?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ARM: OMAP: GPIO: IRQSTATUS2 workaround for retention state
[linux-2.6/kvm.git] / net / sunrpc / xprtsock.c
blob9b62923a9c0689fd75569fb494e66604f06ce4e7
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 Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
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
18 #include <linux/types.h>
19 #include <linux/slab.h>
20 #include <linux/capability.h>
21 #include <linux/sched.h>
22 #include <linux/pagemap.h>
23 #include <linux/errno.h>
24 #include <linux/socket.h>
25 #include <linux/in.h>
26 #include <linux/net.h>
27 #include <linux/mm.h>
28 #include <linux/udp.h>
29 #include <linux/tcp.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/sunrpc/sched.h>
32 #include <linux/file.h>
33
34 #include <net/sock.h>
35 #include <net/checksum.h>
36 #include <net/udp.h>
37 #include <net/tcp.h>
38
39 /*
40 * xprtsock tunables
41 */
42 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
43 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
44
45 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
46 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
47
48 /*
49 * How many times to try sending a request on a socket before waiting
50 * for the socket buffer to clear.
51 */
52 #define XS_SENDMSG_RETRY (10U)
53
54 /*
55 * Time out for an RPC UDP socket connect. UDP socket connects are
56 * synchronous, but we set a timeout anyway in case of resource
57 * exhaustion on the local host.
58 */
59 #define XS_UDP_CONN_TO (5U * HZ)
60
61 /*
62 * Wait duration for an RPC TCP connection to be established. Solaris
63 * NFS over TCP uses 60 seconds, for example, which is in line with how
64 * long a server takes to reboot.
65 */
66 #define XS_TCP_CONN_TO (60U * HZ)
67
68 /*
69 * Wait duration for a reply from the RPC portmapper.
70 */
71 #define XS_BIND_TO (60U * HZ)
72
73 /*
74 * Delay if a UDP socket connect error occurs. This is most likely some
75 * kind of resource problem on the local host.
76 */
77 #define XS_UDP_REEST_TO (2U * HZ)
78
79 /*
80 * The reestablish timeout allows clients to delay for a bit before attempting
81 * to reconnect to a server that just dropped our connection.
82 *
83 * We implement an exponential backoff when trying to reestablish a TCP
84 * transport connection with the server. Some servers like to drop a TCP
85 * connection when they are overworked, so we start with a short timeout and
86 * increase over time if the server is down or not responding.
87 */
88 #define XS_TCP_INIT_REEST_TO (3U * HZ)
89 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
90
91 /*
92 * TCP idle timeout; client drops the transport socket if it is idle
93 * for this long. Note that we also timeout UDP sockets to prevent
94 * holding port numbers when there is no RPC traffic.
95 */
96 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
97
98 #ifdef RPC_DEBUG
99 # undef RPC_DEBUG_DATA
100 # define RPCDBG_FACILITY RPCDBG_TRANS
101 #endif
102
103 #ifdef RPC_DEBUG_DATA
104 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
105 {
106 u8 *buf = (u8 *) packet;
107 int j;
108
109 dprintk("RPC: %s\n", msg);
110 for (j = 0; j < count && j < 128; j += 4) {
111 if (!(j & 31)) {
112 if (j)
113 dprintk("\n");
114 dprintk("0x%04x ", j);
115 }
116 dprintk("%02x%02x%02x%02x ",
117 buf[j], buf[j+1], buf[j+2], buf[j+3]);
118 }
119 dprintk("\n");
120 }
121 #else
122 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
123 {
124 /* NOP */
125 }
126 #endif
127
128 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
129 {
130 struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
131 char *buf;
132
133 buf = kzalloc(20, GFP_KERNEL);
134 if (buf) {
135 snprintf(buf, 20, "%u.%u.%u.%u",
136 NIPQUAD(addr->sin_addr.s_addr));
137 }
138 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
139
140 buf = kzalloc(8, GFP_KERNEL);
141 if (buf) {
142 snprintf(buf, 8, "%u",
143 ntohs(addr->sin_port));
144 }
145 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
146
147 if (xprt->prot == IPPROTO_UDP)
148 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
149 else
150 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
151
152 buf = kzalloc(48, GFP_KERNEL);
153 if (buf) {
154 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
155 NIPQUAD(addr->sin_addr.s_addr),
156 ntohs(addr->sin_port),
157 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
158 }
159 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
160 }
161
162 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
163 {
164 kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
165 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
166 kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
167 }
168
169 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
170
171 static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
172 {
173 struct kvec iov = {
174 .iov_base = xdr->head[0].iov_base + base,
175 .iov_len = len - base,
176 };
177 struct msghdr msg = {
178 .msg_name = addr,
179 .msg_namelen = addrlen,
180 .msg_flags = XS_SENDMSG_FLAGS,
181 };
182
183 if (xdr->len > len)
184 msg.msg_flags |= MSG_MORE;
185
186 if (likely(iov.iov_len))
187 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
188 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
189 }
190
191 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
192 {
193 struct kvec iov = {
194 .iov_base = xdr->tail[0].iov_base + base,
195 .iov_len = len - base,
196 };
197 struct msghdr msg = {
198 .msg_flags = XS_SENDMSG_FLAGS,
199 };
200
201 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
202 }
203
204 /**
205 * xs_sendpages - write pages directly to a socket
206 * @sock: socket to send on
207 * @addr: UDP only -- address of destination
208 * @addrlen: UDP only -- length of destination address
209 * @xdr: buffer containing this request
210 * @base: starting position in the buffer
211 *
212 */
213 static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
214 {
215 struct page **ppage = xdr->pages;
216 unsigned int len, pglen = xdr->page_len;
217 int err, ret = 0;
218
219 if (unlikely(!sock))
220 return -ENOTCONN;
221
222 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
223
224 len = xdr->head[0].iov_len;
225 if (base < len || (addr != NULL && base == 0)) {
226 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
227 if (ret == 0)
228 ret = err;
229 else if (err > 0)
230 ret += err;
231 if (err != (len - base))
232 goto out;
233 base = 0;
234 } else
235 base -= len;
236
237 if (unlikely(pglen == 0))
238 goto copy_tail;
239 if (unlikely(base >= pglen)) {
240 base -= pglen;
241 goto copy_tail;
242 }
243 if (base || xdr->page_base) {
244 pglen -= base;
245 base += xdr->page_base;
246 ppage += base >> PAGE_CACHE_SHIFT;
247 base &= ~PAGE_CACHE_MASK;
248 }
249
250 do {
251 int flags = XS_SENDMSG_FLAGS;
252
253 len = PAGE_CACHE_SIZE;
254 if (base)
255 len -= base;
256 if (pglen < len)
257 len = pglen;
258
259 if (pglen != len || xdr->tail[0].iov_len != 0)
260 flags |= MSG_MORE;
261
262 err = kernel_sendpage(sock, *ppage, base, len, flags);
263 if (ret == 0)
264 ret = err;
265 else if (err > 0)
266 ret += err;
267 if (err != len)
268 goto out;
269 base = 0;
270 ppage++;
271 } while ((pglen -= len) != 0);
272 copy_tail:
273 len = xdr->tail[0].iov_len;
274 if (base < len) {
275 err = xs_send_tail(sock, xdr, base, len);
276 if (ret == 0)
277 ret = err;
278 else if (err > 0)
279 ret += err;
280 }
281 out:
282 return ret;
283 }
284
285 /**
286 * xs_nospace - place task on wait queue if transmit was incomplete
287 * @task: task to put to sleep
288 *
289 */
290 static void xs_nospace(struct rpc_task *task)
291 {
292 struct rpc_rqst *req = task->tk_rqstp;
293 struct rpc_xprt *xprt = req->rq_xprt;
294
295 dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
296 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
297 req->rq_slen);
298
299 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
300 /* Protect against races with write_space */
301 spin_lock_bh(&xprt->transport_lock);
302
303 /* Don't race with disconnect */
304 if (!xprt_connected(xprt))
305 task->tk_status = -ENOTCONN;
306 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
307 xprt_wait_for_buffer_space(task);
308
309 spin_unlock_bh(&xprt->transport_lock);
310 } else
311 /* Keep holding the socket if it is blocked */
312 rpc_delay(task, HZ>>4);
313 }
314
315 /**
316 * xs_udp_send_request - write an RPC request to a UDP socket
317 * @task: address of RPC task that manages the state of an RPC request
318 *
319 * Return values:
320 * 0: The request has been sent
321 * EAGAIN: The socket was blocked, please call again later to
322 * complete the request
323 * ENOTCONN: Caller needs to invoke connect logic then call again
324 * other: Some other error occured, the request was not sent
325 */
326 static int xs_udp_send_request(struct rpc_task *task)
327 {
328 struct rpc_rqst *req = task->tk_rqstp;
329 struct rpc_xprt *xprt = req->rq_xprt;
330 struct xdr_buf *xdr = &req->rq_snd_buf;
331 int status;
332
333 xs_pktdump("packet data:",
334 req->rq_svec->iov_base,
335 req->rq_svec->iov_len);
336
337 req->rq_xtime = jiffies;
338 status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
339 xprt->addrlen, xdr, req->rq_bytes_sent);
340
341 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
342 xdr->len - req->rq_bytes_sent, status);
343
344 if (likely(status >= (int) req->rq_slen))
345 return 0;
346
347 /* Still some bytes left; set up for a retry later. */
348 if (status > 0)
349 status = -EAGAIN;
350
351 switch (status) {
352 case -ENETUNREACH:
353 case -EPIPE:
354 case -ECONNREFUSED:
355 /* When the server has died, an ICMP port unreachable message
356 * prompts ECONNREFUSED. */
357 break;
358 case -EAGAIN:
359 xs_nospace(task);
360 break;
361 default:
362 dprintk("RPC: sendmsg returned unrecognized error %d\n",
363 -status);
364 break;
365 }
366
367 return status;
368 }
369
370 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
371 {
372 u32 reclen = buf->len - sizeof(rpc_fraghdr);
373 rpc_fraghdr *base = buf->head[0].iov_base;
374 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
375 }
376
377 /**
378 * xs_tcp_send_request - write an RPC request to a TCP socket
379 * @task: address of RPC task that manages the state of an RPC request
380 *
381 * Return values:
382 * 0: The request has been sent
383 * EAGAIN: The socket was blocked, please call again later to
384 * complete the request
385 * ENOTCONN: Caller needs to invoke connect logic then call again
386 * other: Some other error occured, the request was not sent
387 *
388 * XXX: In the case of soft timeouts, should we eventually give up
389 * if sendmsg is not able to make progress?
390 */
391 static int xs_tcp_send_request(struct rpc_task *task)
392 {
393 struct rpc_rqst *req = task->tk_rqstp;
394 struct rpc_xprt *xprt = req->rq_xprt;
395 struct xdr_buf *xdr = &req->rq_snd_buf;
396 int status, retry = 0;
397
398 xs_encode_tcp_record_marker(&req->rq_snd_buf);
399
400 xs_pktdump("packet data:",
401 req->rq_svec->iov_base,
402 req->rq_svec->iov_len);
403
404 /* Continue transmitting the packet/record. We must be careful
405 * to cope with writespace callbacks arriving _after_ we have
406 * called sendmsg(). */
407 while (1) {
408 req->rq_xtime = jiffies;
409 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
410 req->rq_bytes_sent);
411
412 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
413 xdr->len - req->rq_bytes_sent, status);
414
415 if (unlikely(status < 0))
416 break;
417
418 /* If we've sent the entire packet, immediately
419 * reset the count of bytes sent. */
420 req->rq_bytes_sent += status;
421 task->tk_bytes_sent += status;
422 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
423 req->rq_bytes_sent = 0;
424 return 0;
425 }
426
427 status = -EAGAIN;
428 if (retry++ > XS_SENDMSG_RETRY)
429 break;
430 }
431
432 switch (status) {
433 case -EAGAIN:
434 xs_nospace(task);
435 break;
436 case -ECONNREFUSED:
437 case -ECONNRESET:
438 case -ENOTCONN:
439 case -EPIPE:
440 status = -ENOTCONN;
441 break;
442 default:
443 dprintk("RPC: sendmsg returned unrecognized error %d\n",
444 -status);
445 xprt_disconnect(xprt);
446 break;
447 }
448
449 return status;
450 }
451
452 /**
453 * xs_tcp_release_xprt - clean up after a tcp transmission
454 * @xprt: transport
455 * @task: rpc task
456 *
457 * This cleans up if an error causes us to abort the transmission of a request.
458 * In this case, the socket may need to be reset in order to avoid confusing
459 * the server.
460 */
461 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
462 {
463 struct rpc_rqst *req;
464
465 if (task != xprt->snd_task)
466 return;
467 if (task == NULL)
468 goto out_release;
469 req = task->tk_rqstp;
470 if (req->rq_bytes_sent == 0)
471 goto out_release;
472 if (req->rq_bytes_sent == req->rq_snd_buf.len)
473 goto out_release;
474 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
475 out_release:
476 xprt_release_xprt(xprt, task);
477 }
478
479 /**
480 * xs_close - close a socket
481 * @xprt: transport
482 *
483 * This is used when all requests are complete; ie, no DRC state remains
484 * on the server we want to save.
485 */
486 static void xs_close(struct rpc_xprt *xprt)
487 {
488 struct socket *sock = xprt->sock;
489 struct sock *sk = xprt->inet;
490
491 if (!sk)
492 goto clear_close_wait;
493
494 dprintk("RPC: xs_close xprt %p\n", xprt);
495
496 write_lock_bh(&sk->sk_callback_lock);
497 xprt->inet = NULL;
498 xprt->sock = NULL;
499
500 sk->sk_user_data = NULL;
501 sk->sk_data_ready = xprt->old_data_ready;
502 sk->sk_state_change = xprt->old_state_change;
503 sk->sk_write_space = xprt->old_write_space;
504 write_unlock_bh(&sk->sk_callback_lock);
505
506 sk->sk_no_check = 0;
507
508 sock_release(sock);
509 clear_close_wait:
510 smp_mb__before_clear_bit();
511 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
512 smp_mb__after_clear_bit();
513 }
514
515 /**
516 * xs_destroy - prepare to shutdown a transport
517 * @xprt: doomed transport
518 *
519 */
520 static void xs_destroy(struct rpc_xprt *xprt)
521 {
522 dprintk("RPC: xs_destroy xprt %p\n", xprt);
523
524 cancel_delayed_work(&xprt->connect_worker);
525 flush_scheduled_work();
526
527 xprt_disconnect(xprt);
528 xs_close(xprt);
529 xs_free_peer_addresses(xprt);
530 kfree(xprt->slot);
531 }
532
533 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
534 {
535 return (struct rpc_xprt *) sk->sk_user_data;
536 }
537
538 /**
539 * xs_udp_data_ready - "data ready" callback for UDP sockets
540 * @sk: socket with data to read
541 * @len: how much data to read
542 *
543 */
544 static void xs_udp_data_ready(struct sock *sk, int len)
545 {
546 struct rpc_task *task;
547 struct rpc_xprt *xprt;
548 struct rpc_rqst *rovr;
549 struct sk_buff *skb;
550 int err, repsize, copied;
551 u32 _xid, *xp;
552
553 read_lock(&sk->sk_callback_lock);
554 dprintk("RPC: xs_udp_data_ready...\n");
555 if (!(xprt = xprt_from_sock(sk)))
556 goto out;
557
558 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
559 goto out;
560
561 if (xprt->shutdown)
562 goto dropit;
563
564 repsize = skb->len - sizeof(struct udphdr);
565 if (repsize < 4) {
566 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
567 goto dropit;
568 }
569
570 /* Copy the XID from the skb... */
571 xp = skb_header_pointer(skb, sizeof(struct udphdr),
572 sizeof(_xid), &_xid);
573 if (xp == NULL)
574 goto dropit;
575
576 /* Look up and lock the request corresponding to the given XID */
577 spin_lock(&xprt->transport_lock);
578 rovr = xprt_lookup_rqst(xprt, *xp);
579 if (!rovr)
580 goto out_unlock;
581 task = rovr->rq_task;
582
583 if ((copied = rovr->rq_private_buf.buflen) > repsize)
584 copied = repsize;
585
586 /* Suck it into the iovec, verify checksum if not done by hw. */
587 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
588 goto out_unlock;
589
590 /* Something worked... */
591 dst_confirm(skb->dst);
592
593 xprt_adjust_cwnd(task, copied);
594 xprt_update_rtt(task);
595 xprt_complete_rqst(task, copied);
596
597 out_unlock:
598 spin_unlock(&xprt->transport_lock);
599 dropit:
600 skb_free_datagram(sk, skb);
601 out:
602 read_unlock(&sk->sk_callback_lock);
603 }
604
605 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
606 {
607 if (len > desc->count)
608 len = desc->count;
609 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
610 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
611 len, desc->count);
612 return 0;
613 }
614 desc->offset += len;
615 desc->count -= len;
616 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
617 len, desc->count);
618 return len;
619 }
620
621 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
622 {
623 size_t len, used;
624 char *p;
625
626 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
627 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
628 used = xs_tcp_copy_data(desc, p, len);
629 xprt->tcp_offset += used;
630 if (used != len)
631 return;
632
633 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
634 if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
635 xprt->tcp_flags |= XPRT_LAST_FRAG;
636 else
637 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
638 xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
639
640 xprt->tcp_flags &= ~XPRT_COPY_RECM;
641 xprt->tcp_offset = 0;
642
643 /* Sanity check of the record length */
644 if (unlikely(xprt->tcp_reclen < 4)) {
645 dprintk("RPC: invalid TCP record fragment length\n");
646 xprt_disconnect(xprt);
647 return;
648 }
649 dprintk("RPC: reading TCP record fragment of length %d\n",
650 xprt->tcp_reclen);
651 }
652
653 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
654 {
655 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
656 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
657 if (xprt->tcp_offset == xprt->tcp_reclen) {
658 xprt->tcp_flags |= XPRT_COPY_RECM;
659 xprt->tcp_offset = 0;
660 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
661 xprt->tcp_flags &= ~XPRT_COPY_DATA;
662 xprt->tcp_flags |= XPRT_COPY_XID;
663 xprt->tcp_copied = 0;
664 }
665 }
666 }
667
668 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
669 {
670 size_t len, used;
671 char *p;
672
673 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
674 dprintk("RPC: reading XID (%Zu bytes)\n", len);
675 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
676 used = xs_tcp_copy_data(desc, p, len);
677 xprt->tcp_offset += used;
678 if (used != len)
679 return;
680 xprt->tcp_flags &= ~XPRT_COPY_XID;
681 xprt->tcp_flags |= XPRT_COPY_DATA;
682 xprt->tcp_copied = 4;
683 dprintk("RPC: reading reply for XID %08x\n",
684 ntohl(xprt->tcp_xid));
685 xs_tcp_check_recm(xprt);
686 }
687
688 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
689 {
690 struct rpc_rqst *req;
691 struct xdr_buf *rcvbuf;
692 size_t len;
693 ssize_t r;
694
695 /* Find and lock the request corresponding to this xid */
696 spin_lock(&xprt->transport_lock);
697 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
698 if (!req) {
699 xprt->tcp_flags &= ~XPRT_COPY_DATA;
700 dprintk("RPC: XID %08x request not found!\n",
701 ntohl(xprt->tcp_xid));
702 spin_unlock(&xprt->transport_lock);
703 return;
704 }
705
706 rcvbuf = &req->rq_private_buf;
707 len = desc->count;
708 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
709 skb_reader_t my_desc;
710
711 len = xprt->tcp_reclen - xprt->tcp_offset;
712 memcpy(&my_desc, desc, sizeof(my_desc));
713 my_desc.count = len;
714 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
715 &my_desc, xs_tcp_copy_data);
716 desc->count -= r;
717 desc->offset += r;
718 } else
719 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
720 desc, xs_tcp_copy_data);
721
722 if (r > 0) {
723 xprt->tcp_copied += r;
724 xprt->tcp_offset += r;
725 }
726 if (r != len) {
727 /* Error when copying to the receive buffer,
728 * usually because we weren't able to allocate
729 * additional buffer pages. All we can do now
730 * is turn off XPRT_COPY_DATA, so the request
731 * will not receive any additional updates,
732 * and time out.
733 * Any remaining data from this record will
734 * be discarded.
735 */
736 xprt->tcp_flags &= ~XPRT_COPY_DATA;
737 dprintk("RPC: XID %08x truncated request\n",
738 ntohl(xprt->tcp_xid));
739 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
740 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
741 goto out;
742 }
743
744 dprintk("RPC: XID %08x read %Zd bytes\n",
745 ntohl(xprt->tcp_xid), r);
746 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
747 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
748
749 if (xprt->tcp_copied == req->rq_private_buf.buflen)
750 xprt->tcp_flags &= ~XPRT_COPY_DATA;
751 else if (xprt->tcp_offset == xprt->tcp_reclen) {
752 if (xprt->tcp_flags & XPRT_LAST_FRAG)
753 xprt->tcp_flags &= ~XPRT_COPY_DATA;
754 }
755
756 out:
757 if (!(xprt->tcp_flags & XPRT_COPY_DATA))
758 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
759 spin_unlock(&xprt->transport_lock);
760 xs_tcp_check_recm(xprt);
761 }
762
763 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
764 {
765 size_t len;
766
767 len = xprt->tcp_reclen - xprt->tcp_offset;
768 if (len > desc->count)
769 len = desc->count;
770 desc->count -= len;
771 desc->offset += len;
772 xprt->tcp_offset += len;
773 dprintk("RPC: discarded %Zu bytes\n", len);
774 xs_tcp_check_recm(xprt);
775 }
776
777 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
778 {
779 struct rpc_xprt *xprt = rd_desc->arg.data;
780 skb_reader_t desc = {
781 .skb = skb,
782 .offset = offset,
783 .count = len,
784 .csum = 0
785 };
786
787 dprintk("RPC: xs_tcp_data_recv started\n");
788 do {
789 /* Read in a new fragment marker if necessary */
790 /* Can we ever really expect to get completely empty fragments? */
791 if (xprt->tcp_flags & XPRT_COPY_RECM) {
792 xs_tcp_read_fraghdr(xprt, &desc);
793 continue;
794 }
795 /* Read in the xid if necessary */
796 if (xprt->tcp_flags & XPRT_COPY_XID) {
797 xs_tcp_read_xid(xprt, &desc);
798 continue;
799 }
800 /* Read in the request data */
801 if (xprt->tcp_flags & XPRT_COPY_DATA) {
802 xs_tcp_read_request(xprt, &desc);
803 continue;
804 }
805 /* Skip over any trailing bytes on short reads */
806 xs_tcp_read_discard(xprt, &desc);
807 } while (desc.count);
808 dprintk("RPC: xs_tcp_data_recv done\n");
809 return len - desc.count;
810 }
811
812 /**
813 * xs_tcp_data_ready - "data ready" callback for TCP sockets
814 * @sk: socket with data to read
815 * @bytes: how much data to read
816 *
817 */
818 static void xs_tcp_data_ready(struct sock *sk, int bytes)
819 {
820 struct rpc_xprt *xprt;
821 read_descriptor_t rd_desc;
822
823 read_lock(&sk->sk_callback_lock);
824 dprintk("RPC: xs_tcp_data_ready...\n");
825 if (!(xprt = xprt_from_sock(sk)))
826 goto out;
827 if (xprt->shutdown)
828 goto out;
829
830 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
831 rd_desc.arg.data = xprt;
832 rd_desc.count = 65536;
833 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
834 out:
835 read_unlock(&sk->sk_callback_lock);
836 }
837
838 /**
839 * xs_tcp_state_change - callback to handle TCP socket state changes
840 * @sk: socket whose state has changed
841 *
842 */
843 static void xs_tcp_state_change(struct sock *sk)
844 {
845 struct rpc_xprt *xprt;
846
847 read_lock(&sk->sk_callback_lock);
848 if (!(xprt = xprt_from_sock(sk)))
849 goto out;
850 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
851 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
852 sk->sk_state, xprt_connected(xprt),
853 sock_flag(sk, SOCK_DEAD),
854 sock_flag(sk, SOCK_ZAPPED));
855
856 switch (sk->sk_state) {
857 case TCP_ESTABLISHED:
858 spin_lock_bh(&xprt->transport_lock);
859 if (!xprt_test_and_set_connected(xprt)) {
860 /* Reset TCP record info */
861 xprt->tcp_offset = 0;
862 xprt->tcp_reclen = 0;
863 xprt->tcp_copied = 0;
864 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
865 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
866 xprt_wake_pending_tasks(xprt, 0);
867 }
868 spin_unlock_bh(&xprt->transport_lock);
869 break;
870 case TCP_SYN_SENT:
871 case TCP_SYN_RECV:
872 break;
873 case TCP_CLOSE_WAIT:
874 /* Try to schedule an autoclose RPC calls */
875 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
876 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
877 schedule_work(&xprt->task_cleanup);
878 default:
879 xprt_disconnect(xprt);
880 }
881 out:
882 read_unlock(&sk->sk_callback_lock);
883 }
884
885 /**
886 * xs_udp_write_space - callback invoked when socket buffer space
887 * becomes available
888 * @sk: socket whose state has changed
889 *
890 * Called when more output buffer space is available for this socket.
891 * We try not to wake our writers until they can make "significant"
892 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
893 * with a bunch of small requests.
894 */
895 static void xs_udp_write_space(struct sock *sk)
896 {
897 read_lock(&sk->sk_callback_lock);
898
899 /* from net/core/sock.c:sock_def_write_space */
900 if (sock_writeable(sk)) {
901 struct socket *sock;
902 struct rpc_xprt *xprt;
903
904 if (unlikely(!(sock = sk->sk_socket)))
905 goto out;
906 if (unlikely(!(xprt = xprt_from_sock(sk))))
907 goto out;
908 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
909 goto out;
910
911 xprt_write_space(xprt);
912 }
913
914 out:
915 read_unlock(&sk->sk_callback_lock);
916 }
917
918 /**
919 * xs_tcp_write_space - callback invoked when socket buffer space
920 * becomes available
921 * @sk: socket whose state has changed
922 *
923 * Called when more output buffer space is available for this socket.
924 * We try not to wake our writers until they can make "significant"
925 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
926 * with a bunch of small requests.
927 */
928 static void xs_tcp_write_space(struct sock *sk)
929 {
930 read_lock(&sk->sk_callback_lock);
931
932 /* from net/core/stream.c:sk_stream_write_space */
933 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
934 struct socket *sock;
935 struct rpc_xprt *xprt;
936
937 if (unlikely(!(sock = sk->sk_socket)))
938 goto out;
939 if (unlikely(!(xprt = xprt_from_sock(sk))))
940 goto out;
941 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
942 goto out;
943
944 xprt_write_space(xprt);
945 }
946
947 out:
948 read_unlock(&sk->sk_callback_lock);
949 }
950
951 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
952 {
953 struct sock *sk = xprt->inet;
954
955 if (xprt->rcvsize) {
956 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
957 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
958 }
959 if (xprt->sndsize) {
960 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
961 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
962 sk->sk_write_space(sk);
963 }
964 }
965
966 /**
967 * xs_udp_set_buffer_size - set send and receive limits
968 * @xprt: generic transport
969 * @sndsize: requested size of send buffer, in bytes
970 * @rcvsize: requested size of receive buffer, in bytes
971 *
972 * Set socket send and receive buffer size limits.
973 */
974 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
975 {
976 xprt->sndsize = 0;
977 if (sndsize)
978 xprt->sndsize = sndsize + 1024;
979 xprt->rcvsize = 0;
980 if (rcvsize)
981 xprt->rcvsize = rcvsize + 1024;
982
983 xs_udp_do_set_buffer_size(xprt);
984 }
985
986 /**
987 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
988 * @task: task that timed out
989 *
990 * Adjust the congestion window after a retransmit timeout has occurred.
991 */
992 static void xs_udp_timer(struct rpc_task *task)
993 {
994 xprt_adjust_cwnd(task, -ETIMEDOUT);
995 }
996
997 static unsigned short xs_get_random_port(void)
998 {
999 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1000 unsigned short rand = (unsigned short) net_random() % range;
1001 return rand + xprt_min_resvport;
1002 }
1003
1004 /**
1005 * xs_print_peer_address - format an IPv4 address for printing
1006 * @xprt: generic transport
1007 * @format: flags field indicating which parts of the address to render
1008 */
1009 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1010 {
1011 if (xprt->address_strings[format] != NULL)
1012 return xprt->address_strings[format];
1013 else
1014 return "unprintable";
1015 }
1016
1017 /**
1018 * xs_set_port - reset the port number in the remote endpoint address
1019 * @xprt: generic transport
1020 * @port: new port number
1021 *
1022 */
1023 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1024 {
1025 struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1026
1027 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1028
1029 sap->sin_port = htons(port);
1030 }
1031
1032 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
1033 {
1034 struct sockaddr_in myaddr = {
1035 .sin_family = AF_INET,
1036 };
1037 int err;
1038 unsigned short port = xprt->port;
1039
1040 do {
1041 myaddr.sin_port = htons(port);
1042 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1043 sizeof(myaddr));
1044 if (err == 0) {
1045 xprt->port = port;
1046 dprintk("RPC: xs_bindresvport bound to port %u\n",
1047 port);
1048 return 0;
1049 }
1050 if (port <= xprt_min_resvport)
1051 port = xprt_max_resvport;
1052 else
1053 port--;
1054 } while (err == -EADDRINUSE && port != xprt->port);
1055
1056 dprintk("RPC: can't bind to reserved port (%d).\n", -err);
1057 return err;
1058 }
1059
1060 /**
1061 * xs_udp_connect_worker - set up a UDP socket
1062 * @args: RPC transport to connect
1063 *
1064 * Invoked by a work queue tasklet.
1065 */
1066 static void xs_udp_connect_worker(void *args)
1067 {
1068 struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1069 struct socket *sock = xprt->sock;
1070 int err, status = -EIO;
1071
1072 if (xprt->shutdown || !xprt_bound(xprt))
1073 goto out;
1074
1075 /* Start by resetting any existing state */
1076 xs_close(xprt);
1077
1078 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1079 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1080 goto out;
1081 }
1082
1083 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1084 sock_release(sock);
1085 goto out;
1086 }
1087
1088 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1089 xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1090
1091 if (!xprt->inet) {
1092 struct sock *sk = sock->sk;
1093
1094 write_lock_bh(&sk->sk_callback_lock);
1095
1096 sk->sk_user_data = xprt;
1097 xprt->old_data_ready = sk->sk_data_ready;
1098 xprt->old_state_change = sk->sk_state_change;
1099 xprt->old_write_space = sk->sk_write_space;
1100 sk->sk_data_ready = xs_udp_data_ready;
1101 sk->sk_write_space = xs_udp_write_space;
1102 sk->sk_no_check = UDP_CSUM_NORCV;
1103 sk->sk_allocation = GFP_ATOMIC;
1104
1105 xprt_set_connected(xprt);
1106
1107 /* Reset to new socket */
1108 xprt->sock = sock;
1109 xprt->inet = sk;
1110
1111 write_unlock_bh(&sk->sk_callback_lock);
1112 }
1113 xs_udp_do_set_buffer_size(xprt);
1114 status = 0;
1115 out:
1116 xprt_wake_pending_tasks(xprt, status);
1117 xprt_clear_connecting(xprt);
1118 }
1119
1120 /*
1121 * We need to preserve the port number so the reply cache on the server can
1122 * find our cached RPC replies when we get around to reconnecting.
1123 */
1124 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1125 {
1126 int result;
1127 struct socket *sock = xprt->sock;
1128 struct sockaddr any;
1129
1130 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1131
1132 /*
1133 * Disconnect the transport socket by doing a connect operation
1134 * with AF_UNSPEC. This should return immediately...
1135 */
1136 memset(&any, 0, sizeof(any));
1137 any.sa_family = AF_UNSPEC;
1138 result = kernel_connect(sock, &any, sizeof(any), 0);
1139 if (result)
1140 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1141 result);
1142 }
1143
1144 /**
1145 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1146 * @args: RPC transport to connect
1147 *
1148 * Invoked by a work queue tasklet.
1149 */
1150 static void xs_tcp_connect_worker(void *args)
1151 {
1152 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1153 struct socket *sock = xprt->sock;
1154 int err, status = -EIO;
1155
1156 if (xprt->shutdown || !xprt_bound(xprt))
1157 goto out;
1158
1159 if (!xprt->sock) {
1160 /* start from scratch */
1161 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1162 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1163 goto out;
1164 }
1165
1166 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1167 sock_release(sock);
1168 goto out;
1169 }
1170 } else
1171 /* "close" the socket, preserving the local port */
1172 xs_tcp_reuse_connection(xprt);
1173
1174 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1175 xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1176
1177 if (!xprt->inet) {
1178 struct sock *sk = sock->sk;
1179
1180 write_lock_bh(&sk->sk_callback_lock);
1181
1182 sk->sk_user_data = xprt;
1183 xprt->old_data_ready = sk->sk_data_ready;
1184 xprt->old_state_change = sk->sk_state_change;
1185 xprt->old_write_space = sk->sk_write_space;
1186 sk->sk_data_ready = xs_tcp_data_ready;
1187 sk->sk_state_change = xs_tcp_state_change;
1188 sk->sk_write_space = xs_tcp_write_space;
1189 sk->sk_allocation = GFP_ATOMIC;
1190
1191 /* socket options */
1192 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1193 sock_reset_flag(sk, SOCK_LINGER);
1194 tcp_sk(sk)->linger2 = 0;
1195 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1196
1197 xprt_clear_connected(xprt);
1198
1199 /* Reset to new socket */
1200 xprt->sock = sock;
1201 xprt->inet = sk;
1202
1203 write_unlock_bh(&sk->sk_callback_lock);
1204 }
1205
1206 /* Tell the socket layer to start connecting... */
1207 xprt->stat.connect_count++;
1208 xprt->stat.connect_start = jiffies;
1209 status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1210 xprt->addrlen, O_NONBLOCK);
1211 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1212 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1213 if (status < 0) {
1214 switch (status) {
1215 case -EINPROGRESS:
1216 case -EALREADY:
1217 goto out_clear;
1218 case -ECONNREFUSED:
1219 case -ECONNRESET:
1220 /* retry with existing socket, after a delay */
1221 break;
1222 default:
1223 /* get rid of existing socket, and retry */
1224 xs_close(xprt);
1225 break;
1226 }
1227 }
1228 out:
1229 xprt_wake_pending_tasks(xprt, status);
1230 out_clear:
1231 xprt_clear_connecting(xprt);
1232 }
1233
1234 /**
1235 * xs_connect - connect a socket to a remote endpoint
1236 * @task: address of RPC task that manages state of connect request
1237 *
1238 * TCP: If the remote end dropped the connection, delay reconnecting.
1239 *
1240 * UDP socket connects are synchronous, but we use a work queue anyway
1241 * to guarantee that even unprivileged user processes can set up a
1242 * socket on a privileged port.
1243 *
1244 * If a UDP socket connect fails, the delay behavior here prevents
1245 * retry floods (hard mounts).
1246 */
1247 static void xs_connect(struct rpc_task *task)
1248 {
1249 struct rpc_xprt *xprt = task->tk_xprt;
1250
1251 if (xprt_test_and_set_connecting(xprt))
1252 return;
1253
1254 if (xprt->sock != NULL) {
1255 dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n",
1256 xprt, xprt->reestablish_timeout / HZ);
1257 schedule_delayed_work(&xprt->connect_worker,
1258 xprt->reestablish_timeout);
1259 xprt->reestablish_timeout <<= 1;
1260 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1261 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1262 } else {
1263 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1264 schedule_work(&xprt->connect_worker);
1265
1266 /* flush_scheduled_work can sleep... */
1267 if (!RPC_IS_ASYNC(task))
1268 flush_scheduled_work();
1269 }
1270 }
1271
1272 /**
1273 * xs_udp_print_stats - display UDP socket-specifc stats
1274 * @xprt: rpc_xprt struct containing statistics
1275 * @seq: output file
1276 *
1277 */
1278 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1279 {
1280 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1281 xprt->port,
1282 xprt->stat.bind_count,
1283 xprt->stat.sends,
1284 xprt->stat.recvs,
1285 xprt->stat.bad_xids,
1286 xprt->stat.req_u,
1287 xprt->stat.bklog_u);
1288 }
1289
1290 /**
1291 * xs_tcp_print_stats - display TCP socket-specifc stats
1292 * @xprt: rpc_xprt struct containing statistics
1293 * @seq: output file
1294 *
1295 */
1296 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1297 {
1298 long idle_time = 0;
1299
1300 if (xprt_connected(xprt))
1301 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1302
1303 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1304 xprt->port,
1305 xprt->stat.bind_count,
1306 xprt->stat.connect_count,
1307 xprt->stat.connect_time,
1308 idle_time,
1309 xprt->stat.sends,
1310 xprt->stat.recvs,
1311 xprt->stat.bad_xids,
1312 xprt->stat.req_u,
1313 xprt->stat.bklog_u);
1314 }
1315
1316 static struct rpc_xprt_ops xs_udp_ops = {
1317 .set_buffer_size = xs_udp_set_buffer_size,
1318 .print_addr = xs_print_peer_address,
1319 .reserve_xprt = xprt_reserve_xprt_cong,
1320 .release_xprt = xprt_release_xprt_cong,
1321 .rpcbind = rpc_getport,
1322 .set_port = xs_set_port,
1323 .connect = xs_connect,
1324 .buf_alloc = rpc_malloc,
1325 .buf_free = rpc_free,
1326 .send_request = xs_udp_send_request,
1327 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1328 .timer = xs_udp_timer,
1329 .release_request = xprt_release_rqst_cong,
1330 .close = xs_close,
1331 .destroy = xs_destroy,
1332 .print_stats = xs_udp_print_stats,
1333 };
1334
1335 static struct rpc_xprt_ops xs_tcp_ops = {
1336 .print_addr = xs_print_peer_address,
1337 .reserve_xprt = xprt_reserve_xprt,
1338 .release_xprt = xs_tcp_release_xprt,
1339 .rpcbind = rpc_getport,
1340 .set_port = xs_set_port,
1341 .connect = xs_connect,
1342 .buf_alloc = rpc_malloc,
1343 .buf_free = rpc_free,
1344 .send_request = xs_tcp_send_request,
1345 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1346 .close = xs_close,
1347 .destroy = xs_destroy,
1348 .print_stats = xs_tcp_print_stats,
1349 };
1350
1351 /**
1352 * xs_setup_udp - Set up transport to use a UDP socket
1353 * @xprt: transport to set up
1354 * @to: timeout parameters
1355 *
1356 */
1357 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1358 {
1359 size_t slot_table_size;
1360 struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
1361
1362 xprt->max_reqs = xprt_udp_slot_table_entries;
1363 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1364 xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1365 if (xprt->slot == NULL)
1366 return -ENOMEM;
1367
1368 if (ntohs(addr->sin_port != 0))
1369 xprt_set_bound(xprt);
1370 xprt->port = xs_get_random_port();
1371
1372 xprt->prot = IPPROTO_UDP;
1373 xprt->tsh_size = 0;
1374 /* XXX: header size can vary due to auth type, IPv6, etc. */
1375 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1376
1377 INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1378 xprt->bind_timeout = XS_BIND_TO;
1379 xprt->connect_timeout = XS_UDP_CONN_TO;
1380 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1381 xprt->idle_timeout = XS_IDLE_DISC_TO;
1382
1383 xprt->ops = &xs_udp_ops;
1384
1385 if (to)
1386 xprt->timeout = *to;
1387 else
1388 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1389
1390 xs_format_peer_addresses(xprt);
1391 dprintk("RPC: set up transport to address %s\n",
1392 xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1393
1394 return 0;
1395 }
1396
1397 /**
1398 * xs_setup_tcp - Set up transport to use a TCP socket
1399 * @xprt: transport to set up
1400 * @to: timeout parameters
1401 *
1402 */
1403 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1404 {
1405 size_t slot_table_size;
1406 struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
1407
1408 xprt->max_reqs = xprt_tcp_slot_table_entries;
1409 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1410 xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1411 if (xprt->slot == NULL)
1412 return -ENOMEM;
1413
1414 if (ntohs(addr->sin_port) != 0)
1415 xprt_set_bound(xprt);
1416 xprt->port = xs_get_random_port();
1417
1418 xprt->prot = IPPROTO_TCP;
1419 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1420 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1421
1422 INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1423 xprt->bind_timeout = XS_BIND_TO;
1424 xprt->connect_timeout = XS_TCP_CONN_TO;
1425 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1426 xprt->idle_timeout = XS_IDLE_DISC_TO;
1427
1428 xprt->ops = &xs_tcp_ops;
1429
1430 if (to)
1431 xprt->timeout = *to;
1432 else
1433 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1434
1435 xs_format_peer_addresses(xprt);
1436 dprintk("RPC: set up transport to address %s\n",
1437 xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1438
1439 return 0;
1440 }
kernel-based virtual machine