\n",
1064 ntohl(transport->tcp_xid), r);
1065 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1066 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1067 transport->tcp_offset, transport->tcp_reclen);
1068
1069 if (transport->tcp_copied == req->rq_private_buf.buflen)
1070 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1071 else if (transport->tcp_offset == transport->tcp_reclen) {
1072 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1073 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1074 }
1075
1076 return;
1077 }
1078
1079 /*
1080 * Finds the request corresponding to the RPC xid and invokes the common
1081 * tcp read code to read the data.
1082 */
1083 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1084 struct xdr_skb_reader *desc)
1085 {
1086 struct sock_xprt *transport =
1087 container_of(xprt, struct sock_xprt, xprt);
1088 struct rpc_rqst *req;
1089
1090 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1091
1092 /* Find and lock the request corresponding to this xid */
1093 spin_lock(&xprt->transport_lock);
1094 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1095 if (!req) {
1096 dprintk("RPC: XID %08x request not found!\n",
1097 ntohl(transport->tcp_xid));
1098 spin_unlock(&xprt->transport_lock);
1099 return -1;
1100 }
1101
1102 xs_tcp_read_common(xprt, desc, req);
1103
1104 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1105 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1106
1107 spin_unlock(&xprt->transport_lock);
1108 return 0;
1109 }
1110
1111 #if defined(CONFIG_NFS_V4_1)
1112 /*
1113 * Obtains an rpc_rqst previously allocated and invokes the common
1114 * tcp read code to read the data. The result is placed in the callback
1115 * queue.
1116 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1117 * connection and return -1.
1118 */
1119 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1120 struct xdr_skb_reader *desc)
1121 {
1122 struct sock_xprt *transport =
1123 container_of(xprt, struct sock_xprt, xprt);
1124 struct rpc_rqst *req;
1125
1126 req = xprt_alloc_bc_request(xprt);
1127 if (req == NULL) {
1128 printk(KERN_WARNING "Callback slot table overflowed\n");
1129 xprt_force_disconnect(xprt);
1130 return -1;
1131 }
1132
1133 req->rq_xid = transport->tcp_xid;
1134 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1135 xs_tcp_read_common(xprt, desc, req);
1136
1137 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1138 struct svc_serv *bc_serv = xprt->bc_serv;
1139
1140 /*
1141 * Add callback request to callback list. The callback
1142 * service sleeps on the sv_cb_waitq waiting for new
1143 * requests. Wake it up after adding enqueing the
1144 * request.
1145 */
1146 dprintk("RPC: add callback request to list\n");
1147 spin_lock(&bc_serv->sv_cb_lock);
1148 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1149 spin_unlock(&bc_serv->sv_cb_lock);
1150 wake_up(&bc_serv->sv_cb_waitq);
1151 }
1152
1153 req->rq_private_buf.len = transport->tcp_copied;
1154
1155 return 0;
1156 }
1157
1158 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1159 struct xdr_skb_reader *desc)
1160 {
1161 struct sock_xprt *transport =
1162 container_of(xprt, struct sock_xprt, xprt);
1163
1164 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1165 xs_tcp_read_reply(xprt, desc) :
1166 xs_tcp_read_callback(xprt, desc);
1167 }
1168 #else
1169 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1170 struct xdr_skb_reader *desc)
1171 {
1172 return xs_tcp_read_reply(xprt, desc);
1173 }
1174 #endif /* CONFIG_NFS_V4_1 */
1175
1176 /*
1177 * Read data off the transport. This can be either an RPC_CALL or an
1178 * RPC_REPLY. Relay the processing to helper functions.
1179 */
1180 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1181 struct xdr_skb_reader *desc)
1182 {
1183 struct sock_xprt *transport =
1184 container_of(xprt, struct sock_xprt, xprt);
1185
1186 if (_xs_tcp_read_data(xprt, desc) == 0)
1187 xs_tcp_check_fraghdr(transport);
1188 else {
1189 /*
1190 * The transport_lock protects the request handling.
1191 * There's no need to hold it to update the tcp_flags.
1192 */
1193 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1194 }
1195 }
1196
1197 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1198 {
1199 size_t len;
1200
1201 len = transport->tcp_reclen - transport->tcp_offset;
1202 if (len > desc->count)
1203 len = desc->count;
1204 desc->count -= len;
1205 desc->offset += len;
1206 transport->tcp_offset += len;
1207 dprintk("RPC: discarded %Zu bytes\n", len);
1208 xs_tcp_check_fraghdr(transport);
1209 }
1210
1211 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1212 {
1213 struct rpc_xprt *xprt = rd_desc->arg.data;
1214 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1215 struct xdr_skb_reader desc = {
1216 .skb = skb,
1217 .offset = offset,
1218 .count = len,
1219 };
1220
1221 dprintk("RPC: xs_tcp_data_recv started\n");
1222 do {
1223 /* Read in a new fragment marker if necessary */
1224 /* Can we ever really expect to get completely empty fragments? */
1225 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1226 xs_tcp_read_fraghdr(xprt, &desc);
1227 continue;
1228 }
1229 /* Read in the xid if necessary */
1230 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1231 xs_tcp_read_xid(transport, &desc);
1232 continue;
1233 }
1234 /* Read in the call/reply flag */
1235 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1236 xs_tcp_read_calldir(transport, &desc);
1237 continue;
1238 }
1239 /* Read in the request data */
1240 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1241 xs_tcp_read_data(xprt, &desc);
1242 continue;
1243 }
1244 /* Skip over any trailing bytes on short reads */
1245 xs_tcp_read_discard(transport, &desc);
1246 } while (desc.count);
1247 dprintk("RPC: xs_tcp_data_recv done\n");
1248 return len - desc.count;
1249 }
1250
1251 /**
1252 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1253 * @sk: socket with data to read
1254 * @bytes: how much data to read
1255 *
1256 */
1257 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1258 {
1259 struct rpc_xprt *xprt;
1260 read_descriptor_t rd_desc;
1261 int read;
1262
1263 dprintk("RPC: xs_tcp_data_ready...\n");
1264
1265 read_lock(&sk->sk_callback_lock);
1266 if (!(xprt = xprt_from_sock(sk)))
1267 goto out;
1268 if (xprt->shutdown)
1269 goto out;
1270
1271 /* Any data means we had a useful conversation, so
1272 * the we don't need to delay the next reconnect
1273 */
1274 if (xprt->reestablish_timeout)
1275 xprt->reestablish_timeout = 0;
1276
1277 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1278 rd_desc.arg.data = xprt;
1279 do {
1280 rd_desc.count = 65536;
1281 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1282 } while (read > 0);
1283 out:
1284 read_unlock(&sk->sk_callback_lock);
1285 }
1286
1287 /*
1288 * Do the equivalent of linger/linger2 handling for dealing with
1289 * broken servers that don't close the socket in a timely
1290 * fashion
1291 */
1292 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1293 unsigned long timeout)
1294 {
1295 struct sock_xprt *transport;
1296
1297 if (xprt_test_and_set_connecting(xprt))
1298 return;
1299 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1300 transport = container_of(xprt, struct sock_xprt, xprt);
1301 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1302 timeout);
1303 }
1304
1305 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1306 {
1307 struct sock_xprt *transport;
1308
1309 transport = container_of(xprt, struct sock_xprt, xprt);
1310
1311 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1312 !cancel_delayed_work(&transport->connect_worker))
1313 return;
1314 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1315 xprt_clear_connecting(xprt);
1316 }
1317
1318 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1319 {
1320 smp_mb__before_clear_bit();
1321 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1322 clear_bit(XPRT_CLOSING, &xprt->state);
1323 smp_mb__after_clear_bit();
1324 /* Mark transport as closed and wake up all pending tasks */
1325 xprt_disconnect_done(xprt);
1326 }
1327
1328 /**
1329 * xs_tcp_state_change - callback to handle TCP socket state changes
1330 * @sk: socket whose state has changed
1331 *
1332 */
1333 static void xs_tcp_state_change(struct sock *sk)
1334 {
1335 struct rpc_xprt *xprt;
1336
1337 read_lock(&sk->sk_callback_lock);
1338 if (!(xprt = xprt_from_sock(sk)))
1339 goto out;
1340 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1341 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1342 sk->sk_state, xprt_connected(xprt),
1343 sock_flag(sk, SOCK_DEAD),
1344 sock_flag(sk, SOCK_ZAPPED),
1345 sk->sk_shutdown);
1346
1347 switch (sk->sk_state) {
1348 case TCP_ESTABLISHED:
1349 spin_lock_bh(&xprt->transport_lock);
1350 if (!xprt_test_and_set_connected(xprt)) {
1351 struct sock_xprt *transport = container_of(xprt,
1352 struct sock_xprt, xprt);
1353
1354 /* Reset TCP record info */
1355 transport->tcp_offset = 0;
1356 transport->tcp_reclen = 0;
1357 transport->tcp_copied = 0;
1358 transport->tcp_flags =
1359 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1360
1361 xprt_wake_pending_tasks(xprt, -EAGAIN);
1362 }
1363 spin_unlock_bh(&xprt->transport_lock);
1364 break;
1365 case TCP_FIN_WAIT1:
1366 /* The client initiated a shutdown of the socket */
1367 xprt->connect_cookie++;
1368 xprt->reestablish_timeout = 0;
1369 set_bit(XPRT_CLOSING, &xprt->state);
1370 smp_mb__before_clear_bit();
1371 clear_bit(XPRT_CONNECTED, &xprt->state);
1372 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1373 smp_mb__after_clear_bit();
1374 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1375 break;
1376 case TCP_CLOSE_WAIT:
1377 /* The server initiated a shutdown of the socket */
1378 xprt_force_disconnect(xprt);
1379 xprt->connect_cookie++;
1380 case TCP_CLOSING:
1381 /*
1382 * If the server closed down the connection, make sure that
1383 * we back off before reconnecting
1384 */
1385 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1386 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1387 break;
1388 case TCP_LAST_ACK:
1389 set_bit(XPRT_CLOSING, &xprt->state);
1390 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1391 smp_mb__before_clear_bit();
1392 clear_bit(XPRT_CONNECTED, &xprt->state);
1393 smp_mb__after_clear_bit();
1394 break;
1395 case TCP_CLOSE:
1396 xs_tcp_cancel_linger_timeout(xprt);
1397 xs_sock_mark_closed(xprt);
1398 }
1399 out:
1400 read_unlock(&sk->sk_callback_lock);
1401 }
1402
1403 /**
1404 * xs_error_report - callback mainly for catching socket errors
1405 * @sk: socket
1406 */
1407 static void xs_error_report(struct sock *sk)
1408 {
1409 struct rpc_xprt *xprt;
1410
1411 read_lock(&sk->sk_callback_lock);
1412 if (!(xprt = xprt_from_sock(sk)))
1413 goto out;
1414 dprintk("RPC: %s client %p...\n"
1415 "RPC: error %d\n",
1416 __func__, xprt, sk->sk_err);
1417 xprt_wake_pending_tasks(xprt, -EAGAIN);
1418 out:
1419 read_unlock(&sk->sk_callback_lock);
1420 }
1421
1422 static void xs_write_space(struct sock *sk)
1423 {
1424 struct socket *sock;
1425 struct rpc_xprt *xprt;
1426
1427 if (unlikely(!(sock = sk->sk_socket)))
1428 return;
1429 clear_bit(SOCK_NOSPACE, &sock->flags);
1430
1431 if (unlikely(!(xprt = xprt_from_sock(sk))))
1432 return;
1433 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1434 return;
1435
1436 xprt_write_space(xprt);
1437 }
1438
1439 /**
1440 * xs_udp_write_space - callback invoked when socket buffer space
1441 * becomes available
1442 * @sk: socket whose state has changed
1443 *
1444 * Called when more output buffer space is available for this socket.
1445 * We try not to wake our writers until they can make "significant"
1446 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1447 * with a bunch of small requests.
1448 */
1449 static void xs_udp_write_space(struct sock *sk)
1450 {
1451 read_lock(&sk->sk_callback_lock);
1452
1453 /* from net/core/sock.c:sock_def_write_space */
1454 if (sock_writeable(sk))
1455 xs_write_space(sk);
1456
1457 read_unlock(&sk->sk_callback_lock);
1458 }
1459
1460 /**
1461 * xs_tcp_write_space - callback invoked when socket buffer space
1462 * becomes available
1463 * @sk: socket whose state has changed
1464 *
1465 * Called when more output buffer space is available for this socket.
1466 * We try not to wake our writers until they can make "significant"
1467 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1468 * with a bunch of small requests.
1469 */
1470 static void xs_tcp_write_space(struct sock *sk)
1471 {
1472 read_lock(&sk->sk_callback_lock);
1473
1474 /* from net/core/stream.c:sk_stream_write_space */
1475 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1476 xs_write_space(sk);
1477
1478 read_unlock(&sk->sk_callback_lock);
1479 }
1480
1481 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1482 {
1483 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1484 struct sock *sk = transport->inet;
1485
1486 if (transport->rcvsize) {
1487 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1488 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1489 }
1490 if (transport->sndsize) {
1491 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1492 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1493 sk->sk_write_space(sk);
1494 }
1495 }
1496
1497 /**
1498 * xs_udp_set_buffer_size - set send and receive limits
1499 * @xprt: generic transport
1500 * @sndsize: requested size of send buffer, in bytes
1501 * @rcvsize: requested size of receive buffer, in bytes
1502 *
1503 * Set socket send and receive buffer size limits.
1504 */
1505 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1506 {
1507 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1508
1509 transport->sndsize = 0;
1510 if (sndsize)
1511 transport->sndsize = sndsize + 1024;
1512 transport->rcvsize = 0;
1513 if (rcvsize)
1514 transport->rcvsize = rcvsize + 1024;
1515
1516 xs_udp_do_set_buffer_size(xprt);
1517 }
1518
1519 /**
1520 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1521 * @task: task that timed out
1522 *
1523 * Adjust the congestion window after a retransmit timeout has occurred.
1524 */
1525 static void xs_udp_timer(struct rpc_task *task)
1526 {
1527 xprt_adjust_cwnd(task, -ETIMEDOUT);
1528 }
1529
1530 static unsigned short xs_get_random_port(void)
1531 {
1532 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1533 unsigned short rand = (unsigned short) net_random() % range;
1534 return rand + xprt_min_resvport;
1535 }
1536
1537 /**
1538 * xs_set_port - reset the port number in the remote endpoint address
1539 * @xprt: generic transport
1540 * @port: new port number
1541 *
1542 */
1543 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1544 {
1545 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1546
1547 rpc_set_port(xs_addr(xprt), port);
1548 xs_update_peer_port(xprt);
1549 }
1550
1551 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1552 {
1553 unsigned short port = transport->srcport;
1554
1555 if (port == 0 && transport->xprt.resvport)
1556 port = xs_get_random_port();
1557 return port;
1558 }
1559
1560 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1561 {
1562 if (transport->srcport != 0)
1563 transport->srcport = 0;
1564 if (!transport->xprt.resvport)
1565 return 0;
1566 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1567 return xprt_max_resvport;
1568 return --port;
1569 }
1570
1571 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1572 {
1573 struct sockaddr_in myaddr = {
1574 .sin_family = AF_INET,
1575 };
1576 struct sockaddr_in *sa;
1577 int err, nloop = 0;
1578 unsigned short port = xs_get_srcport(transport, sock);
1579 unsigned short last;
1580
1581 sa = (struct sockaddr_in *)&transport->srcaddr;
1582 myaddr.sin_addr = sa->sin_addr;
1583 do {
1584 myaddr.sin_port = htons(port);
1585 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1586 sizeof(myaddr));
1587 if (port == 0)
1588 break;
1589 if (err == 0) {
1590 transport->srcport = port;
1591 break;
1592 }
1593 last = port;
1594 port = xs_next_srcport(transport, sock, port);
1595 if (port > last)
1596 nloop++;
1597 } while (err == -EADDRINUSE && nloop != 2);
1598 dprintk("RPC: %s %pI4:%u: %s (%d)\n",
1599 __func__, &myaddr.sin_addr,
1600 port, err ? "failed" : "ok", err);
1601 return err;
1602 }
1603
1604 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1605 {
1606 struct sockaddr_in6 myaddr = {
1607 .sin6_family = AF_INET6,
1608 };
1609 struct sockaddr_in6 *sa;
1610 int err, nloop = 0;
1611 unsigned short port = xs_get_srcport(transport, sock);
1612 unsigned short last;
1613
1614 sa = (struct sockaddr_in6 *)&transport->srcaddr;
1615 myaddr.sin6_addr = sa->sin6_addr;
1616 do {
1617 myaddr.sin6_port = htons(port);
1618 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1619 sizeof(myaddr));
1620 if (port == 0)
1621 break;
1622 if (err == 0) {
1623 transport->srcport = port;
1624 break;
1625 }
1626 last = port;
1627 port = xs_next_srcport(transport, sock, port);
1628 if (port > last)
1629 nloop++;
1630 } while (err == -EADDRINUSE && nloop != 2);
1631 dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n",
1632 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1633 return err;
1634 }
1635
1636 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1637 static struct lock_class_key xs_key[2];
1638 static struct lock_class_key xs_slock_key[2];
1639
1640 static inline void xs_reclassify_socket4(struct socket *sock)
1641 {
1642 struct sock *sk = sock->sk;
1643
1644 BUG_ON(sock_owned_by_user(sk));
1645 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1646 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1647 }
1648
1649 static inline void xs_reclassify_socket6(struct socket *sock)
1650 {
1651 struct sock *sk = sock->sk;
1652
1653 BUG_ON(sock_owned_by_user(sk));
1654 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1655 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1656 }
1657 #else
1658 static inline void xs_reclassify_socket4(struct socket *sock)
1659 {
1660 }
1661
1662 static inline void xs_reclassify_socket6(struct socket *sock)
1663 {
1664 }
1665 #endif
1666
1667 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1668 {
1669 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1670
1671 if (!transport->inet) {
1672 struct sock *sk = sock->sk;
1673
1674 write_lock_bh(&sk->sk_callback_lock);
1675
1676 xs_save_old_callbacks(transport, sk);
1677
1678 sk->sk_user_data = xprt;
1679 sk->sk_data_ready = xs_udp_data_ready;
1680 sk->sk_write_space = xs_udp_write_space;
1681 sk->sk_error_report = xs_error_report;
1682 sk->sk_no_check = UDP_CSUM_NORCV;
1683 sk->sk_allocation = GFP_ATOMIC;
1684
1685 xprt_set_connected(xprt);
1686
1687 /* Reset to new socket */
1688 transport->sock = sock;
1689 transport->inet = sk;
1690
1691 write_unlock_bh(&sk->sk_callback_lock);
1692 }
1693 xs_udp_do_set_buffer_size(xprt);
1694 }
1695
1696 /**
1697 * xs_udp_connect_worker4 - set up a UDP socket
1698 * @work: RPC transport to connect
1699 *
1700 * Invoked by a work queue tasklet.
1701 */
1702 static void xs_udp_connect_worker4(struct work_struct *work)
1703 {
1704 struct sock_xprt *transport =
1705 container_of(work, struct sock_xprt, connect_worker.work);
1706 struct rpc_xprt *xprt = &transport->xprt;
1707 struct socket *sock = transport->sock;
1708 int err, status = -EIO;
1709
1710 if (xprt->shutdown)
1711 goto out;
1712
1713 /* Start by resetting any existing state */
1714 xs_reset_transport(transport);
1715
1716 err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1717 if (err < 0) {
1718 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1719 goto out;
1720 }
1721 xs_reclassify_socket4(sock);
1722
1723 if (xs_bind4(transport, sock)) {
1724 sock_release(sock);
1725 goto out;
1726 }
1727
1728 dprintk("RPC: worker connecting xprt %p via %s to "
1729 "%s (port %s)\n", xprt,
1730 xprt->address_strings[RPC_DISPLAY_PROTO],
1731 xprt->address_strings[RPC_DISPLAY_ADDR],
1732 xprt->address_strings[RPC_DISPLAY_PORT]);
1733
1734 xs_udp_finish_connecting(xprt, sock);
1735 status = 0;
1736 out:
1737 xprt_clear_connecting(xprt);
1738 xprt_wake_pending_tasks(xprt, status);
1739 }
1740
1741 /**
1742 * xs_udp_connect_worker6 - set up a UDP socket
1743 * @work: RPC transport to connect
1744 *
1745 * Invoked by a work queue tasklet.
1746 */
1747 static void xs_udp_connect_worker6(struct work_struct *work)
1748 {
1749 struct sock_xprt *transport =
1750 container_of(work, struct sock_xprt, connect_worker.work);
1751 struct rpc_xprt *xprt = &transport->xprt;
1752 struct socket *sock = transport->sock;
1753 int err, status = -EIO;
1754
1755 if (xprt->shutdown)
1756 goto out;
1757
1758 /* Start by resetting any existing state */
1759 xs_reset_transport(transport);
1760
1761 err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1762 if (err < 0) {
1763 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1764 goto out;
1765 }
1766 xs_reclassify_socket6(sock);
1767
1768 if (xs_bind6(transport, sock) < 0) {
1769 sock_release(sock);
1770 goto out;
1771 }
1772
1773 dprintk("RPC: worker connecting xprt %p via %s to "
1774 "%s (port %s)\n", xprt,
1775 xprt->address_strings[RPC_DISPLAY_PROTO],
1776 xprt->address_strings[RPC_DISPLAY_ADDR],
1777 xprt->address_strings[RPC_DISPLAY_PORT]);
1778
1779 xs_udp_finish_connecting(xprt, sock);
1780 status = 0;
1781 out:
1782 xprt_clear_connecting(xprt);
1783 xprt_wake_pending_tasks(xprt, status);
1784 }
1785
1786 /*
1787 * We need to preserve the port number so the reply cache on the server can
1788 * find our cached RPC replies when we get around to reconnecting.
1789 */
1790 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1791 {
1792 int result;
1793 struct sockaddr any;
1794
1795 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1796
1797 /*
1798 * Disconnect the transport socket by doing a connect operation
1799 * with AF_UNSPEC. This should return immediately...
1800 */
1801 memset(&any, 0, sizeof(any));
1802 any.sa_family = AF_UNSPEC;
1803 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1804 if (!result)
1805 xs_sock_mark_closed(xprt);
1806 else
1807 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1808 result);
1809 }
1810
1811 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1812 {
1813 unsigned int state = transport->inet->sk_state;
1814
1815 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
1816 /* we don't need to abort the connection if the socket
1817 * hasn't undergone a shutdown
1818 */
1819 if (transport->inet->sk_shutdown == 0)
1820 return;
1821 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
1822 __func__, transport->inet->sk_shutdown);
1823 }
1824 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
1825 /* we don't need to abort the connection if the socket
1826 * hasn't undergone a shutdown
1827 */
1828 if (transport->inet->sk_shutdown == 0)
1829 return;
1830 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
1831 "sk_shutdown set to %d\n",
1832 __func__, transport->inet->sk_shutdown);
1833 }
1834 xs_abort_connection(xprt, transport);
1835 }
1836
1837 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1838 {
1839 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1840 int ret = -ENOTCONN;
1841
1842 if (!transport->inet) {
1843 struct sock *sk = sock->sk;
1844
1845 write_lock_bh(&sk->sk_callback_lock);
1846
1847 xs_save_old_callbacks(transport, sk);
1848
1849 sk->sk_user_data = xprt;
1850 sk->sk_data_ready = xs_tcp_data_ready;
1851 sk->sk_state_change = xs_tcp_state_change;
1852 sk->sk_write_space = xs_tcp_write_space;
1853 sk->sk_error_report = xs_error_report;
1854 sk->sk_allocation = GFP_ATOMIC;
1855
1856 /* socket options */
1857 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1858 sock_reset_flag(sk, SOCK_LINGER);
1859 tcp_sk(sk)->linger2 = 0;
1860 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1861
1862 xprt_clear_connected(xprt);
1863
1864 /* Reset to new socket */
1865 transport->sock = sock;
1866 transport->inet = sk;
1867
1868 write_unlock_bh(&sk->sk_callback_lock);
1869 }
1870
1871 if (!xprt_bound(xprt))
1872 goto out;
1873
1874 /* Tell the socket layer to start connecting... */
1875 xprt->stat.connect_count++;
1876 xprt->stat.connect_start = jiffies;
1877 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1878 switch (ret) {
1879 case 0:
1880 case -EINPROGRESS:
1881 /* SYN_SENT! */
1882 xprt->connect_cookie++;
1883 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1884 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1885 }
1886 out:
1887 return ret;
1888 }
1889
1890 /**
1891 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1892 * @xprt: RPC transport to connect
1893 * @transport: socket transport to connect
1894 * @create_sock: function to create a socket of the correct type
1895 *
1896 * Invoked by a work queue tasklet.
1897 */
1898 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1899 struct sock_xprt *transport,
1900 struct socket *(*create_sock)(struct rpc_xprt *,
1901 struct sock_xprt *))
1902 {
1903 struct socket *sock = transport->sock;
1904 int status = -EIO;
1905
1906 if (xprt->shutdown)
1907 goto out;
1908
1909 if (!sock) {
1910 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1911 sock = create_sock(xprt, transport);
1912 if (IS_ERR(sock)) {
1913 status = PTR_ERR(sock);
1914 goto out;
1915 }
1916 } else {
1917 int abort_and_exit;
1918
1919 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1920 &xprt->state);
1921 /* "close" the socket, preserving the local port */
1922 xs_tcp_reuse_connection(xprt, transport);
1923
1924 if (abort_and_exit)
1925 goto out_eagain;
1926 }
1927
1928 dprintk("RPC: worker connecting xprt %p via %s to "
1929 "%s (port %s)\n", xprt,
1930 xprt->address_strings[RPC_DISPLAY_PROTO],
1931 xprt->address_strings[RPC_DISPLAY_ADDR],
1932 xprt->address_strings[RPC_DISPLAY_PORT]);
1933
1934 status = xs_tcp_finish_connecting(xprt, sock);
1935 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1936 xprt, -status, xprt_connected(xprt),
1937 sock->sk->sk_state);
1938 switch (status) {
1939 default:
1940 printk("%s: connect returned unhandled error %d\n",
1941 __func__, status);
1942 case -EADDRNOTAVAIL:
1943 /* We're probably in TIME_WAIT. Get rid of existing socket,
1944 * and retry
1945 */
1946 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1947 xprt_force_disconnect(xprt);
1948 break;
1949 case -ECONNREFUSED:
1950 case -ECONNRESET:
1951 case -ENETUNREACH:
1952 /* retry with existing socket, after a delay */
1953 case 0:
1954 case -EINPROGRESS:
1955 case -EALREADY:
1956 xprt_clear_connecting(xprt);
1957 return;
1958 case -EINVAL:
1959 /* Happens, for instance, if the user specified a link
1960 * local IPv6 address without a scope-id.
1961 */
1962 goto out;
1963 }
1964 out_eagain:
1965 status = -EAGAIN;
1966 out:
1967 xprt_clear_connecting(xprt);
1968 xprt_wake_pending_tasks(xprt, status);
1969 }
1970
1971 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1972 struct sock_xprt *transport)
1973 {
1974 struct socket *sock;
1975 int err;
1976
1977 /* start from scratch */
1978 err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1979 if (err < 0) {
1980 dprintk("RPC: can't create TCP transport socket (%d).\n",
1981 -err);
1982 goto out_err;
1983 }
1984 xs_reclassify_socket4(sock);
1985
1986 if (xs_bind4(transport, sock) < 0) {
1987 sock_release(sock);
1988 goto out_err;
1989 }
1990 return sock;
1991 out_err:
1992 return ERR_PTR(-EIO);
1993 }
1994
1995 /**
1996 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1997 * @work: RPC transport to connect
1998 *
1999 * Invoked by a work queue tasklet.
2000 */
2001 static void xs_tcp_connect_worker4(struct work_struct *work)
2002 {
2003 struct sock_xprt *transport =
2004 container_of(work, struct sock_xprt, connect_worker.work);
2005 struct rpc_xprt *xprt = &transport->xprt;
2006
2007 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
2008 }
2009
2010 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
2011 struct sock_xprt *transport)
2012 {
2013 struct socket *sock;
2014 int err;
2015
2016 /* start from scratch */
2017 err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
2018 if (err < 0) {
2019 dprintk("RPC: can't create TCP transport socket (%d).\n",
2020 -err);
2021 goto out_err;
2022 }
2023 xs_reclassify_socket6(sock);
2024
2025 if (xs_bind6(transport, sock) < 0) {
2026 sock_release(sock);
2027 goto out_err;
2028 }
2029 return sock;
2030 out_err:
2031 return ERR_PTR(-EIO);
2032 }
2033
2034 /**
2035 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
2036 * @work: RPC transport to connect
2037 *
2038 * Invoked by a work queue tasklet.
2039 */
2040 static void xs_tcp_connect_worker6(struct work_struct *work)
2041 {
2042 struct sock_xprt *transport =
2043 container_of(work, struct sock_xprt, connect_worker.work);
2044 struct rpc_xprt *xprt = &transport->xprt;
2045
2046 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
2047 }
2048
2049 /**
2050 * xs_connect - connect a socket to a remote endpoint
2051 * @task: address of RPC task that manages state of connect request
2052 *
2053 * TCP: If the remote end dropped the connection, delay reconnecting.
2054 *
2055 * UDP socket connects are synchronous, but we use a work queue anyway
2056 * to guarantee that even unprivileged user processes can set up a
2057 * socket on a privileged port.
2058 *
2059 * If a UDP socket connect fails, the delay behavior here prevents
2060 * retry floods (hard mounts).
2061 */
2062 static void xs_connect(struct rpc_task *task)
2063 {
2064 struct rpc_xprt *xprt = task->tk_xprt;
2065 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2066
2067 if (xprt_test_and_set_connecting(xprt))
2068 return;
2069
2070 if (transport->sock != NULL) {
2071 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2072 "seconds\n",
2073 xprt, xprt->reestablish_timeout / HZ);
2074 queue_delayed_work(rpciod_workqueue,
2075 &transport->connect_worker,
2076 xprt->reestablish_timeout);
2077 xprt->reestablish_timeout <<= 1;
2078 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2079 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2080 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2081 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2082 } else {
2083 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2084 queue_delayed_work(rpciod_workqueue,
2085 &transport->connect_worker, 0);
2086 }
2087 }
2088
2089 static void xs_tcp_connect(struct rpc_task *task)
2090 {
2091 struct rpc_xprt *xprt = task->tk_xprt;
2092
2093 /* Exit if we need to wait for socket shutdown to complete */
2094 if (test_bit(XPRT_CLOSING, &xprt->state))
2095 return;
2096 xs_connect(task);
2097 }
2098
2099 /**
2100 * xs_udp_print_stats - display UDP socket-specifc stats
2101 * @xprt: rpc_xprt struct containing statistics
2102 * @seq: output file
2103 *
2104 */
2105 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2106 {
2107 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2108
2109 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2110 transport->srcport,
2111 xprt->stat.bind_count,
2112 xprt->stat.sends,
2113 xprt->stat.recvs,
2114 xprt->stat.bad_xids,
2115 xprt->stat.req_u,
2116 xprt->stat.bklog_u);
2117 }
2118
2119 /**
2120 * xs_tcp_print_stats - display TCP socket-specifc stats
2121 * @xprt: rpc_xprt struct containing statistics
2122 * @seq: output file
2123 *
2124 */
2125 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2126 {
2127 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2128 long idle_time = 0;
2129
2130 if (xprt_connected(xprt))
2131 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2132
2133 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2134 transport->srcport,
2135 xprt->stat.bind_count,
2136 xprt->stat.connect_count,
2137 xprt->stat.connect_time,
2138 idle_time,
2139 xprt->stat.sends,
2140 xprt->stat.recvs,
2141 xprt->stat.bad_xids,
2142 xprt->stat.req_u,
2143 xprt->stat.bklog_u);
2144 }
2145
2146 /*
2147 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2148 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2149 * to use the server side send routines.
2150 */
2151 void *bc_malloc(struct rpc_task *task, size_t size)
2152 {
2153 struct page *page;
2154 struct rpc_buffer *buf;
2155
2156 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2157 page = alloc_page(GFP_KERNEL);
2158
2159 if (!page)
2160 return NULL;
2161
2162 buf = page_address(page);
2163 buf->len = PAGE_SIZE;
2164
2165 return buf->data;
2166 }
2167
2168 /*
2169 * Free the space allocated in the bc_alloc routine
2170 */
2171 void bc_free(void *buffer)
2172 {
2173 struct rpc_buffer *buf;
2174
2175 if (!buffer)
2176 return;
2177
2178 buf = container_of(buffer, struct rpc_buffer, data);
2179 free_page((unsigned long)buf);
2180 }
2181
2182 /*
2183 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2184 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2185 */
2186 static int bc_sendto(struct rpc_rqst *req)
2187 {
2188 int len;
2189 struct xdr_buf *xbufp = &req->rq_snd_buf;
2190 struct rpc_xprt *xprt = req->rq_xprt;
2191 struct sock_xprt *transport =
2192 container_of(xprt, struct sock_xprt, xprt);
2193 struct socket *sock = transport->sock;
2194 unsigned long headoff;
2195 unsigned long tailoff;
2196
2197 /*
2198 * Set up the rpc header and record marker stuff
2199 */
2200 xs_encode_tcp_record_marker(xbufp);
2201
2202 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2203 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2204 len = svc_send_common(sock, xbufp,
2205 virt_to_page(xbufp->head[0].iov_base), headoff,
2206 xbufp->tail[0].iov_base, tailoff);
2207
2208 if (len != xbufp->len) {
2209 printk(KERN_NOTICE "Error sending entire callback!\n");
2210 len = -EAGAIN;
2211 }
2212
2213 return len;
2214 }
2215
2216 /*
2217 * The send routine. Borrows from svc_send
2218 */
2219 static int bc_send_request(struct rpc_task *task)
2220 {
2221 struct rpc_rqst *req = task->tk_rqstp;
2222 struct svc_xprt *xprt;
2223 struct svc_sock *svsk;
2224 u32 len;
2225
2226 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2227 /*
2228 * Get the server socket associated with this callback xprt
2229 */
2230 xprt = req->rq_xprt->bc_xprt;
2231 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2232
2233 /*
2234 * Grab the mutex to serialize data as the connection is shared
2235 * with the fore channel
2236 */
2237 if (!mutex_trylock(&xprt->xpt_mutex)) {
2238 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2239 if (!mutex_trylock(&xprt->xpt_mutex))
2240 return -EAGAIN;
2241 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2242 }
2243 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2244 len = -ENOTCONN;
2245 else
2246 len = bc_sendto(req);
2247 mutex_unlock(&xprt->xpt_mutex);
2248
2249 if (len > 0)
2250 len = 0;
2251
2252 return len;
2253 }
2254
2255 /*
2256 * The close routine. Since this is client initiated, we do nothing
2257 */
2258
2259 static void bc_close(struct rpc_xprt *xprt)
2260 {
2261 return;
2262 }
2263
2264 /*
2265 * The xprt destroy routine. Again, because this connection is client
2266 * initiated, we do nothing
2267 */
2268
2269 static void bc_destroy(struct rpc_xprt *xprt)
2270 {
2271 return;
2272 }
2273
2274 static struct rpc_xprt_ops xs_udp_ops = {
2275 .set_buffer_size = xs_udp_set_buffer_size,
2276 .reserve_xprt = xprt_reserve_xprt_cong,
2277 .release_xprt = xprt_release_xprt_cong,
2278 .rpcbind = rpcb_getport_async,
2279 .set_port = xs_set_port,
2280 .connect = xs_connect,
2281 .buf_alloc = rpc_malloc,
2282 .buf_free = rpc_free,
2283 .send_request = xs_udp_send_request,
2284 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2285 .timer = xs_udp_timer,
2286 .release_request = xprt_release_rqst_cong,
2287 .close = xs_close,
2288 .destroy = xs_destroy,
2289 .print_stats = xs_udp_print_stats,
2290 };
2291
2292 static struct rpc_xprt_ops xs_tcp_ops = {
2293 .reserve_xprt = xprt_reserve_xprt,
2294 .release_xprt = xs_tcp_release_xprt,
2295 .rpcbind = rpcb_getport_async,
2296 .set_port = xs_set_port,
2297 .connect = xs_tcp_connect,
2298 .buf_alloc = rpc_malloc,
2299 .buf_free = rpc_free,
2300 .send_request = xs_tcp_send_request,
2301 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2302 #if defined(CONFIG_NFS_V4_1)
2303 .release_request = bc_release_request,
2304 #endif /* CONFIG_NFS_V4_1 */
2305 .close = xs_tcp_close,
2306 .destroy = xs_destroy,
2307 .print_stats = xs_tcp_print_stats,
2308 };
2309
2310 /*
2311 * The rpc_xprt_ops for the server backchannel
2312 */
2313
2314 static struct rpc_xprt_ops bc_tcp_ops = {
2315 .reserve_xprt = xprt_reserve_xprt,
2316 .release_xprt = xprt_release_xprt,
2317 .buf_alloc = bc_malloc,
2318 .buf_free = bc_free,
2319 .send_request = bc_send_request,
2320 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2321 .close = bc_close,
2322 .destroy = bc_destroy,
2323 .print_stats = xs_tcp_print_stats,
2324 };
2325
2326 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2327 unsigned int slot_table_size)
2328 {
2329 struct rpc_xprt *xprt;
2330 struct sock_xprt *new;
2331
2332 if (args->addrlen > sizeof(xprt->addr)) {
2333 dprintk("RPC: xs_setup_xprt: address too large\n");
2334 return ERR_PTR(-EBADF);
2335 }
2336
2337 new = kzalloc(sizeof(*new), GFP_KERNEL);
2338 if (new == NULL) {
2339 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2340 "rpc_xprt\n");
2341 return ERR_PTR(-ENOMEM);
2342 }
2343 xprt = &new->xprt;
2344
2345 xprt->max_reqs = slot_table_size;
2346 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2347 if (xprt->slot == NULL) {
2348 kfree(xprt);
2349 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
2350 "table\n");
2351 return ERR_PTR(-ENOMEM);
2352 }
2353
2354 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2355 xprt->addrlen = args->addrlen;
2356 if (args->srcaddr)
2357 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2358
2359 return xprt;
2360 }
2361
2362 static const struct rpc_timeout xs_udp_default_timeout = {
2363 .to_initval = 5 * HZ,
2364 .to_maxval = 30 * HZ,
2365 .to_increment = 5 * HZ,
2366 .to_retries = 5,
2367 };
2368
2369 /**
2370 * xs_setup_udp - Set up transport to use a UDP socket
2371 * @args: rpc transport creation arguments
2372 *
2373 */
2374 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2375 {
2376 struct sockaddr *addr = args->dstaddr;
2377 struct rpc_xprt *xprt;
2378 struct sock_xprt *transport;
2379
2380 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2381 if (IS_ERR(xprt))
2382 return xprt;
2383 transport = container_of(xprt, struct sock_xprt, xprt);
2384
2385 xprt->prot = IPPROTO_UDP;
2386 xprt->tsh_size = 0;
2387 /* XXX: header size can vary due to auth type, IPv6, etc. */
2388 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2389
2390 xprt->bind_timeout = XS_BIND_TO;
2391 xprt->connect_timeout = XS_UDP_CONN_TO;
2392 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2393 xprt->idle_timeout = XS_IDLE_DISC_TO;
2394
2395 xprt->ops = &xs_udp_ops;
2396
2397 xprt->timeout = &xs_udp_default_timeout;
2398
2399 switch (addr->sa_family) {
2400 case AF_INET:
2401 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2402 xprt_set_bound(xprt);
2403
2404 INIT_DELAYED_WORK(&transport->connect_worker,
2405 xs_udp_connect_worker4);
2406 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2407 break;
2408 case AF_INET6:
2409 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2410 xprt_set_bound(xprt);
2411
2412 INIT_DELAYED_WORK(&transport->connect_worker,
2413 xs_udp_connect_worker6);
2414 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2415 break;
2416 default:
2417 kfree(xprt);
2418 return ERR_PTR(-EAFNOSUPPORT);
2419 }
2420
2421 if (xprt_bound(xprt))
2422 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2423 xprt->address_strings[RPC_DISPLAY_ADDR],
2424 xprt->address_strings[RPC_DISPLAY_PORT],
2425 xprt->address_strings[RPC_DISPLAY_PROTO]);
2426 else
2427 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2428 xprt->address_strings[RPC_DISPLAY_ADDR],
2429 xprt->address_strings[RPC_DISPLAY_PROTO]);
2430
2431 if (try_module_get(THIS_MODULE))
2432 return xprt;
2433
2434 kfree(xprt->slot);
2435 kfree(xprt);
2436 return ERR_PTR(-EINVAL);
2437 }
2438
2439 static const struct rpc_timeout xs_tcp_default_timeout = {
2440 .to_initval = 60 * HZ,
2441 .to_maxval = 60 * HZ,
2442 .to_retries = 2,
2443 };
2444
2445 /**
2446 * xs_setup_tcp - Set up transport to use a TCP socket
2447 * @args: rpc transport creation arguments
2448 *
2449 */
2450 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2451 {
2452 struct sockaddr *addr = args->dstaddr;
2453 struct rpc_xprt *xprt;
2454 struct sock_xprt *transport;
2455
2456 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2457 if (IS_ERR(xprt))
2458 return xprt;
2459 transport = container_of(xprt, struct sock_xprt, xprt);
2460
2461 xprt->prot = IPPROTO_TCP;
2462 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2463 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2464
2465 xprt->bind_timeout = XS_BIND_TO;
2466 xprt->connect_timeout = XS_TCP_CONN_TO;
2467 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2468 xprt->idle_timeout = XS_IDLE_DISC_TO;
2469
2470 xprt->ops = &xs_tcp_ops;
2471 xprt->timeout = &xs_tcp_default_timeout;
2472
2473 switch (addr->sa_family) {
2474 case AF_INET:
2475 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2476 xprt_set_bound(xprt);
2477
2478 INIT_DELAYED_WORK(&transport->connect_worker,
2479 xs_tcp_connect_worker4);
2480 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2481 break;
2482 case AF_INET6:
2483 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2484 xprt_set_bound(xprt);
2485
2486 INIT_DELAYED_WORK(&transport->connect_worker,
2487 xs_tcp_connect_worker6);
2488 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2489 break;
2490 default:
2491 kfree(xprt);
2492 return ERR_PTR(-EAFNOSUPPORT);
2493 }
2494
2495 if (xprt_bound(xprt))
2496 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2497 xprt->address_strings[RPC_DISPLAY_ADDR],
2498 xprt->address_strings[RPC_DISPLAY_PORT],
2499 xprt->address_strings[RPC_DISPLAY_PROTO]);
2500 else
2501 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2502 xprt->address_strings[RPC_DISPLAY_ADDR],
2503 xprt->address_strings[RPC_DISPLAY_PROTO]);
2504
2505
2506 if (try_module_get(THIS_MODULE))
2507 return xprt;
2508
2509 kfree(xprt->slot);
2510 kfree(xprt);
2511 return ERR_PTR(-EINVAL);
2512 }
2513
2514 /**
2515 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2516 * @args: rpc transport creation arguments
2517 *
2518 */
2519 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2520 {
2521 struct sockaddr *addr = args->dstaddr;
2522 struct rpc_xprt *xprt;
2523 struct sock_xprt *transport;
2524 struct svc_sock *bc_sock;
2525
2526 if (!args->bc_xprt)
2527 ERR_PTR(-EINVAL);
2528
2529 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2530 if (IS_ERR(xprt))
2531 return xprt;
2532 transport = container_of(xprt, struct sock_xprt, xprt);
2533
2534 xprt->prot = IPPROTO_TCP;
2535 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2536 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2537 xprt->timeout = &xs_tcp_default_timeout;
2538
2539 /* backchannel */
2540 xprt_set_bound(xprt);
2541 xprt->bind_timeout = 0;
2542 xprt->connect_timeout = 0;
2543 xprt->reestablish_timeout = 0;
2544 xprt->idle_timeout = 0;
2545
2546 /*
2547 * The backchannel uses the same socket connection as the
2548 * forechannel
2549 */
2550 xprt->bc_xprt = args->bc_xprt;
2551 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2552 bc_sock->sk_bc_xprt = xprt;
2553 transport->sock = bc_sock->sk_sock;
2554 transport->inet = bc_sock->sk_sk;
2555
2556 xprt->ops = &bc_tcp_ops;
2557
2558 switch (addr->sa_family) {
2559 case AF_INET:
2560 xs_format_peer_addresses(xprt, "tcp",
2561 RPCBIND_NETID_TCP);
2562 break;
2563 case AF_INET6:
2564 xs_format_peer_addresses(xprt, "tcp",
2565 RPCBIND_NETID_TCP6);
2566 break;
2567 default:
2568 kfree(xprt);
2569 return ERR_PTR(-EAFNOSUPPORT);
2570 }
2571
2572 if (xprt_bound(xprt))
2573 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2574 xprt->address_strings[RPC_DISPLAY_ADDR],
2575 xprt->address_strings[RPC_DISPLAY_PORT],
2576 xprt->address_strings[RPC_DISPLAY_PROTO]);
2577 else
2578 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2579 xprt->address_strings[RPC_DISPLAY_ADDR],
2580 xprt->address_strings[RPC_DISPLAY_PROTO]);
2581
2582 /*
2583 * Since we don't want connections for the backchannel, we set
2584 * the xprt status to connected
2585 */
2586 xprt_set_connected(xprt);
2587
2588
2589 if (try_module_get(THIS_MODULE))
2590 return xprt;
2591 kfree(xprt->slot);
2592 kfree(xprt);
2593 return ERR_PTR(-EINVAL);
2594 }
2595
2596 static struct xprt_class xs_udp_transport = {
2597 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2598 .name = "udp",
2599 .owner = THIS_MODULE,
2600 .ident = XPRT_TRANSPORT_UDP,
2601 .setup = xs_setup_udp,
2602 };
2603
2604 static struct xprt_class xs_tcp_transport = {
2605 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2606 .name = "tcp",
2607 .owner = THIS_MODULE,
2608 .ident = XPRT_TRANSPORT_TCP,
2609 .setup = xs_setup_tcp,
2610 };
2611
2612 static struct xprt_class xs_bc_tcp_transport = {
2613 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2614 .name = "tcp NFSv4.1 backchannel",
2615 .owner = THIS_MODULE,
2616 .ident = XPRT_TRANSPORT_BC_TCP,
2617 .setup = xs_setup_bc_tcp,
2618 };
2619
2620 /**
2621 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2622 *
2623 */
2624 int init_socket_xprt(void)
2625 {
2626 #ifdef RPC_DEBUG
2627 if (!sunrpc_table_header)
2628 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2629 #endif
2630
2631 xprt_register_transport(&xs_udp_transport);
2632 xprt_register_transport(&xs_tcp_transport);
2633 xprt_register_transport(&xs_bc_tcp_transport);
2634
2635 return 0;
2636 }
2637
2638 /**
2639 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2640 *
2641 */
2642 void cleanup_socket_xprt(void)
2643 {
2644 #ifdef RPC_DEBUG
2645 if (sunrpc_table_header) {
2646 unregister_sysctl_table(sunrpc_table_header);
2647 sunrpc_table_header = NULL;
2648 }
2649 #endif
2650
2651 xprt_unregister_transport(&xs_udp_transport);
2652 xprt_unregister_transport(&xs_tcp_transport);
2653 xprt_unregister_transport(&xs_bc_tcp_transport);
2654 }
2655
2656 static int param_set_uint_minmax(const char *val, struct kernel_param *kp,
2657 unsigned int min, unsigned int max)
2658 {
2659 unsigned long num;
2660 int ret;
2661
2662 if (!val)
2663 return -EINVAL;
2664 ret = strict_strtoul(val, 0, &num);
2665 if (ret == -EINVAL || num < min || num > max)
2666 return -EINVAL;
2667 *((unsigned int *)kp->arg) = num;
2668 return 0;
2669 }
2670
2671 static int param_set_portnr(const char *val, struct kernel_param *kp)
2672 {
2673 return param_set_uint_minmax(val, kp,
2674 RPC_MIN_RESVPORT,
2675 RPC_MAX_RESVPORT);
2676 }
2677
2678 static int param_get_portnr(char *buffer, struct kernel_param *kp)
2679 {
2680 return param_get_uint(buffer, kp);
2681 }
2682 #define param_check_portnr(name, p) \
2683 __param_check(name, p, unsigned int);
2684
2685 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2686 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2687
2688 static int param_set_slot_table_size(const char *val, struct kernel_param *kp)
2689 {
2690 return param_set_uint_minmax(val, kp,
2691 RPC_MIN_SLOT_TABLE,
2692 RPC_MAX_SLOT_TABLE);
2693 }
2694
2695 static int param_get_slot_table_size(char *buffer, struct kernel_param *kp)
2696 {
2697 return param_get_uint(buffer, kp);
2698 }
2699 #define param_check_slot_table_size(name, p) \
2700 __param_check(name, p, unsigned int);
2701
2702 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2703 slot_table_size, 0644);
2704 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2705 slot_table_size, 0644);
2706
Linux with added FPC-III support