?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
USB: cdc-acm: fix memory corruption / panic
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sunrpc / xprtsock.c
blob2334279b3268e4e54ada21979b4c800aa0e51058
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 * Time out for an RPC UDP socket connect. UDP socket connects are
142 * synchronous, but we set a timeout anyway in case of resource
143 * exhaustion on the local host.
144 */
145 #define XS_UDP_CONN_TO (5U * HZ)
146
147 /*
148 * Wait duration for an RPC TCP connection to be established. Solaris
149 * NFS over TCP uses 60 seconds, for example, which is in line with how
150 * long a server takes to reboot.
151 */
152 #define XS_TCP_CONN_TO (60U * HZ)
153
154 /*
155 * Wait duration for a reply from the RPC portmapper.
156 */
157 #define XS_BIND_TO (60U * HZ)
158
159 /*
160 * Delay if a UDP socket connect error occurs. This is most likely some
161 * kind of resource problem on the local host.
162 */
163 #define XS_UDP_REEST_TO (2U * HZ)
164
165 /*
166 * The reestablish timeout allows clients to delay for a bit before attempting
167 * to reconnect to a server that just dropped our connection.
168 *
169 * We implement an exponential backoff when trying to reestablish a TCP
170 * transport connection with the server. Some servers like to drop a TCP
171 * connection when they are overworked, so we start with a short timeout and
172 * increase over time if the server is down or not responding.
173 */
174 #define XS_TCP_INIT_REEST_TO (3U * HZ)
175 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
176
177 /*
178 * TCP idle timeout; client drops the transport socket if it is idle
179 * for this long. Note that we also timeout UDP sockets to prevent
180 * holding port numbers when there is no RPC traffic.
181 */
182 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
183
184 #ifdef RPC_DEBUG
185 # undef RPC_DEBUG_DATA
186 # define RPCDBG_FACILITY RPCDBG_TRANS
187 #endif
188
189 #ifdef RPC_DEBUG_DATA
190 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
191 {
192 u8 *buf = (u8 *) packet;
193 int j;
194
195 dprintk("RPC: %s\n", msg);
196 for (j = 0; j < count && j < 128; j += 4) {
197 if (!(j & 31)) {
198 if (j)
199 dprintk("\n");
200 dprintk("0x%04x ", j);
201 }
202 dprintk("%02x%02x%02x%02x ",
203 buf[j], buf[j+1], buf[j+2], buf[j+3]);
204 }
205 dprintk("\n");
206 }
207 #else
208 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
209 {
210 /* NOP */
211 }
212 #endif
213
214 struct sock_xprt {
215 struct rpc_xprt xprt;
216
217 /*
218 * Network layer
219 */
220 struct socket * sock;
221 struct sock * inet;
222
223 /*
224 * State of TCP reply receive
225 */
226 __be32 tcp_fraghdr,
227 tcp_xid,
228 tcp_calldir;
229
230 u32 tcp_offset,
231 tcp_reclen;
232
233 unsigned long tcp_copied,
234 tcp_flags;
235
236 /*
237 * Connection of transports
238 */
239 struct delayed_work connect_worker;
240 struct sockaddr_storage srcaddr;
241 unsigned short srcport;
242
243 /*
244 * UDP socket buffer size parameters
245 */
246 size_t rcvsize,
247 sndsize;
248
249 /*
250 * Saved socket callback addresses
251 */
252 void (*old_data_ready)(struct sock *, int);
253 void (*old_state_change)(struct sock *);
254 void (*old_write_space)(struct sock *);
255 void (*old_error_report)(struct sock *);
256 };
257
258 /*
259 * TCP receive state flags
260 */
261 #define TCP_RCV_LAST_FRAG (1UL << 0)
262 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
263 #define TCP_RCV_COPY_XID (1UL << 2)
264 #define TCP_RCV_COPY_DATA (1UL << 3)
265 #define TCP_RCV_READ_CALLDIR (1UL << 4)
266 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
267
268 /*
269 * TCP RPC flags
270 */
271 #define TCP_RPC_REPLY (1UL << 6)
272
273 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
274 {
275 return (struct sockaddr *) &xprt->addr;
276 }
277
278 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
279 {
280 return (struct sockaddr_in *) &xprt->addr;
281 }
282
283 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
284 {
285 return (struct sockaddr_in6 *) &xprt->addr;
286 }
287
288 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
289 {
290 struct sockaddr *sap = xs_addr(xprt);
291 struct sockaddr_in6 *sin6;
292 struct sockaddr_in *sin;
293 char buf[128];
294
295 (void)rpc_ntop(sap, buf, sizeof(buf));
296 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
297
298 switch (sap->sa_family) {
299 case AF_INET:
300 sin = xs_addr_in(xprt);
301 (void)snprintf(buf, sizeof(buf), "%02x%02x%02x%02x",
302 NIPQUAD(sin->sin_addr.s_addr));
303 break;
304 case AF_INET6:
305 sin6 = xs_addr_in6(xprt);
306 (void)snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
307 break;
308 default:
309 BUG();
310 }
311 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
312 }
313
314 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
315 {
316 struct sockaddr *sap = xs_addr(xprt);
317 char buf[128];
318
319 (void)snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
320 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
321
322 (void)snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
323 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
324 }
325
326 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
327 const char *protocol,
328 const char *netid)
329 {
330 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
331 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
332 xs_format_common_peer_addresses(xprt);
333 xs_format_common_peer_ports(xprt);
334 }
335
336 static void xs_update_peer_port(struct rpc_xprt *xprt)
337 {
338 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
339 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
340
341 xs_format_common_peer_ports(xprt);
342 }
343
344 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
345 {
346 unsigned int i;
347
348 for (i = 0; i < RPC_DISPLAY_MAX; i++)
349 switch (i) {
350 case RPC_DISPLAY_PROTO:
351 case RPC_DISPLAY_NETID:
352 continue;
353 default:
354 kfree(xprt->address_strings[i]);
355 }
356 }
357
358 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
359
360 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
361 {
362 struct msghdr msg = {
363 .msg_name = addr,
364 .msg_namelen = addrlen,
365 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
366 };
367 struct kvec iov = {
368 .iov_base = vec->iov_base + base,
369 .iov_len = vec->iov_len - base,
370 };
371
372 if (iov.iov_len != 0)
373 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
374 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
375 }
376
377 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
378 {
379 struct page **ppage;
380 unsigned int remainder;
381 int err, sent = 0;
382
383 remainder = xdr->page_len - base;
384 base += xdr->page_base;
385 ppage = xdr->pages + (base >> PAGE_SHIFT);
386 base &= ~PAGE_MASK;
387 for(;;) {
388 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
389 int flags = XS_SENDMSG_FLAGS;
390
391 remainder -= len;
392 if (remainder != 0 || more)
393 flags |= MSG_MORE;
394 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
395 if (remainder == 0 || err != len)
396 break;
397 sent += err;
398 ppage++;
399 base = 0;
400 }
401 if (sent == 0)
402 return err;
403 if (err > 0)
404 sent += err;
405 return sent;
406 }
407
408 /**
409 * xs_sendpages - write pages directly to a socket
410 * @sock: socket to send on
411 * @addr: UDP only -- address of destination
412 * @addrlen: UDP only -- length of destination address
413 * @xdr: buffer containing this request
414 * @base: starting position in the buffer
415 *
416 */
417 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
418 {
419 unsigned int remainder = xdr->len - base;
420 int err, sent = 0;
421
422 if (unlikely(!sock))
423 return -ENOTSOCK;
424
425 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
426 if (base != 0) {
427 addr = NULL;
428 addrlen = 0;
429 }
430
431 if (base < xdr->head[0].iov_len || addr != NULL) {
432 unsigned int len = xdr->head[0].iov_len - base;
433 remainder -= len;
434 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
435 if (remainder == 0 || err != len)
436 goto out;
437 sent += err;
438 base = 0;
439 } else
440 base -= xdr->head[0].iov_len;
441
442 if (base < xdr->page_len) {
443 unsigned int len = xdr->page_len - base;
444 remainder -= len;
445 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
446 if (remainder == 0 || err != len)
447 goto out;
448 sent += err;
449 base = 0;
450 } else
451 base -= xdr->page_len;
452
453 if (base >= xdr->tail[0].iov_len)
454 return sent;
455 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
456 out:
457 if (sent == 0)
458 return err;
459 if (err > 0)
460 sent += err;
461 return sent;
462 }
463
464 static void xs_nospace_callback(struct rpc_task *task)
465 {
466 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
467
468 transport->inet->sk_write_pending--;
469 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
470 }
471
472 /**
473 * xs_nospace - place task on wait queue if transmit was incomplete
474 * @task: task to put to sleep
475 *
476 */
477 static int xs_nospace(struct rpc_task *task)
478 {
479 struct rpc_rqst *req = task->tk_rqstp;
480 struct rpc_xprt *xprt = req->rq_xprt;
481 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
482 int ret = 0;
483
484 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
485 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
486 req->rq_slen);
487
488 /* Protect against races with write_space */
489 spin_lock_bh(&xprt->transport_lock);
490
491 /* Don't race with disconnect */
492 if (xprt_connected(xprt)) {
493 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
494 ret = -EAGAIN;
495 /*
496 * Notify TCP that we're limited by the application
497 * window size
498 */
499 set_bit(SOCK_NOSPACE, &transport->sock->flags);
500 transport->inet->sk_write_pending++;
501 /* ...and wait for more buffer space */
502 xprt_wait_for_buffer_space(task, xs_nospace_callback);
503 }
504 } else {
505 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
506 ret = -ENOTCONN;
507 }
508
509 spin_unlock_bh(&xprt->transport_lock);
510 return ret;
511 }
512
513 /**
514 * xs_udp_send_request - write an RPC request to a UDP socket
515 * @task: address of RPC task that manages the state of an RPC request
516 *
517 * Return values:
518 * 0: The request has been sent
519 * EAGAIN: The socket was blocked, please call again later to
520 * complete the request
521 * ENOTCONN: Caller needs to invoke connect logic then call again
522 * other: Some other error occured, the request was not sent
523 */
524 static int xs_udp_send_request(struct rpc_task *task)
525 {
526 struct rpc_rqst *req = task->tk_rqstp;
527 struct rpc_xprt *xprt = req->rq_xprt;
528 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
529 struct xdr_buf *xdr = &req->rq_snd_buf;
530 int status;
531
532 xs_pktdump("packet data:",
533 req->rq_svec->iov_base,
534 req->rq_svec->iov_len);
535
536 if (!xprt_bound(xprt))
537 return -ENOTCONN;
538 status = xs_sendpages(transport->sock,
539 xs_addr(xprt),
540 xprt->addrlen, xdr,
541 req->rq_bytes_sent);
542
543 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
544 xdr->len - req->rq_bytes_sent, status);
545
546 if (status >= 0) {
547 task->tk_bytes_sent += status;
548 if (status >= req->rq_slen)
549 return 0;
550 /* Still some bytes left; set up for a retry later. */
551 status = -EAGAIN;
552 }
553 if (!transport->sock)
554 goto out;
555
556 switch (status) {
557 case -ENOTSOCK:
558 status = -ENOTCONN;
559 /* Should we call xs_close() here? */
560 break;
561 case -EAGAIN:
562 status = xs_nospace(task);
563 break;
564 default:
565 dprintk("RPC: sendmsg returned unrecognized error %d\n",
566 -status);
567 case -ENETUNREACH:
568 case -EPIPE:
569 case -ECONNREFUSED:
570 /* When the server has died, an ICMP port unreachable message
571 * prompts ECONNREFUSED. */
572 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
573 }
574 out:
575 return status;
576 }
577
578 /**
579 * xs_tcp_shutdown - gracefully shut down a TCP socket
580 * @xprt: transport
581 *
582 * Initiates a graceful shutdown of the TCP socket by calling the
583 * equivalent of shutdown(SHUT_WR);
584 */
585 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
586 {
587 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
588 struct socket *sock = transport->sock;
589
590 if (sock != NULL)
591 kernel_sock_shutdown(sock, SHUT_WR);
592 }
593
594 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
595 {
596 u32 reclen = buf->len - sizeof(rpc_fraghdr);
597 rpc_fraghdr *base = buf->head[0].iov_base;
598 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
599 }
600
601 /**
602 * xs_tcp_send_request - write an RPC request to a TCP socket
603 * @task: address of RPC task that manages the state of an RPC request
604 *
605 * Return values:
606 * 0: The request has been sent
607 * EAGAIN: The socket was blocked, please call again later to
608 * complete the request
609 * ENOTCONN: Caller needs to invoke connect logic then call again
610 * other: Some other error occured, the request was not sent
611 *
612 * XXX: In the case of soft timeouts, should we eventually give up
613 * if sendmsg is not able to make progress?
614 */
615 static int xs_tcp_send_request(struct rpc_task *task)
616 {
617 struct rpc_rqst *req = task->tk_rqstp;
618 struct rpc_xprt *xprt = req->rq_xprt;
619 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
620 struct xdr_buf *xdr = &req->rq_snd_buf;
621 int status;
622
623 xs_encode_tcp_record_marker(&req->rq_snd_buf);
624
625 xs_pktdump("packet data:",
626 req->rq_svec->iov_base,
627 req->rq_svec->iov_len);
628
629 /* Continue transmitting the packet/record. We must be careful
630 * to cope with writespace callbacks arriving _after_ we have
631 * called sendmsg(). */
632 while (1) {
633 status = xs_sendpages(transport->sock,
634 NULL, 0, xdr, req->rq_bytes_sent);
635
636 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
637 xdr->len - req->rq_bytes_sent, status);
638
639 if (unlikely(status < 0))
640 break;
641
642 /* If we've sent the entire packet, immediately
643 * reset the count of bytes sent. */
644 req->rq_bytes_sent += status;
645 task->tk_bytes_sent += status;
646 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
647 req->rq_bytes_sent = 0;
648 return 0;
649 }
650
651 if (status != 0)
652 continue;
653 status = -EAGAIN;
654 break;
655 }
656 if (!transport->sock)
657 goto out;
658
659 switch (status) {
660 case -ENOTSOCK:
661 status = -ENOTCONN;
662 /* Should we call xs_close() here? */
663 break;
664 case -EAGAIN:
665 status = xs_nospace(task);
666 break;
667 default:
668 dprintk("RPC: sendmsg returned unrecognized error %d\n",
669 -status);
670 case -ECONNRESET:
671 case -EPIPE:
672 xs_tcp_shutdown(xprt);
673 case -ECONNREFUSED:
674 case -ENOTCONN:
675 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
676 }
677 out:
678 return status;
679 }
680
681 /**
682 * xs_tcp_release_xprt - clean up after a tcp transmission
683 * @xprt: transport
684 * @task: rpc task
685 *
686 * This cleans up if an error causes us to abort the transmission of a request.
687 * In this case, the socket may need to be reset in order to avoid confusing
688 * the server.
689 */
690 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
691 {
692 struct rpc_rqst *req;
693
694 if (task != xprt->snd_task)
695 return;
696 if (task == NULL)
697 goto out_release;
698 req = task->tk_rqstp;
699 if (req->rq_bytes_sent == 0)
700 goto out_release;
701 if (req->rq_bytes_sent == req->rq_snd_buf.len)
702 goto out_release;
703 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
704 out_release:
705 xprt_release_xprt(xprt, task);
706 }
707
708 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
709 {
710 transport->old_data_ready = sk->sk_data_ready;
711 transport->old_state_change = sk->sk_state_change;
712 transport->old_write_space = sk->sk_write_space;
713 transport->old_error_report = sk->sk_error_report;
714 }
715
716 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
717 {
718 sk->sk_data_ready = transport->old_data_ready;
719 sk->sk_state_change = transport->old_state_change;
720 sk->sk_write_space = transport->old_write_space;
721 sk->sk_error_report = transport->old_error_report;
722 }
723
724 static void xs_reset_transport(struct sock_xprt *transport)
725 {
726 struct socket *sock = transport->sock;
727 struct sock *sk = transport->inet;
728
729 if (sk == NULL)
730 return;
731
732 write_lock_bh(&sk->sk_callback_lock);
733 transport->inet = NULL;
734 transport->sock = NULL;
735
736 sk->sk_user_data = NULL;
737
738 xs_restore_old_callbacks(transport, sk);
739 write_unlock_bh(&sk->sk_callback_lock);
740
741 sk->sk_no_check = 0;
742
743 sock_release(sock);
744 }
745
746 /**
747 * xs_close - close a socket
748 * @xprt: transport
749 *
750 * This is used when all requests are complete; ie, no DRC state remains
751 * on the server we want to save.
752 *
753 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
754 * xs_reset_transport() zeroing the socket from underneath a writer.
755 */
756 static void xs_close(struct rpc_xprt *xprt)
757 {
758 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
759
760 dprintk("RPC: xs_close xprt %p\n", xprt);
761
762 xs_reset_transport(transport);
763 xprt->reestablish_timeout = 0;
764
765 smp_mb__before_clear_bit();
766 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
767 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
768 clear_bit(XPRT_CLOSING, &xprt->state);
769 smp_mb__after_clear_bit();
770 xprt_disconnect_done(xprt);
771 }
772
773 static void xs_tcp_close(struct rpc_xprt *xprt)
774 {
775 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
776 xs_close(xprt);
777 else
778 xs_tcp_shutdown(xprt);
779 }
780
781 /**
782 * xs_destroy - prepare to shutdown a transport
783 * @xprt: doomed transport
784 *
785 */
786 static void xs_destroy(struct rpc_xprt *xprt)
787 {
788 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
789
790 dprintk("RPC: xs_destroy xprt %p\n", xprt);
791
792 cancel_rearming_delayed_work(&transport->connect_worker);
793
794 xs_close(xprt);
795 xs_free_peer_addresses(xprt);
796 kfree(xprt->slot);
797 kfree(xprt);
798 module_put(THIS_MODULE);
799 }
800
801 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
802 {
803 return (struct rpc_xprt *) sk->sk_user_data;
804 }
805
806 /**
807 * xs_udp_data_ready - "data ready" callback for UDP sockets
808 * @sk: socket with data to read
809 * @len: how much data to read
810 *
811 */
812 static void xs_udp_data_ready(struct sock *sk, int len)
813 {
814 struct rpc_task *task;
815 struct rpc_xprt *xprt;
816 struct rpc_rqst *rovr;
817 struct sk_buff *skb;
818 int err, repsize, copied;
819 u32 _xid;
820 __be32 *xp;
821
822 read_lock(&sk->sk_callback_lock);
823 dprintk("RPC: xs_udp_data_ready...\n");
824 if (!(xprt = xprt_from_sock(sk)))
825 goto out;
826
827 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
828 goto out;
829
830 if (xprt->shutdown)
831 goto dropit;
832
833 repsize = skb->len - sizeof(struct udphdr);
834 if (repsize < 4) {
835 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
836 goto dropit;
837 }
838
839 /* Copy the XID from the skb... */
840 xp = skb_header_pointer(skb, sizeof(struct udphdr),
841 sizeof(_xid), &_xid);
842 if (xp == NULL)
843 goto dropit;
844
845 /* Look up and lock the request corresponding to the given XID */
846 spin_lock(&xprt->transport_lock);
847 rovr = xprt_lookup_rqst(xprt, *xp);
848 if (!rovr)
849 goto out_unlock;
850 task = rovr->rq_task;
851
852 if ((copied = rovr->rq_private_buf.buflen) > repsize)
853 copied = repsize;
854
855 /* Suck it into the iovec, verify checksum if not done by hw. */
856 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
857 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
858 goto out_unlock;
859 }
860
861 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
862
863 /* Something worked... */
864 dst_confirm(skb_dst(skb));
865
866 xprt_adjust_cwnd(task, copied);
867 xprt_update_rtt(task);
868 xprt_complete_rqst(task, copied);
869
870 out_unlock:
871 spin_unlock(&xprt->transport_lock);
872 dropit:
873 skb_free_datagram(sk, skb);
874 out:
875 read_unlock(&sk->sk_callback_lock);
876 }
877
878 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
879 {
880 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
881 size_t len, used;
882 char *p;
883
884 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
885 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
886 used = xdr_skb_read_bits(desc, p, len);
887 transport->tcp_offset += used;
888 if (used != len)
889 return;
890
891 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
892 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
893 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
894 else
895 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
896 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
897
898 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
899 transport->tcp_offset = 0;
900
901 /* Sanity check of the record length */
902 if (unlikely(transport->tcp_reclen < 8)) {
903 dprintk("RPC: invalid TCP record fragment length\n");
904 xprt_force_disconnect(xprt);
905 return;
906 }
907 dprintk("RPC: reading TCP record fragment of length %d\n",
908 transport->tcp_reclen);
909 }
910
911 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
912 {
913 if (transport->tcp_offset == transport->tcp_reclen) {
914 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
915 transport->tcp_offset = 0;
916 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
917 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
918 transport->tcp_flags |= TCP_RCV_COPY_XID;
919 transport->tcp_copied = 0;
920 }
921 }
922 }
923
924 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
925 {
926 size_t len, used;
927 char *p;
928
929 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
930 dprintk("RPC: reading XID (%Zu bytes)\n", len);
931 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
932 used = xdr_skb_read_bits(desc, p, len);
933 transport->tcp_offset += used;
934 if (used != len)
935 return;
936 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
937 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
938 transport->tcp_copied = 4;
939 dprintk("RPC: reading %s XID %08x\n",
940 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
941 : "request with",
942 ntohl(transport->tcp_xid));
943 xs_tcp_check_fraghdr(transport);
944 }
945
946 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
947 struct xdr_skb_reader *desc)
948 {
949 size_t len, used;
950 u32 offset;
951 char *p;
952
953 /*
954 * We want transport->tcp_offset to be 8 at the end of this routine
955 * (4 bytes for the xid and 4 bytes for the call/reply flag).
956 * When this function is called for the first time,
957 * transport->tcp_offset is 4 (after having already read the xid).
958 */
959 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
960 len = sizeof(transport->tcp_calldir) - offset;
961 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
962 p = ((char *) &transport->tcp_calldir) + offset;
963 used = xdr_skb_read_bits(desc, p, len);
964 transport->tcp_offset += used;
965 if (used != len)
966 return;
967 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
968 /*
969 * We don't yet have the XDR buffer, so we will write the calldir
970 * out after we get the buffer from the 'struct rpc_rqst'
971 */
972 switch (ntohl(transport->tcp_calldir)) {
973 case RPC_REPLY:
974 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
975 transport->tcp_flags |= TCP_RCV_COPY_DATA;
976 transport->tcp_flags |= TCP_RPC_REPLY;
977 break;
978 case RPC_CALL:
979 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
980 transport->tcp_flags |= TCP_RCV_COPY_DATA;
981 transport->tcp_flags &= ~TCP_RPC_REPLY;
982 break;
983 default:
984 dprintk("RPC: invalid request message type\n");
985 xprt_force_disconnect(&transport->xprt);
986 }
987 xs_tcp_check_fraghdr(transport);
988 }
989
990 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
991 struct xdr_skb_reader *desc,
992 struct rpc_rqst *req)
993 {
994 struct sock_xprt *transport =
995 container_of(xprt, struct sock_xprt, xprt);
996 struct xdr_buf *rcvbuf;
997 size_t len;
998 ssize_t r;
999
1000 rcvbuf = &req->rq_private_buf;
1001
1002 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1003 /*
1004 * Save the RPC direction in the XDR buffer
1005 */
1006 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1007 &transport->tcp_calldir,
1008 sizeof(transport->tcp_calldir));
1009 transport->tcp_copied += sizeof(transport->tcp_calldir);
1010 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1011 }
1012
1013 len = desc->count;
1014 if (len > transport->tcp_reclen - transport->tcp_offset) {
1015 struct xdr_skb_reader my_desc;
1016
1017 len = transport->tcp_reclen - transport->tcp_offset;
1018 memcpy(&my_desc, desc, sizeof(my_desc));
1019 my_desc.count = len;
1020 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1021 &my_desc, xdr_skb_read_bits);
1022 desc->count -= r;
1023 desc->offset += r;
1024 } else
1025 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1026 desc, xdr_skb_read_bits);
1027
1028 if (r > 0) {
1029 transport->tcp_copied += r;
1030 transport->tcp_offset += r;
1031 }
1032 if (r != len) {
1033 /* Error when copying to the receive buffer,
1034 * usually because we weren't able to allocate
1035 * additional buffer pages. All we can do now
1036 * is turn off TCP_RCV_COPY_DATA, so the request
1037 * will not receive any additional updates,
1038 * and time out.
1039 * Any remaining data from this record will
1040 * be discarded.
1041 */
1042 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1043 dprintk("RPC: XID %08x truncated request\n",
1044 ntohl(transport->tcp_xid));
1045 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1046 "tcp_offset = %u, tcp_reclen = %u\n",
1047 xprt, transport->tcp_copied,
1048 transport->tcp_offset, transport->tcp_reclen);
1049 return;
1050 }
1051
1052 dprintk("RPC: XID %08x read %Zd bytes\n",
1053 ntohl(transport->tcp_xid), r);
1054 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1055 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1056 transport->tcp_offset, transport->tcp_reclen);
1057
1058 if (transport->tcp_copied == req->rq_private_buf.buflen)
1059 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1060 else if (transport->tcp_offset == transport->tcp_reclen) {
1061 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1062 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1063 }
1064
1065 return;
1066 }
1067
1068 /*
1069 * Finds the request corresponding to the RPC xid and invokes the common
1070 * tcp read code to read the data.
1071 */
1072 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1073 struct xdr_skb_reader *desc)
1074 {
1075 struct sock_xprt *transport =
1076 container_of(xprt, struct sock_xprt, xprt);
1077 struct rpc_rqst *req;
1078
1079 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1080
1081 /* Find and lock the request corresponding to this xid */
1082 spin_lock(&xprt->transport_lock);
1083 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1084 if (!req) {
1085 dprintk("RPC: XID %08x request not found!\n",
1086 ntohl(transport->tcp_xid));
1087 spin_unlock(&xprt->transport_lock);
1088 return -1;
1089 }
1090
1091 xs_tcp_read_common(xprt, desc, req);
1092
1093 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1094 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1095
1096 spin_unlock(&xprt->transport_lock);
1097 return 0;
1098 }
1099
1100 #if defined(CONFIG_NFS_V4_1)
1101 /*
1102 * Obtains an rpc_rqst previously allocated and invokes the common
1103 * tcp read code to read the data. The result is placed in the callback
1104 * queue.
1105 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1106 * connection and return -1.
1107 */
1108 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1109 struct xdr_skb_reader *desc)
1110 {
1111 struct sock_xprt *transport =
1112 container_of(xprt, struct sock_xprt, xprt);
1113 struct rpc_rqst *req;
1114
1115 req = xprt_alloc_bc_request(xprt);
1116 if (req == NULL) {
1117 printk(KERN_WARNING "Callback slot table overflowed\n");
1118 xprt_force_disconnect(xprt);
1119 return -1;
1120 }
1121
1122 req->rq_xid = transport->tcp_xid;
1123 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1124 xs_tcp_read_common(xprt, desc, req);
1125
1126 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1127 struct svc_serv *bc_serv = xprt->bc_serv;
1128
1129 /*
1130 * Add callback request to callback list. The callback
1131 * service sleeps on the sv_cb_waitq waiting for new
1132 * requests. Wake it up after adding enqueing the
1133 * request.
1134 */
1135 dprintk("RPC: add callback request to list\n");
1136 spin_lock(&bc_serv->sv_cb_lock);
1137 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1138 spin_unlock(&bc_serv->sv_cb_lock);
1139 wake_up(&bc_serv->sv_cb_waitq);
1140 }
1141
1142 req->rq_private_buf.len = transport->tcp_copied;
1143
1144 return 0;
1145 }
1146
1147 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1148 struct xdr_skb_reader *desc)
1149 {
1150 struct sock_xprt *transport =
1151 container_of(xprt, struct sock_xprt, xprt);
1152
1153 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1154 xs_tcp_read_reply(xprt, desc) :
1155 xs_tcp_read_callback(xprt, desc);
1156 }
1157 #else
1158 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1159 struct xdr_skb_reader *desc)
1160 {
1161 return xs_tcp_read_reply(xprt, desc);
1162 }
1163 #endif /* CONFIG_NFS_V4_1 */
1164
1165 /*
1166 * Read data off the transport. This can be either an RPC_CALL or an
1167 * RPC_REPLY. Relay the processing to helper functions.
1168 */
1169 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1170 struct xdr_skb_reader *desc)
1171 {
1172 struct sock_xprt *transport =
1173 container_of(xprt, struct sock_xprt, xprt);
1174
1175 if (_xs_tcp_read_data(xprt, desc) == 0)
1176 xs_tcp_check_fraghdr(transport);
1177 else {
1178 /*
1179 * The transport_lock protects the request handling.
1180 * There's no need to hold it to update the tcp_flags.
1181 */
1182 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1183 }
1184 }
1185
1186 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1187 {
1188 size_t len;
1189
1190 len = transport->tcp_reclen - transport->tcp_offset;
1191 if (len > desc->count)
1192 len = desc->count;
1193 desc->count -= len;
1194 desc->offset += len;
1195 transport->tcp_offset += len;
1196 dprintk("RPC: discarded %Zu bytes\n", len);
1197 xs_tcp_check_fraghdr(transport);
1198 }
1199
1200 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1201 {
1202 struct rpc_xprt *xprt = rd_desc->arg.data;
1203 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1204 struct xdr_skb_reader desc = {
1205 .skb = skb,
1206 .offset = offset,
1207 .count = len,
1208 };
1209
1210 dprintk("RPC: xs_tcp_data_recv started\n");
1211 do {
1212 /* Read in a new fragment marker if necessary */
1213 /* Can we ever really expect to get completely empty fragments? */
1214 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1215 xs_tcp_read_fraghdr(xprt, &desc);
1216 continue;
1217 }
1218 /* Read in the xid if necessary */
1219 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1220 xs_tcp_read_xid(transport, &desc);
1221 continue;
1222 }
1223 /* Read in the call/reply flag */
1224 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1225 xs_tcp_read_calldir(transport, &desc);
1226 continue;
1227 }
1228 /* Read in the request data */
1229 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1230 xs_tcp_read_data(xprt, &desc);
1231 continue;
1232 }
1233 /* Skip over any trailing bytes on short reads */
1234 xs_tcp_read_discard(transport, &desc);
1235 } while (desc.count);
1236 dprintk("RPC: xs_tcp_data_recv done\n");
1237 return len - desc.count;
1238 }
1239
1240 /**
1241 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1242 * @sk: socket with data to read
1243 * @bytes: how much data to read
1244 *
1245 */
1246 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1247 {
1248 struct rpc_xprt *xprt;
1249 read_descriptor_t rd_desc;
1250 int read;
1251
1252 dprintk("RPC: xs_tcp_data_ready...\n");
1253
1254 read_lock(&sk->sk_callback_lock);
1255 if (!(xprt = xprt_from_sock(sk)))
1256 goto out;
1257 if (xprt->shutdown)
1258 goto out;
1259
1260 /* Any data means we had a useful conversation, so
1261 * the we don't need to delay the next reconnect
1262 */
1263 if (xprt->reestablish_timeout)
1264 xprt->reestablish_timeout = 0;
1265
1266 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1267 rd_desc.arg.data = xprt;
1268 do {
1269 rd_desc.count = 65536;
1270 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1271 } while (read > 0);
1272 out:
1273 read_unlock(&sk->sk_callback_lock);
1274 }
1275
1276 /*
1277 * Do the equivalent of linger/linger2 handling for dealing with
1278 * broken servers that don't close the socket in a timely
1279 * fashion
1280 */
1281 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1282 unsigned long timeout)
1283 {
1284 struct sock_xprt *transport;
1285
1286 if (xprt_test_and_set_connecting(xprt))
1287 return;
1288 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1289 transport = container_of(xprt, struct sock_xprt, xprt);
1290 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1291 timeout);
1292 }
1293
1294 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1295 {
1296 struct sock_xprt *transport;
1297
1298 transport = container_of(xprt, struct sock_xprt, xprt);
1299
1300 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1301 !cancel_delayed_work(&transport->connect_worker))
1302 return;
1303 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1304 xprt_clear_connecting(xprt);
1305 }
1306
1307 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1308 {
1309 smp_mb__before_clear_bit();
1310 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1311 clear_bit(XPRT_CLOSING, &xprt->state);
1312 smp_mb__after_clear_bit();
1313 /* Mark transport as closed and wake up all pending tasks */
1314 xprt_disconnect_done(xprt);
1315 }
1316
1317 /**
1318 * xs_tcp_state_change - callback to handle TCP socket state changes
1319 * @sk: socket whose state has changed
1320 *
1321 */
1322 static void xs_tcp_state_change(struct sock *sk)
1323 {
1324 struct rpc_xprt *xprt;
1325
1326 read_lock(&sk->sk_callback_lock);
1327 if (!(xprt = xprt_from_sock(sk)))
1328 goto out;
1329 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1330 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1331 sk->sk_state, xprt_connected(xprt),
1332 sock_flag(sk, SOCK_DEAD),
1333 sock_flag(sk, SOCK_ZAPPED));
1334
1335 switch (sk->sk_state) {
1336 case TCP_ESTABLISHED:
1337 spin_lock_bh(&xprt->transport_lock);
1338 if (!xprt_test_and_set_connected(xprt)) {
1339 struct sock_xprt *transport = container_of(xprt,
1340 struct sock_xprt, xprt);
1341
1342 /* Reset TCP record info */
1343 transport->tcp_offset = 0;
1344 transport->tcp_reclen = 0;
1345 transport->tcp_copied = 0;
1346 transport->tcp_flags =
1347 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1348
1349 xprt_wake_pending_tasks(xprt, -EAGAIN);
1350 }
1351 spin_unlock_bh(&xprt->transport_lock);
1352 break;
1353 case TCP_FIN_WAIT1:
1354 /* The client initiated a shutdown of the socket */
1355 xprt->connect_cookie++;
1356 xprt->reestablish_timeout = 0;
1357 set_bit(XPRT_CLOSING, &xprt->state);
1358 smp_mb__before_clear_bit();
1359 clear_bit(XPRT_CONNECTED, &xprt->state);
1360 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1361 smp_mb__after_clear_bit();
1362 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1363 break;
1364 case TCP_CLOSE_WAIT:
1365 /* The server initiated a shutdown of the socket */
1366 xprt_force_disconnect(xprt);
1367 case TCP_SYN_SENT:
1368 xprt->connect_cookie++;
1369 case TCP_CLOSING:
1370 /*
1371 * If the server closed down the connection, make sure that
1372 * we back off before reconnecting
1373 */
1374 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1375 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1376 break;
1377 case TCP_LAST_ACK:
1378 set_bit(XPRT_CLOSING, &xprt->state);
1379 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1380 smp_mb__before_clear_bit();
1381 clear_bit(XPRT_CONNECTED, &xprt->state);
1382 smp_mb__after_clear_bit();
1383 break;
1384 case TCP_CLOSE:
1385 xs_tcp_cancel_linger_timeout(xprt);
1386 xs_sock_mark_closed(xprt);
1387 }
1388 out:
1389 read_unlock(&sk->sk_callback_lock);
1390 }
1391
1392 /**
1393 * xs_error_report - callback mainly for catching socket errors
1394 * @sk: socket
1395 */
1396 static void xs_error_report(struct sock *sk)
1397 {
1398 struct rpc_xprt *xprt;
1399
1400 read_lock(&sk->sk_callback_lock);
1401 if (!(xprt = xprt_from_sock(sk)))
1402 goto out;
1403 dprintk("RPC: %s client %p...\n"
1404 "RPC: error %d\n",
1405 __func__, xprt, sk->sk_err);
1406 xprt_wake_pending_tasks(xprt, -EAGAIN);
1407 out:
1408 read_unlock(&sk->sk_callback_lock);
1409 }
1410
1411 static void xs_write_space(struct sock *sk)
1412 {
1413 struct socket *sock;
1414 struct rpc_xprt *xprt;
1415
1416 if (unlikely(!(sock = sk->sk_socket)))
1417 return;
1418 clear_bit(SOCK_NOSPACE, &sock->flags);
1419
1420 if (unlikely(!(xprt = xprt_from_sock(sk))))
1421 return;
1422 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1423 return;
1424
1425 xprt_write_space(xprt);
1426 }
1427
1428 /**
1429 * xs_udp_write_space - callback invoked when socket buffer space
1430 * becomes available
1431 * @sk: socket whose state has changed
1432 *
1433 * Called when more output buffer space is available for this socket.
1434 * We try not to wake our writers until they can make "significant"
1435 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1436 * with a bunch of small requests.
1437 */
1438 static void xs_udp_write_space(struct sock *sk)
1439 {
1440 read_lock(&sk->sk_callback_lock);
1441
1442 /* from net/core/sock.c:sock_def_write_space */
1443 if (sock_writeable(sk))
1444 xs_write_space(sk);
1445
1446 read_unlock(&sk->sk_callback_lock);
1447 }
1448
1449 /**
1450 * xs_tcp_write_space - callback invoked when socket buffer space
1451 * becomes available
1452 * @sk: socket whose state has changed
1453 *
1454 * Called when more output buffer space is available for this socket.
1455 * We try not to wake our writers until they can make "significant"
1456 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1457 * with a bunch of small requests.
1458 */
1459 static void xs_tcp_write_space(struct sock *sk)
1460 {
1461 read_lock(&sk->sk_callback_lock);
1462
1463 /* from net/core/stream.c:sk_stream_write_space */
1464 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1465 xs_write_space(sk);
1466
1467 read_unlock(&sk->sk_callback_lock);
1468 }
1469
1470 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1471 {
1472 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1473 struct sock *sk = transport->inet;
1474
1475 if (transport->rcvsize) {
1476 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1477 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1478 }
1479 if (transport->sndsize) {
1480 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1481 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1482 sk->sk_write_space(sk);
1483 }
1484 }
1485
1486 /**
1487 * xs_udp_set_buffer_size - set send and receive limits
1488 * @xprt: generic transport
1489 * @sndsize: requested size of send buffer, in bytes
1490 * @rcvsize: requested size of receive buffer, in bytes
1491 *
1492 * Set socket send and receive buffer size limits.
1493 */
1494 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1495 {
1496 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1497
1498 transport->sndsize = 0;
1499 if (sndsize)
1500 transport->sndsize = sndsize + 1024;
1501 transport->rcvsize = 0;
1502 if (rcvsize)
1503 transport->rcvsize = rcvsize + 1024;
1504
1505 xs_udp_do_set_buffer_size(xprt);
1506 }
1507
1508 /**
1509 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1510 * @task: task that timed out
1511 *
1512 * Adjust the congestion window after a retransmit timeout has occurred.
1513 */
1514 static void xs_udp_timer(struct rpc_task *task)
1515 {
1516 xprt_adjust_cwnd(task, -ETIMEDOUT);
1517 }
1518
1519 static unsigned short xs_get_random_port(void)
1520 {
1521 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1522 unsigned short rand = (unsigned short) net_random() % range;
1523 return rand + xprt_min_resvport;
1524 }
1525
1526 /**
1527 * xs_set_port - reset the port number in the remote endpoint address
1528 * @xprt: generic transport
1529 * @port: new port number
1530 *
1531 */
1532 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1533 {
1534 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1535
1536 rpc_set_port(xs_addr(xprt), port);
1537 xs_update_peer_port(xprt);
1538 }
1539
1540 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1541 {
1542 unsigned short port = transport->srcport;
1543
1544 if (port == 0 && transport->xprt.resvport)
1545 port = xs_get_random_port();
1546 return port;
1547 }
1548
1549 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1550 {
1551 if (transport->srcport != 0)
1552 transport->srcport = 0;
1553 if (!transport->xprt.resvport)
1554 return 0;
1555 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1556 return xprt_max_resvport;
1557 return --port;
1558 }
1559
1560 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1561 {
1562 struct sockaddr_in myaddr = {
1563 .sin_family = AF_INET,
1564 };
1565 struct sockaddr_in *sa;
1566 int err, nloop = 0;
1567 unsigned short port = xs_get_srcport(transport, sock);
1568 unsigned short last;
1569
1570 sa = (struct sockaddr_in *)&transport->srcaddr;
1571 myaddr.sin_addr = sa->sin_addr;
1572 do {
1573 myaddr.sin_port = htons(port);
1574 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1575 sizeof(myaddr));
1576 if (port == 0)
1577 break;
1578 if (err == 0) {
1579 transport->srcport = port;
1580 break;
1581 }
1582 last = port;
1583 port = xs_next_srcport(transport, sock, port);
1584 if (port > last)
1585 nloop++;
1586 } while (err == -EADDRINUSE && nloop != 2);
1587 dprintk("RPC: %s %pI4:%u: %s (%d)\n",
1588 __func__, &myaddr.sin_addr,
1589 port, err ? "failed" : "ok", err);
1590 return err;
1591 }
1592
1593 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1594 {
1595 struct sockaddr_in6 myaddr = {
1596 .sin6_family = AF_INET6,
1597 };
1598 struct sockaddr_in6 *sa;
1599 int err, nloop = 0;
1600 unsigned short port = xs_get_srcport(transport, sock);
1601 unsigned short last;
1602
1603 sa = (struct sockaddr_in6 *)&transport->srcaddr;
1604 myaddr.sin6_addr = sa->sin6_addr;
1605 do {
1606 myaddr.sin6_port = htons(port);
1607 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1608 sizeof(myaddr));
1609 if (port == 0)
1610 break;
1611 if (err == 0) {
1612 transport->srcport = port;
1613 break;
1614 }
1615 last = port;
1616 port = xs_next_srcport(transport, sock, port);
1617 if (port > last)
1618 nloop++;
1619 } while (err == -EADDRINUSE && nloop != 2);
1620 dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n",
1621 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1622 return err;
1623 }
1624
1625 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1626 static struct lock_class_key xs_key[2];
1627 static struct lock_class_key xs_slock_key[2];
1628
1629 static inline void xs_reclassify_socket4(struct socket *sock)
1630 {
1631 struct sock *sk = sock->sk;
1632
1633 BUG_ON(sock_owned_by_user(sk));
1634 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1635 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1636 }
1637
1638 static inline void xs_reclassify_socket6(struct socket *sock)
1639 {
1640 struct sock *sk = sock->sk;
1641
1642 BUG_ON(sock_owned_by_user(sk));
1643 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1644 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1645 }
1646 #else
1647 static inline void xs_reclassify_socket4(struct socket *sock)
1648 {
1649 }
1650
1651 static inline void xs_reclassify_socket6(struct socket *sock)
1652 {
1653 }
1654 #endif
1655
1656 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1657 {
1658 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1659
1660 if (!transport->inet) {
1661 struct sock *sk = sock->sk;
1662
1663 write_lock_bh(&sk->sk_callback_lock);
1664
1665 xs_save_old_callbacks(transport, sk);
1666
1667 sk->sk_user_data = xprt;
1668 sk->sk_data_ready = xs_udp_data_ready;
1669 sk->sk_write_space = xs_udp_write_space;
1670 sk->sk_error_report = xs_error_report;
1671 sk->sk_no_check = UDP_CSUM_NORCV;
1672 sk->sk_allocation = GFP_ATOMIC;
1673
1674 xprt_set_connected(xprt);
1675
1676 /* Reset to new socket */
1677 transport->sock = sock;
1678 transport->inet = sk;
1679
1680 write_unlock_bh(&sk->sk_callback_lock);
1681 }
1682 xs_udp_do_set_buffer_size(xprt);
1683 }
1684
1685 /**
1686 * xs_udp_connect_worker4 - set up a UDP socket
1687 * @work: RPC transport to connect
1688 *
1689 * Invoked by a work queue tasklet.
1690 */
1691 static void xs_udp_connect_worker4(struct work_struct *work)
1692 {
1693 struct sock_xprt *transport =
1694 container_of(work, struct sock_xprt, connect_worker.work);
1695 struct rpc_xprt *xprt = &transport->xprt;
1696 struct socket *sock = transport->sock;
1697 int err, status = -EIO;
1698
1699 if (xprt->shutdown)
1700 goto out;
1701
1702 /* Start by resetting any existing state */
1703 xs_reset_transport(transport);
1704
1705 err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1706 if (err < 0) {
1707 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1708 goto out;
1709 }
1710 xs_reclassify_socket4(sock);
1711
1712 if (xs_bind4(transport, sock)) {
1713 sock_release(sock);
1714 goto out;
1715 }
1716
1717 dprintk("RPC: worker connecting xprt %p via %s to "
1718 "%s (port %s)\n", xprt,
1719 xprt->address_strings[RPC_DISPLAY_PROTO],
1720 xprt->address_strings[RPC_DISPLAY_ADDR],
1721 xprt->address_strings[RPC_DISPLAY_PORT]);
1722
1723 xs_udp_finish_connecting(xprt, sock);
1724 status = 0;
1725 out:
1726 xprt_clear_connecting(xprt);
1727 xprt_wake_pending_tasks(xprt, status);
1728 }
1729
1730 /**
1731 * xs_udp_connect_worker6 - set up a UDP socket
1732 * @work: RPC transport to connect
1733 *
1734 * Invoked by a work queue tasklet.
1735 */
1736 static void xs_udp_connect_worker6(struct work_struct *work)
1737 {
1738 struct sock_xprt *transport =
1739 container_of(work, struct sock_xprt, connect_worker.work);
1740 struct rpc_xprt *xprt = &transport->xprt;
1741 struct socket *sock = transport->sock;
1742 int err, status = -EIO;
1743
1744 if (xprt->shutdown)
1745 goto out;
1746
1747 /* Start by resetting any existing state */
1748 xs_reset_transport(transport);
1749
1750 err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1751 if (err < 0) {
1752 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1753 goto out;
1754 }
1755 xs_reclassify_socket6(sock);
1756
1757 if (xs_bind6(transport, sock) < 0) {
1758 sock_release(sock);
1759 goto out;
1760 }
1761
1762 dprintk("RPC: worker connecting xprt %p via %s to "
1763 "%s (port %s)\n", xprt,
1764 xprt->address_strings[RPC_DISPLAY_PROTO],
1765 xprt->address_strings[RPC_DISPLAY_ADDR],
1766 xprt->address_strings[RPC_DISPLAY_PORT]);
1767
1768 xs_udp_finish_connecting(xprt, sock);
1769 status = 0;
1770 out:
1771 xprt_clear_connecting(xprt);
1772 xprt_wake_pending_tasks(xprt, status);
1773 }
1774
1775 /*
1776 * We need to preserve the port number so the reply cache on the server can
1777 * find our cached RPC replies when we get around to reconnecting.
1778 */
1779 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1780 {
1781 int result;
1782 struct sockaddr any;
1783
1784 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1785
1786 /*
1787 * Disconnect the transport socket by doing a connect operation
1788 * with AF_UNSPEC. This should return immediately...
1789 */
1790 memset(&any, 0, sizeof(any));
1791 any.sa_family = AF_UNSPEC;
1792 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1793 if (!result)
1794 xs_sock_mark_closed(xprt);
1795 else
1796 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1797 result);
1798 }
1799
1800 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1801 {
1802 unsigned int state = transport->inet->sk_state;
1803
1804 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1805 return;
1806 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1807 return;
1808 xs_abort_connection(xprt, transport);
1809 }
1810
1811 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1812 {
1813 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1814
1815 if (!transport->inet) {
1816 struct sock *sk = sock->sk;
1817
1818 write_lock_bh(&sk->sk_callback_lock);
1819
1820 xs_save_old_callbacks(transport, sk);
1821
1822 sk->sk_user_data = xprt;
1823 sk->sk_data_ready = xs_tcp_data_ready;
1824 sk->sk_state_change = xs_tcp_state_change;
1825 sk->sk_write_space = xs_tcp_write_space;
1826 sk->sk_error_report = xs_error_report;
1827 sk->sk_allocation = GFP_ATOMIC;
1828
1829 /* socket options */
1830 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1831 sock_reset_flag(sk, SOCK_LINGER);
1832 tcp_sk(sk)->linger2 = 0;
1833 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1834
1835 xprt_clear_connected(xprt);
1836
1837 /* Reset to new socket */
1838 transport->sock = sock;
1839 transport->inet = sk;
1840
1841 write_unlock_bh(&sk->sk_callback_lock);
1842 }
1843
1844 if (!xprt_bound(xprt))
1845 return -ENOTCONN;
1846
1847 /* Tell the socket layer to start connecting... */
1848 xprt->stat.connect_count++;
1849 xprt->stat.connect_start = jiffies;
1850 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1851 }
1852
1853 /**
1854 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1855 * @xprt: RPC transport to connect
1856 * @transport: socket transport to connect
1857 * @create_sock: function to create a socket of the correct type
1858 *
1859 * Invoked by a work queue tasklet.
1860 */
1861 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1862 struct sock_xprt *transport,
1863 struct socket *(*create_sock)(struct rpc_xprt *,
1864 struct sock_xprt *))
1865 {
1866 struct socket *sock = transport->sock;
1867 int status = -EIO;
1868
1869 if (xprt->shutdown)
1870 goto out;
1871
1872 if (!sock) {
1873 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1874 sock = create_sock(xprt, transport);
1875 if (IS_ERR(sock)) {
1876 status = PTR_ERR(sock);
1877 goto out;
1878 }
1879 } else {
1880 int abort_and_exit;
1881
1882 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1883 &xprt->state);
1884 /* "close" the socket, preserving the local port */
1885 xs_tcp_reuse_connection(xprt, transport);
1886
1887 if (abort_and_exit)
1888 goto out_eagain;
1889 }
1890
1891 dprintk("RPC: worker connecting xprt %p via %s to "
1892 "%s (port %s)\n", xprt,
1893 xprt->address_strings[RPC_DISPLAY_PROTO],
1894 xprt->address_strings[RPC_DISPLAY_ADDR],
1895 xprt->address_strings[RPC_DISPLAY_PORT]);
1896
1897 status = xs_tcp_finish_connecting(xprt, sock);
1898 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1899 xprt, -status, xprt_connected(xprt),
1900 sock->sk->sk_state);
1901 switch (status) {
1902 default:
1903 printk("%s: connect returned unhandled error %d\n",
1904 __func__, status);
1905 case -EADDRNOTAVAIL:
1906 /* We're probably in TIME_WAIT. Get rid of existing socket,
1907 * and retry
1908 */
1909 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1910 xprt_force_disconnect(xprt);
1911 break;
1912 case -ECONNREFUSED:
1913 case -ECONNRESET:
1914 case -ENETUNREACH:
1915 /* retry with existing socket, after a delay */
1916 case 0:
1917 case -EINPROGRESS:
1918 case -EALREADY:
1919 xprt_clear_connecting(xprt);
1920 return;
1921 case -EINVAL:
1922 /* Happens, for instance, if the user specified a link
1923 * local IPv6 address without a scope-id.
1924 */
1925 goto out;
1926 }
1927 out_eagain:
1928 status = -EAGAIN;
1929 out:
1930 xprt_clear_connecting(xprt);
1931 xprt_wake_pending_tasks(xprt, status);
1932 }
1933
1934 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1935 struct sock_xprt *transport)
1936 {
1937 struct socket *sock;
1938 int err;
1939
1940 /* start from scratch */
1941 err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1942 if (err < 0) {
1943 dprintk("RPC: can't create TCP transport socket (%d).\n",
1944 -err);
1945 goto out_err;
1946 }
1947 xs_reclassify_socket4(sock);
1948
1949 if (xs_bind4(transport, sock) < 0) {
1950 sock_release(sock);
1951 goto out_err;
1952 }
1953 return sock;
1954 out_err:
1955 return ERR_PTR(-EIO);
1956 }
1957
1958 /**
1959 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1960 * @work: RPC transport to connect
1961 *
1962 * Invoked by a work queue tasklet.
1963 */
1964 static void xs_tcp_connect_worker4(struct work_struct *work)
1965 {
1966 struct sock_xprt *transport =
1967 container_of(work, struct sock_xprt, connect_worker.work);
1968 struct rpc_xprt *xprt = &transport->xprt;
1969
1970 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
1971 }
1972
1973 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
1974 struct sock_xprt *transport)
1975 {
1976 struct socket *sock;
1977 int err;
1978
1979 /* start from scratch */
1980 err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
1981 if (err < 0) {
1982 dprintk("RPC: can't create TCP transport socket (%d).\n",
1983 -err);
1984 goto out_err;
1985 }
1986 xs_reclassify_socket6(sock);
1987
1988 if (xs_bind6(transport, sock) < 0) {
1989 sock_release(sock);
1990 goto out_err;
1991 }
1992 return sock;
1993 out_err:
1994 return ERR_PTR(-EIO);
1995 }
1996
1997 /**
1998 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1999 * @work: RPC transport to connect
2000 *
2001 * Invoked by a work queue tasklet.
2002 */
2003 static void xs_tcp_connect_worker6(struct work_struct *work)
2004 {
2005 struct sock_xprt *transport =
2006 container_of(work, struct sock_xprt, connect_worker.work);
2007 struct rpc_xprt *xprt = &transport->xprt;
2008
2009 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
2010 }
2011
2012 /**
2013 * xs_connect - connect a socket to a remote endpoint
2014 * @task: address of RPC task that manages state of connect request
2015 *
2016 * TCP: If the remote end dropped the connection, delay reconnecting.
2017 *
2018 * UDP socket connects are synchronous, but we use a work queue anyway
2019 * to guarantee that even unprivileged user processes can set up a
2020 * socket on a privileged port.
2021 *
2022 * If a UDP socket connect fails, the delay behavior here prevents
2023 * retry floods (hard mounts).
2024 */
2025 static void xs_connect(struct rpc_task *task)
2026 {
2027 struct rpc_xprt *xprt = task->tk_xprt;
2028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2029
2030 if (xprt_test_and_set_connecting(xprt))
2031 return;
2032
2033 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2034 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2035 "seconds\n",
2036 xprt, xprt->reestablish_timeout / HZ);
2037 queue_delayed_work(rpciod_workqueue,
2038 &transport->connect_worker,
2039 xprt->reestablish_timeout);
2040 xprt->reestablish_timeout <<= 1;
2041 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2042 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2043 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2044 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2045 } else {
2046 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2047 queue_delayed_work(rpciod_workqueue,
2048 &transport->connect_worker, 0);
2049 }
2050 }
2051
2052 static void xs_tcp_connect(struct rpc_task *task)
2053 {
2054 struct rpc_xprt *xprt = task->tk_xprt;
2055
2056 /* Exit if we need to wait for socket shutdown to complete */
2057 if (test_bit(XPRT_CLOSING, &xprt->state))
2058 return;
2059 xs_connect(task);
2060 }
2061
2062 /**
2063 * xs_udp_print_stats - display UDP socket-specifc stats
2064 * @xprt: rpc_xprt struct containing statistics
2065 * @seq: output file
2066 *
2067 */
2068 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2069 {
2070 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2071
2072 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2073 transport->srcport,
2074 xprt->stat.bind_count,
2075 xprt->stat.sends,
2076 xprt->stat.recvs,
2077 xprt->stat.bad_xids,
2078 xprt->stat.req_u,
2079 xprt->stat.bklog_u);
2080 }
2081
2082 /**
2083 * xs_tcp_print_stats - display TCP socket-specifc stats
2084 * @xprt: rpc_xprt struct containing statistics
2085 * @seq: output file
2086 *
2087 */
2088 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2089 {
2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2091 long idle_time = 0;
2092
2093 if (xprt_connected(xprt))
2094 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2095
2096 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2097 transport->srcport,
2098 xprt->stat.bind_count,
2099 xprt->stat.connect_count,
2100 xprt->stat.connect_time,
2101 idle_time,
2102 xprt->stat.sends,
2103 xprt->stat.recvs,
2104 xprt->stat.bad_xids,
2105 xprt->stat.req_u,
2106 xprt->stat.bklog_u);
2107 }
2108
2109 /*
2110 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2111 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2112 * to use the server side send routines.
2113 */
2114 void *bc_malloc(struct rpc_task *task, size_t size)
2115 {
2116 struct page *page;
2117 struct rpc_buffer *buf;
2118
2119 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2120 page = alloc_page(GFP_KERNEL);
2121
2122 if (!page)
2123 return NULL;
2124
2125 buf = page_address(page);
2126 buf->len = PAGE_SIZE;
2127
2128 return buf->data;
2129 }
2130
2131 /*
2132 * Free the space allocated in the bc_alloc routine
2133 */
2134 void bc_free(void *buffer)
2135 {
2136 struct rpc_buffer *buf;
2137
2138 if (!buffer)
2139 return;
2140
2141 buf = container_of(buffer, struct rpc_buffer, data);
2142 free_page((unsigned long)buf);
2143 }
2144
2145 /*
2146 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2147 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2148 */
2149 static int bc_sendto(struct rpc_rqst *req)
2150 {
2151 int len;
2152 struct xdr_buf *xbufp = &req->rq_snd_buf;
2153 struct rpc_xprt *xprt = req->rq_xprt;
2154 struct sock_xprt *transport =
2155 container_of(xprt, struct sock_xprt, xprt);
2156 struct socket *sock = transport->sock;
2157 unsigned long headoff;
2158 unsigned long tailoff;
2159
2160 /*
2161 * Set up the rpc header and record marker stuff
2162 */
2163 xs_encode_tcp_record_marker(xbufp);
2164
2165 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2166 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2167 len = svc_send_common(sock, xbufp,
2168 virt_to_page(xbufp->head[0].iov_base), headoff,
2169 xbufp->tail[0].iov_base, tailoff);
2170
2171 if (len != xbufp->len) {
2172 printk(KERN_NOTICE "Error sending entire callback!\n");
2173 len = -EAGAIN;
2174 }
2175
2176 return len;
2177 }
2178
2179 /*
2180 * The send routine. Borrows from svc_send
2181 */
2182 static int bc_send_request(struct rpc_task *task)
2183 {
2184 struct rpc_rqst *req = task->tk_rqstp;
2185 struct svc_xprt *xprt;
2186 struct svc_sock *svsk;
2187 u32 len;
2188
2189 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2190 /*
2191 * Get the server socket associated with this callback xprt
2192 */
2193 xprt = req->rq_xprt->bc_xprt;
2194 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2195
2196 /*
2197 * Grab the mutex to serialize data as the connection is shared
2198 * with the fore channel
2199 */
2200 if (!mutex_trylock(&xprt->xpt_mutex)) {
2201 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2202 if (!mutex_trylock(&xprt->xpt_mutex))
2203 return -EAGAIN;
2204 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2205 }
2206 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2207 len = -ENOTCONN;
2208 else
2209 len = bc_sendto(req);
2210 mutex_unlock(&xprt->xpt_mutex);
2211
2212 if (len > 0)
2213 len = 0;
2214
2215 return len;
2216 }
2217
2218 /*
2219 * The close routine. Since this is client initiated, we do nothing
2220 */
2221
2222 static void bc_close(struct rpc_xprt *xprt)
2223 {
2224 return;
2225 }
2226
2227 /*
2228 * The xprt destroy routine. Again, because this connection is client
2229 * initiated, we do nothing
2230 */
2231
2232 static void bc_destroy(struct rpc_xprt *xprt)
2233 {
2234 return;
2235 }
2236
2237 static struct rpc_xprt_ops xs_udp_ops = {
2238 .set_buffer_size = xs_udp_set_buffer_size,
2239 .reserve_xprt = xprt_reserve_xprt_cong,
2240 .release_xprt = xprt_release_xprt_cong,
2241 .rpcbind = rpcb_getport_async,
2242 .set_port = xs_set_port,
2243 .connect = xs_connect,
2244 .buf_alloc = rpc_malloc,
2245 .buf_free = rpc_free,
2246 .send_request = xs_udp_send_request,
2247 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2248 .timer = xs_udp_timer,
2249 .release_request = xprt_release_rqst_cong,
2250 .close = xs_close,
2251 .destroy = xs_destroy,
2252 .print_stats = xs_udp_print_stats,
2253 };
2254
2255 static struct rpc_xprt_ops xs_tcp_ops = {
2256 .reserve_xprt = xprt_reserve_xprt,
2257 .release_xprt = xs_tcp_release_xprt,
2258 .rpcbind = rpcb_getport_async,
2259 .set_port = xs_set_port,
2260 .connect = xs_tcp_connect,
2261 .buf_alloc = rpc_malloc,
2262 .buf_free = rpc_free,
2263 .send_request = xs_tcp_send_request,
2264 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2265 #if defined(CONFIG_NFS_V4_1)
2266 .release_request = bc_release_request,
2267 #endif /* CONFIG_NFS_V4_1 */
2268 .close = xs_tcp_close,
2269 .destroy = xs_destroy,
2270 .print_stats = xs_tcp_print_stats,
2271 };
2272
2273 /*
2274 * The rpc_xprt_ops for the server backchannel
2275 */
2276
2277 static struct rpc_xprt_ops bc_tcp_ops = {
2278 .reserve_xprt = xprt_reserve_xprt,
2279 .release_xprt = xprt_release_xprt,
2280 .buf_alloc = bc_malloc,
2281 .buf_free = bc_free,
2282 .send_request = bc_send_request,
2283 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2284 .close = bc_close,
2285 .destroy = bc_destroy,
2286 .print_stats = xs_tcp_print_stats,
2287 };
2288
2289 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2290 unsigned int slot_table_size)
2291 {
2292 struct rpc_xprt *xprt;
2293 struct sock_xprt *new;
2294
2295 if (args->addrlen > sizeof(xprt->addr)) {
2296 dprintk("RPC: xs_setup_xprt: address too large\n");
2297 return ERR_PTR(-EBADF);
2298 }
2299
2300 new = kzalloc(sizeof(*new), GFP_KERNEL);
2301 if (new == NULL) {
2302 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2303 "rpc_xprt\n");
2304 return ERR_PTR(-ENOMEM);
2305 }
2306 xprt = &new->xprt;
2307
2308 xprt->max_reqs = slot_table_size;
2309 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2310 if (xprt->slot == NULL) {
2311 kfree(xprt);
2312 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
2313 "table\n");
2314 return ERR_PTR(-ENOMEM);
2315 }
2316
2317 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2318 xprt->addrlen = args->addrlen;
2319 if (args->srcaddr)
2320 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2321
2322 return xprt;
2323 }
2324
2325 static const struct rpc_timeout xs_udp_default_timeout = {
2326 .to_initval = 5 * HZ,
2327 .to_maxval = 30 * HZ,
2328 .to_increment = 5 * HZ,
2329 .to_retries = 5,
2330 };
2331
2332 /**
2333 * xs_setup_udp - Set up transport to use a UDP socket
2334 * @args: rpc transport creation arguments
2335 *
2336 */
2337 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2338 {
2339 struct sockaddr *addr = args->dstaddr;
2340 struct rpc_xprt *xprt;
2341 struct sock_xprt *transport;
2342
2343 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2344 if (IS_ERR(xprt))
2345 return xprt;
2346 transport = container_of(xprt, struct sock_xprt, xprt);
2347
2348 xprt->prot = IPPROTO_UDP;
2349 xprt->tsh_size = 0;
2350 /* XXX: header size can vary due to auth type, IPv6, etc. */
2351 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2352
2353 xprt->bind_timeout = XS_BIND_TO;
2354 xprt->connect_timeout = XS_UDP_CONN_TO;
2355 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2356 xprt->idle_timeout = XS_IDLE_DISC_TO;
2357
2358 xprt->ops = &xs_udp_ops;
2359
2360 xprt->timeout = &xs_udp_default_timeout;
2361
2362 switch (addr->sa_family) {
2363 case AF_INET:
2364 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2365 xprt_set_bound(xprt);
2366
2367 INIT_DELAYED_WORK(&transport->connect_worker,
2368 xs_udp_connect_worker4);
2369 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2370 break;
2371 case AF_INET6:
2372 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2373 xprt_set_bound(xprt);
2374
2375 INIT_DELAYED_WORK(&transport->connect_worker,
2376 xs_udp_connect_worker6);
2377 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2378 break;
2379 default:
2380 kfree(xprt);
2381 return ERR_PTR(-EAFNOSUPPORT);
2382 }
2383
2384 if (xprt_bound(xprt))
2385 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2386 xprt->address_strings[RPC_DISPLAY_ADDR],
2387 xprt->address_strings[RPC_DISPLAY_PORT],
2388 xprt->address_strings[RPC_DISPLAY_PROTO]);
2389 else
2390 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2391 xprt->address_strings[RPC_DISPLAY_ADDR],
2392 xprt->address_strings[RPC_DISPLAY_PROTO]);
2393
2394 if (try_module_get(THIS_MODULE))
2395 return xprt;
2396
2397 kfree(xprt->slot);
2398 kfree(xprt);
2399 return ERR_PTR(-EINVAL);
2400 }
2401
2402 static const struct rpc_timeout xs_tcp_default_timeout = {
2403 .to_initval = 60 * HZ,
2404 .to_maxval = 60 * HZ,
2405 .to_retries = 2,
2406 };
2407
2408 /**
2409 * xs_setup_tcp - Set up transport to use a TCP socket
2410 * @args: rpc transport creation arguments
2411 *
2412 */
2413 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2414 {
2415 struct sockaddr *addr = args->dstaddr;
2416 struct rpc_xprt *xprt;
2417 struct sock_xprt *transport;
2418
2419 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2420 if (IS_ERR(xprt))
2421 return xprt;
2422 transport = container_of(xprt, struct sock_xprt, xprt);
2423
2424 xprt->prot = IPPROTO_TCP;
2425 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2426 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2427
2428 xprt->bind_timeout = XS_BIND_TO;
2429 xprt->connect_timeout = XS_TCP_CONN_TO;
2430 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2431 xprt->idle_timeout = XS_IDLE_DISC_TO;
2432
2433 xprt->ops = &xs_tcp_ops;
2434 xprt->timeout = &xs_tcp_default_timeout;
2435
2436 switch (addr->sa_family) {
2437 case AF_INET:
2438 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2439 xprt_set_bound(xprt);
2440
2441 INIT_DELAYED_WORK(&transport->connect_worker,
2442 xs_tcp_connect_worker4);
2443 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2444 break;
2445 case AF_INET6:
2446 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2447 xprt_set_bound(xprt);
2448
2449 INIT_DELAYED_WORK(&transport->connect_worker,
2450 xs_tcp_connect_worker6);
2451 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2452 break;
2453 default:
2454 kfree(xprt);
2455 return ERR_PTR(-EAFNOSUPPORT);
2456 }
2457
2458 if (xprt_bound(xprt))
2459 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2460 xprt->address_strings[RPC_DISPLAY_ADDR],
2461 xprt->address_strings[RPC_DISPLAY_PORT],
2462 xprt->address_strings[RPC_DISPLAY_PROTO]);
2463 else
2464 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2465 xprt->address_strings[RPC_DISPLAY_ADDR],
2466 xprt->address_strings[RPC_DISPLAY_PROTO]);
2467
2468
2469 if (try_module_get(THIS_MODULE))
2470 return xprt;
2471
2472 kfree(xprt->slot);
2473 kfree(xprt);
2474 return ERR_PTR(-EINVAL);
2475 }
2476
2477 /**
2478 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2479 * @args: rpc transport creation arguments
2480 *
2481 */
2482 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2483 {
2484 struct sockaddr *addr = args->dstaddr;
2485 struct rpc_xprt *xprt;
2486 struct sock_xprt *transport;
2487 struct svc_sock *bc_sock;
2488
2489 if (!args->bc_xprt)
2490 ERR_PTR(-EINVAL);
2491
2492 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2493 if (IS_ERR(xprt))
2494 return xprt;
2495 transport = container_of(xprt, struct sock_xprt, xprt);
2496
2497 xprt->prot = IPPROTO_TCP;
2498 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2499 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2500 xprt->timeout = &xs_tcp_default_timeout;
2501
2502 /* backchannel */
2503 xprt_set_bound(xprt);
2504 xprt->bind_timeout = 0;
2505 xprt->connect_timeout = 0;
2506 xprt->reestablish_timeout = 0;
2507 xprt->idle_timeout = 0;
2508
2509 /*
2510 * The backchannel uses the same socket connection as the
2511 * forechannel
2512 */
2513 xprt->bc_xprt = args->bc_xprt;
2514 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2515 bc_sock->sk_bc_xprt = xprt;
2516 transport->sock = bc_sock->sk_sock;
2517 transport->inet = bc_sock->sk_sk;
2518
2519 xprt->ops = &bc_tcp_ops;
2520
2521 switch (addr->sa_family) {
2522 case AF_INET:
2523 xs_format_peer_addresses(xprt, "tcp",
2524 RPCBIND_NETID_TCP);
2525 break;
2526 case AF_INET6:
2527 xs_format_peer_addresses(xprt, "tcp",
2528 RPCBIND_NETID_TCP6);
2529 break;
2530 default:
2531 kfree(xprt);
2532 return ERR_PTR(-EAFNOSUPPORT);
2533 }
2534
2535 if (xprt_bound(xprt))
2536 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2537 xprt->address_strings[RPC_DISPLAY_ADDR],
2538 xprt->address_strings[RPC_DISPLAY_PORT],
2539 xprt->address_strings[RPC_DISPLAY_PROTO]);
2540 else
2541 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2542 xprt->address_strings[RPC_DISPLAY_ADDR],
2543 xprt->address_strings[RPC_DISPLAY_PROTO]);
2544
2545 /*
2546 * Since we don't want connections for the backchannel, we set
2547 * the xprt status to connected
2548 */
2549 xprt_set_connected(xprt);
2550
2551
2552 if (try_module_get(THIS_MODULE))
2553 return xprt;
2554 kfree(xprt->slot);
2555 kfree(xprt);
2556 return ERR_PTR(-EINVAL);
2557 }
2558
2559 static struct xprt_class xs_udp_transport = {
2560 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2561 .name = "udp",
2562 .owner = THIS_MODULE,
2563 .ident = XPRT_TRANSPORT_UDP,
2564 .setup = xs_setup_udp,
2565 };
2566
2567 static struct xprt_class xs_tcp_transport = {
2568 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2569 .name = "tcp",
2570 .owner = THIS_MODULE,
2571 .ident = XPRT_TRANSPORT_TCP,
2572 .setup = xs_setup_tcp,
2573 };
2574
2575 static struct xprt_class xs_bc_tcp_transport = {
2576 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2577 .name = "tcp NFSv4.1 backchannel",
2578 .owner = THIS_MODULE,
2579 .ident = XPRT_TRANSPORT_BC_TCP,
2580 .setup = xs_setup_bc_tcp,
2581 };
2582
2583 /**
2584 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2585 *
2586 */
2587 int init_socket_xprt(void)
2588 {
2589 #ifdef RPC_DEBUG
2590 if (!sunrpc_table_header)
2591 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2592 #endif
2593
2594 xprt_register_transport(&xs_udp_transport);
2595 xprt_register_transport(&xs_tcp_transport);
2596 xprt_register_transport(&xs_bc_tcp_transport);
2597
2598 return 0;
2599 }
2600
2601 /**
2602 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2603 *
2604 */
2605 void cleanup_socket_xprt(void)
2606 {
2607 #ifdef RPC_DEBUG
2608 if (sunrpc_table_header) {
2609 unregister_sysctl_table(sunrpc_table_header);
2610 sunrpc_table_header = NULL;
2611 }
2612 #endif
2613
2614 xprt_unregister_transport(&xs_udp_transport);
2615 xprt_unregister_transport(&xs_tcp_transport);
2616 xprt_unregister_transport(&xs_bc_tcp_transport);
2617 }
2618
2619 static int param_set_uint_minmax(const char *val, struct kernel_param *kp,
2620 unsigned int min, unsigned int max)
2621 {
2622 unsigned long num;
2623 int ret;
2624
2625 if (!val)
2626 return -EINVAL;
2627 ret = strict_strtoul(val, 0, &num);
2628 if (ret == -EINVAL || num < min || num > max)
2629 return -EINVAL;
2630 *((unsigned int *)kp->arg) = num;
2631 return 0;
2632 }
2633
2634 static int param_set_portnr(const char *val, struct kernel_param *kp)
2635 {
2636 return param_set_uint_minmax(val, kp,
2637 RPC_MIN_RESVPORT,
2638 RPC_MAX_RESVPORT);
2639 }
2640
2641 static int param_get_portnr(char *buffer, struct kernel_param *kp)
2642 {
2643 return param_get_uint(buffer, kp);
2644 }
2645 #define param_check_portnr(name, p) \
2646 __param_check(name, p, unsigned int);
2647
2648 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2649 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2650
2651 static int param_set_slot_table_size(const char *val, struct kernel_param *kp)
2652 {
2653 return param_set_uint_minmax(val, kp,
2654 RPC_MIN_SLOT_TABLE,
2655 RPC_MAX_SLOT_TABLE);
2656 }
2657
2658 static int param_get_slot_table_size(char *buffer, struct kernel_param *kp)
2659 {
2660 return param_get_uint(buffer, kp);
2661 }
2662 #define param_check_slot_table_size(name, p) \
2663 __param_check(name, p, unsigned int);
2664
2665 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2666 slot_table_size, 0644);
2667 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2668 slot_table_size, 0644);
2669
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree