?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
SUNRPC: Rename xs_encode_tcp_fragment_header()
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sunrpc / xprtsock.c
blob079366974e1fef57417e18d55d64483960ff32f4
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 tcp_calldir;
215
216 u32 tcp_offset,
217 tcp_reclen;
218
219 unsigned long tcp_copied,
220 tcp_flags;
221
222 /*
223 * Connection of transports
224 */
225 struct delayed_work connect_worker;
226 struct sockaddr_storage srcaddr;
227 unsigned short srcport;
228
229 /*
230 * UDP socket buffer size parameters
231 */
232 size_t rcvsize,
233 sndsize;
234
235 /*
236 * Saved socket callback addresses
237 */
238 void (*old_data_ready)(struct sock *, int);
239 void (*old_state_change)(struct sock *);
240 void (*old_write_space)(struct sock *);
241 void (*old_error_report)(struct sock *);
242 };
243
244 /*
245 * TCP receive state flags
246 */
247 #define TCP_RCV_LAST_FRAG (1UL << 0)
248 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
249 #define TCP_RCV_COPY_XID (1UL << 2)
250 #define TCP_RCV_COPY_DATA (1UL << 3)
251 #define TCP_RCV_READ_CALLDIR (1UL << 4)
252 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
253
254 /*
255 * TCP RPC flags
256 */
257 #define TCP_RPC_REPLY (1UL << 6)
258
259 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
260 {
261 return (struct sockaddr *) &xprt->addr;
262 }
263
264 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
265 {
266 return (struct sockaddr_in *) &xprt->addr;
267 }
268
269 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
270 {
271 return (struct sockaddr_in6 *) &xprt->addr;
272 }
273
274 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
275 {
276 struct sockaddr *sap = xs_addr(xprt);
277 struct sockaddr_in6 *sin6;
278 struct sockaddr_in *sin;
279 char buf[128];
280
281 (void)rpc_ntop(sap, buf, sizeof(buf));
282 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
283
284 switch (sap->sa_family) {
285 case AF_INET:
286 sin = xs_addr_in(xprt);
287 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
288 break;
289 case AF_INET6:
290 sin6 = xs_addr_in6(xprt);
291 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
292 break;
293 default:
294 BUG();
295 }
296 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
297 }
298
299 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
300 {
301 struct sockaddr *sap = xs_addr(xprt);
302 char buf[128];
303
304 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
305 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
306
307 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
308 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
309 }
310
311 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
312 const char *protocol,
313 const char *netid)
314 {
315 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
316 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
317 xs_format_common_peer_addresses(xprt);
318 xs_format_common_peer_ports(xprt);
319 }
320
321 static void xs_update_peer_port(struct rpc_xprt *xprt)
322 {
323 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
324 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
325
326 xs_format_common_peer_ports(xprt);
327 }
328
329 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
330 {
331 unsigned int i;
332
333 for (i = 0; i < RPC_DISPLAY_MAX; i++)
334 switch (i) {
335 case RPC_DISPLAY_PROTO:
336 case RPC_DISPLAY_NETID:
337 continue;
338 default:
339 kfree(xprt->address_strings[i]);
340 }
341 }
342
343 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
344
345 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
346 {
347 struct msghdr msg = {
348 .msg_name = addr,
349 .msg_namelen = addrlen,
350 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
351 };
352 struct kvec iov = {
353 .iov_base = vec->iov_base + base,
354 .iov_len = vec->iov_len - base,
355 };
356
357 if (iov.iov_len != 0)
358 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
359 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
360 }
361
362 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
363 {
364 struct page **ppage;
365 unsigned int remainder;
366 int err, sent = 0;
367
368 remainder = xdr->page_len - base;
369 base += xdr->page_base;
370 ppage = xdr->pages + (base >> PAGE_SHIFT);
371 base &= ~PAGE_MASK;
372 for(;;) {
373 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
374 int flags = XS_SENDMSG_FLAGS;
375
376 remainder -= len;
377 if (remainder != 0 || more)
378 flags |= MSG_MORE;
379 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
380 if (remainder == 0 || err != len)
381 break;
382 sent += err;
383 ppage++;
384 base = 0;
385 }
386 if (sent == 0)
387 return err;
388 if (err > 0)
389 sent += err;
390 return sent;
391 }
392
393 /**
394 * xs_sendpages - write pages directly to a socket
395 * @sock: socket to send on
396 * @addr: UDP only -- address of destination
397 * @addrlen: UDP only -- length of destination address
398 * @xdr: buffer containing this request
399 * @base: starting position in the buffer
400 *
401 */
402 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
403 {
404 unsigned int remainder = xdr->len - base;
405 int err, sent = 0;
406
407 if (unlikely(!sock))
408 return -ENOTSOCK;
409
410 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
411 if (base != 0) {
412 addr = NULL;
413 addrlen = 0;
414 }
415
416 if (base < xdr->head[0].iov_len || addr != NULL) {
417 unsigned int len = xdr->head[0].iov_len - base;
418 remainder -= len;
419 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
420 if (remainder == 0 || err != len)
421 goto out;
422 sent += err;
423 base = 0;
424 } else
425 base -= xdr->head[0].iov_len;
426
427 if (base < xdr->page_len) {
428 unsigned int len = xdr->page_len - base;
429 remainder -= len;
430 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
431 if (remainder == 0 || err != len)
432 goto out;
433 sent += err;
434 base = 0;
435 } else
436 base -= xdr->page_len;
437
438 if (base >= xdr->tail[0].iov_len)
439 return sent;
440 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
441 out:
442 if (sent == 0)
443 return err;
444 if (err > 0)
445 sent += err;
446 return sent;
447 }
448
449 static void xs_nospace_callback(struct rpc_task *task)
450 {
451 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
452
453 transport->inet->sk_write_pending--;
454 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
455 }
456
457 /**
458 * xs_nospace - place task on wait queue if transmit was incomplete
459 * @task: task to put to sleep
460 *
461 */
462 static int xs_nospace(struct rpc_task *task)
463 {
464 struct rpc_rqst *req = task->tk_rqstp;
465 struct rpc_xprt *xprt = req->rq_xprt;
466 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
467 int ret = 0;
468
469 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
470 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
471 req->rq_slen);
472
473 /* Protect against races with write_space */
474 spin_lock_bh(&xprt->transport_lock);
475
476 /* Don't race with disconnect */
477 if (xprt_connected(xprt)) {
478 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
479 ret = -EAGAIN;
480 /*
481 * Notify TCP that we're limited by the application
482 * window size
483 */
484 set_bit(SOCK_NOSPACE, &transport->sock->flags);
485 transport->inet->sk_write_pending++;
486 /* ...and wait for more buffer space */
487 xprt_wait_for_buffer_space(task, xs_nospace_callback);
488 }
489 } else {
490 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
491 ret = -ENOTCONN;
492 }
493
494 spin_unlock_bh(&xprt->transport_lock);
495 return ret;
496 }
497
498 /*
499 * Construct a stream transport record marker in @buf.
500 */
501 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
502 {
503 u32 reclen = buf->len - sizeof(rpc_fraghdr);
504 rpc_fraghdr *base = buf->head[0].iov_base;
505 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
506 }
507
508 /**
509 * xs_udp_send_request - write an RPC request to a UDP socket
510 * @task: address of RPC task that manages the state of an RPC request
511 *
512 * Return values:
513 * 0: The request has been sent
514 * EAGAIN: The socket was blocked, please call again later to
515 * complete the request
516 * ENOTCONN: Caller needs to invoke connect logic then call again
517 * other: Some other error occurred, the request was not sent
518 */
519 static int xs_udp_send_request(struct rpc_task *task)
520 {
521 struct rpc_rqst *req = task->tk_rqstp;
522 struct rpc_xprt *xprt = req->rq_xprt;
523 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
524 struct xdr_buf *xdr = &req->rq_snd_buf;
525 int status;
526
527 xs_pktdump("packet data:",
528 req->rq_svec->iov_base,
529 req->rq_svec->iov_len);
530
531 if (!xprt_bound(xprt))
532 return -ENOTCONN;
533 status = xs_sendpages(transport->sock,
534 xs_addr(xprt),
535 xprt->addrlen, xdr,
536 req->rq_bytes_sent);
537
538 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
539 xdr->len - req->rq_bytes_sent, status);
540
541 if (status >= 0) {
542 req->rq_xmit_bytes_sent += status;
543 if (status >= req->rq_slen)
544 return 0;
545 /* Still some bytes left; set up for a retry later. */
546 status = -EAGAIN;
547 }
548
549 switch (status) {
550 case -ENOTSOCK:
551 status = -ENOTCONN;
552 /* Should we call xs_close() here? */
553 break;
554 case -EAGAIN:
555 status = xs_nospace(task);
556 break;
557 default:
558 dprintk("RPC: sendmsg returned unrecognized error %d\n",
559 -status);
560 case -ENETUNREACH:
561 case -EPIPE:
562 case -ECONNREFUSED:
563 /* When the server has died, an ICMP port unreachable message
564 * prompts ECONNREFUSED. */
565 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
566 }
567
568 return status;
569 }
570
571 /**
572 * xs_tcp_shutdown - gracefully shut down a TCP socket
573 * @xprt: transport
574 *
575 * Initiates a graceful shutdown of the TCP socket by calling the
576 * equivalent of shutdown(SHUT_WR);
577 */
578 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
579 {
580 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
581 struct socket *sock = transport->sock;
582
583 if (sock != NULL)
584 kernel_sock_shutdown(sock, SHUT_WR);
585 }
586
587 /**
588 * xs_tcp_send_request - write an RPC request to a TCP socket
589 * @task: address of RPC task that manages the state of an RPC request
590 *
591 * Return values:
592 * 0: The request has been sent
593 * EAGAIN: The socket was blocked, please call again later to
594 * complete the request
595 * ENOTCONN: Caller needs to invoke connect logic then call again
596 * other: Some other error occurred, the request was not sent
597 *
598 * XXX: In the case of soft timeouts, should we eventually give up
599 * if sendmsg is not able to make progress?
600 */
601 static int xs_tcp_send_request(struct rpc_task *task)
602 {
603 struct rpc_rqst *req = task->tk_rqstp;
604 struct rpc_xprt *xprt = req->rq_xprt;
605 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
606 struct xdr_buf *xdr = &req->rq_snd_buf;
607 int status;
608
609 xs_encode_stream_record_marker(&req->rq_snd_buf);
610
611 xs_pktdump("packet data:",
612 req->rq_svec->iov_base,
613 req->rq_svec->iov_len);
614
615 /* Continue transmitting the packet/record. We must be careful
616 * to cope with writespace callbacks arriving _after_ we have
617 * called sendmsg(). */
618 while (1) {
619 status = xs_sendpages(transport->sock,
620 NULL, 0, xdr, req->rq_bytes_sent);
621
622 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
623 xdr->len - req->rq_bytes_sent, status);
624
625 if (unlikely(status < 0))
626 break;
627
628 /* If we've sent the entire packet, immediately
629 * reset the count of bytes sent. */
630 req->rq_bytes_sent += status;
631 req->rq_xmit_bytes_sent += status;
632 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
633 req->rq_bytes_sent = 0;
634 return 0;
635 }
636
637 if (status != 0)
638 continue;
639 status = -EAGAIN;
640 break;
641 }
642
643 switch (status) {
644 case -ENOTSOCK:
645 status = -ENOTCONN;
646 /* Should we call xs_close() here? */
647 break;
648 case -EAGAIN:
649 status = xs_nospace(task);
650 break;
651 default:
652 dprintk("RPC: sendmsg returned unrecognized error %d\n",
653 -status);
654 case -ECONNRESET:
655 case -EPIPE:
656 xs_tcp_shutdown(xprt);
657 case -ECONNREFUSED:
658 case -ENOTCONN:
659 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
660 }
661
662 return status;
663 }
664
665 /**
666 * xs_tcp_release_xprt - clean up after a tcp transmission
667 * @xprt: transport
668 * @task: rpc task
669 *
670 * This cleans up if an error causes us to abort the transmission of a request.
671 * In this case, the socket may need to be reset in order to avoid confusing
672 * the server.
673 */
674 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
675 {
676 struct rpc_rqst *req;
677
678 if (task != xprt->snd_task)
679 return;
680 if (task == NULL)
681 goto out_release;
682 req = task->tk_rqstp;
683 if (req->rq_bytes_sent == 0)
684 goto out_release;
685 if (req->rq_bytes_sent == req->rq_snd_buf.len)
686 goto out_release;
687 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
688 out_release:
689 xprt_release_xprt(xprt, task);
690 }
691
692 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
693 {
694 transport->old_data_ready = sk->sk_data_ready;
695 transport->old_state_change = sk->sk_state_change;
696 transport->old_write_space = sk->sk_write_space;
697 transport->old_error_report = sk->sk_error_report;
698 }
699
700 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
701 {
702 sk->sk_data_ready = transport->old_data_ready;
703 sk->sk_state_change = transport->old_state_change;
704 sk->sk_write_space = transport->old_write_space;
705 sk->sk_error_report = transport->old_error_report;
706 }
707
708 static void xs_reset_transport(struct sock_xprt *transport)
709 {
710 struct socket *sock = transport->sock;
711 struct sock *sk = transport->inet;
712
713 if (sk == NULL)
714 return;
715
716 transport->srcport = 0;
717
718 write_lock_bh(&sk->sk_callback_lock);
719 transport->inet = NULL;
720 transport->sock = NULL;
721
722 sk->sk_user_data = NULL;
723
724 xs_restore_old_callbacks(transport, sk);
725 write_unlock_bh(&sk->sk_callback_lock);
726
727 sk->sk_no_check = 0;
728
729 sock_release(sock);
730 }
731
732 /**
733 * xs_close - close a socket
734 * @xprt: transport
735 *
736 * This is used when all requests are complete; ie, no DRC state remains
737 * on the server we want to save.
738 *
739 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
740 * xs_reset_transport() zeroing the socket from underneath a writer.
741 */
742 static void xs_close(struct rpc_xprt *xprt)
743 {
744 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
745
746 dprintk("RPC: xs_close xprt %p\n", xprt);
747
748 xs_reset_transport(transport);
749 xprt->reestablish_timeout = 0;
750
751 smp_mb__before_clear_bit();
752 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
753 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
754 clear_bit(XPRT_CLOSING, &xprt->state);
755 smp_mb__after_clear_bit();
756 xprt_disconnect_done(xprt);
757 }
758
759 static void xs_tcp_close(struct rpc_xprt *xprt)
760 {
761 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
762 xs_close(xprt);
763 else
764 xs_tcp_shutdown(xprt);
765 }
766
767 /**
768 * xs_destroy - prepare to shutdown a transport
769 * @xprt: doomed transport
770 *
771 */
772 static void xs_destroy(struct rpc_xprt *xprt)
773 {
774 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
775
776 dprintk("RPC: xs_destroy xprt %p\n", xprt);
777
778 cancel_delayed_work_sync(&transport->connect_worker);
779
780 xs_close(xprt);
781 xs_free_peer_addresses(xprt);
782 xprt_free(xprt);
783 module_put(THIS_MODULE);
784 }
785
786 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
787 {
788 return (struct rpc_xprt *) sk->sk_user_data;
789 }
790
791 /**
792 * xs_udp_data_ready - "data ready" callback for UDP sockets
793 * @sk: socket with data to read
794 * @len: how much data to read
795 *
796 */
797 static void xs_udp_data_ready(struct sock *sk, int len)
798 {
799 struct rpc_task *task;
800 struct rpc_xprt *xprt;
801 struct rpc_rqst *rovr;
802 struct sk_buff *skb;
803 int err, repsize, copied;
804 u32 _xid;
805 __be32 *xp;
806
807 read_lock_bh(&sk->sk_callback_lock);
808 dprintk("RPC: xs_udp_data_ready...\n");
809 if (!(xprt = xprt_from_sock(sk)))
810 goto out;
811
812 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
813 goto out;
814
815 if (xprt->shutdown)
816 goto dropit;
817
818 repsize = skb->len - sizeof(struct udphdr);
819 if (repsize < 4) {
820 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
821 goto dropit;
822 }
823
824 /* Copy the XID from the skb... */
825 xp = skb_header_pointer(skb, sizeof(struct udphdr),
826 sizeof(_xid), &_xid);
827 if (xp == NULL)
828 goto dropit;
829
830 /* Look up and lock the request corresponding to the given XID */
831 spin_lock(&xprt->transport_lock);
832 rovr = xprt_lookup_rqst(xprt, *xp);
833 if (!rovr)
834 goto out_unlock;
835 task = rovr->rq_task;
836
837 if ((copied = rovr->rq_private_buf.buflen) > repsize)
838 copied = repsize;
839
840 /* Suck it into the iovec, verify checksum if not done by hw. */
841 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
842 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
843 goto out_unlock;
844 }
845
846 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
847
848 /* Something worked... */
849 dst_confirm(skb_dst(skb));
850
851 xprt_adjust_cwnd(task, copied);
852 xprt_complete_rqst(task, copied);
853
854 out_unlock:
855 spin_unlock(&xprt->transport_lock);
856 dropit:
857 skb_free_datagram(sk, skb);
858 out:
859 read_unlock_bh(&sk->sk_callback_lock);
860 }
861
862 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
863 {
864 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
865 size_t len, used;
866 char *p;
867
868 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
869 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
870 used = xdr_skb_read_bits(desc, p, len);
871 transport->tcp_offset += used;
872 if (used != len)
873 return;
874
875 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
876 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
877 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
878 else
879 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
880 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
881
882 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
883 transport->tcp_offset = 0;
884
885 /* Sanity check of the record length */
886 if (unlikely(transport->tcp_reclen < 8)) {
887 dprintk("RPC: invalid TCP record fragment length\n");
888 xprt_force_disconnect(xprt);
889 return;
890 }
891 dprintk("RPC: reading TCP record fragment of length %d\n",
892 transport->tcp_reclen);
893 }
894
895 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
896 {
897 if (transport->tcp_offset == transport->tcp_reclen) {
898 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
899 transport->tcp_offset = 0;
900 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
901 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
902 transport->tcp_flags |= TCP_RCV_COPY_XID;
903 transport->tcp_copied = 0;
904 }
905 }
906 }
907
908 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
909 {
910 size_t len, used;
911 char *p;
912
913 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
914 dprintk("RPC: reading XID (%Zu bytes)\n", len);
915 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
916 used = xdr_skb_read_bits(desc, p, len);
917 transport->tcp_offset += used;
918 if (used != len)
919 return;
920 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
921 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
922 transport->tcp_copied = 4;
923 dprintk("RPC: reading %s XID %08x\n",
924 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
925 : "request with",
926 ntohl(transport->tcp_xid));
927 xs_tcp_check_fraghdr(transport);
928 }
929
930 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
931 struct xdr_skb_reader *desc)
932 {
933 size_t len, used;
934 u32 offset;
935 char *p;
936
937 /*
938 * We want transport->tcp_offset to be 8 at the end of this routine
939 * (4 bytes for the xid and 4 bytes for the call/reply flag).
940 * When this function is called for the first time,
941 * transport->tcp_offset is 4 (after having already read the xid).
942 */
943 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
944 len = sizeof(transport->tcp_calldir) - offset;
945 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
946 p = ((char *) &transport->tcp_calldir) + offset;
947 used = xdr_skb_read_bits(desc, p, len);
948 transport->tcp_offset += used;
949 if (used != len)
950 return;
951 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
952 /*
953 * We don't yet have the XDR buffer, so we will write the calldir
954 * out after we get the buffer from the 'struct rpc_rqst'
955 */
956 switch (ntohl(transport->tcp_calldir)) {
957 case RPC_REPLY:
958 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
959 transport->tcp_flags |= TCP_RCV_COPY_DATA;
960 transport->tcp_flags |= TCP_RPC_REPLY;
961 break;
962 case RPC_CALL:
963 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
964 transport->tcp_flags |= TCP_RCV_COPY_DATA;
965 transport->tcp_flags &= ~TCP_RPC_REPLY;
966 break;
967 default:
968 dprintk("RPC: invalid request message type\n");
969 xprt_force_disconnect(&transport->xprt);
970 }
971 xs_tcp_check_fraghdr(transport);
972 }
973
974 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
975 struct xdr_skb_reader *desc,
976 struct rpc_rqst *req)
977 {
978 struct sock_xprt *transport =
979 container_of(xprt, struct sock_xprt, xprt);
980 struct xdr_buf *rcvbuf;
981 size_t len;
982 ssize_t r;
983
984 rcvbuf = &req->rq_private_buf;
985
986 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
987 /*
988 * Save the RPC direction in the XDR buffer
989 */
990 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
991 &transport->tcp_calldir,
992 sizeof(transport->tcp_calldir));
993 transport->tcp_copied += sizeof(transport->tcp_calldir);
994 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
995 }
996
997 len = desc->count;
998 if (len > transport->tcp_reclen - transport->tcp_offset) {
999 struct xdr_skb_reader my_desc;
1000
1001 len = transport->tcp_reclen - transport->tcp_offset;
1002 memcpy(&my_desc, desc, sizeof(my_desc));
1003 my_desc.count = len;
1004 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1005 &my_desc, xdr_skb_read_bits);
1006 desc->count -= r;
1007 desc->offset += r;
1008 } else
1009 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1010 desc, xdr_skb_read_bits);
1011
1012 if (r > 0) {
1013 transport->tcp_copied += r;
1014 transport->tcp_offset += r;
1015 }
1016 if (r != len) {
1017 /* Error when copying to the receive buffer,
1018 * usually because we weren't able to allocate
1019 * additional buffer pages. All we can do now
1020 * is turn off TCP_RCV_COPY_DATA, so the request
1021 * will not receive any additional updates,
1022 * and time out.
1023 * Any remaining data from this record will
1024 * be discarded.
1025 */
1026 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1027 dprintk("RPC: XID %08x truncated request\n",
1028 ntohl(transport->tcp_xid));
1029 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1030 "tcp_offset = %u, tcp_reclen = %u\n",
1031 xprt, transport->tcp_copied,
1032 transport->tcp_offset, transport->tcp_reclen);
1033 return;
1034 }
1035
1036 dprintk("RPC: XID %08x read %Zd bytes\n",
1037 ntohl(transport->tcp_xid), r);
1038 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1039 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1040 transport->tcp_offset, transport->tcp_reclen);
1041
1042 if (transport->tcp_copied == req->rq_private_buf.buflen)
1043 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1044 else if (transport->tcp_offset == transport->tcp_reclen) {
1045 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1046 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1047 }
1048 }
1049
1050 /*
1051 * Finds the request corresponding to the RPC xid and invokes the common
1052 * tcp read code to read the data.
1053 */
1054 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1055 struct xdr_skb_reader *desc)
1056 {
1057 struct sock_xprt *transport =
1058 container_of(xprt, struct sock_xprt, xprt);
1059 struct rpc_rqst *req;
1060
1061 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1062
1063 /* Find and lock the request corresponding to this xid */
1064 spin_lock(&xprt->transport_lock);
1065 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1066 if (!req) {
1067 dprintk("RPC: XID %08x request not found!\n",
1068 ntohl(transport->tcp_xid));
1069 spin_unlock(&xprt->transport_lock);
1070 return -1;
1071 }
1072
1073 xs_tcp_read_common(xprt, desc, req);
1074
1075 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1076 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1077
1078 spin_unlock(&xprt->transport_lock);
1079 return 0;
1080 }
1081
1082 #if defined(CONFIG_NFS_V4_1)
1083 /*
1084 * Obtains an rpc_rqst previously allocated and invokes the common
1085 * tcp read code to read the data. The result is placed in the callback
1086 * queue.
1087 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1088 * connection and return -1.
1089 */
1090 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1091 struct xdr_skb_reader *desc)
1092 {
1093 struct sock_xprt *transport =
1094 container_of(xprt, struct sock_xprt, xprt);
1095 struct rpc_rqst *req;
1096
1097 req = xprt_alloc_bc_request(xprt);
1098 if (req == NULL) {
1099 printk(KERN_WARNING "Callback slot table overflowed\n");
1100 xprt_force_disconnect(xprt);
1101 return -1;
1102 }
1103
1104 req->rq_xid = transport->tcp_xid;
1105 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1106 xs_tcp_read_common(xprt, desc, req);
1107
1108 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1109 struct svc_serv *bc_serv = xprt->bc_serv;
1110
1111 /*
1112 * Add callback request to callback list. The callback
1113 * service sleeps on the sv_cb_waitq waiting for new
1114 * requests. Wake it up after adding enqueing the
1115 * request.
1116 */
1117 dprintk("RPC: add callback request to list\n");
1118 spin_lock(&bc_serv->sv_cb_lock);
1119 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1120 spin_unlock(&bc_serv->sv_cb_lock);
1121 wake_up(&bc_serv->sv_cb_waitq);
1122 }
1123
1124 req->rq_private_buf.len = transport->tcp_copied;
1125
1126 return 0;
1127 }
1128
1129 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1130 struct xdr_skb_reader *desc)
1131 {
1132 struct sock_xprt *transport =
1133 container_of(xprt, struct sock_xprt, xprt);
1134
1135 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1136 xs_tcp_read_reply(xprt, desc) :
1137 xs_tcp_read_callback(xprt, desc);
1138 }
1139 #else
1140 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1141 struct xdr_skb_reader *desc)
1142 {
1143 return xs_tcp_read_reply(xprt, desc);
1144 }
1145 #endif /* CONFIG_NFS_V4_1 */
1146
1147 /*
1148 * Read data off the transport. This can be either an RPC_CALL or an
1149 * RPC_REPLY. Relay the processing to helper functions.
1150 */
1151 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1152 struct xdr_skb_reader *desc)
1153 {
1154 struct sock_xprt *transport =
1155 container_of(xprt, struct sock_xprt, xprt);
1156
1157 if (_xs_tcp_read_data(xprt, desc) == 0)
1158 xs_tcp_check_fraghdr(transport);
1159 else {
1160 /*
1161 * The transport_lock protects the request handling.
1162 * There's no need to hold it to update the tcp_flags.
1163 */
1164 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1165 }
1166 }
1167
1168 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1169 {
1170 size_t len;
1171
1172 len = transport->tcp_reclen - transport->tcp_offset;
1173 if (len > desc->count)
1174 len = desc->count;
1175 desc->count -= len;
1176 desc->offset += len;
1177 transport->tcp_offset += len;
1178 dprintk("RPC: discarded %Zu bytes\n", len);
1179 xs_tcp_check_fraghdr(transport);
1180 }
1181
1182 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1183 {
1184 struct rpc_xprt *xprt = rd_desc->arg.data;
1185 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1186 struct xdr_skb_reader desc = {
1187 .skb = skb,
1188 .offset = offset,
1189 .count = len,
1190 };
1191
1192 dprintk("RPC: xs_tcp_data_recv started\n");
1193 do {
1194 /* Read in a new fragment marker if necessary */
1195 /* Can we ever really expect to get completely empty fragments? */
1196 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1197 xs_tcp_read_fraghdr(xprt, &desc);
1198 continue;
1199 }
1200 /* Read in the xid if necessary */
1201 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1202 xs_tcp_read_xid(transport, &desc);
1203 continue;
1204 }
1205 /* Read in the call/reply flag */
1206 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1207 xs_tcp_read_calldir(transport, &desc);
1208 continue;
1209 }
1210 /* Read in the request data */
1211 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1212 xs_tcp_read_data(xprt, &desc);
1213 continue;
1214 }
1215 /* Skip over any trailing bytes on short reads */
1216 xs_tcp_read_discard(transport, &desc);
1217 } while (desc.count);
1218 dprintk("RPC: xs_tcp_data_recv done\n");
1219 return len - desc.count;
1220 }
1221
1222 /**
1223 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1224 * @sk: socket with data to read
1225 * @bytes: how much data to read
1226 *
1227 */
1228 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1229 {
1230 struct rpc_xprt *xprt;
1231 read_descriptor_t rd_desc;
1232 int read;
1233
1234 dprintk("RPC: xs_tcp_data_ready...\n");
1235
1236 read_lock_bh(&sk->sk_callback_lock);
1237 if (!(xprt = xprt_from_sock(sk)))
1238 goto out;
1239 if (xprt->shutdown)
1240 goto out;
1241
1242 /* Any data means we had a useful conversation, so
1243 * the we don't need to delay the next reconnect
1244 */
1245 if (xprt->reestablish_timeout)
1246 xprt->reestablish_timeout = 0;
1247
1248 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1249 rd_desc.arg.data = xprt;
1250 do {
1251 rd_desc.count = 65536;
1252 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1253 } while (read > 0);
1254 out:
1255 read_unlock_bh(&sk->sk_callback_lock);
1256 }
1257
1258 /*
1259 * Do the equivalent of linger/linger2 handling for dealing with
1260 * broken servers that don't close the socket in a timely
1261 * fashion
1262 */
1263 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1264 unsigned long timeout)
1265 {
1266 struct sock_xprt *transport;
1267
1268 if (xprt_test_and_set_connecting(xprt))
1269 return;
1270 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1271 transport = container_of(xprt, struct sock_xprt, xprt);
1272 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1273 timeout);
1274 }
1275
1276 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1277 {
1278 struct sock_xprt *transport;
1279
1280 transport = container_of(xprt, struct sock_xprt, xprt);
1281
1282 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1283 !cancel_delayed_work(&transport->connect_worker))
1284 return;
1285 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1286 xprt_clear_connecting(xprt);
1287 }
1288
1289 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1290 {
1291 smp_mb__before_clear_bit();
1292 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1293 clear_bit(XPRT_CLOSING, &xprt->state);
1294 smp_mb__after_clear_bit();
1295 /* Mark transport as closed and wake up all pending tasks */
1296 xprt_disconnect_done(xprt);
1297 }
1298
1299 /**
1300 * xs_tcp_state_change - callback to handle TCP socket state changes
1301 * @sk: socket whose state has changed
1302 *
1303 */
1304 static void xs_tcp_state_change(struct sock *sk)
1305 {
1306 struct rpc_xprt *xprt;
1307
1308 read_lock_bh(&sk->sk_callback_lock);
1309 if (!(xprt = xprt_from_sock(sk)))
1310 goto out;
1311 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1312 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1313 sk->sk_state, xprt_connected(xprt),
1314 sock_flag(sk, SOCK_DEAD),
1315 sock_flag(sk, SOCK_ZAPPED),
1316 sk->sk_shutdown);
1317
1318 switch (sk->sk_state) {
1319 case TCP_ESTABLISHED:
1320 spin_lock(&xprt->transport_lock);
1321 if (!xprt_test_and_set_connected(xprt)) {
1322 struct sock_xprt *transport = container_of(xprt,
1323 struct sock_xprt, xprt);
1324
1325 /* Reset TCP record info */
1326 transport->tcp_offset = 0;
1327 transport->tcp_reclen = 0;
1328 transport->tcp_copied = 0;
1329 transport->tcp_flags =
1330 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1331
1332 xprt_wake_pending_tasks(xprt, -EAGAIN);
1333 }
1334 spin_unlock(&xprt->transport_lock);
1335 break;
1336 case TCP_FIN_WAIT1:
1337 /* The client initiated a shutdown of the socket */
1338 xprt->connect_cookie++;
1339 xprt->reestablish_timeout = 0;
1340 set_bit(XPRT_CLOSING, &xprt->state);
1341 smp_mb__before_clear_bit();
1342 clear_bit(XPRT_CONNECTED, &xprt->state);
1343 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1344 smp_mb__after_clear_bit();
1345 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1346 break;
1347 case TCP_CLOSE_WAIT:
1348 /* The server initiated a shutdown of the socket */
1349 xprt_force_disconnect(xprt);
1350 xprt->connect_cookie++;
1351 case TCP_CLOSING:
1352 /*
1353 * If the server closed down the connection, make sure that
1354 * we back off before reconnecting
1355 */
1356 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1357 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1358 break;
1359 case TCP_LAST_ACK:
1360 set_bit(XPRT_CLOSING, &xprt->state);
1361 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1362 smp_mb__before_clear_bit();
1363 clear_bit(XPRT_CONNECTED, &xprt->state);
1364 smp_mb__after_clear_bit();
1365 break;
1366 case TCP_CLOSE:
1367 xs_tcp_cancel_linger_timeout(xprt);
1368 xs_sock_mark_closed(xprt);
1369 }
1370 out:
1371 read_unlock_bh(&sk->sk_callback_lock);
1372 }
1373
1374 /**
1375 * xs_error_report - callback mainly for catching socket errors
1376 * @sk: socket
1377 */
1378 static void xs_error_report(struct sock *sk)
1379 {
1380 struct rpc_xprt *xprt;
1381
1382 read_lock_bh(&sk->sk_callback_lock);
1383 if (!(xprt = xprt_from_sock(sk)))
1384 goto out;
1385 dprintk("RPC: %s client %p...\n"
1386 "RPC: error %d\n",
1387 __func__, xprt, sk->sk_err);
1388 xprt_wake_pending_tasks(xprt, -EAGAIN);
1389 out:
1390 read_unlock_bh(&sk->sk_callback_lock);
1391 }
1392
1393 static void xs_write_space(struct sock *sk)
1394 {
1395 struct socket *sock;
1396 struct rpc_xprt *xprt;
1397
1398 if (unlikely(!(sock = sk->sk_socket)))
1399 return;
1400 clear_bit(SOCK_NOSPACE, &sock->flags);
1401
1402 if (unlikely(!(xprt = xprt_from_sock(sk))))
1403 return;
1404 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1405 return;
1406
1407 xprt_write_space(xprt);
1408 }
1409
1410 /**
1411 * xs_udp_write_space - callback invoked when socket buffer space
1412 * becomes available
1413 * @sk: socket whose state has changed
1414 *
1415 * Called when more output buffer space is available for this socket.
1416 * We try not to wake our writers until they can make "significant"
1417 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1418 * with a bunch of small requests.
1419 */
1420 static void xs_udp_write_space(struct sock *sk)
1421 {
1422 read_lock_bh(&sk->sk_callback_lock);
1423
1424 /* from net/core/sock.c:sock_def_write_space */
1425 if (sock_writeable(sk))
1426 xs_write_space(sk);
1427
1428 read_unlock_bh(&sk->sk_callback_lock);
1429 }
1430
1431 /**
1432 * xs_tcp_write_space - callback invoked when socket buffer space
1433 * becomes available
1434 * @sk: socket whose state has changed
1435 *
1436 * Called when more output buffer space is available for this socket.
1437 * We try not to wake our writers until they can make "significant"
1438 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1439 * with a bunch of small requests.
1440 */
1441 static void xs_tcp_write_space(struct sock *sk)
1442 {
1443 read_lock_bh(&sk->sk_callback_lock);
1444
1445 /* from net/core/stream.c:sk_stream_write_space */
1446 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1447 xs_write_space(sk);
1448
1449 read_unlock_bh(&sk->sk_callback_lock);
1450 }
1451
1452 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1453 {
1454 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1455 struct sock *sk = transport->inet;
1456
1457 if (transport->rcvsize) {
1458 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1459 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1460 }
1461 if (transport->sndsize) {
1462 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1463 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1464 sk->sk_write_space(sk);
1465 }
1466 }
1467
1468 /**
1469 * xs_udp_set_buffer_size - set send and receive limits
1470 * @xprt: generic transport
1471 * @sndsize: requested size of send buffer, in bytes
1472 * @rcvsize: requested size of receive buffer, in bytes
1473 *
1474 * Set socket send and receive buffer size limits.
1475 */
1476 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1477 {
1478 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1479
1480 transport->sndsize = 0;
1481 if (sndsize)
1482 transport->sndsize = sndsize + 1024;
1483 transport->rcvsize = 0;
1484 if (rcvsize)
1485 transport->rcvsize = rcvsize + 1024;
1486
1487 xs_udp_do_set_buffer_size(xprt);
1488 }
1489
1490 /**
1491 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1492 * @task: task that timed out
1493 *
1494 * Adjust the congestion window after a retransmit timeout has occurred.
1495 */
1496 static void xs_udp_timer(struct rpc_task *task)
1497 {
1498 xprt_adjust_cwnd(task, -ETIMEDOUT);
1499 }
1500
1501 static unsigned short xs_get_random_port(void)
1502 {
1503 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1504 unsigned short rand = (unsigned short) net_random() % range;
1505 return rand + xprt_min_resvport;
1506 }
1507
1508 /**
1509 * xs_set_port - reset the port number in the remote endpoint address
1510 * @xprt: generic transport
1511 * @port: new port number
1512 *
1513 */
1514 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1515 {
1516 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1517
1518 rpc_set_port(xs_addr(xprt), port);
1519 xs_update_peer_port(xprt);
1520 }
1521
1522 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1523 {
1524 unsigned short port = transport->srcport;
1525
1526 if (port == 0 && transport->xprt.resvport)
1527 port = xs_get_random_port();
1528 return port;
1529 }
1530
1531 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1532 {
1533 if (transport->srcport != 0)
1534 transport->srcport = 0;
1535 if (!transport->xprt.resvport)
1536 return 0;
1537 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1538 return xprt_max_resvport;
1539 return --port;
1540 }
1541 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1542 {
1543 struct sockaddr_storage myaddr;
1544 int err, nloop = 0;
1545 unsigned short port = xs_get_srcport(transport);
1546 unsigned short last;
1547
1548 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1549 do {
1550 rpc_set_port((struct sockaddr *)&myaddr, port);
1551 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1552 transport->xprt.addrlen);
1553 if (port == 0)
1554 break;
1555 if (err == 0) {
1556 transport->srcport = port;
1557 break;
1558 }
1559 last = port;
1560 port = xs_next_srcport(transport, port);
1561 if (port > last)
1562 nloop++;
1563 } while (err == -EADDRINUSE && nloop != 2);
1564
1565 if (myaddr.ss_family == AF_INET)
1566 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1567 &((struct sockaddr_in *)&myaddr)->sin_addr,
1568 port, err ? "failed" : "ok", err);
1569 else
1570 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1571 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1572 port, err ? "failed" : "ok", err);
1573 return err;
1574 }
1575
1576
1577 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1578 static struct lock_class_key xs_key[2];
1579 static struct lock_class_key xs_slock_key[2];
1580
1581 static inline void xs_reclassify_socket4(struct socket *sock)
1582 {
1583 struct sock *sk = sock->sk;
1584
1585 BUG_ON(sock_owned_by_user(sk));
1586 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1587 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1588 }
1589
1590 static inline void xs_reclassify_socket6(struct socket *sock)
1591 {
1592 struct sock *sk = sock->sk;
1593
1594 BUG_ON(sock_owned_by_user(sk));
1595 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1596 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1597 }
1598
1599 static inline void xs_reclassify_socket(int family, struct socket *sock)
1600 {
1601 switch (family) {
1602 case AF_INET:
1603 xs_reclassify_socket4(sock);
1604 break;
1605 case AF_INET6:
1606 xs_reclassify_socket6(sock);
1607 break;
1608 }
1609 }
1610 #else
1611 static inline void xs_reclassify_socket4(struct socket *sock)
1612 {
1613 }
1614
1615 static inline void xs_reclassify_socket6(struct socket *sock)
1616 {
1617 }
1618
1619 static inline void xs_reclassify_socket(int family, struct socket *sock)
1620 {
1621 }
1622 #endif
1623
1624 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1625 struct sock_xprt *transport, int family, int type, int protocol)
1626 {
1627 struct socket *sock;
1628 int err;
1629
1630 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1631 if (err < 0) {
1632 dprintk("RPC: can't create %d transport socket (%d).\n",
1633 protocol, -err);
1634 goto out;
1635 }
1636 xs_reclassify_socket(family, sock);
1637
1638 err = xs_bind(transport, sock);
1639 if (err) {
1640 sock_release(sock);
1641 goto out;
1642 }
1643
1644 return sock;
1645 out:
1646 return ERR_PTR(err);
1647 }
1648
1649 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1650 {
1651 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1652
1653 if (!transport->inet) {
1654 struct sock *sk = sock->sk;
1655
1656 write_lock_bh(&sk->sk_callback_lock);
1657
1658 xs_save_old_callbacks(transport, sk);
1659
1660 sk->sk_user_data = xprt;
1661 sk->sk_data_ready = xs_udp_data_ready;
1662 sk->sk_write_space = xs_udp_write_space;
1663 sk->sk_error_report = xs_error_report;
1664 sk->sk_no_check = UDP_CSUM_NORCV;
1665 sk->sk_allocation = GFP_ATOMIC;
1666
1667 xprt_set_connected(xprt);
1668
1669 /* Reset to new socket */
1670 transport->sock = sock;
1671 transport->inet = sk;
1672
1673 write_unlock_bh(&sk->sk_callback_lock);
1674 }
1675 xs_udp_do_set_buffer_size(xprt);
1676 }
1677
1678 static void xs_udp_setup_socket(struct work_struct *work)
1679 {
1680 struct sock_xprt *transport =
1681 container_of(work, struct sock_xprt, connect_worker.work);
1682 struct rpc_xprt *xprt = &transport->xprt;
1683 struct socket *sock = transport->sock;
1684 int status = -EIO;
1685
1686 if (xprt->shutdown)
1687 goto out;
1688
1689 /* Start by resetting any existing state */
1690 xs_reset_transport(transport);
1691 sock = xs_create_sock(xprt, transport,
1692 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
1693 if (IS_ERR(sock))
1694 goto out;
1695
1696 dprintk("RPC: worker connecting xprt %p via %s to "
1697 "%s (port %s)\n", xprt,
1698 xprt->address_strings[RPC_DISPLAY_PROTO],
1699 xprt->address_strings[RPC_DISPLAY_ADDR],
1700 xprt->address_strings[RPC_DISPLAY_PORT]);
1701
1702 xs_udp_finish_connecting(xprt, sock);
1703 status = 0;
1704 out:
1705 xprt_clear_connecting(xprt);
1706 xprt_wake_pending_tasks(xprt, status);
1707 }
1708
1709 /*
1710 * We need to preserve the port number so the reply cache on the server can
1711 * find our cached RPC replies when we get around to reconnecting.
1712 */
1713 static void xs_abort_connection(struct sock_xprt *transport)
1714 {
1715 int result;
1716 struct sockaddr any;
1717
1718 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
1719
1720 /*
1721 * Disconnect the transport socket by doing a connect operation
1722 * with AF_UNSPEC. This should return immediately...
1723 */
1724 memset(&any, 0, sizeof(any));
1725 any.sa_family = AF_UNSPEC;
1726 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1727 if (!result)
1728 xs_sock_mark_closed(&transport->xprt);
1729 else
1730 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1731 result);
1732 }
1733
1734 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
1735 {
1736 unsigned int state = transport->inet->sk_state;
1737
1738 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
1739 /* we don't need to abort the connection if the socket
1740 * hasn't undergone a shutdown
1741 */
1742 if (transport->inet->sk_shutdown == 0)
1743 return;
1744 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
1745 __func__, transport->inet->sk_shutdown);
1746 }
1747 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
1748 /* we don't need to abort the connection if the socket
1749 * hasn't undergone a shutdown
1750 */
1751 if (transport->inet->sk_shutdown == 0)
1752 return;
1753 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
1754 "sk_shutdown set to %d\n",
1755 __func__, transport->inet->sk_shutdown);
1756 }
1757 xs_abort_connection(transport);
1758 }
1759
1760 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1761 {
1762 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1763 int ret = -ENOTCONN;
1764
1765 if (!transport->inet) {
1766 struct sock *sk = sock->sk;
1767
1768 write_lock_bh(&sk->sk_callback_lock);
1769
1770 xs_save_old_callbacks(transport, sk);
1771
1772 sk->sk_user_data = xprt;
1773 sk->sk_data_ready = xs_tcp_data_ready;
1774 sk->sk_state_change = xs_tcp_state_change;
1775 sk->sk_write_space = xs_tcp_write_space;
1776 sk->sk_error_report = xs_error_report;
1777 sk->sk_allocation = GFP_ATOMIC;
1778
1779 /* socket options */
1780 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1781 sock_reset_flag(sk, SOCK_LINGER);
1782 tcp_sk(sk)->linger2 = 0;
1783 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1784
1785 xprt_clear_connected(xprt);
1786
1787 /* Reset to new socket */
1788 transport->sock = sock;
1789 transport->inet = sk;
1790
1791 write_unlock_bh(&sk->sk_callback_lock);
1792 }
1793
1794 if (!xprt_bound(xprt))
1795 goto out;
1796
1797 /* Tell the socket layer to start connecting... */
1798 xprt->stat.connect_count++;
1799 xprt->stat.connect_start = jiffies;
1800 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1801 switch (ret) {
1802 case 0:
1803 case -EINPROGRESS:
1804 /* SYN_SENT! */
1805 xprt->connect_cookie++;
1806 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1807 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1808 }
1809 out:
1810 return ret;
1811 }
1812
1813 /**
1814 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1815 * @xprt: RPC transport to connect
1816 * @transport: socket transport to connect
1817 * @create_sock: function to create a socket of the correct type
1818 *
1819 * Invoked by a work queue tasklet.
1820 */
1821 static void xs_tcp_setup_socket(struct work_struct *work)
1822 {
1823 struct sock_xprt *transport =
1824 container_of(work, struct sock_xprt, connect_worker.work);
1825 struct socket *sock = transport->sock;
1826 struct rpc_xprt *xprt = &transport->xprt;
1827 int status = -EIO;
1828
1829 if (xprt->shutdown)
1830 goto out;
1831
1832 if (!sock) {
1833 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1834 sock = xs_create_sock(xprt, transport,
1835 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
1836 if (IS_ERR(sock)) {
1837 status = PTR_ERR(sock);
1838 goto out;
1839 }
1840 } else {
1841 int abort_and_exit;
1842
1843 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1844 &xprt->state);
1845 /* "close" the socket, preserving the local port */
1846 xs_tcp_reuse_connection(transport);
1847
1848 if (abort_and_exit)
1849 goto out_eagain;
1850 }
1851
1852 dprintk("RPC: worker connecting xprt %p via %s to "
1853 "%s (port %s)\n", xprt,
1854 xprt->address_strings[RPC_DISPLAY_PROTO],
1855 xprt->address_strings[RPC_DISPLAY_ADDR],
1856 xprt->address_strings[RPC_DISPLAY_PORT]);
1857
1858 status = xs_tcp_finish_connecting(xprt, sock);
1859 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1860 xprt, -status, xprt_connected(xprt),
1861 sock->sk->sk_state);
1862 switch (status) {
1863 default:
1864 printk("%s: connect returned unhandled error %d\n",
1865 __func__, status);
1866 case -EADDRNOTAVAIL:
1867 /* We're probably in TIME_WAIT. Get rid of existing socket,
1868 * and retry
1869 */
1870 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1871 xprt_force_disconnect(xprt);
1872 break;
1873 case -ECONNREFUSED:
1874 case -ECONNRESET:
1875 case -ENETUNREACH:
1876 /* retry with existing socket, after a delay */
1877 case 0:
1878 case -EINPROGRESS:
1879 case -EALREADY:
1880 xprt_clear_connecting(xprt);
1881 return;
1882 case -EINVAL:
1883 /* Happens, for instance, if the user specified a link
1884 * local IPv6 address without a scope-id.
1885 */
1886 goto out;
1887 }
1888 out_eagain:
1889 status = -EAGAIN;
1890 out:
1891 xprt_clear_connecting(xprt);
1892 xprt_wake_pending_tasks(xprt, status);
1893 }
1894
1895 /**
1896 * xs_connect - connect a socket to a remote endpoint
1897 * @task: address of RPC task that manages state of connect request
1898 *
1899 * TCP: If the remote end dropped the connection, delay reconnecting.
1900 *
1901 * UDP socket connects are synchronous, but we use a work queue anyway
1902 * to guarantee that even unprivileged user processes can set up a
1903 * socket on a privileged port.
1904 *
1905 * If a UDP socket connect fails, the delay behavior here prevents
1906 * retry floods (hard mounts).
1907 */
1908 static void xs_connect(struct rpc_task *task)
1909 {
1910 struct rpc_xprt *xprt = task->tk_xprt;
1911 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1912
1913 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
1914 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1915 "seconds\n",
1916 xprt, xprt->reestablish_timeout / HZ);
1917 queue_delayed_work(rpciod_workqueue,
1918 &transport->connect_worker,
1919 xprt->reestablish_timeout);
1920 xprt->reestablish_timeout <<= 1;
1921 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1922 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1923 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1924 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1925 } else {
1926 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1927 queue_delayed_work(rpciod_workqueue,
1928 &transport->connect_worker, 0);
1929 }
1930 }
1931
1932 /**
1933 * xs_udp_print_stats - display UDP socket-specifc stats
1934 * @xprt: rpc_xprt struct containing statistics
1935 * @seq: output file
1936 *
1937 */
1938 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1939 {
1940 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1941
1942 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1943 transport->srcport,
1944 xprt->stat.bind_count,
1945 xprt->stat.sends,
1946 xprt->stat.recvs,
1947 xprt->stat.bad_xids,
1948 xprt->stat.req_u,
1949 xprt->stat.bklog_u);
1950 }
1951
1952 /**
1953 * xs_tcp_print_stats - display TCP socket-specifc stats
1954 * @xprt: rpc_xprt struct containing statistics
1955 * @seq: output file
1956 *
1957 */
1958 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1959 {
1960 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1961 long idle_time = 0;
1962
1963 if (xprt_connected(xprt))
1964 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1965
1966 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1967 transport->srcport,
1968 xprt->stat.bind_count,
1969 xprt->stat.connect_count,
1970 xprt->stat.connect_time,
1971 idle_time,
1972 xprt->stat.sends,
1973 xprt->stat.recvs,
1974 xprt->stat.bad_xids,
1975 xprt->stat.req_u,
1976 xprt->stat.bklog_u);
1977 }
1978
1979 /*
1980 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
1981 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
1982 * to use the server side send routines.
1983 */
1984 static void *bc_malloc(struct rpc_task *task, size_t size)
1985 {
1986 struct page *page;
1987 struct rpc_buffer *buf;
1988
1989 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
1990 page = alloc_page(GFP_KERNEL);
1991
1992 if (!page)
1993 return NULL;
1994
1995 buf = page_address(page);
1996 buf->len = PAGE_SIZE;
1997
1998 return buf->data;
1999 }
2000
2001 /*
2002 * Free the space allocated in the bc_alloc routine
2003 */
2004 static void bc_free(void *buffer)
2005 {
2006 struct rpc_buffer *buf;
2007
2008 if (!buffer)
2009 return;
2010
2011 buf = container_of(buffer, struct rpc_buffer, data);
2012 free_page((unsigned long)buf);
2013 }
2014
2015 /*
2016 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2017 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2018 */
2019 static int bc_sendto(struct rpc_rqst *req)
2020 {
2021 int len;
2022 struct xdr_buf *xbufp = &req->rq_snd_buf;
2023 struct rpc_xprt *xprt = req->rq_xprt;
2024 struct sock_xprt *transport =
2025 container_of(xprt, struct sock_xprt, xprt);
2026 struct socket *sock = transport->sock;
2027 unsigned long headoff;
2028 unsigned long tailoff;
2029
2030 xs_encode_stream_record_marker(xbufp);
2031
2032 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2033 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2034 len = svc_send_common(sock, xbufp,
2035 virt_to_page(xbufp->head[0].iov_base), headoff,
2036 xbufp->tail[0].iov_base, tailoff);
2037
2038 if (len != xbufp->len) {
2039 printk(KERN_NOTICE "Error sending entire callback!\n");
2040 len = -EAGAIN;
2041 }
2042
2043 return len;
2044 }
2045
2046 /*
2047 * The send routine. Borrows from svc_send
2048 */
2049 static int bc_send_request(struct rpc_task *task)
2050 {
2051 struct rpc_rqst *req = task->tk_rqstp;
2052 struct svc_xprt *xprt;
2053 struct svc_sock *svsk;
2054 u32 len;
2055
2056 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2057 /*
2058 * Get the server socket associated with this callback xprt
2059 */
2060 xprt = req->rq_xprt->bc_xprt;
2061 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2062
2063 /*
2064 * Grab the mutex to serialize data as the connection is shared
2065 * with the fore channel
2066 */
2067 if (!mutex_trylock(&xprt->xpt_mutex)) {
2068 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2069 if (!mutex_trylock(&xprt->xpt_mutex))
2070 return -EAGAIN;
2071 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2072 }
2073 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2074 len = -ENOTCONN;
2075 else
2076 len = bc_sendto(req);
2077 mutex_unlock(&xprt->xpt_mutex);
2078
2079 if (len > 0)
2080 len = 0;
2081
2082 return len;
2083 }
2084
2085 /*
2086 * The close routine. Since this is client initiated, we do nothing
2087 */
2088
2089 static void bc_close(struct rpc_xprt *xprt)
2090 {
2091 }
2092
2093 /*
2094 * The xprt destroy routine. Again, because this connection is client
2095 * initiated, we do nothing
2096 */
2097
2098 static void bc_destroy(struct rpc_xprt *xprt)
2099 {
2100 }
2101
2102 static struct rpc_xprt_ops xs_udp_ops = {
2103 .set_buffer_size = xs_udp_set_buffer_size,
2104 .reserve_xprt = xprt_reserve_xprt_cong,
2105 .release_xprt = xprt_release_xprt_cong,
2106 .rpcbind = rpcb_getport_async,
2107 .set_port = xs_set_port,
2108 .connect = xs_connect,
2109 .buf_alloc = rpc_malloc,
2110 .buf_free = rpc_free,
2111 .send_request = xs_udp_send_request,
2112 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2113 .timer = xs_udp_timer,
2114 .release_request = xprt_release_rqst_cong,
2115 .close = xs_close,
2116 .destroy = xs_destroy,
2117 .print_stats = xs_udp_print_stats,
2118 };
2119
2120 static struct rpc_xprt_ops xs_tcp_ops = {
2121 .reserve_xprt = xprt_reserve_xprt,
2122 .release_xprt = xs_tcp_release_xprt,
2123 .rpcbind = rpcb_getport_async,
2124 .set_port = xs_set_port,
2125 .connect = xs_connect,
2126 .buf_alloc = rpc_malloc,
2127 .buf_free = rpc_free,
2128 .send_request = xs_tcp_send_request,
2129 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2130 .close = xs_tcp_close,
2131 .destroy = xs_destroy,
2132 .print_stats = xs_tcp_print_stats,
2133 };
2134
2135 /*
2136 * The rpc_xprt_ops for the server backchannel
2137 */
2138
2139 static struct rpc_xprt_ops bc_tcp_ops = {
2140 .reserve_xprt = xprt_reserve_xprt,
2141 .release_xprt = xprt_release_xprt,
2142 .buf_alloc = bc_malloc,
2143 .buf_free = bc_free,
2144 .send_request = bc_send_request,
2145 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2146 .close = bc_close,
2147 .destroy = bc_destroy,
2148 .print_stats = xs_tcp_print_stats,
2149 };
2150
2151 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2152 {
2153 static const struct sockaddr_in sin = {
2154 .sin_family = AF_INET,
2155 .sin_addr.s_addr = htonl(INADDR_ANY),
2156 };
2157 static const struct sockaddr_in6 sin6 = {
2158 .sin6_family = AF_INET6,
2159 .sin6_addr = IN6ADDR_ANY_INIT,
2160 };
2161
2162 switch (family) {
2163 case AF_INET:
2164 memcpy(sap, &sin, sizeof(sin));
2165 break;
2166 case AF_INET6:
2167 memcpy(sap, &sin6, sizeof(sin6));
2168 break;
2169 default:
2170 dprintk("RPC: %s: Bad address family\n", __func__);
2171 return -EAFNOSUPPORT;
2172 }
2173 return 0;
2174 }
2175
2176 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2177 unsigned int slot_table_size)
2178 {
2179 struct rpc_xprt *xprt;
2180 struct sock_xprt *new;
2181
2182 if (args->addrlen > sizeof(xprt->addr)) {
2183 dprintk("RPC: xs_setup_xprt: address too large\n");
2184 return ERR_PTR(-EBADF);
2185 }
2186
2187 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size);
2188 if (xprt == NULL) {
2189 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2190 "rpc_xprt\n");
2191 return ERR_PTR(-ENOMEM);
2192 }
2193
2194 new = container_of(xprt, struct sock_xprt, xprt);
2195 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2196 xprt->addrlen = args->addrlen;
2197 if (args->srcaddr)
2198 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2199 else {
2200 int err;
2201 err = xs_init_anyaddr(args->dstaddr->sa_family,
2202 (struct sockaddr *)&new->srcaddr);
2203 if (err != 0)
2204 return ERR_PTR(err);
2205 }
2206
2207 return xprt;
2208 }
2209
2210 static const struct rpc_timeout xs_udp_default_timeout = {
2211 .to_initval = 5 * HZ,
2212 .to_maxval = 30 * HZ,
2213 .to_increment = 5 * HZ,
2214 .to_retries = 5,
2215 };
2216
2217 /**
2218 * xs_setup_udp - Set up transport to use a UDP socket
2219 * @args: rpc transport creation arguments
2220 *
2221 */
2222 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2223 {
2224 struct sockaddr *addr = args->dstaddr;
2225 struct rpc_xprt *xprt;
2226 struct sock_xprt *transport;
2227 struct rpc_xprt *ret;
2228
2229 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2230 if (IS_ERR(xprt))
2231 return xprt;
2232 transport = container_of(xprt, struct sock_xprt, xprt);
2233
2234 xprt->prot = IPPROTO_UDP;
2235 xprt->tsh_size = 0;
2236 /* XXX: header size can vary due to auth type, IPv6, etc. */
2237 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2238
2239 xprt->bind_timeout = XS_BIND_TO;
2240 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2241 xprt->idle_timeout = XS_IDLE_DISC_TO;
2242
2243 xprt->ops = &xs_udp_ops;
2244
2245 xprt->timeout = &xs_udp_default_timeout;
2246
2247 switch (addr->sa_family) {
2248 case AF_INET:
2249 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2250 xprt_set_bound(xprt);
2251
2252 INIT_DELAYED_WORK(&transport->connect_worker,
2253 xs_udp_setup_socket);
2254 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2255 break;
2256 case AF_INET6:
2257 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2258 xprt_set_bound(xprt);
2259
2260 INIT_DELAYED_WORK(&transport->connect_worker,
2261 xs_udp_setup_socket);
2262 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2263 break;
2264 default:
2265 ret = ERR_PTR(-EAFNOSUPPORT);
2266 goto out_err;
2267 }
2268
2269 if (xprt_bound(xprt))
2270 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2271 xprt->address_strings[RPC_DISPLAY_ADDR],
2272 xprt->address_strings[RPC_DISPLAY_PORT],
2273 xprt->address_strings[RPC_DISPLAY_PROTO]);
2274 else
2275 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2276 xprt->address_strings[RPC_DISPLAY_ADDR],
2277 xprt->address_strings[RPC_DISPLAY_PROTO]);
2278
2279 if (try_module_get(THIS_MODULE))
2280 return xprt;
2281 ret = ERR_PTR(-EINVAL);
2282 out_err:
2283 xprt_free(xprt);
2284 return ret;
2285 }
2286
2287 static const struct rpc_timeout xs_tcp_default_timeout = {
2288 .to_initval = 60 * HZ,
2289 .to_maxval = 60 * HZ,
2290 .to_retries = 2,
2291 };
2292
2293 /**
2294 * xs_setup_tcp - Set up transport to use a TCP socket
2295 * @args: rpc transport creation arguments
2296 *
2297 */
2298 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2299 {
2300 struct sockaddr *addr = args->dstaddr;
2301 struct rpc_xprt *xprt;
2302 struct sock_xprt *transport;
2303 struct rpc_xprt *ret;
2304
2305 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2306 if (IS_ERR(xprt))
2307 return xprt;
2308 transport = container_of(xprt, struct sock_xprt, xprt);
2309
2310 xprt->prot = IPPROTO_TCP;
2311 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2312 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2313
2314 xprt->bind_timeout = XS_BIND_TO;
2315 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2316 xprt->idle_timeout = XS_IDLE_DISC_TO;
2317
2318 xprt->ops = &xs_tcp_ops;
2319 xprt->timeout = &xs_tcp_default_timeout;
2320
2321 switch (addr->sa_family) {
2322 case AF_INET:
2323 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2324 xprt_set_bound(xprt);
2325
2326 INIT_DELAYED_WORK(&transport->connect_worker,
2327 xs_tcp_setup_socket);
2328 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2329 break;
2330 case AF_INET6:
2331 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2332 xprt_set_bound(xprt);
2333
2334 INIT_DELAYED_WORK(&transport->connect_worker,
2335 xs_tcp_setup_socket);
2336 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2337 break;
2338 default:
2339 ret = ERR_PTR(-EAFNOSUPPORT);
2340 goto out_err;
2341 }
2342
2343 if (xprt_bound(xprt))
2344 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2345 xprt->address_strings[RPC_DISPLAY_ADDR],
2346 xprt->address_strings[RPC_DISPLAY_PORT],
2347 xprt->address_strings[RPC_DISPLAY_PROTO]);
2348 else
2349 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2350 xprt->address_strings[RPC_DISPLAY_ADDR],
2351 xprt->address_strings[RPC_DISPLAY_PROTO]);
2352
2353
2354 if (try_module_get(THIS_MODULE))
2355 return xprt;
2356 ret = ERR_PTR(-EINVAL);
2357 out_err:
2358 xprt_free(xprt);
2359 return ret;
2360 }
2361
2362 /**
2363 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2364 * @args: rpc transport creation arguments
2365 *
2366 */
2367 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2368 {
2369 struct sockaddr *addr = args->dstaddr;
2370 struct rpc_xprt *xprt;
2371 struct sock_xprt *transport;
2372 struct svc_sock *bc_sock;
2373 struct rpc_xprt *ret;
2374
2375 if (args->bc_xprt->xpt_bc_xprt) {
2376 /*
2377 * This server connection already has a backchannel
2378 * export; we can't create a new one, as we wouldn't be
2379 * able to match replies based on xid any more. So,
2380 * reuse the already-existing one:
2381 */
2382 return args->bc_xprt->xpt_bc_xprt;
2383 }
2384 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2385 if (IS_ERR(xprt))
2386 return xprt;
2387 transport = container_of(xprt, struct sock_xprt, xprt);
2388
2389 xprt->prot = IPPROTO_TCP;
2390 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2391 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2392 xprt->timeout = &xs_tcp_default_timeout;
2393
2394 /* backchannel */
2395 xprt_set_bound(xprt);
2396 xprt->bind_timeout = 0;
2397 xprt->reestablish_timeout = 0;
2398 xprt->idle_timeout = 0;
2399
2400 xprt->ops = &bc_tcp_ops;
2401
2402 switch (addr->sa_family) {
2403 case AF_INET:
2404 xs_format_peer_addresses(xprt, "tcp",
2405 RPCBIND_NETID_TCP);
2406 break;
2407 case AF_INET6:
2408 xs_format_peer_addresses(xprt, "tcp",
2409 RPCBIND_NETID_TCP6);
2410 break;
2411 default:
2412 ret = ERR_PTR(-EAFNOSUPPORT);
2413 goto out_err;
2414 }
2415
2416 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2417 xprt->address_strings[RPC_DISPLAY_ADDR],
2418 xprt->address_strings[RPC_DISPLAY_PORT],
2419 xprt->address_strings[RPC_DISPLAY_PROTO]);
2420
2421 /*
2422 * Once we've associated a backchannel xprt with a connection,
2423 * we want to keep it around as long as long as the connection
2424 * lasts, in case we need to start using it for a backchannel
2425 * again; this reference won't be dropped until bc_xprt is
2426 * destroyed.
2427 */
2428 xprt_get(xprt);
2429 args->bc_xprt->xpt_bc_xprt = xprt;
2430 xprt->bc_xprt = args->bc_xprt;
2431 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2432 transport->sock = bc_sock->sk_sock;
2433 transport->inet = bc_sock->sk_sk;
2434
2435 /*
2436 * Since we don't want connections for the backchannel, we set
2437 * the xprt status to connected
2438 */
2439 xprt_set_connected(xprt);
2440
2441
2442 if (try_module_get(THIS_MODULE))
2443 return xprt;
2444 xprt_put(xprt);
2445 ret = ERR_PTR(-EINVAL);
2446 out_err:
2447 xprt_free(xprt);
2448 return ret;
2449 }
2450
2451 static struct xprt_class xs_udp_transport = {
2452 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2453 .name = "udp",
2454 .owner = THIS_MODULE,
2455 .ident = XPRT_TRANSPORT_UDP,
2456 .setup = xs_setup_udp,
2457 };
2458
2459 static struct xprt_class xs_tcp_transport = {
2460 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2461 .name = "tcp",
2462 .owner = THIS_MODULE,
2463 .ident = XPRT_TRANSPORT_TCP,
2464 .setup = xs_setup_tcp,
2465 };
2466
2467 static struct xprt_class xs_bc_tcp_transport = {
2468 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2469 .name = "tcp NFSv4.1 backchannel",
2470 .owner = THIS_MODULE,
2471 .ident = XPRT_TRANSPORT_BC_TCP,
2472 .setup = xs_setup_bc_tcp,
2473 };
2474
2475 /**
2476 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2477 *
2478 */
2479 int init_socket_xprt(void)
2480 {
2481 #ifdef RPC_DEBUG
2482 if (!sunrpc_table_header)
2483 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2484 #endif
2485
2486 xprt_register_transport(&xs_udp_transport);
2487 xprt_register_transport(&xs_tcp_transport);
2488 xprt_register_transport(&xs_bc_tcp_transport);
2489
2490 return 0;
2491 }
2492
2493 /**
2494 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2495 *
2496 */
2497 void cleanup_socket_xprt(void)
2498 {
2499 #ifdef RPC_DEBUG
2500 if (sunrpc_table_header) {
2501 unregister_sysctl_table(sunrpc_table_header);
2502 sunrpc_table_header = NULL;
2503 }
2504 #endif
2505
2506 xprt_unregister_transport(&xs_udp_transport);
2507 xprt_unregister_transport(&xs_tcp_transport);
2508 xprt_unregister_transport(&xs_bc_tcp_transport);
2509 }
2510
2511 static int param_set_uint_minmax(const char *val,
2512 const struct kernel_param *kp,
2513 unsigned int min, unsigned int max)
2514 {
2515 unsigned long num;
2516 int ret;
2517
2518 if (!val)
2519 return -EINVAL;
2520 ret = strict_strtoul(val, 0, &num);
2521 if (ret == -EINVAL || num < min || num > max)
2522 return -EINVAL;
2523 *((unsigned int *)kp->arg) = num;
2524 return 0;
2525 }
2526
2527 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2528 {
2529 return param_set_uint_minmax(val, kp,
2530 RPC_MIN_RESVPORT,
2531 RPC_MAX_RESVPORT);
2532 }
2533
2534 static struct kernel_param_ops param_ops_portnr = {
2535 .set = param_set_portnr,
2536 .get = param_get_uint,
2537 };
2538
2539 #define param_check_portnr(name, p) \
2540 __param_check(name, p, unsigned int);
2541
2542 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2543 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2544
2545 static int param_set_slot_table_size(const char *val,
2546 const struct kernel_param *kp)
2547 {
2548 return param_set_uint_minmax(val, kp,
2549 RPC_MIN_SLOT_TABLE,
2550 RPC_MAX_SLOT_TABLE);
2551 }
2552
2553 static struct kernel_param_ops param_ops_slot_table_size = {
2554 .set = param_set_slot_table_size,
2555 .get = param_get_uint,
2556 };
2557
2558 #define param_check_slot_table_size(name, p) \
2559 __param_check(name, p, unsigned int);
2560
2561 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2562 slot_table_size, 0644);
2563 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2564 slot_table_size, 0644);
2565
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree