?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
SUNRPC: Fix xs_setup_bc_tcp()
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sunrpc / xprtsock.c
blobbeefa7a3a90e519421fb617382fe6d757e778116
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 .procname = "udp_slot_table_entries",
85 .data = &xprt_udp_slot_table_entries,
86 .maxlen = sizeof(unsigned int),
87 .mode = 0644,
88 .proc_handler = proc_dointvec_minmax,
89 .extra1 = &min_slot_table_size,
90 .extra2 = &max_slot_table_size
91 },
92 {
93 .procname = "tcp_slot_table_entries",
94 .data = &xprt_tcp_slot_table_entries,
95 .maxlen = sizeof(unsigned int),
96 .mode = 0644,
97 .proc_handler = proc_dointvec_minmax,
98 .extra1 = &min_slot_table_size,
99 .extra2 = &max_slot_table_size
100 },
101 {
102 .procname = "min_resvport",
103 .data = &xprt_min_resvport,
104 .maxlen = sizeof(unsigned int),
105 .mode = 0644,
106 .proc_handler = proc_dointvec_minmax,
107 .extra1 = &xprt_min_resvport_limit,
108 .extra2 = &xprt_max_resvport_limit
109 },
110 {
111 .procname = "max_resvport",
112 .data = &xprt_max_resvport,
113 .maxlen = sizeof(unsigned int),
114 .mode = 0644,
115 .proc_handler = proc_dointvec_minmax,
116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
118 },
119 {
120 .procname = "tcp_fin_timeout",
121 .data = &xs_tcp_fin_timeout,
122 .maxlen = sizeof(xs_tcp_fin_timeout),
123 .mode = 0644,
124 .proc_handler = proc_dointvec_jiffies,
125 },
126 { },
127 };
128
129 static ctl_table sunrpc_table[] = {
130 {
131 .procname = "sunrpc",
132 .mode = 0555,
133 .child = xs_tunables_table
134 },
135 { },
136 };
137
138 #endif
139
140 /*
141 * Wait duration for a reply from the RPC portmapper.
142 */
143 #define XS_BIND_TO (60U * HZ)
144
145 /*
146 * Delay if a UDP socket connect error occurs. This is most likely some
147 * kind of resource problem on the local host.
148 */
149 #define XS_UDP_REEST_TO (2U * HZ)
150
151 /*
152 * The reestablish timeout allows clients to delay for a bit before attempting
153 * to reconnect to a server that just dropped our connection.
154 *
155 * We implement an exponential backoff when trying to reestablish a TCP
156 * transport connection with the server. Some servers like to drop a TCP
157 * connection when they are overworked, so we start with a short timeout and
158 * increase over time if the server is down or not responding.
159 */
160 #define XS_TCP_INIT_REEST_TO (3U * HZ)
161 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
162
163 /*
164 * TCP idle timeout; client drops the transport socket if it is idle
165 * for this long. Note that we also timeout UDP sockets to prevent
166 * holding port numbers when there is no RPC traffic.
167 */
168 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
169
170 #ifdef RPC_DEBUG
171 # undef RPC_DEBUG_DATA
172 # define RPCDBG_FACILITY RPCDBG_TRANS
173 #endif
174
175 #ifdef RPC_DEBUG_DATA
176 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
177 {
178 u8 *buf = (u8 *) packet;
179 int j;
180
181 dprintk("RPC: %s\n", msg);
182 for (j = 0; j < count && j < 128; j += 4) {
183 if (!(j & 31)) {
184 if (j)
185 dprintk("\n");
186 dprintk("0x%04x ", j);
187 }
188 dprintk("%02x%02x%02x%02x ",
189 buf[j], buf[j+1], buf[j+2], buf[j+3]);
190 }
191 dprintk("\n");
192 }
193 #else
194 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
195 {
196 /* NOP */
197 }
198 #endif
199
200 struct sock_xprt {
201 struct rpc_xprt xprt;
202
203 /*
204 * Network layer
205 */
206 struct socket * sock;
207 struct sock * inet;
208
209 /*
210 * State of TCP reply receive
211 */
212 __be32 tcp_fraghdr,
213 tcp_xid;
214
215 u32 tcp_offset,
216 tcp_reclen;
217
218 unsigned long tcp_copied,
219 tcp_flags;
220
221 /*
222 * Connection of transports
223 */
224 struct delayed_work connect_worker;
225 struct sockaddr_storage srcaddr;
226 unsigned short srcport;
227
228 /*
229 * UDP socket buffer size parameters
230 */
231 size_t rcvsize,
232 sndsize;
233
234 /*
235 * Saved socket callback addresses
236 */
237 void (*old_data_ready)(struct sock *, int);
238 void (*old_state_change)(struct sock *);
239 void (*old_write_space)(struct sock *);
240 void (*old_error_report)(struct sock *);
241 };
242
243 /*
244 * TCP receive state flags
245 */
246 #define TCP_RCV_LAST_FRAG (1UL << 0)
247 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
248 #define TCP_RCV_COPY_XID (1UL << 2)
249 #define TCP_RCV_COPY_DATA (1UL << 3)
250 #define TCP_RCV_READ_CALLDIR (1UL << 4)
251 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
252
253 /*
254 * TCP RPC flags
255 */
256 #define TCP_RPC_REPLY (1UL << 6)
257
258 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
259 {
260 return (struct sockaddr *) &xprt->addr;
261 }
262
263 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
264 {
265 return (struct sockaddr_in *) &xprt->addr;
266 }
267
268 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
269 {
270 return (struct sockaddr_in6 *) &xprt->addr;
271 }
272
273 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
274 {
275 struct sockaddr *sap = xs_addr(xprt);
276 struct sockaddr_in6 *sin6;
277 struct sockaddr_in *sin;
278 char buf[128];
279
280 (void)rpc_ntop(sap, buf, sizeof(buf));
281 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
282
283 switch (sap->sa_family) {
284 case AF_INET:
285 sin = xs_addr_in(xprt);
286 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
287 break;
288 case AF_INET6:
289 sin6 = xs_addr_in6(xprt);
290 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
291 break;
292 default:
293 BUG();
294 }
295 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
296 }
297
298 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
299 {
300 struct sockaddr *sap = xs_addr(xprt);
301 char buf[128];
302
303 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
304 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
305
306 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
307 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
308 }
309
310 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
311 const char *protocol,
312 const char *netid)
313 {
314 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
315 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
316 xs_format_common_peer_addresses(xprt);
317 xs_format_common_peer_ports(xprt);
318 }
319
320 static void xs_update_peer_port(struct rpc_xprt *xprt)
321 {
322 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
323 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
324
325 xs_format_common_peer_ports(xprt);
326 }
327
328 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
329 {
330 unsigned int i;
331
332 for (i = 0; i < RPC_DISPLAY_MAX; i++)
333 switch (i) {
334 case RPC_DISPLAY_PROTO:
335 case RPC_DISPLAY_NETID:
336 continue;
337 default:
338 kfree(xprt->address_strings[i]);
339 }
340 }
341
342 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
343
344 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
345 {
346 struct msghdr msg = {
347 .msg_name = addr,
348 .msg_namelen = addrlen,
349 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
350 };
351 struct kvec iov = {
352 .iov_base = vec->iov_base + base,
353 .iov_len = vec->iov_len - base,
354 };
355
356 if (iov.iov_len != 0)
357 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
358 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
359 }
360
361 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
362 {
363 struct page **ppage;
364 unsigned int remainder;
365 int err, sent = 0;
366
367 remainder = xdr->page_len - base;
368 base += xdr->page_base;
369 ppage = xdr->pages + (base >> PAGE_SHIFT);
370 base &= ~PAGE_MASK;
371 for(;;) {
372 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
373 int flags = XS_SENDMSG_FLAGS;
374
375 remainder -= len;
376 if (remainder != 0 || more)
377 flags |= MSG_MORE;
378 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
379 if (remainder == 0 || err != len)
380 break;
381 sent += err;
382 ppage++;
383 base = 0;
384 }
385 if (sent == 0)
386 return err;
387 if (err > 0)
388 sent += err;
389 return sent;
390 }
391
392 /**
393 * xs_sendpages - write pages directly to a socket
394 * @sock: socket to send on
395 * @addr: UDP only -- address of destination
396 * @addrlen: UDP only -- length of destination address
397 * @xdr: buffer containing this request
398 * @base: starting position in the buffer
399 *
400 */
401 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
402 {
403 unsigned int remainder = xdr->len - base;
404 int err, sent = 0;
405
406 if (unlikely(!sock))
407 return -ENOTSOCK;
408
409 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
410 if (base != 0) {
411 addr = NULL;
412 addrlen = 0;
413 }
414
415 if (base < xdr->head[0].iov_len || addr != NULL) {
416 unsigned int len = xdr->head[0].iov_len - base;
417 remainder -= len;
418 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
419 if (remainder == 0 || err != len)
420 goto out;
421 sent += err;
422 base = 0;
423 } else
424 base -= xdr->head[0].iov_len;
425
426 if (base < xdr->page_len) {
427 unsigned int len = xdr->page_len - base;
428 remainder -= len;
429 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
430 if (remainder == 0 || err != len)
431 goto out;
432 sent += err;
433 base = 0;
434 } else
435 base -= xdr->page_len;
436
437 if (base >= xdr->tail[0].iov_len)
438 return sent;
439 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
440 out:
441 if (sent == 0)
442 return err;
443 if (err > 0)
444 sent += err;
445 return sent;
446 }
447
448 static void xs_nospace_callback(struct rpc_task *task)
449 {
450 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
451
452 transport->inet->sk_write_pending--;
453 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
454 }
455
456 /**
457 * xs_nospace - place task on wait queue if transmit was incomplete
458 * @task: task to put to sleep
459 *
460 */
461 static int xs_nospace(struct rpc_task *task)
462 {
463 struct rpc_rqst *req = task->tk_rqstp;
464 struct rpc_xprt *xprt = req->rq_xprt;
465 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
466 int ret = 0;
467
468 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
469 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
470 req->rq_slen);
471
472 /* Protect against races with write_space */
473 spin_lock_bh(&xprt->transport_lock);
474
475 /* Don't race with disconnect */
476 if (xprt_connected(xprt)) {
477 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
478 ret = -EAGAIN;
479 /*
480 * Notify TCP that we're limited by the application
481 * window size
482 */
483 set_bit(SOCK_NOSPACE, &transport->sock->flags);
484 transport->inet->sk_write_pending++;
485 /* ...and wait for more buffer space */
486 xprt_wait_for_buffer_space(task, xs_nospace_callback);
487 }
488 } else {
489 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
490 ret = -ENOTCONN;
491 }
492
493 spin_unlock_bh(&xprt->transport_lock);
494 return ret;
495 }
496
497 /**
498 * xs_udp_send_request - write an RPC request to a UDP socket
499 * @task: address of RPC task that manages the state of an RPC request
500 *
501 * Return values:
502 * 0: The request has been sent
503 * EAGAIN: The socket was blocked, please call again later to
504 * complete the request
505 * ENOTCONN: Caller needs to invoke connect logic then call again
506 * other: Some other error occured, the request was not sent
507 */
508 static int xs_udp_send_request(struct rpc_task *task)
509 {
510 struct rpc_rqst *req = task->tk_rqstp;
511 struct rpc_xprt *xprt = req->rq_xprt;
512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
513 struct xdr_buf *xdr = &req->rq_snd_buf;
514 int status;
515
516 xs_pktdump("packet data:",
517 req->rq_svec->iov_base,
518 req->rq_svec->iov_len);
519
520 if (!xprt_bound(xprt))
521 return -ENOTCONN;
522 status = xs_sendpages(transport->sock,
523 xs_addr(xprt),
524 xprt->addrlen, xdr,
525 req->rq_bytes_sent);
526
527 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
528 xdr->len - req->rq_bytes_sent, status);
529
530 if (status >= 0) {
531 task->tk_bytes_sent += status;
532 if (status >= req->rq_slen)
533 return 0;
534 /* Still some bytes left; set up for a retry later. */
535 status = -EAGAIN;
536 }
537
538 switch (status) {
539 case -ENOTSOCK:
540 status = -ENOTCONN;
541 /* Should we call xs_close() here? */
542 break;
543 case -EAGAIN:
544 status = xs_nospace(task);
545 break;
546 default:
547 dprintk("RPC: sendmsg returned unrecognized error %d\n",
548 -status);
549 case -ENETUNREACH:
550 case -EPIPE:
551 case -ECONNREFUSED:
552 /* When the server has died, an ICMP port unreachable message
553 * prompts ECONNREFUSED. */
554 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
555 }
556
557 return status;
558 }
559
560 /**
561 * xs_tcp_shutdown - gracefully shut down a TCP socket
562 * @xprt: transport
563 *
564 * Initiates a graceful shutdown of the TCP socket by calling the
565 * equivalent of shutdown(SHUT_WR);
566 */
567 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
568 {
569 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
570 struct socket *sock = transport->sock;
571
572 if (sock != NULL)
573 kernel_sock_shutdown(sock, SHUT_WR);
574 }
575
576 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
577 {
578 u32 reclen = buf->len - sizeof(rpc_fraghdr);
579 rpc_fraghdr *base = buf->head[0].iov_base;
580 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
581 }
582
583 /**
584 * xs_tcp_send_request - write an RPC request to a TCP socket
585 * @task: address of RPC task that manages the state of an RPC request
586 *
587 * Return values:
588 * 0: The request has been sent
589 * EAGAIN: The socket was blocked, please call again later to
590 * complete the request
591 * ENOTCONN: Caller needs to invoke connect logic then call again
592 * other: Some other error occured, the request was not sent
593 *
594 * XXX: In the case of soft timeouts, should we eventually give up
595 * if sendmsg is not able to make progress?
596 */
597 static int xs_tcp_send_request(struct rpc_task *task)
598 {
599 struct rpc_rqst *req = task->tk_rqstp;
600 struct rpc_xprt *xprt = req->rq_xprt;
601 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
602 struct xdr_buf *xdr = &req->rq_snd_buf;
603 int status;
604
605 xs_encode_tcp_record_marker(&req->rq_snd_buf);
606
607 xs_pktdump("packet data:",
608 req->rq_svec->iov_base,
609 req->rq_svec->iov_len);
610
611 /* Continue transmitting the packet/record. We must be careful
612 * to cope with writespace callbacks arriving _after_ we have
613 * called sendmsg(). */
614 while (1) {
615 status = xs_sendpages(transport->sock,
616 NULL, 0, xdr, req->rq_bytes_sent);
617
618 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
619 xdr->len - req->rq_bytes_sent, status);
620
621 if (unlikely(status < 0))
622 break;
623
624 /* If we've sent the entire packet, immediately
625 * reset the count of bytes sent. */
626 req->rq_bytes_sent += status;
627 task->tk_bytes_sent += status;
628 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
629 req->rq_bytes_sent = 0;
630 return 0;
631 }
632
633 if (status != 0)
634 continue;
635 status = -EAGAIN;
636 break;
637 }
638
639 switch (status) {
640 case -ENOTSOCK:
641 status = -ENOTCONN;
642 /* Should we call xs_close() here? */
643 break;
644 case -EAGAIN:
645 status = xs_nospace(task);
646 break;
647 default:
648 dprintk("RPC: sendmsg returned unrecognized error %d\n",
649 -status);
650 case -ECONNRESET:
651 case -EPIPE:
652 xs_tcp_shutdown(xprt);
653 case -ECONNREFUSED:
654 case -ENOTCONN:
655 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
656 }
657
658 return status;
659 }
660
661 /**
662 * xs_tcp_release_xprt - clean up after a tcp transmission
663 * @xprt: transport
664 * @task: rpc task
665 *
666 * This cleans up if an error causes us to abort the transmission of a request.
667 * In this case, the socket may need to be reset in order to avoid confusing
668 * the server.
669 */
670 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
671 {
672 struct rpc_rqst *req;
673
674 if (task != xprt->snd_task)
675 return;
676 if (task == NULL)
677 goto out_release;
678 req = task->tk_rqstp;
679 if (req->rq_bytes_sent == 0)
680 goto out_release;
681 if (req->rq_bytes_sent == req->rq_snd_buf.len)
682 goto out_release;
683 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
684 out_release:
685 xprt_release_xprt(xprt, task);
686 }
687
688 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
689 {
690 transport->old_data_ready = sk->sk_data_ready;
691 transport->old_state_change = sk->sk_state_change;
692 transport->old_write_space = sk->sk_write_space;
693 transport->old_error_report = sk->sk_error_report;
694 }
695
696 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
697 {
698 sk->sk_data_ready = transport->old_data_ready;
699 sk->sk_state_change = transport->old_state_change;
700 sk->sk_write_space = transport->old_write_space;
701 sk->sk_error_report = transport->old_error_report;
702 }
703
704 static void xs_reset_transport(struct sock_xprt *transport)
705 {
706 struct socket *sock = transport->sock;
707 struct sock *sk = transport->inet;
708
709 if (sk == NULL)
710 return;
711
712 write_lock_bh(&sk->sk_callback_lock);
713 transport->inet = NULL;
714 transport->sock = NULL;
715
716 sk->sk_user_data = NULL;
717
718 xs_restore_old_callbacks(transport, sk);
719 write_unlock_bh(&sk->sk_callback_lock);
720
721 sk->sk_no_check = 0;
722
723 sock_release(sock);
724 }
725
726 /**
727 * xs_close - close a socket
728 * @xprt: transport
729 *
730 * This is used when all requests are complete; ie, no DRC state remains
731 * on the server we want to save.
732 *
733 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
734 * xs_reset_transport() zeroing the socket from underneath a writer.
735 */
736 static void xs_close(struct rpc_xprt *xprt)
737 {
738 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
739
740 dprintk("RPC: xs_close xprt %p\n", xprt);
741
742 xs_reset_transport(transport);
743 xprt->reestablish_timeout = 0;
744
745 smp_mb__before_clear_bit();
746 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
747 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
748 clear_bit(XPRT_CLOSING, &xprt->state);
749 smp_mb__after_clear_bit();
750 xprt_disconnect_done(xprt);
751 }
752
753 static void xs_tcp_close(struct rpc_xprt *xprt)
754 {
755 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
756 xs_close(xprt);
757 else
758 xs_tcp_shutdown(xprt);
759 }
760
761 /**
762 * xs_destroy - prepare to shutdown a transport
763 * @xprt: doomed transport
764 *
765 */
766 static void xs_destroy(struct rpc_xprt *xprt)
767 {
768 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
769
770 dprintk("RPC: xs_destroy xprt %p\n", xprt);
771
772 cancel_rearming_delayed_work(&transport->connect_worker);
773
774 xs_close(xprt);
775 xs_free_peer_addresses(xprt);
776 kfree(xprt->slot);
777 kfree(xprt);
778 module_put(THIS_MODULE);
779 }
780
781 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
782 {
783 return (struct rpc_xprt *) sk->sk_user_data;
784 }
785
786 /**
787 * xs_udp_data_ready - "data ready" callback for UDP sockets
788 * @sk: socket with data to read
789 * @len: how much data to read
790 *
791 */
792 static void xs_udp_data_ready(struct sock *sk, int len)
793 {
794 struct rpc_task *task;
795 struct rpc_xprt *xprt;
796 struct rpc_rqst *rovr;
797 struct sk_buff *skb;
798 int err, repsize, copied;
799 u32 _xid;
800 __be32 *xp;
801
802 read_lock(&sk->sk_callback_lock);
803 dprintk("RPC: xs_udp_data_ready...\n");
804 if (!(xprt = xprt_from_sock(sk)))
805 goto out;
806
807 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
808 goto out;
809
810 if (xprt->shutdown)
811 goto dropit;
812
813 repsize = skb->len - sizeof(struct udphdr);
814 if (repsize < 4) {
815 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
816 goto dropit;
817 }
818
819 /* Copy the XID from the skb... */
820 xp = skb_header_pointer(skb, sizeof(struct udphdr),
821 sizeof(_xid), &_xid);
822 if (xp == NULL)
823 goto dropit;
824
825 /* Look up and lock the request corresponding to the given XID */
826 spin_lock(&xprt->transport_lock);
827 rovr = xprt_lookup_rqst(xprt, *xp);
828 if (!rovr)
829 goto out_unlock;
830 task = rovr->rq_task;
831
832 if ((copied = rovr->rq_private_buf.buflen) > repsize)
833 copied = repsize;
834
835 /* Suck it into the iovec, verify checksum if not done by hw. */
836 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
837 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
838 goto out_unlock;
839 }
840
841 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
842
843 /* Something worked... */
844 dst_confirm(skb_dst(skb));
845
846 xprt_adjust_cwnd(task, copied);
847 xprt_complete_rqst(task, copied);
848
849 out_unlock:
850 spin_unlock(&xprt->transport_lock);
851 dropit:
852 skb_free_datagram(sk, skb);
853 out:
854 read_unlock(&sk->sk_callback_lock);
855 }
856
857 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
858 {
859 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
860 size_t len, used;
861 char *p;
862
863 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
864 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
865 used = xdr_skb_read_bits(desc, p, len);
866 transport->tcp_offset += used;
867 if (used != len)
868 return;
869
870 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
871 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
872 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
873 else
874 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
875 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
876
877 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
878 transport->tcp_offset = 0;
879
880 /* Sanity check of the record length */
881 if (unlikely(transport->tcp_reclen < 8)) {
882 dprintk("RPC: invalid TCP record fragment length\n");
883 xprt_force_disconnect(xprt);
884 return;
885 }
886 dprintk("RPC: reading TCP record fragment of length %d\n",
887 transport->tcp_reclen);
888 }
889
890 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
891 {
892 if (transport->tcp_offset == transport->tcp_reclen) {
893 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
894 transport->tcp_offset = 0;
895 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
896 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
897 transport->tcp_flags |= TCP_RCV_COPY_XID;
898 transport->tcp_copied = 0;
899 }
900 }
901 }
902
903 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
904 {
905 size_t len, used;
906 char *p;
907
908 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
909 dprintk("RPC: reading XID (%Zu bytes)\n", len);
910 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
911 used = xdr_skb_read_bits(desc, p, len);
912 transport->tcp_offset += used;
913 if (used != len)
914 return;
915 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
916 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
917 transport->tcp_copied = 4;
918 dprintk("RPC: reading %s XID %08x\n",
919 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
920 : "request with",
921 ntohl(transport->tcp_xid));
922 xs_tcp_check_fraghdr(transport);
923 }
924
925 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
926 struct xdr_skb_reader *desc)
927 {
928 size_t len, used;
929 u32 offset;
930 __be32 calldir;
931
932 /*
933 * We want transport->tcp_offset to be 8 at the end of this routine
934 * (4 bytes for the xid and 4 bytes for the call/reply flag).
935 * When this function is called for the first time,
936 * transport->tcp_offset is 4 (after having already read the xid).
937 */
938 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
939 len = sizeof(calldir) - offset;
940 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
941 used = xdr_skb_read_bits(desc, &calldir, len);
942 transport->tcp_offset += used;
943 if (used != len)
944 return;
945 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
946 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
947 transport->tcp_flags |= TCP_RCV_COPY_DATA;
948 /*
949 * We don't yet have the XDR buffer, so we will write the calldir
950 * out after we get the buffer from the 'struct rpc_rqst'
951 */
952 if (ntohl(calldir) == RPC_REPLY)
953 transport->tcp_flags |= TCP_RPC_REPLY;
954 else
955 transport->tcp_flags &= ~TCP_RPC_REPLY;
956 dprintk("RPC: reading %s CALL/REPLY flag %08x\n",
957 (transport->tcp_flags & TCP_RPC_REPLY) ?
958 "reply for" : "request with", calldir);
959 xs_tcp_check_fraghdr(transport);
960 }
961
962 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
963 struct xdr_skb_reader *desc,
964 struct rpc_rqst *req)
965 {
966 struct sock_xprt *transport =
967 container_of(xprt, struct sock_xprt, xprt);
968 struct xdr_buf *rcvbuf;
969 size_t len;
970 ssize_t r;
971
972 rcvbuf = &req->rq_private_buf;
973
974 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
975 /*
976 * Save the RPC direction in the XDR buffer
977 */
978 __be32 calldir = transport->tcp_flags & TCP_RPC_REPLY ?
979 htonl(RPC_REPLY) : 0;
980
981 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
982 &calldir, sizeof(calldir));
983 transport->tcp_copied += sizeof(calldir);
984 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
985 }
986
987 len = desc->count;
988 if (len > transport->tcp_reclen - transport->tcp_offset) {
989 struct xdr_skb_reader my_desc;
990
991 len = transport->tcp_reclen - transport->tcp_offset;
992 memcpy(&my_desc, desc, sizeof(my_desc));
993 my_desc.count = len;
994 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
995 &my_desc, xdr_skb_read_bits);
996 desc->count -= r;
997 desc->offset += r;
998 } else
999 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1000 desc, xdr_skb_read_bits);
1001
1002 if (r > 0) {
1003 transport->tcp_copied += r;
1004 transport->tcp_offset += r;
1005 }
1006 if (r != len) {
1007 /* Error when copying to the receive buffer,
1008 * usually because we weren't able to allocate
1009 * additional buffer pages. All we can do now
1010 * is turn off TCP_RCV_COPY_DATA, so the request
1011 * will not receive any additional updates,
1012 * and time out.
1013 * Any remaining data from this record will
1014 * be discarded.
1015 */
1016 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1017 dprintk("RPC: XID %08x truncated request\n",
1018 ntohl(transport->tcp_xid));
1019 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1020 "tcp_offset = %u, tcp_reclen = %u\n",
1021 xprt, transport->tcp_copied,
1022 transport->tcp_offset, transport->tcp_reclen);
1023 return;
1024 }
1025
1026 dprintk("RPC: XID %08x read %Zd bytes\n",
1027 ntohl(transport->tcp_xid), r);
1028 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1029 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1030 transport->tcp_offset, transport->tcp_reclen);
1031
1032 if (transport->tcp_copied == req->rq_private_buf.buflen)
1033 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1034 else if (transport->tcp_offset == transport->tcp_reclen) {
1035 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1036 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1037 }
1038
1039 return;
1040 }
1041
1042 /*
1043 * Finds the request corresponding to the RPC xid and invokes the common
1044 * tcp read code to read the data.
1045 */
1046 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1047 struct xdr_skb_reader *desc)
1048 {
1049 struct sock_xprt *transport =
1050 container_of(xprt, struct sock_xprt, xprt);
1051 struct rpc_rqst *req;
1052
1053 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1054
1055 /* Find and lock the request corresponding to this xid */
1056 spin_lock(&xprt->transport_lock);
1057 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1058 if (!req) {
1059 dprintk("RPC: XID %08x request not found!\n",
1060 ntohl(transport->tcp_xid));
1061 spin_unlock(&xprt->transport_lock);
1062 return -1;
1063 }
1064
1065 xs_tcp_read_common(xprt, desc, req);
1066
1067 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1068 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1069
1070 spin_unlock(&xprt->transport_lock);
1071 return 0;
1072 }
1073
1074 #if defined(CONFIG_NFS_V4_1)
1075 /*
1076 * Obtains an rpc_rqst previously allocated and invokes the common
1077 * tcp read code to read the data. The result is placed in the callback
1078 * queue.
1079 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1080 * connection and return -1.
1081 */
1082 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1083 struct xdr_skb_reader *desc)
1084 {
1085 struct sock_xprt *transport =
1086 container_of(xprt, struct sock_xprt, xprt);
1087 struct rpc_rqst *req;
1088
1089 req = xprt_alloc_bc_request(xprt);
1090 if (req == NULL) {
1091 printk(KERN_WARNING "Callback slot table overflowed\n");
1092 xprt_force_disconnect(xprt);
1093 return -1;
1094 }
1095
1096 req->rq_xid = transport->tcp_xid;
1097 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1098 xs_tcp_read_common(xprt, desc, req);
1099
1100 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1101 struct svc_serv *bc_serv = xprt->bc_serv;
1102
1103 /*
1104 * Add callback request to callback list. The callback
1105 * service sleeps on the sv_cb_waitq waiting for new
1106 * requests. Wake it up after adding enqueing the
1107 * request.
1108 */
1109 dprintk("RPC: add callback request to list\n");
1110 spin_lock(&bc_serv->sv_cb_lock);
1111 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1112 spin_unlock(&bc_serv->sv_cb_lock);
1113 wake_up(&bc_serv->sv_cb_waitq);
1114 }
1115
1116 req->rq_private_buf.len = transport->tcp_copied;
1117
1118 return 0;
1119 }
1120
1121 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1122 struct xdr_skb_reader *desc)
1123 {
1124 struct sock_xprt *transport =
1125 container_of(xprt, struct sock_xprt, xprt);
1126
1127 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1128 xs_tcp_read_reply(xprt, desc) :
1129 xs_tcp_read_callback(xprt, desc);
1130 }
1131 #else
1132 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1133 struct xdr_skb_reader *desc)
1134 {
1135 return xs_tcp_read_reply(xprt, desc);
1136 }
1137 #endif /* CONFIG_NFS_V4_1 */
1138
1139 /*
1140 * Read data off the transport. This can be either an RPC_CALL or an
1141 * RPC_REPLY. Relay the processing to helper functions.
1142 */
1143 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1144 struct xdr_skb_reader *desc)
1145 {
1146 struct sock_xprt *transport =
1147 container_of(xprt, struct sock_xprt, xprt);
1148
1149 if (_xs_tcp_read_data(xprt, desc) == 0)
1150 xs_tcp_check_fraghdr(transport);
1151 else {
1152 /*
1153 * The transport_lock protects the request handling.
1154 * There's no need to hold it to update the tcp_flags.
1155 */
1156 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1157 }
1158 }
1159
1160 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1161 {
1162 size_t len;
1163
1164 len = transport->tcp_reclen - transport->tcp_offset;
1165 if (len > desc->count)
1166 len = desc->count;
1167 desc->count -= len;
1168 desc->offset += len;
1169 transport->tcp_offset += len;
1170 dprintk("RPC: discarded %Zu bytes\n", len);
1171 xs_tcp_check_fraghdr(transport);
1172 }
1173
1174 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1175 {
1176 struct rpc_xprt *xprt = rd_desc->arg.data;
1177 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1178 struct xdr_skb_reader desc = {
1179 .skb = skb,
1180 .offset = offset,
1181 .count = len,
1182 };
1183
1184 dprintk("RPC: xs_tcp_data_recv started\n");
1185 do {
1186 /* Read in a new fragment marker if necessary */
1187 /* Can we ever really expect to get completely empty fragments? */
1188 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1189 xs_tcp_read_fraghdr(xprt, &desc);
1190 continue;
1191 }
1192 /* Read in the xid if necessary */
1193 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1194 xs_tcp_read_xid(transport, &desc);
1195 continue;
1196 }
1197 /* Read in the call/reply flag */
1198 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1199 xs_tcp_read_calldir(transport, &desc);
1200 continue;
1201 }
1202 /* Read in the request data */
1203 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1204 xs_tcp_read_data(xprt, &desc);
1205 continue;
1206 }
1207 /* Skip over any trailing bytes on short reads */
1208 xs_tcp_read_discard(transport, &desc);
1209 } while (desc.count);
1210 dprintk("RPC: xs_tcp_data_recv done\n");
1211 return len - desc.count;
1212 }
1213
1214 /**
1215 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1216 * @sk: socket with data to read
1217 * @bytes: how much data to read
1218 *
1219 */
1220 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1221 {
1222 struct rpc_xprt *xprt;
1223 read_descriptor_t rd_desc;
1224 int read;
1225
1226 dprintk("RPC: xs_tcp_data_ready...\n");
1227
1228 read_lock(&sk->sk_callback_lock);
1229 if (!(xprt = xprt_from_sock(sk)))
1230 goto out;
1231 if (xprt->shutdown)
1232 goto out;
1233
1234 /* Any data means we had a useful conversation, so
1235 * the we don't need to delay the next reconnect
1236 */
1237 if (xprt->reestablish_timeout)
1238 xprt->reestablish_timeout = 0;
1239
1240 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1241 rd_desc.arg.data = xprt;
1242 do {
1243 rd_desc.count = 65536;
1244 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1245 } while (read > 0);
1246 out:
1247 read_unlock(&sk->sk_callback_lock);
1248 }
1249
1250 /*
1251 * Do the equivalent of linger/linger2 handling for dealing with
1252 * broken servers that don't close the socket in a timely
1253 * fashion
1254 */
1255 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1256 unsigned long timeout)
1257 {
1258 struct sock_xprt *transport;
1259
1260 if (xprt_test_and_set_connecting(xprt))
1261 return;
1262 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1263 transport = container_of(xprt, struct sock_xprt, xprt);
1264 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1265 timeout);
1266 }
1267
1268 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1269 {
1270 struct sock_xprt *transport;
1271
1272 transport = container_of(xprt, struct sock_xprt, xprt);
1273
1274 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1275 !cancel_delayed_work(&transport->connect_worker))
1276 return;
1277 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1278 xprt_clear_connecting(xprt);
1279 }
1280
1281 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1282 {
1283 smp_mb__before_clear_bit();
1284 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1285 clear_bit(XPRT_CLOSING, &xprt->state);
1286 smp_mb__after_clear_bit();
1287 /* Mark transport as closed and wake up all pending tasks */
1288 xprt_disconnect_done(xprt);
1289 }
1290
1291 /**
1292 * xs_tcp_state_change - callback to handle TCP socket state changes
1293 * @sk: socket whose state has changed
1294 *
1295 */
1296 static void xs_tcp_state_change(struct sock *sk)
1297 {
1298 struct rpc_xprt *xprt;
1299
1300 read_lock(&sk->sk_callback_lock);
1301 if (!(xprt = xprt_from_sock(sk)))
1302 goto out;
1303 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1304 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1305 sk->sk_state, xprt_connected(xprt),
1306 sock_flag(sk, SOCK_DEAD),
1307 sock_flag(sk, SOCK_ZAPPED));
1308
1309 switch (sk->sk_state) {
1310 case TCP_ESTABLISHED:
1311 spin_lock_bh(&xprt->transport_lock);
1312 if (!xprt_test_and_set_connected(xprt)) {
1313 struct sock_xprt *transport = container_of(xprt,
1314 struct sock_xprt, xprt);
1315
1316 /* Reset TCP record info */
1317 transport->tcp_offset = 0;
1318 transport->tcp_reclen = 0;
1319 transport->tcp_copied = 0;
1320 transport->tcp_flags =
1321 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1322
1323 xprt_wake_pending_tasks(xprt, -EAGAIN);
1324 }
1325 spin_unlock_bh(&xprt->transport_lock);
1326 break;
1327 case TCP_FIN_WAIT1:
1328 /* The client initiated a shutdown of the socket */
1329 xprt->connect_cookie++;
1330 xprt->reestablish_timeout = 0;
1331 set_bit(XPRT_CLOSING, &xprt->state);
1332 smp_mb__before_clear_bit();
1333 clear_bit(XPRT_CONNECTED, &xprt->state);
1334 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1335 smp_mb__after_clear_bit();
1336 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1337 break;
1338 case TCP_CLOSE_WAIT:
1339 /* The server initiated a shutdown of the socket */
1340 xprt_force_disconnect(xprt);
1341 case TCP_SYN_SENT:
1342 xprt->connect_cookie++;
1343 case TCP_CLOSING:
1344 /*
1345 * If the server closed down the connection, make sure that
1346 * we back off before reconnecting
1347 */
1348 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1349 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1350 break;
1351 case TCP_LAST_ACK:
1352 set_bit(XPRT_CLOSING, &xprt->state);
1353 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1354 smp_mb__before_clear_bit();
1355 clear_bit(XPRT_CONNECTED, &xprt->state);
1356 smp_mb__after_clear_bit();
1357 break;
1358 case TCP_CLOSE:
1359 xs_tcp_cancel_linger_timeout(xprt);
1360 xs_sock_mark_closed(xprt);
1361 }
1362 out:
1363 read_unlock(&sk->sk_callback_lock);
1364 }
1365
1366 /**
1367 * xs_error_report - callback mainly for catching socket errors
1368 * @sk: socket
1369 */
1370 static void xs_error_report(struct sock *sk)
1371 {
1372 struct rpc_xprt *xprt;
1373
1374 read_lock(&sk->sk_callback_lock);
1375 if (!(xprt = xprt_from_sock(sk)))
1376 goto out;
1377 dprintk("RPC: %s client %p...\n"
1378 "RPC: error %d\n",
1379 __func__, xprt, sk->sk_err);
1380 xprt_wake_pending_tasks(xprt, -EAGAIN);
1381 out:
1382 read_unlock(&sk->sk_callback_lock);
1383 }
1384
1385 static void xs_write_space(struct sock *sk)
1386 {
1387 struct socket *sock;
1388 struct rpc_xprt *xprt;
1389
1390 if (unlikely(!(sock = sk->sk_socket)))
1391 return;
1392 clear_bit(SOCK_NOSPACE, &sock->flags);
1393
1394 if (unlikely(!(xprt = xprt_from_sock(sk))))
1395 return;
1396 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1397 return;
1398
1399 xprt_write_space(xprt);
1400 }
1401
1402 /**
1403 * xs_udp_write_space - callback invoked when socket buffer space
1404 * becomes available
1405 * @sk: socket whose state has changed
1406 *
1407 * Called when more output buffer space is available for this socket.
1408 * We try not to wake our writers until they can make "significant"
1409 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1410 * with a bunch of small requests.
1411 */
1412 static void xs_udp_write_space(struct sock *sk)
1413 {
1414 read_lock(&sk->sk_callback_lock);
1415
1416 /* from net/core/sock.c:sock_def_write_space */
1417 if (sock_writeable(sk))
1418 xs_write_space(sk);
1419
1420 read_unlock(&sk->sk_callback_lock);
1421 }
1422
1423 /**
1424 * xs_tcp_write_space - callback invoked when socket buffer space
1425 * becomes available
1426 * @sk: socket whose state has changed
1427 *
1428 * Called when more output buffer space is available for this socket.
1429 * We try not to wake our writers until they can make "significant"
1430 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1431 * with a bunch of small requests.
1432 */
1433 static void xs_tcp_write_space(struct sock *sk)
1434 {
1435 read_lock(&sk->sk_callback_lock);
1436
1437 /* from net/core/stream.c:sk_stream_write_space */
1438 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1439 xs_write_space(sk);
1440
1441 read_unlock(&sk->sk_callback_lock);
1442 }
1443
1444 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1445 {
1446 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1447 struct sock *sk = transport->inet;
1448
1449 if (transport->rcvsize) {
1450 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1451 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1452 }
1453 if (transport->sndsize) {
1454 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1455 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1456 sk->sk_write_space(sk);
1457 }
1458 }
1459
1460 /**
1461 * xs_udp_set_buffer_size - set send and receive limits
1462 * @xprt: generic transport
1463 * @sndsize: requested size of send buffer, in bytes
1464 * @rcvsize: requested size of receive buffer, in bytes
1465 *
1466 * Set socket send and receive buffer size limits.
1467 */
1468 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1469 {
1470 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1471
1472 transport->sndsize = 0;
1473 if (sndsize)
1474 transport->sndsize = sndsize + 1024;
1475 transport->rcvsize = 0;
1476 if (rcvsize)
1477 transport->rcvsize = rcvsize + 1024;
1478
1479 xs_udp_do_set_buffer_size(xprt);
1480 }
1481
1482 /**
1483 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1484 * @task: task that timed out
1485 *
1486 * Adjust the congestion window after a retransmit timeout has occurred.
1487 */
1488 static void xs_udp_timer(struct rpc_task *task)
1489 {
1490 xprt_adjust_cwnd(task, -ETIMEDOUT);
1491 }
1492
1493 static unsigned short xs_get_random_port(void)
1494 {
1495 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1496 unsigned short rand = (unsigned short) net_random() % range;
1497 return rand + xprt_min_resvport;
1498 }
1499
1500 /**
1501 * xs_set_port - reset the port number in the remote endpoint address
1502 * @xprt: generic transport
1503 * @port: new port number
1504 *
1505 */
1506 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1507 {
1508 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1509
1510 rpc_set_port(xs_addr(xprt), port);
1511 xs_update_peer_port(xprt);
1512 }
1513
1514 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1515 {
1516 unsigned short port = transport->srcport;
1517
1518 if (port == 0 && transport->xprt.resvport)
1519 port = xs_get_random_port();
1520 return port;
1521 }
1522
1523 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1524 {
1525 if (transport->srcport != 0)
1526 transport->srcport = 0;
1527 if (!transport->xprt.resvport)
1528 return 0;
1529 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1530 return xprt_max_resvport;
1531 return --port;
1532 }
1533
1534 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1535 {
1536 struct sockaddr_in myaddr = {
1537 .sin_family = AF_INET,
1538 };
1539 struct sockaddr_in *sa;
1540 int err, nloop = 0;
1541 unsigned short port = xs_get_srcport(transport, sock);
1542 unsigned short last;
1543
1544 sa = (struct sockaddr_in *)&transport->srcaddr;
1545 myaddr.sin_addr = sa->sin_addr;
1546 do {
1547 myaddr.sin_port = htons(port);
1548 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1549 sizeof(myaddr));
1550 if (port == 0)
1551 break;
1552 if (err == 0) {
1553 transport->srcport = port;
1554 break;
1555 }
1556 last = port;
1557 port = xs_next_srcport(transport, sock, port);
1558 if (port > last)
1559 nloop++;
1560 } while (err == -EADDRINUSE && nloop != 2);
1561 dprintk("RPC: %s %pI4:%u: %s (%d)\n",
1562 __func__, &myaddr.sin_addr,
1563 port, err ? "failed" : "ok", err);
1564 return err;
1565 }
1566
1567 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1568 {
1569 struct sockaddr_in6 myaddr = {
1570 .sin6_family = AF_INET6,
1571 };
1572 struct sockaddr_in6 *sa;
1573 int err, nloop = 0;
1574 unsigned short port = xs_get_srcport(transport, sock);
1575 unsigned short last;
1576
1577 sa = (struct sockaddr_in6 *)&transport->srcaddr;
1578 myaddr.sin6_addr = sa->sin6_addr;
1579 do {
1580 myaddr.sin6_port = htons(port);
1581 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1582 sizeof(myaddr));
1583 if (port == 0)
1584 break;
1585 if (err == 0) {
1586 transport->srcport = port;
1587 break;
1588 }
1589 last = port;
1590 port = xs_next_srcport(transport, sock, port);
1591 if (port > last)
1592 nloop++;
1593 } while (err == -EADDRINUSE && nloop != 2);
1594 dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n",
1595 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1596 return err;
1597 }
1598
1599 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1600 static struct lock_class_key xs_key[2];
1601 static struct lock_class_key xs_slock_key[2];
1602
1603 static inline void xs_reclassify_socket4(struct socket *sock)
1604 {
1605 struct sock *sk = sock->sk;
1606
1607 BUG_ON(sock_owned_by_user(sk));
1608 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1609 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1610 }
1611
1612 static inline void xs_reclassify_socket6(struct socket *sock)
1613 {
1614 struct sock *sk = sock->sk;
1615
1616 BUG_ON(sock_owned_by_user(sk));
1617 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1618 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1619 }
1620 #else
1621 static inline void xs_reclassify_socket4(struct socket *sock)
1622 {
1623 }
1624
1625 static inline void xs_reclassify_socket6(struct socket *sock)
1626 {
1627 }
1628 #endif
1629
1630 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1631 {
1632 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1633
1634 if (!transport->inet) {
1635 struct sock *sk = sock->sk;
1636
1637 write_lock_bh(&sk->sk_callback_lock);
1638
1639 xs_save_old_callbacks(transport, sk);
1640
1641 sk->sk_user_data = xprt;
1642 sk->sk_data_ready = xs_udp_data_ready;
1643 sk->sk_write_space = xs_udp_write_space;
1644 sk->sk_error_report = xs_error_report;
1645 sk->sk_no_check = UDP_CSUM_NORCV;
1646 sk->sk_allocation = GFP_ATOMIC;
1647
1648 xprt_set_connected(xprt);
1649
1650 /* Reset to new socket */
1651 transport->sock = sock;
1652 transport->inet = sk;
1653
1654 write_unlock_bh(&sk->sk_callback_lock);
1655 }
1656 xs_udp_do_set_buffer_size(xprt);
1657 }
1658
1659 /**
1660 * xs_udp_connect_worker4 - set up a UDP socket
1661 * @work: RPC transport to connect
1662 *
1663 * Invoked by a work queue tasklet.
1664 */
1665 static void xs_udp_connect_worker4(struct work_struct *work)
1666 {
1667 struct sock_xprt *transport =
1668 container_of(work, struct sock_xprt, connect_worker.work);
1669 struct rpc_xprt *xprt = &transport->xprt;
1670 struct socket *sock = transport->sock;
1671 int err, status = -EIO;
1672
1673 if (xprt->shutdown)
1674 goto out;
1675
1676 /* Start by resetting any existing state */
1677 xs_reset_transport(transport);
1678
1679 err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1680 if (err < 0) {
1681 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1682 goto out;
1683 }
1684 xs_reclassify_socket4(sock);
1685
1686 if (xs_bind4(transport, sock)) {
1687 sock_release(sock);
1688 goto out;
1689 }
1690
1691 dprintk("RPC: worker connecting xprt %p via %s to "
1692 "%s (port %s)\n", xprt,
1693 xprt->address_strings[RPC_DISPLAY_PROTO],
1694 xprt->address_strings[RPC_DISPLAY_ADDR],
1695 xprt->address_strings[RPC_DISPLAY_PORT]);
1696
1697 xs_udp_finish_connecting(xprt, sock);
1698 status = 0;
1699 out:
1700 xprt_clear_connecting(xprt);
1701 xprt_wake_pending_tasks(xprt, status);
1702 }
1703
1704 /**
1705 * xs_udp_connect_worker6 - set up a UDP socket
1706 * @work: RPC transport to connect
1707 *
1708 * Invoked by a work queue tasklet.
1709 */
1710 static void xs_udp_connect_worker6(struct work_struct *work)
1711 {
1712 struct sock_xprt *transport =
1713 container_of(work, struct sock_xprt, connect_worker.work);
1714 struct rpc_xprt *xprt = &transport->xprt;
1715 struct socket *sock = transport->sock;
1716 int err, status = -EIO;
1717
1718 if (xprt->shutdown)
1719 goto out;
1720
1721 /* Start by resetting any existing state */
1722 xs_reset_transport(transport);
1723
1724 err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1725 if (err < 0) {
1726 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1727 goto out;
1728 }
1729 xs_reclassify_socket6(sock);
1730
1731 if (xs_bind6(transport, sock) < 0) {
1732 sock_release(sock);
1733 goto out;
1734 }
1735
1736 dprintk("RPC: worker connecting xprt %p via %s to "
1737 "%s (port %s)\n", xprt,
1738 xprt->address_strings[RPC_DISPLAY_PROTO],
1739 xprt->address_strings[RPC_DISPLAY_ADDR],
1740 xprt->address_strings[RPC_DISPLAY_PORT]);
1741
1742 xs_udp_finish_connecting(xprt, sock);
1743 status = 0;
1744 out:
1745 xprt_clear_connecting(xprt);
1746 xprt_wake_pending_tasks(xprt, status);
1747 }
1748
1749 /*
1750 * We need to preserve the port number so the reply cache on the server can
1751 * find our cached RPC replies when we get around to reconnecting.
1752 */
1753 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1754 {
1755 int result;
1756 struct sockaddr any;
1757
1758 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1759
1760 /*
1761 * Disconnect the transport socket by doing a connect operation
1762 * with AF_UNSPEC. This should return immediately...
1763 */
1764 memset(&any, 0, sizeof(any));
1765 any.sa_family = AF_UNSPEC;
1766 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1767 if (!result)
1768 xs_sock_mark_closed(xprt);
1769 else
1770 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1771 result);
1772 }
1773
1774 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1775 {
1776 unsigned int state = transport->inet->sk_state;
1777
1778 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1779 return;
1780 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1781 return;
1782 xs_abort_connection(xprt, transport);
1783 }
1784
1785 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1786 {
1787 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1788
1789 if (!transport->inet) {
1790 struct sock *sk = sock->sk;
1791
1792 write_lock_bh(&sk->sk_callback_lock);
1793
1794 xs_save_old_callbacks(transport, sk);
1795
1796 sk->sk_user_data = xprt;
1797 sk->sk_data_ready = xs_tcp_data_ready;
1798 sk->sk_state_change = xs_tcp_state_change;
1799 sk->sk_write_space = xs_tcp_write_space;
1800 sk->sk_error_report = xs_error_report;
1801 sk->sk_allocation = GFP_ATOMIC;
1802
1803 /* socket options */
1804 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1805 sock_reset_flag(sk, SOCK_LINGER);
1806 tcp_sk(sk)->linger2 = 0;
1807 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1808
1809 xprt_clear_connected(xprt);
1810
1811 /* Reset to new socket */
1812 transport->sock = sock;
1813 transport->inet = sk;
1814
1815 write_unlock_bh(&sk->sk_callback_lock);
1816 }
1817
1818 if (!xprt_bound(xprt))
1819 return -ENOTCONN;
1820
1821 /* Tell the socket layer to start connecting... */
1822 xprt->stat.connect_count++;
1823 xprt->stat.connect_start = jiffies;
1824 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1825 }
1826
1827 /**
1828 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1829 * @xprt: RPC transport to connect
1830 * @transport: socket transport to connect
1831 * @create_sock: function to create a socket of the correct type
1832 *
1833 * Invoked by a work queue tasklet.
1834 */
1835 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1836 struct sock_xprt *transport,
1837 struct socket *(*create_sock)(struct rpc_xprt *,
1838 struct sock_xprt *))
1839 {
1840 struct socket *sock = transport->sock;
1841 int status = -EIO;
1842
1843 if (xprt->shutdown)
1844 goto out;
1845
1846 if (!sock) {
1847 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1848 sock = create_sock(xprt, transport);
1849 if (IS_ERR(sock)) {
1850 status = PTR_ERR(sock);
1851 goto out;
1852 }
1853 } else {
1854 int abort_and_exit;
1855
1856 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1857 &xprt->state);
1858 /* "close" the socket, preserving the local port */
1859 xs_tcp_reuse_connection(xprt, transport);
1860
1861 if (abort_and_exit)
1862 goto out_eagain;
1863 }
1864
1865 dprintk("RPC: worker connecting xprt %p via %s to "
1866 "%s (port %s)\n", xprt,
1867 xprt->address_strings[RPC_DISPLAY_PROTO],
1868 xprt->address_strings[RPC_DISPLAY_ADDR],
1869 xprt->address_strings[RPC_DISPLAY_PORT]);
1870
1871 status = xs_tcp_finish_connecting(xprt, sock);
1872 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1873 xprt, -status, xprt_connected(xprt),
1874 sock->sk->sk_state);
1875 switch (status) {
1876 default:
1877 printk("%s: connect returned unhandled error %d\n",
1878 __func__, status);
1879 case -EADDRNOTAVAIL:
1880 /* We're probably in TIME_WAIT. Get rid of existing socket,
1881 * and retry
1882 */
1883 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1884 xprt_force_disconnect(xprt);
1885 break;
1886 case -ECONNREFUSED:
1887 case -ECONNRESET:
1888 case -ENETUNREACH:
1889 /* retry with existing socket, after a delay */
1890 case 0:
1891 case -EINPROGRESS:
1892 case -EALREADY:
1893 xprt_clear_connecting(xprt);
1894 return;
1895 case -EINVAL:
1896 /* Happens, for instance, if the user specified a link
1897 * local IPv6 address without a scope-id.
1898 */
1899 goto out;
1900 }
1901 out_eagain:
1902 status = -EAGAIN;
1903 out:
1904 xprt_clear_connecting(xprt);
1905 xprt_wake_pending_tasks(xprt, status);
1906 }
1907
1908 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1909 struct sock_xprt *transport)
1910 {
1911 struct socket *sock;
1912 int err;
1913
1914 /* start from scratch */
1915 err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1916 if (err < 0) {
1917 dprintk("RPC: can't create TCP transport socket (%d).\n",
1918 -err);
1919 goto out_err;
1920 }
1921 xs_reclassify_socket4(sock);
1922
1923 if (xs_bind4(transport, sock) < 0) {
1924 sock_release(sock);
1925 goto out_err;
1926 }
1927 return sock;
1928 out_err:
1929 return ERR_PTR(-EIO);
1930 }
1931
1932 /**
1933 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1934 * @work: RPC transport to connect
1935 *
1936 * Invoked by a work queue tasklet.
1937 */
1938 static void xs_tcp_connect_worker4(struct work_struct *work)
1939 {
1940 struct sock_xprt *transport =
1941 container_of(work, struct sock_xprt, connect_worker.work);
1942 struct rpc_xprt *xprt = &transport->xprt;
1943
1944 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
1945 }
1946
1947 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
1948 struct sock_xprt *transport)
1949 {
1950 struct socket *sock;
1951 int err;
1952
1953 /* start from scratch */
1954 err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
1955 if (err < 0) {
1956 dprintk("RPC: can't create TCP transport socket (%d).\n",
1957 -err);
1958 goto out_err;
1959 }
1960 xs_reclassify_socket6(sock);
1961
1962 if (xs_bind6(transport, sock) < 0) {
1963 sock_release(sock);
1964 goto out_err;
1965 }
1966 return sock;
1967 out_err:
1968 return ERR_PTR(-EIO);
1969 }
1970
1971 /**
1972 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1973 * @work: RPC transport to connect
1974 *
1975 * Invoked by a work queue tasklet.
1976 */
1977 static void xs_tcp_connect_worker6(struct work_struct *work)
1978 {
1979 struct sock_xprt *transport =
1980 container_of(work, struct sock_xprt, connect_worker.work);
1981 struct rpc_xprt *xprt = &transport->xprt;
1982
1983 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
1984 }
1985
1986 /**
1987 * xs_connect - connect a socket to a remote endpoint
1988 * @task: address of RPC task that manages state of connect request
1989 *
1990 * TCP: If the remote end dropped the connection, delay reconnecting.
1991 *
1992 * UDP socket connects are synchronous, but we use a work queue anyway
1993 * to guarantee that even unprivileged user processes can set up a
1994 * socket on a privileged port.
1995 *
1996 * If a UDP socket connect fails, the delay behavior here prevents
1997 * retry floods (hard mounts).
1998 */
1999 static void xs_connect(struct rpc_task *task)
2000 {
2001 struct rpc_xprt *xprt = task->tk_xprt;
2002 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2003
2004 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2005 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2006 "seconds\n",
2007 xprt, xprt->reestablish_timeout / HZ);
2008 queue_delayed_work(rpciod_workqueue,
2009 &transport->connect_worker,
2010 xprt->reestablish_timeout);
2011 xprt->reestablish_timeout <<= 1;
2012 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2013 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2014 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2015 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2016 } else {
2017 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2018 queue_delayed_work(rpciod_workqueue,
2019 &transport->connect_worker, 0);
2020 }
2021 }
2022
2023 /**
2024 * xs_udp_print_stats - display UDP socket-specifc stats
2025 * @xprt: rpc_xprt struct containing statistics
2026 * @seq: output file
2027 *
2028 */
2029 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2030 {
2031 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2032
2033 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2034 transport->srcport,
2035 xprt->stat.bind_count,
2036 xprt->stat.sends,
2037 xprt->stat.recvs,
2038 xprt->stat.bad_xids,
2039 xprt->stat.req_u,
2040 xprt->stat.bklog_u);
2041 }
2042
2043 /**
2044 * xs_tcp_print_stats - display TCP socket-specifc stats
2045 * @xprt: rpc_xprt struct containing statistics
2046 * @seq: output file
2047 *
2048 */
2049 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2050 {
2051 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2052 long idle_time = 0;
2053
2054 if (xprt_connected(xprt))
2055 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2056
2057 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2058 transport->srcport,
2059 xprt->stat.bind_count,
2060 xprt->stat.connect_count,
2061 xprt->stat.connect_time,
2062 idle_time,
2063 xprt->stat.sends,
2064 xprt->stat.recvs,
2065 xprt->stat.bad_xids,
2066 xprt->stat.req_u,
2067 xprt->stat.bklog_u);
2068 }
2069
2070 /*
2071 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2072 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2073 * to use the server side send routines.
2074 */
2075 static void *bc_malloc(struct rpc_task *task, size_t size)
2076 {
2077 struct page *page;
2078 struct rpc_buffer *buf;
2079
2080 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2081 page = alloc_page(GFP_KERNEL);
2082
2083 if (!page)
2084 return NULL;
2085
2086 buf = page_address(page);
2087 buf->len = PAGE_SIZE;
2088
2089 return buf->data;
2090 }
2091
2092 /*
2093 * Free the space allocated in the bc_alloc routine
2094 */
2095 static void bc_free(void *buffer)
2096 {
2097 struct rpc_buffer *buf;
2098
2099 if (!buffer)
2100 return;
2101
2102 buf = container_of(buffer, struct rpc_buffer, data);
2103 free_page((unsigned long)buf);
2104 }
2105
2106 /*
2107 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2108 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2109 */
2110 static int bc_sendto(struct rpc_rqst *req)
2111 {
2112 int len;
2113 struct xdr_buf *xbufp = &req->rq_snd_buf;
2114 struct rpc_xprt *xprt = req->rq_xprt;
2115 struct sock_xprt *transport =
2116 container_of(xprt, struct sock_xprt, xprt);
2117 struct socket *sock = transport->sock;
2118 unsigned long headoff;
2119 unsigned long tailoff;
2120
2121 /*
2122 * Set up the rpc header and record marker stuff
2123 */
2124 xs_encode_tcp_record_marker(xbufp);
2125
2126 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2127 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2128 len = svc_send_common(sock, xbufp,
2129 virt_to_page(xbufp->head[0].iov_base), headoff,
2130 xbufp->tail[0].iov_base, tailoff);
2131
2132 if (len != xbufp->len) {
2133 printk(KERN_NOTICE "Error sending entire callback!\n");
2134 len = -EAGAIN;
2135 }
2136
2137 return len;
2138 }
2139
2140 /*
2141 * The send routine. Borrows from svc_send
2142 */
2143 static int bc_send_request(struct rpc_task *task)
2144 {
2145 struct rpc_rqst *req = task->tk_rqstp;
2146 struct svc_xprt *xprt;
2147 struct svc_sock *svsk;
2148 u32 len;
2149
2150 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2151 /*
2152 * Get the server socket associated with this callback xprt
2153 */
2154 xprt = req->rq_xprt->bc_xprt;
2155 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2156
2157 /*
2158 * Grab the mutex to serialize data as the connection is shared
2159 * with the fore channel
2160 */
2161 if (!mutex_trylock(&xprt->xpt_mutex)) {
2162 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2163 if (!mutex_trylock(&xprt->xpt_mutex))
2164 return -EAGAIN;
2165 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2166 }
2167 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2168 len = -ENOTCONN;
2169 else
2170 len = bc_sendto(req);
2171 mutex_unlock(&xprt->xpt_mutex);
2172
2173 if (len > 0)
2174 len = 0;
2175
2176 return len;
2177 }
2178
2179 /*
2180 * The close routine. Since this is client initiated, we do nothing
2181 */
2182
2183 static void bc_close(struct rpc_xprt *xprt)
2184 {
2185 return;
2186 }
2187
2188 /*
2189 * The xprt destroy routine. Again, because this connection is client
2190 * initiated, we do nothing
2191 */
2192
2193 static void bc_destroy(struct rpc_xprt *xprt)
2194 {
2195 return;
2196 }
2197
2198 static struct rpc_xprt_ops xs_udp_ops = {
2199 .set_buffer_size = xs_udp_set_buffer_size,
2200 .reserve_xprt = xprt_reserve_xprt_cong,
2201 .release_xprt = xprt_release_xprt_cong,
2202 .rpcbind = rpcb_getport_async,
2203 .set_port = xs_set_port,
2204 .connect = xs_connect,
2205 .buf_alloc = rpc_malloc,
2206 .buf_free = rpc_free,
2207 .send_request = xs_udp_send_request,
2208 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2209 .timer = xs_udp_timer,
2210 .release_request = xprt_release_rqst_cong,
2211 .close = xs_close,
2212 .destroy = xs_destroy,
2213 .print_stats = xs_udp_print_stats,
2214 };
2215
2216 static struct rpc_xprt_ops xs_tcp_ops = {
2217 .reserve_xprt = xprt_reserve_xprt,
2218 .release_xprt = xs_tcp_release_xprt,
2219 .rpcbind = rpcb_getport_async,
2220 .set_port = xs_set_port,
2221 .connect = xs_connect,
2222 .buf_alloc = rpc_malloc,
2223 .buf_free = rpc_free,
2224 .send_request = xs_tcp_send_request,
2225 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2226 .close = xs_tcp_close,
2227 .destroy = xs_destroy,
2228 .print_stats = xs_tcp_print_stats,
2229 };
2230
2231 /*
2232 * The rpc_xprt_ops for the server backchannel
2233 */
2234
2235 static struct rpc_xprt_ops bc_tcp_ops = {
2236 .reserve_xprt = xprt_reserve_xprt,
2237 .release_xprt = xprt_release_xprt,
2238 .buf_alloc = bc_malloc,
2239 .buf_free = bc_free,
2240 .send_request = bc_send_request,
2241 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2242 .close = bc_close,
2243 .destroy = bc_destroy,
2244 .print_stats = xs_tcp_print_stats,
2245 };
2246
2247 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2248 unsigned int slot_table_size)
2249 {
2250 struct rpc_xprt *xprt;
2251 struct sock_xprt *new;
2252
2253 if (args->addrlen > sizeof(xprt->addr)) {
2254 dprintk("RPC: xs_setup_xprt: address too large\n");
2255 return ERR_PTR(-EBADF);
2256 }
2257
2258 new = kzalloc(sizeof(*new), GFP_KERNEL);
2259 if (new == NULL) {
2260 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2261 "rpc_xprt\n");
2262 return ERR_PTR(-ENOMEM);
2263 }
2264 xprt = &new->xprt;
2265
2266 xprt->max_reqs = slot_table_size;
2267 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2268 if (xprt->slot == NULL) {
2269 kfree(xprt);
2270 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
2271 "table\n");
2272 return ERR_PTR(-ENOMEM);
2273 }
2274
2275 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2276 xprt->addrlen = args->addrlen;
2277 if (args->srcaddr)
2278 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2279
2280 return xprt;
2281 }
2282
2283 static const struct rpc_timeout xs_udp_default_timeout = {
2284 .to_initval = 5 * HZ,
2285 .to_maxval = 30 * HZ,
2286 .to_increment = 5 * HZ,
2287 .to_retries = 5,
2288 };
2289
2290 /**
2291 * xs_setup_udp - Set up transport to use a UDP socket
2292 * @args: rpc transport creation arguments
2293 *
2294 */
2295 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2296 {
2297 struct sockaddr *addr = args->dstaddr;
2298 struct rpc_xprt *xprt;
2299 struct sock_xprt *transport;
2300
2301 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2302 if (IS_ERR(xprt))
2303 return xprt;
2304 transport = container_of(xprt, struct sock_xprt, xprt);
2305
2306 xprt->prot = IPPROTO_UDP;
2307 xprt->tsh_size = 0;
2308 /* XXX: header size can vary due to auth type, IPv6, etc. */
2309 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2310
2311 xprt->bind_timeout = XS_BIND_TO;
2312 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2313 xprt->idle_timeout = XS_IDLE_DISC_TO;
2314
2315 xprt->ops = &xs_udp_ops;
2316
2317 xprt->timeout = &xs_udp_default_timeout;
2318
2319 switch (addr->sa_family) {
2320 case AF_INET:
2321 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2322 xprt_set_bound(xprt);
2323
2324 INIT_DELAYED_WORK(&transport->connect_worker,
2325 xs_udp_connect_worker4);
2326 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2327 break;
2328 case AF_INET6:
2329 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2330 xprt_set_bound(xprt);
2331
2332 INIT_DELAYED_WORK(&transport->connect_worker,
2333 xs_udp_connect_worker6);
2334 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2335 break;
2336 default:
2337 kfree(xprt);
2338 return ERR_PTR(-EAFNOSUPPORT);
2339 }
2340
2341 if (xprt_bound(xprt))
2342 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2343 xprt->address_strings[RPC_DISPLAY_ADDR],
2344 xprt->address_strings[RPC_DISPLAY_PORT],
2345 xprt->address_strings[RPC_DISPLAY_PROTO]);
2346 else
2347 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2348 xprt->address_strings[RPC_DISPLAY_ADDR],
2349 xprt->address_strings[RPC_DISPLAY_PROTO]);
2350
2351 if (try_module_get(THIS_MODULE))
2352 return xprt;
2353
2354 kfree(xprt->slot);
2355 kfree(xprt);
2356 return ERR_PTR(-EINVAL);
2357 }
2358
2359 static const struct rpc_timeout xs_tcp_default_timeout = {
2360 .to_initval = 60 * HZ,
2361 .to_maxval = 60 * HZ,
2362 .to_retries = 2,
2363 };
2364
2365 /**
2366 * xs_setup_tcp - Set up transport to use a TCP socket
2367 * @args: rpc transport creation arguments
2368 *
2369 */
2370 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2371 {
2372 struct sockaddr *addr = args->dstaddr;
2373 struct rpc_xprt *xprt;
2374 struct sock_xprt *transport;
2375
2376 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2377 if (IS_ERR(xprt))
2378 return xprt;
2379 transport = container_of(xprt, struct sock_xprt, xprt);
2380
2381 xprt->prot = IPPROTO_TCP;
2382 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2383 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2384
2385 xprt->bind_timeout = XS_BIND_TO;
2386 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2387 xprt->idle_timeout = XS_IDLE_DISC_TO;
2388
2389 xprt->ops = &xs_tcp_ops;
2390 xprt->timeout = &xs_tcp_default_timeout;
2391
2392 switch (addr->sa_family) {
2393 case AF_INET:
2394 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2395 xprt_set_bound(xprt);
2396
2397 INIT_DELAYED_WORK(&transport->connect_worker,
2398 xs_tcp_connect_worker4);
2399 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2400 break;
2401 case AF_INET6:
2402 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2403 xprt_set_bound(xprt);
2404
2405 INIT_DELAYED_WORK(&transport->connect_worker,
2406 xs_tcp_connect_worker6);
2407 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2408 break;
2409 default:
2410 kfree(xprt);
2411 return ERR_PTR(-EAFNOSUPPORT);
2412 }
2413
2414 if (xprt_bound(xprt))
2415 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2416 xprt->address_strings[RPC_DISPLAY_ADDR],
2417 xprt->address_strings[RPC_DISPLAY_PORT],
2418 xprt->address_strings[RPC_DISPLAY_PROTO]);
2419 else
2420 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2421 xprt->address_strings[RPC_DISPLAY_ADDR],
2422 xprt->address_strings[RPC_DISPLAY_PROTO]);
2423
2424
2425 if (try_module_get(THIS_MODULE))
2426 return xprt;
2427
2428 kfree(xprt->slot);
2429 kfree(xprt);
2430 return ERR_PTR(-EINVAL);
2431 }
2432
2433 /**
2434 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2435 * @args: rpc transport creation arguments
2436 *
2437 */
2438 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2439 {
2440 struct sockaddr *addr = args->dstaddr;
2441 struct rpc_xprt *xprt;
2442 struct sock_xprt *transport;
2443 struct svc_sock *bc_sock;
2444
2445 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2446 if (IS_ERR(xprt))
2447 return xprt;
2448 transport = container_of(xprt, struct sock_xprt, xprt);
2449
2450 xprt->prot = IPPROTO_TCP;
2451 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2452 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2453 xprt->timeout = &xs_tcp_default_timeout;
2454
2455 /* backchannel */
2456 xprt_set_bound(xprt);
2457 xprt->bind_timeout = 0;
2458 xprt->reestablish_timeout = 0;
2459 xprt->idle_timeout = 0;
2460
2461 /*
2462 * The backchannel uses the same socket connection as the
2463 * forechannel
2464 */
2465 xprt->bc_xprt = args->bc_xprt;
2466 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2467 bc_sock->sk_bc_xprt = xprt;
2468 transport->sock = bc_sock->sk_sock;
2469 transport->inet = bc_sock->sk_sk;
2470
2471 xprt->ops = &bc_tcp_ops;
2472
2473 switch (addr->sa_family) {
2474 case AF_INET:
2475 xs_format_peer_addresses(xprt, "tcp",
2476 RPCBIND_NETID_TCP);
2477 break;
2478 case AF_INET6:
2479 xs_format_peer_addresses(xprt, "tcp",
2480 RPCBIND_NETID_TCP6);
2481 break;
2482 default:
2483 kfree(xprt);
2484 return ERR_PTR(-EAFNOSUPPORT);
2485 }
2486
2487 if (xprt_bound(xprt))
2488 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2489 xprt->address_strings[RPC_DISPLAY_ADDR],
2490 xprt->address_strings[RPC_DISPLAY_PORT],
2491 xprt->address_strings[RPC_DISPLAY_PROTO]);
2492 else
2493 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2494 xprt->address_strings[RPC_DISPLAY_ADDR],
2495 xprt->address_strings[RPC_DISPLAY_PROTO]);
2496
2497 /*
2498 * Since we don't want connections for the backchannel, we set
2499 * the xprt status to connected
2500 */
2501 xprt_set_connected(xprt);
2502
2503
2504 if (try_module_get(THIS_MODULE))
2505 return xprt;
2506 kfree(xprt->slot);
2507 kfree(xprt);
2508 return ERR_PTR(-EINVAL);
2509 }
2510
2511 static struct xprt_class xs_udp_transport = {
2512 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2513 .name = "udp",
2514 .owner = THIS_MODULE,
2515 .ident = XPRT_TRANSPORT_UDP,
2516 .setup = xs_setup_udp,
2517 };
2518
2519 static struct xprt_class xs_tcp_transport = {
2520 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2521 .name = "tcp",
2522 .owner = THIS_MODULE,
2523 .ident = XPRT_TRANSPORT_TCP,
2524 .setup = xs_setup_tcp,
2525 };
2526
2527 static struct xprt_class xs_bc_tcp_transport = {
2528 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2529 .name = "tcp NFSv4.1 backchannel",
2530 .owner = THIS_MODULE,
2531 .ident = XPRT_TRANSPORT_BC_TCP,
2532 .setup = xs_setup_bc_tcp,
2533 };
2534
2535 /**
2536 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2537 *
2538 */
2539 int init_socket_xprt(void)
2540 {
2541 #ifdef RPC_DEBUG
2542 if (!sunrpc_table_header)
2543 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2544 #endif
2545
2546 xprt_register_transport(&xs_udp_transport);
2547 xprt_register_transport(&xs_tcp_transport);
2548 xprt_register_transport(&xs_bc_tcp_transport);
2549
2550 return 0;
2551 }
2552
2553 /**
2554 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2555 *
2556 */
2557 void cleanup_socket_xprt(void)
2558 {
2559 #ifdef RPC_DEBUG
2560 if (sunrpc_table_header) {
2561 unregister_sysctl_table(sunrpc_table_header);
2562 sunrpc_table_header = NULL;
2563 }
2564 #endif
2565
2566 xprt_unregister_transport(&xs_udp_transport);
2567 xprt_unregister_transport(&xs_tcp_transport);
2568 xprt_unregister_transport(&xs_bc_tcp_transport);
2569 }
2570
2571 static int param_set_uint_minmax(const char *val, struct kernel_param *kp,
2572 unsigned int min, unsigned int max)
2573 {
2574 unsigned long num;
2575 int ret;
2576
2577 if (!val)
2578 return -EINVAL;
2579 ret = strict_strtoul(val, 0, &num);
2580 if (ret == -EINVAL || num < min || num > max)
2581 return -EINVAL;
2582 *((unsigned int *)kp->arg) = num;
2583 return 0;
2584 }
2585
2586 static int param_set_portnr(const char *val, struct kernel_param *kp)
2587 {
2588 return param_set_uint_minmax(val, kp,
2589 RPC_MIN_RESVPORT,
2590 RPC_MAX_RESVPORT);
2591 }
2592
2593 static int param_get_portnr(char *buffer, struct kernel_param *kp)
2594 {
2595 return param_get_uint(buffer, kp);
2596 }
2597 #define param_check_portnr(name, p) \
2598 __param_check(name, p, unsigned int);
2599
2600 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2601 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2602
2603 static int param_set_slot_table_size(const char *val, struct kernel_param *kp)
2604 {
2605 return param_set_uint_minmax(val, kp,
2606 RPC_MIN_SLOT_TABLE,
2607 RPC_MAX_SLOT_TABLE);
2608 }
2609
2610 static int param_get_slot_table_size(char *buffer, struct kernel_param *kp)
2611 {
2612 return param_get_uint(buffer, kp);
2613 }
2614 #define param_check_slot_table_size(name, p) \
2615 __param_check(name, p, unsigned int);
2616
2617 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2618 slot_table_size, 0644);
2619 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2620 slot_table_size, 0644);
2621
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree