blob: c458f8d1d6d1090e193c1c1cd66a4c7552c92986 [file] [log] [blame]
Chuck Levera246b012005-08-11 16:25:23 -04001/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
Chuck Lever55aa4f52005-08-11 16:25:47 -040014 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
Chuck Levera246b012005-08-11 16:25:23 -040016 */
17
18#include <linux/types.h>
19#include <linux/slab.h>
20#include <linux/capability.h>
21#include <linux/sched.h>
22#include <linux/pagemap.h>
23#include <linux/errno.h>
24#include <linux/socket.h>
25#include <linux/in.h>
26#include <linux/net.h>
27#include <linux/mm.h>
28#include <linux/udp.h>
29#include <linux/tcp.h>
30#include <linux/sunrpc/clnt.h>
Chuck Lever02107142006-01-03 09:55:49 +010031#include <linux/sunrpc/sched.h>
Chuck Levera246b012005-08-11 16:25:23 -040032#include <linux/file.h>
33
34#include <net/sock.h>
35#include <net/checksum.h>
36#include <net/udp.h>
37#include <net/tcp.h>
38
Chuck Lever9903cd12005-08-11 16:25:26 -040039/*
Chuck Leverc556b752005-11-01 12:24:48 -050040 * xprtsock tunables
41 */
42unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
43unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
44
45unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
46unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
47
48/*
Chuck Lever262965f2005-08-11 16:25:56 -040049 * How many times to try sending a request on a socket before waiting
50 * for the socket buffer to clear.
51 */
52#define XS_SENDMSG_RETRY (10U)
53
Chuck Lever03bf4b72005-08-25 16:25:55 -070054/*
55 * Time out for an RPC UDP socket connect. UDP socket connects are
56 * synchronous, but we set a timeout anyway in case of resource
57 * exhaustion on the local host.
58 */
59#define XS_UDP_CONN_TO (5U * HZ)
60
61/*
62 * Wait duration for an RPC TCP connection to be established. Solaris
63 * NFS over TCP uses 60 seconds, for example, which is in line with how
64 * long a server takes to reboot.
65 */
66#define XS_TCP_CONN_TO (60U * HZ)
67
68/*
69 * Wait duration for a reply from the RPC portmapper.
70 */
71#define XS_BIND_TO (60U * HZ)
72
73/*
74 * Delay if a UDP socket connect error occurs. This is most likely some
75 * kind of resource problem on the local host.
76 */
77#define XS_UDP_REEST_TO (2U * HZ)
78
79/*
80 * The reestablish timeout allows clients to delay for a bit before attempting
81 * to reconnect to a server that just dropped our connection.
82 *
83 * We implement an exponential backoff when trying to reestablish a TCP
84 * transport connection with the server. Some servers like to drop a TCP
85 * connection when they are overworked, so we start with a short timeout and
86 * increase over time if the server is down or not responding.
87 */
88#define XS_TCP_INIT_REEST_TO (3U * HZ)
89#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
90
91/*
92 * TCP idle timeout; client drops the transport socket if it is idle
93 * for this long. Note that we also timeout UDP sockets to prevent
94 * holding port numbers when there is no RPC traffic.
95 */
96#define XS_IDLE_DISC_TO (5U * 60 * HZ)
97
Chuck Levera246b012005-08-11 16:25:23 -040098#ifdef RPC_DEBUG
99# undef RPC_DEBUG_DATA
Chuck Lever9903cd12005-08-11 16:25:26 -0400100# define RPCDBG_FACILITY RPCDBG_TRANS
Chuck Levera246b012005-08-11 16:25:23 -0400101#endif
102
Chuck Levera246b012005-08-11 16:25:23 -0400103#ifdef RPC_DEBUG_DATA
Chuck Lever9903cd12005-08-11 16:25:26 -0400104static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
Chuck Levera246b012005-08-11 16:25:23 -0400105{
Chuck Lever9903cd12005-08-11 16:25:26 -0400106 u8 *buf = (u8 *) packet;
107 int j;
Chuck Levera246b012005-08-11 16:25:23 -0400108
109 dprintk("RPC: %s\n", msg);
110 for (j = 0; j < count && j < 128; j += 4) {
111 if (!(j & 31)) {
112 if (j)
113 dprintk("\n");
114 dprintk("0x%04x ", j);
115 }
116 dprintk("%02x%02x%02x%02x ",
117 buf[j], buf[j+1], buf[j+2], buf[j+3]);
118 }
119 dprintk("\n");
120}
121#else
Chuck Lever9903cd12005-08-11 16:25:26 -0400122static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
Chuck Levera246b012005-08-11 16:25:23 -0400123{
124 /* NOP */
125}
126#endif
127
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400128#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
129
130static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
131{
132 struct kvec iov = {
133 .iov_base = xdr->head[0].iov_base + base,
134 .iov_len = len - base,
135 };
136 struct msghdr msg = {
137 .msg_name = addr,
138 .msg_namelen = addrlen,
139 .msg_flags = XS_SENDMSG_FLAGS,
140 };
141
142 if (xdr->len > len)
143 msg.msg_flags |= MSG_MORE;
144
145 if (likely(iov.iov_len))
146 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
147 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
148}
149
150static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
151{
152 struct kvec iov = {
153 .iov_base = xdr->tail[0].iov_base + base,
154 .iov_len = len - base,
155 };
156 struct msghdr msg = {
157 .msg_flags = XS_SENDMSG_FLAGS,
158 };
159
160 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
161}
162
Chuck Lever9903cd12005-08-11 16:25:26 -0400163/**
164 * xs_sendpages - write pages directly to a socket
165 * @sock: socket to send on
166 * @addr: UDP only -- address of destination
167 * @addrlen: UDP only -- length of destination address
168 * @xdr: buffer containing this request
169 * @base: starting position in the buffer
170 *
Chuck Levera246b012005-08-11 16:25:23 -0400171 */
Chuck Lever262965f2005-08-11 16:25:56 -0400172static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
Chuck Levera246b012005-08-11 16:25:23 -0400173{
174 struct page **ppage = xdr->pages;
175 unsigned int len, pglen = xdr->page_len;
176 int err, ret = 0;
177 ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
178
Chuck Lever262965f2005-08-11 16:25:56 -0400179 if (unlikely(!sock))
180 return -ENOTCONN;
181
182 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
183
Chuck Levera246b012005-08-11 16:25:23 -0400184 len = xdr->head[0].iov_len;
185 if (base < len || (addr != NULL && base == 0)) {
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400186 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
Chuck Levera246b012005-08-11 16:25:23 -0400187 if (ret == 0)
188 ret = err;
189 else if (err > 0)
190 ret += err;
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400191 if (err != (len - base))
Chuck Levera246b012005-08-11 16:25:23 -0400192 goto out;
193 base = 0;
194 } else
195 base -= len;
196
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400197 if (unlikely(pglen == 0))
Chuck Levera246b012005-08-11 16:25:23 -0400198 goto copy_tail;
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400199 if (unlikely(base >= pglen)) {
Chuck Levera246b012005-08-11 16:25:23 -0400200 base -= pglen;
201 goto copy_tail;
202 }
203 if (base || xdr->page_base) {
204 pglen -= base;
Chuck Lever9903cd12005-08-11 16:25:26 -0400205 base += xdr->page_base;
Chuck Levera246b012005-08-11 16:25:23 -0400206 ppage += base >> PAGE_CACHE_SHIFT;
207 base &= ~PAGE_CACHE_MASK;
208 }
209
210 sendpage = sock->ops->sendpage ? : sock_no_sendpage;
211 do {
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400212 int flags = XS_SENDMSG_FLAGS;
Chuck Levera246b012005-08-11 16:25:23 -0400213
214 len = PAGE_CACHE_SIZE;
215 if (base)
216 len -= base;
217 if (pglen < len)
218 len = pglen;
219
220 if (pglen != len || xdr->tail[0].iov_len != 0)
221 flags |= MSG_MORE;
222
223 /* Hmm... We might be dealing with highmem pages */
224 if (PageHighMem(*ppage))
225 sendpage = sock_no_sendpage;
226 err = sendpage(sock, *ppage, base, len, flags);
227 if (ret == 0)
228 ret = err;
229 else if (err > 0)
230 ret += err;
231 if (err != len)
232 goto out;
233 base = 0;
234 ppage++;
235 } while ((pglen -= len) != 0);
236copy_tail:
237 len = xdr->tail[0].iov_len;
238 if (base < len) {
Chuck Leverb4b5cc82005-08-11 16:25:29 -0400239 err = xs_send_tail(sock, xdr, base, len);
Chuck Levera246b012005-08-11 16:25:23 -0400240 if (ret == 0)
241 ret = err;
242 else if (err > 0)
243 ret += err;
244 }
245out:
246 return ret;
247}
248
Chuck Lever9903cd12005-08-11 16:25:26 -0400249/**
Chuck Lever262965f2005-08-11 16:25:56 -0400250 * xs_nospace - place task on wait queue if transmit was incomplete
251 * @task: task to put to sleep
Chuck Lever9903cd12005-08-11 16:25:26 -0400252 *
Chuck Levera246b012005-08-11 16:25:23 -0400253 */
Chuck Lever262965f2005-08-11 16:25:56 -0400254static void xs_nospace(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -0400255{
Chuck Lever262965f2005-08-11 16:25:56 -0400256 struct rpc_rqst *req = task->tk_rqstp;
257 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Levera246b012005-08-11 16:25:23 -0400258
Chuck Lever262965f2005-08-11 16:25:56 -0400259 dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
260 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
261 req->rq_slen);
262
263 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
264 /* Protect against races with write_space */
265 spin_lock_bh(&xprt->transport_lock);
266
267 /* Don't race with disconnect */
268 if (!xprt_connected(xprt))
269 task->tk_status = -ENOTCONN;
270 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
271 xprt_wait_for_buffer_space(task);
272
273 spin_unlock_bh(&xprt->transport_lock);
274 } else
275 /* Keep holding the socket if it is blocked */
276 rpc_delay(task, HZ>>4);
277}
278
279/**
280 * xs_udp_send_request - write an RPC request to a UDP socket
281 * @task: address of RPC task that manages the state of an RPC request
282 *
283 * Return values:
284 * 0: The request has been sent
285 * EAGAIN: The socket was blocked, please call again later to
286 * complete the request
287 * ENOTCONN: Caller needs to invoke connect logic then call again
288 * other: Some other error occured, the request was not sent
289 */
290static int xs_udp_send_request(struct rpc_task *task)
291{
292 struct rpc_rqst *req = task->tk_rqstp;
293 struct rpc_xprt *xprt = req->rq_xprt;
294 struct xdr_buf *xdr = &req->rq_snd_buf;
295 int status;
Chuck Levera246b012005-08-11 16:25:23 -0400296
Chuck Lever9903cd12005-08-11 16:25:26 -0400297 xs_pktdump("packet data:",
Chuck Levera246b012005-08-11 16:25:23 -0400298 req->rq_svec->iov_base,
299 req->rq_svec->iov_len);
300
Chuck Lever262965f2005-08-11 16:25:56 -0400301 req->rq_xtime = jiffies;
302 status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
303 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
Chuck Levera246b012005-08-11 16:25:23 -0400304
Chuck Lever262965f2005-08-11 16:25:56 -0400305 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
306 xdr->len - req->rq_bytes_sent, status);
Chuck Levera246b012005-08-11 16:25:23 -0400307
Chuck Lever262965f2005-08-11 16:25:56 -0400308 if (likely(status >= (int) req->rq_slen))
309 return 0;
Chuck Levera246b012005-08-11 16:25:23 -0400310
Chuck Lever262965f2005-08-11 16:25:56 -0400311 /* Still some bytes left; set up for a retry later. */
312 if (status > 0)
313 status = -EAGAIN;
Chuck Levera246b012005-08-11 16:25:23 -0400314
Chuck Lever262965f2005-08-11 16:25:56 -0400315 switch (status) {
316 case -ENETUNREACH:
317 case -EPIPE:
Chuck Levera246b012005-08-11 16:25:23 -0400318 case -ECONNREFUSED:
319 /* When the server has died, an ICMP port unreachable message
Chuck Lever9903cd12005-08-11 16:25:26 -0400320 * prompts ECONNREFUSED. */
Chuck Levera246b012005-08-11 16:25:23 -0400321 break;
Chuck Lever262965f2005-08-11 16:25:56 -0400322 case -EAGAIN:
323 xs_nospace(task);
Chuck Levera246b012005-08-11 16:25:23 -0400324 break;
325 default:
Chuck Lever262965f2005-08-11 16:25:56 -0400326 dprintk("RPC: sendmsg returned unrecognized error %d\n",
327 -status);
Chuck Lever9903cd12005-08-11 16:25:26 -0400328 break;
Chuck Levera246b012005-08-11 16:25:23 -0400329 }
Chuck Lever262965f2005-08-11 16:25:56 -0400330
331 return status;
Chuck Levera246b012005-08-11 16:25:23 -0400332}
333
Chuck Lever808012f2005-08-25 16:25:49 -0700334static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
335{
336 u32 reclen = buf->len - sizeof(rpc_fraghdr);
337 rpc_fraghdr *base = buf->head[0].iov_base;
338 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
339}
340
Chuck Lever9903cd12005-08-11 16:25:26 -0400341/**
Chuck Lever262965f2005-08-11 16:25:56 -0400342 * xs_tcp_send_request - write an RPC request to a TCP socket
Chuck Lever9903cd12005-08-11 16:25:26 -0400343 * @task: address of RPC task that manages the state of an RPC request
344 *
345 * Return values:
Chuck Lever262965f2005-08-11 16:25:56 -0400346 * 0: The request has been sent
347 * EAGAIN: The socket was blocked, please call again later to
348 * complete the request
349 * ENOTCONN: Caller needs to invoke connect logic then call again
350 * other: Some other error occured, the request was not sent
Chuck Lever9903cd12005-08-11 16:25:26 -0400351 *
352 * XXX: In the case of soft timeouts, should we eventually give up
Chuck Lever262965f2005-08-11 16:25:56 -0400353 * if sendmsg is not able to make progress?
Chuck Lever9903cd12005-08-11 16:25:26 -0400354 */
Chuck Lever262965f2005-08-11 16:25:56 -0400355static int xs_tcp_send_request(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -0400356{
357 struct rpc_rqst *req = task->tk_rqstp;
358 struct rpc_xprt *xprt = req->rq_xprt;
Chuck Lever262965f2005-08-11 16:25:56 -0400359 struct xdr_buf *xdr = &req->rq_snd_buf;
Chuck Levera246b012005-08-11 16:25:23 -0400360 int status, retry = 0;
361
Chuck Lever808012f2005-08-25 16:25:49 -0700362 xs_encode_tcp_record_marker(&req->rq_snd_buf);
Chuck Levera246b012005-08-11 16:25:23 -0400363
Chuck Lever262965f2005-08-11 16:25:56 -0400364 xs_pktdump("packet data:",
365 req->rq_svec->iov_base,
366 req->rq_svec->iov_len);
Chuck Levera246b012005-08-11 16:25:23 -0400367
368 /* Continue transmitting the packet/record. We must be careful
369 * to cope with writespace callbacks arriving _after_ we have
Chuck Lever262965f2005-08-11 16:25:56 -0400370 * called sendmsg(). */
Chuck Levera246b012005-08-11 16:25:23 -0400371 while (1) {
372 req->rq_xtime = jiffies;
Chuck Lever262965f2005-08-11 16:25:56 -0400373 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
374 req->rq_bytes_sent);
Chuck Levera246b012005-08-11 16:25:23 -0400375
Chuck Lever262965f2005-08-11 16:25:56 -0400376 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
377 xdr->len - req->rq_bytes_sent, status);
378
379 if (unlikely(status < 0))
Chuck Levera246b012005-08-11 16:25:23 -0400380 break;
381
Chuck Lever262965f2005-08-11 16:25:56 -0400382 /* If we've sent the entire packet, immediately
383 * reset the count of bytes sent. */
384 req->rq_bytes_sent += status;
385 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
386 req->rq_bytes_sent = 0;
387 return 0;
Chuck Levera246b012005-08-11 16:25:23 -0400388 }
389
Chuck Levera246b012005-08-11 16:25:23 -0400390 status = -EAGAIN;
Chuck Lever262965f2005-08-11 16:25:56 -0400391 if (retry++ > XS_SENDMSG_RETRY)
Chuck Levera246b012005-08-11 16:25:23 -0400392 break;
393 }
394
Chuck Lever262965f2005-08-11 16:25:56 -0400395 switch (status) {
396 case -EAGAIN:
397 xs_nospace(task);
398 break;
399 case -ECONNREFUSED:
400 case -ECONNRESET:
401 case -ENOTCONN:
402 case -EPIPE:
403 status = -ENOTCONN;
404 break;
405 default:
406 dprintk("RPC: sendmsg returned unrecognized error %d\n",
407 -status);
Chuck Lever43118c22005-08-25 16:25:49 -0700408 xprt_disconnect(xprt);
Chuck Lever262965f2005-08-11 16:25:56 -0400409 break;
Chuck Levera246b012005-08-11 16:25:23 -0400410 }
Chuck Lever262965f2005-08-11 16:25:56 -0400411
Chuck Levera246b012005-08-11 16:25:23 -0400412 return status;
413}
414
Chuck Lever9903cd12005-08-11 16:25:26 -0400415/**
416 * xs_close - close a socket
417 * @xprt: transport
418 *
Chuck Lever3167e122005-08-25 16:25:55 -0700419 * This is used when all requests are complete; ie, no DRC state remains
420 * on the server we want to save.
Chuck Levera246b012005-08-11 16:25:23 -0400421 */
Chuck Lever9903cd12005-08-11 16:25:26 -0400422static void xs_close(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -0400423{
Chuck Lever9903cd12005-08-11 16:25:26 -0400424 struct socket *sock = xprt->sock;
425 struct sock *sk = xprt->inet;
Chuck Levera246b012005-08-11 16:25:23 -0400426
427 if (!sk)
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100428 goto clear_close_wait;
Chuck Levera246b012005-08-11 16:25:23 -0400429
Chuck Lever9903cd12005-08-11 16:25:26 -0400430 dprintk("RPC: xs_close xprt %p\n", xprt);
431
Chuck Levera246b012005-08-11 16:25:23 -0400432 write_lock_bh(&sk->sk_callback_lock);
433 xprt->inet = NULL;
434 xprt->sock = NULL;
435
Chuck Lever9903cd12005-08-11 16:25:26 -0400436 sk->sk_user_data = NULL;
437 sk->sk_data_ready = xprt->old_data_ready;
Chuck Levera246b012005-08-11 16:25:23 -0400438 sk->sk_state_change = xprt->old_state_change;
Chuck Lever9903cd12005-08-11 16:25:26 -0400439 sk->sk_write_space = xprt->old_write_space;
Chuck Levera246b012005-08-11 16:25:23 -0400440 write_unlock_bh(&sk->sk_callback_lock);
441
Chuck Lever9903cd12005-08-11 16:25:26 -0400442 sk->sk_no_check = 0;
Chuck Levera246b012005-08-11 16:25:23 -0400443
444 sock_release(sock);
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100445clear_close_wait:
446 smp_mb__before_clear_bit();
447 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
448 smp_mb__after_clear_bit();
Chuck Levera246b012005-08-11 16:25:23 -0400449}
450
Chuck Lever9903cd12005-08-11 16:25:26 -0400451/**
452 * xs_destroy - prepare to shutdown a transport
453 * @xprt: doomed transport
454 *
455 */
456static void xs_destroy(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -0400457{
Chuck Lever9903cd12005-08-11 16:25:26 -0400458 dprintk("RPC: xs_destroy xprt %p\n", xprt);
459
Chuck Lever55aa4f52005-08-11 16:25:47 -0400460 cancel_delayed_work(&xprt->connect_worker);
Chuck Levera246b012005-08-11 16:25:23 -0400461 flush_scheduled_work();
462
463 xprt_disconnect(xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400464 xs_close(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400465 kfree(xprt->slot);
466}
467
Chuck Lever9903cd12005-08-11 16:25:26 -0400468static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -0400469{
Chuck Lever9903cd12005-08-11 16:25:26 -0400470 return (struct rpc_xprt *) sk->sk_user_data;
471}
472
473/**
474 * xs_udp_data_ready - "data ready" callback for UDP sockets
475 * @sk: socket with data to read
476 * @len: how much data to read
477 *
478 */
479static void xs_udp_data_ready(struct sock *sk, int len)
480{
481 struct rpc_task *task;
482 struct rpc_xprt *xprt;
Chuck Levera246b012005-08-11 16:25:23 -0400483 struct rpc_rqst *rovr;
Chuck Lever9903cd12005-08-11 16:25:26 -0400484 struct sk_buff *skb;
Chuck Levera246b012005-08-11 16:25:23 -0400485 int err, repsize, copied;
486 u32 _xid, *xp;
487
488 read_lock(&sk->sk_callback_lock);
Chuck Lever9903cd12005-08-11 16:25:26 -0400489 dprintk("RPC: xs_udp_data_ready...\n");
490 if (!(xprt = xprt_from_sock(sk)))
Chuck Levera246b012005-08-11 16:25:23 -0400491 goto out;
Chuck Levera246b012005-08-11 16:25:23 -0400492
493 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
494 goto out;
495
496 if (xprt->shutdown)
497 goto dropit;
498
499 repsize = skb->len - sizeof(struct udphdr);
500 if (repsize < 4) {
Chuck Lever9903cd12005-08-11 16:25:26 -0400501 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
Chuck Levera246b012005-08-11 16:25:23 -0400502 goto dropit;
503 }
504
505 /* Copy the XID from the skb... */
506 xp = skb_header_pointer(skb, sizeof(struct udphdr),
507 sizeof(_xid), &_xid);
508 if (xp == NULL)
509 goto dropit;
510
511 /* Look up and lock the request corresponding to the given XID */
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400512 spin_lock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400513 rovr = xprt_lookup_rqst(xprt, *xp);
514 if (!rovr)
515 goto out_unlock;
516 task = rovr->rq_task;
517
Chuck Levera246b012005-08-11 16:25:23 -0400518 if ((copied = rovr->rq_private_buf.buflen) > repsize)
519 copied = repsize;
520
521 /* Suck it into the iovec, verify checksum if not done by hw. */
522 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
523 goto out_unlock;
524
525 /* Something worked... */
526 dst_confirm(skb->dst);
527
Chuck Lever1570c1e2005-08-25 16:25:52 -0700528 xprt_adjust_cwnd(task, copied);
529 xprt_update_rtt(task);
530 xprt_complete_rqst(task, copied);
Chuck Levera246b012005-08-11 16:25:23 -0400531
532 out_unlock:
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400533 spin_unlock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400534 dropit:
535 skb_free_datagram(sk, skb);
536 out:
537 read_unlock(&sk->sk_callback_lock);
538}
539
Chuck Lever9903cd12005-08-11 16:25:26 -0400540static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
Chuck Levera246b012005-08-11 16:25:23 -0400541{
542 if (len > desc->count)
543 len = desc->count;
544 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
545 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
546 len, desc->count);
547 return 0;
548 }
549 desc->offset += len;
550 desc->count -= len;
551 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
552 len, desc->count);
553 return len;
554}
555
Chuck Lever9903cd12005-08-11 16:25:26 -0400556static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400557{
558 size_t len, used;
559 char *p;
560
561 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
562 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
Chuck Lever9903cd12005-08-11 16:25:26 -0400563 used = xs_tcp_copy_data(desc, p, len);
Chuck Levera246b012005-08-11 16:25:23 -0400564 xprt->tcp_offset += used;
565 if (used != len)
566 return;
Chuck Lever808012f2005-08-25 16:25:49 -0700567
Chuck Levera246b012005-08-11 16:25:23 -0400568 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
Chuck Lever808012f2005-08-25 16:25:49 -0700569 if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
Chuck Levera246b012005-08-11 16:25:23 -0400570 xprt->tcp_flags |= XPRT_LAST_FRAG;
571 else
572 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
Chuck Lever808012f2005-08-25 16:25:49 -0700573 xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
574
Chuck Levera246b012005-08-11 16:25:23 -0400575 xprt->tcp_flags &= ~XPRT_COPY_RECM;
576 xprt->tcp_offset = 0;
Chuck Lever808012f2005-08-25 16:25:49 -0700577
Chuck Levera246b012005-08-11 16:25:23 -0400578 /* Sanity check of the record length */
Chuck Lever808012f2005-08-25 16:25:49 -0700579 if (unlikely(xprt->tcp_reclen < 4)) {
Chuck Lever9903cd12005-08-11 16:25:26 -0400580 dprintk("RPC: invalid TCP record fragment length\n");
Chuck Levera246b012005-08-11 16:25:23 -0400581 xprt_disconnect(xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400582 return;
Chuck Levera246b012005-08-11 16:25:23 -0400583 }
584 dprintk("RPC: reading TCP record fragment of length %d\n",
585 xprt->tcp_reclen);
586}
587
Chuck Lever9903cd12005-08-11 16:25:26 -0400588static void xs_tcp_check_recm(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -0400589{
590 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
591 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
592 if (xprt->tcp_offset == xprt->tcp_reclen) {
593 xprt->tcp_flags |= XPRT_COPY_RECM;
594 xprt->tcp_offset = 0;
595 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
596 xprt->tcp_flags &= ~XPRT_COPY_DATA;
597 xprt->tcp_flags |= XPRT_COPY_XID;
598 xprt->tcp_copied = 0;
599 }
600 }
601}
602
Chuck Lever9903cd12005-08-11 16:25:26 -0400603static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400604{
605 size_t len, used;
606 char *p;
607
608 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
609 dprintk("RPC: reading XID (%Zu bytes)\n", len);
610 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
Chuck Lever9903cd12005-08-11 16:25:26 -0400611 used = xs_tcp_copy_data(desc, p, len);
Chuck Levera246b012005-08-11 16:25:23 -0400612 xprt->tcp_offset += used;
613 if (used != len)
614 return;
615 xprt->tcp_flags &= ~XPRT_COPY_XID;
616 xprt->tcp_flags |= XPRT_COPY_DATA;
617 xprt->tcp_copied = 4;
618 dprintk("RPC: reading reply for XID %08x\n",
619 ntohl(xprt->tcp_xid));
Chuck Lever9903cd12005-08-11 16:25:26 -0400620 xs_tcp_check_recm(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400621}
622
Chuck Lever9903cd12005-08-11 16:25:26 -0400623static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400624{
625 struct rpc_rqst *req;
626 struct xdr_buf *rcvbuf;
627 size_t len;
628 ssize_t r;
629
630 /* Find and lock the request corresponding to this xid */
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400631 spin_lock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400632 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
633 if (!req) {
634 xprt->tcp_flags &= ~XPRT_COPY_DATA;
635 dprintk("RPC: XID %08x request not found!\n",
636 ntohl(xprt->tcp_xid));
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400637 spin_unlock(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400638 return;
639 }
640
641 rcvbuf = &req->rq_private_buf;
642 len = desc->count;
643 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
644 skb_reader_t my_desc;
645
646 len = xprt->tcp_reclen - xprt->tcp_offset;
647 memcpy(&my_desc, desc, sizeof(my_desc));
648 my_desc.count = len;
649 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
Chuck Lever9903cd12005-08-11 16:25:26 -0400650 &my_desc, xs_tcp_copy_data);
Chuck Levera246b012005-08-11 16:25:23 -0400651 desc->count -= r;
652 desc->offset += r;
653 } else
654 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
Chuck Lever9903cd12005-08-11 16:25:26 -0400655 desc, xs_tcp_copy_data);
Chuck Levera246b012005-08-11 16:25:23 -0400656
657 if (r > 0) {
658 xprt->tcp_copied += r;
659 xprt->tcp_offset += r;
660 }
661 if (r != len) {
662 /* Error when copying to the receive buffer,
663 * usually because we weren't able to allocate
664 * additional buffer pages. All we can do now
665 * is turn off XPRT_COPY_DATA, so the request
666 * will not receive any additional updates,
667 * and time out.
668 * Any remaining data from this record will
669 * be discarded.
670 */
671 xprt->tcp_flags &= ~XPRT_COPY_DATA;
672 dprintk("RPC: XID %08x truncated request\n",
673 ntohl(xprt->tcp_xid));
674 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
675 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
676 goto out;
677 }
678
679 dprintk("RPC: XID %08x read %Zd bytes\n",
680 ntohl(xprt->tcp_xid), r);
681 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
682 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
683
684 if (xprt->tcp_copied == req->rq_private_buf.buflen)
685 xprt->tcp_flags &= ~XPRT_COPY_DATA;
686 else if (xprt->tcp_offset == xprt->tcp_reclen) {
687 if (xprt->tcp_flags & XPRT_LAST_FRAG)
688 xprt->tcp_flags &= ~XPRT_COPY_DATA;
689 }
690
691out:
Chuck Lever1570c1e2005-08-25 16:25:52 -0700692 if (!(xprt->tcp_flags & XPRT_COPY_DATA))
693 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400694 spin_unlock(&xprt->transport_lock);
Chuck Lever9903cd12005-08-11 16:25:26 -0400695 xs_tcp_check_recm(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400696}
697
Chuck Lever9903cd12005-08-11 16:25:26 -0400698static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
Chuck Levera246b012005-08-11 16:25:23 -0400699{
700 size_t len;
701
702 len = xprt->tcp_reclen - xprt->tcp_offset;
703 if (len > desc->count)
704 len = desc->count;
705 desc->count -= len;
706 desc->offset += len;
707 xprt->tcp_offset += len;
708 dprintk("RPC: discarded %Zu bytes\n", len);
Chuck Lever9903cd12005-08-11 16:25:26 -0400709 xs_tcp_check_recm(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400710}
711
Chuck Lever9903cd12005-08-11 16:25:26 -0400712static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
Chuck Levera246b012005-08-11 16:25:23 -0400713{
714 struct rpc_xprt *xprt = rd_desc->arg.data;
715 skb_reader_t desc = {
716 .skb = skb,
717 .offset = offset,
718 .count = len,
719 .csum = 0
Chuck Lever9903cd12005-08-11 16:25:26 -0400720 };
Chuck Levera246b012005-08-11 16:25:23 -0400721
Chuck Lever9903cd12005-08-11 16:25:26 -0400722 dprintk("RPC: xs_tcp_data_recv started\n");
Chuck Levera246b012005-08-11 16:25:23 -0400723 do {
724 /* Read in a new fragment marker if necessary */
725 /* Can we ever really expect to get completely empty fragments? */
726 if (xprt->tcp_flags & XPRT_COPY_RECM) {
Chuck Lever9903cd12005-08-11 16:25:26 -0400727 xs_tcp_read_fraghdr(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -0400728 continue;
729 }
730 /* Read in the xid if necessary */
731 if (xprt->tcp_flags & XPRT_COPY_XID) {
Chuck Lever9903cd12005-08-11 16:25:26 -0400732 xs_tcp_read_xid(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -0400733 continue;
734 }
735 /* Read in the request data */
736 if (xprt->tcp_flags & XPRT_COPY_DATA) {
Chuck Lever9903cd12005-08-11 16:25:26 -0400737 xs_tcp_read_request(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -0400738 continue;
739 }
740 /* Skip over any trailing bytes on short reads */
Chuck Lever9903cd12005-08-11 16:25:26 -0400741 xs_tcp_read_discard(xprt, &desc);
Chuck Levera246b012005-08-11 16:25:23 -0400742 } while (desc.count);
Chuck Lever9903cd12005-08-11 16:25:26 -0400743 dprintk("RPC: xs_tcp_data_recv done\n");
Chuck Levera246b012005-08-11 16:25:23 -0400744 return len - desc.count;
745}
746
Chuck Lever9903cd12005-08-11 16:25:26 -0400747/**
748 * xs_tcp_data_ready - "data ready" callback for TCP sockets
749 * @sk: socket with data to read
750 * @bytes: how much data to read
751 *
752 */
753static void xs_tcp_data_ready(struct sock *sk, int bytes)
Chuck Levera246b012005-08-11 16:25:23 -0400754{
755 struct rpc_xprt *xprt;
756 read_descriptor_t rd_desc;
757
758 read_lock(&sk->sk_callback_lock);
Chuck Lever9903cd12005-08-11 16:25:26 -0400759 dprintk("RPC: xs_tcp_data_ready...\n");
760 if (!(xprt = xprt_from_sock(sk)))
Chuck Levera246b012005-08-11 16:25:23 -0400761 goto out;
Chuck Levera246b012005-08-11 16:25:23 -0400762 if (xprt->shutdown)
763 goto out;
764
Chuck Lever9903cd12005-08-11 16:25:26 -0400765 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
Chuck Levera246b012005-08-11 16:25:23 -0400766 rd_desc.arg.data = xprt;
767 rd_desc.count = 65536;
Chuck Lever9903cd12005-08-11 16:25:26 -0400768 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
Chuck Levera246b012005-08-11 16:25:23 -0400769out:
770 read_unlock(&sk->sk_callback_lock);
771}
772
Chuck Lever9903cd12005-08-11 16:25:26 -0400773/**
774 * xs_tcp_state_change - callback to handle TCP socket state changes
775 * @sk: socket whose state has changed
776 *
777 */
778static void xs_tcp_state_change(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -0400779{
Chuck Lever9903cd12005-08-11 16:25:26 -0400780 struct rpc_xprt *xprt;
Chuck Levera246b012005-08-11 16:25:23 -0400781
782 read_lock(&sk->sk_callback_lock);
783 if (!(xprt = xprt_from_sock(sk)))
784 goto out;
Chuck Lever9903cd12005-08-11 16:25:26 -0400785 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400786 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
787 sk->sk_state, xprt_connected(xprt),
788 sock_flag(sk, SOCK_DEAD),
789 sock_flag(sk, SOCK_ZAPPED));
790
791 switch (sk->sk_state) {
792 case TCP_ESTABLISHED:
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400793 spin_lock_bh(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400794 if (!xprt_test_and_set_connected(xprt)) {
795 /* Reset TCP record info */
796 xprt->tcp_offset = 0;
797 xprt->tcp_reclen = 0;
798 xprt->tcp_copied = 0;
799 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
Chuck Lever03bf4b72005-08-25 16:25:55 -0700800 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
Chuck Lever44fbac22005-08-11 16:25:44 -0400801 xprt_wake_pending_tasks(xprt, 0);
Chuck Levera246b012005-08-11 16:25:23 -0400802 }
Chuck Lever4a0f8c02005-08-11 16:25:32 -0400803 spin_unlock_bh(&xprt->transport_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400804 break;
805 case TCP_SYN_SENT:
806 case TCP_SYN_RECV:
807 break;
Trond Myklebust632e3bd2006-01-03 09:55:55 +0100808 case TCP_CLOSE_WAIT:
809 /* Try to schedule an autoclose RPC calls */
810 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
811 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
812 schedule_work(&xprt->task_cleanup);
Chuck Levera246b012005-08-11 16:25:23 -0400813 default:
814 xprt_disconnect(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400815 }
816 out:
817 read_unlock(&sk->sk_callback_lock);
818}
819
Chuck Lever9903cd12005-08-11 16:25:26 -0400820/**
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400821 * xs_udp_write_space - callback invoked when socket buffer space
822 * becomes available
Chuck Lever9903cd12005-08-11 16:25:26 -0400823 * @sk: socket whose state has changed
824 *
Chuck Levera246b012005-08-11 16:25:23 -0400825 * Called when more output buffer space is available for this socket.
826 * We try not to wake our writers until they can make "significant"
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400827 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
Chuck Levera246b012005-08-11 16:25:23 -0400828 * with a bunch of small requests.
829 */
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400830static void xs_udp_write_space(struct sock *sk)
Chuck Levera246b012005-08-11 16:25:23 -0400831{
Chuck Levera246b012005-08-11 16:25:23 -0400832 read_lock(&sk->sk_callback_lock);
Chuck Levera246b012005-08-11 16:25:23 -0400833
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400834 /* from net/core/sock.c:sock_def_write_space */
835 if (sock_writeable(sk)) {
836 struct socket *sock;
837 struct rpc_xprt *xprt;
838
839 if (unlikely(!(sock = sk->sk_socket)))
Chuck Levera246b012005-08-11 16:25:23 -0400840 goto out;
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400841 if (unlikely(!(xprt = xprt_from_sock(sk))))
Chuck Levera246b012005-08-11 16:25:23 -0400842 goto out;
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400843 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
844 goto out;
845
846 xprt_write_space(xprt);
Chuck Levera246b012005-08-11 16:25:23 -0400847 }
848
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400849 out:
850 read_unlock(&sk->sk_callback_lock);
851}
Chuck Levera246b012005-08-11 16:25:23 -0400852
Chuck Leverc7b2cae2005-08-11 16:25:50 -0400853/**
854 * xs_tcp_write_space - callback invoked when socket buffer space
855 * becomes available
856 * @sk: socket whose state has changed
857 *
858 * Called when more output buffer space is available for this socket.
859 * We try not to wake our writers until they can make "significant"
860 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
861 * with a bunch of small requests.
862 */
863static void xs_tcp_write_space(struct sock *sk)
864{
865 read_lock(&sk->sk_callback_lock);
866
867 /* from net/core/stream.c:sk_stream_write_space */
868 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
869 struct socket *sock;
870 struct rpc_xprt *xprt;
871
872 if (unlikely(!(sock = sk->sk_socket)))
873 goto out;
874 if (unlikely(!(xprt = xprt_from_sock(sk))))
875 goto out;
876 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
877 goto out;
878
879 xprt_write_space(xprt);
880 }
881
882 out:
Chuck Levera246b012005-08-11 16:25:23 -0400883 read_unlock(&sk->sk_callback_lock);
884}
885
Chuck Lever470056c2005-08-25 16:25:56 -0700886static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
Chuck Levera246b012005-08-11 16:25:23 -0400887{
888 struct sock *sk = xprt->inet;
889
Chuck Levera246b012005-08-11 16:25:23 -0400890 if (xprt->rcvsize) {
891 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
892 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
893 }
894 if (xprt->sndsize) {
895 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
896 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
897 sk->sk_write_space(sk);
898 }
899}
900
Chuck Lever43118c22005-08-25 16:25:49 -0700901/**
Chuck Lever470056c2005-08-25 16:25:56 -0700902 * xs_udp_set_buffer_size - set send and receive limits
Chuck Lever43118c22005-08-25 16:25:49 -0700903 * @xprt: generic transport
Chuck Lever470056c2005-08-25 16:25:56 -0700904 * @sndsize: requested size of send buffer, in bytes
905 * @rcvsize: requested size of receive buffer, in bytes
Chuck Lever43118c22005-08-25 16:25:49 -0700906 *
Chuck Lever470056c2005-08-25 16:25:56 -0700907 * Set socket send and receive buffer size limits.
Chuck Lever43118c22005-08-25 16:25:49 -0700908 */
Chuck Lever470056c2005-08-25 16:25:56 -0700909static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
Chuck Lever43118c22005-08-25 16:25:49 -0700910{
Chuck Lever470056c2005-08-25 16:25:56 -0700911 xprt->sndsize = 0;
912 if (sndsize)
913 xprt->sndsize = sndsize + 1024;
914 xprt->rcvsize = 0;
915 if (rcvsize)
916 xprt->rcvsize = rcvsize + 1024;
917
918 xs_udp_do_set_buffer_size(xprt);
Chuck Lever43118c22005-08-25 16:25:49 -0700919}
920
Chuck Lever46c0ee82005-08-25 16:25:52 -0700921/**
922 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
923 * @task: task that timed out
924 *
925 * Adjust the congestion window after a retransmit timeout has occurred.
926 */
927static void xs_udp_timer(struct rpc_task *task)
928{
929 xprt_adjust_cwnd(task, -ETIMEDOUT);
930}
931
Chuck Lever92200412006-01-03 09:55:51 +0100932/**
933 * xs_set_port - reset the port number in the remote endpoint address
934 * @xprt: generic transport
935 * @port: new port number
936 *
937 */
938static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
939{
940 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
941 xprt->addr.sin_port = htons(port);
942}
943
Chuck Lever9903cd12005-08-11 16:25:26 -0400944static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
Chuck Levera246b012005-08-11 16:25:23 -0400945{
946 struct sockaddr_in myaddr = {
947 .sin_family = AF_INET,
948 };
Chuck Lever529b33c2005-08-25 16:25:54 -0700949 int err;
950 unsigned short port = xprt->port;
Chuck Levera246b012005-08-11 16:25:23 -0400951
Chuck Levera246b012005-08-11 16:25:23 -0400952 do {
953 myaddr.sin_port = htons(port);
954 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
955 sizeof(myaddr));
956 if (err == 0) {
957 xprt->port = port;
Chuck Lever9903cd12005-08-11 16:25:26 -0400958 dprintk("RPC: xs_bindresvport bound to port %u\n",
959 port);
Chuck Levera246b012005-08-11 16:25:23 -0400960 return 0;
961 }
Chuck Lever529b33c2005-08-25 16:25:54 -0700962 if (port <= xprt_min_resvport)
963 port = xprt_max_resvport;
964 else
965 port--;
Chuck Levera246b012005-08-11 16:25:23 -0400966 } while (err == -EADDRINUSE && port != xprt->port);
967
Chuck Lever9903cd12005-08-11 16:25:26 -0400968 dprintk("RPC: can't bind to reserved port (%d).\n", -err);
Chuck Levera246b012005-08-11 16:25:23 -0400969 return err;
970}
971
Chuck Lever9903cd12005-08-11 16:25:26 -0400972/**
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400973 * xs_udp_connect_worker - set up a UDP socket
Chuck Lever9903cd12005-08-11 16:25:26 -0400974 * @args: RPC transport to connect
975 *
976 * Invoked by a work queue tasklet.
Chuck Levera246b012005-08-11 16:25:23 -0400977 */
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400978static void xs_udp_connect_worker(void *args)
Chuck Levera246b012005-08-11 16:25:23 -0400979{
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400980 struct rpc_xprt *xprt = (struct rpc_xprt *) args;
Chuck Levera246b012005-08-11 16:25:23 -0400981 struct socket *sock = xprt->sock;
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400982 int err, status = -EIO;
Chuck Levera246b012005-08-11 16:25:23 -0400983
984 if (xprt->shutdown || xprt->addr.sin_port == 0)
985 goto out;
986
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400987 dprintk("RPC: xs_udp_connect_worker for xprt %p\n", xprt);
Chuck Lever9903cd12005-08-11 16:25:26 -0400988
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400989 /* Start by resetting any existing state */
Chuck Lever9903cd12005-08-11 16:25:26 -0400990 xs_close(xprt);
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400991
992 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
993 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
Chuck Levera246b012005-08-11 16:25:23 -0400994 goto out;
995 }
Chuck Levera246b012005-08-11 16:25:23 -0400996
Chuck Leverb0d93ad2005-08-11 16:25:53 -0400997 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
998 sock_release(sock);
999 goto out;
1000 }
1001
1002 if (!xprt->inet) {
1003 struct sock *sk = sock->sk;
1004
1005 write_lock_bh(&sk->sk_callback_lock);
1006
1007 sk->sk_user_data = xprt;
1008 xprt->old_data_ready = sk->sk_data_ready;
1009 xprt->old_state_change = sk->sk_state_change;
1010 xprt->old_write_space = sk->sk_write_space;
1011 sk->sk_data_ready = xs_udp_data_ready;
1012 sk->sk_write_space = xs_udp_write_space;
1013 sk->sk_no_check = UDP_CSUM_NORCV;
Trond Myklebustb079fa7b2005-12-13 16:13:52 -05001014 sk->sk_allocation = GFP_ATOMIC;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001015
1016 xprt_set_connected(xprt);
1017
1018 /* Reset to new socket */
1019 xprt->sock = sock;
1020 xprt->inet = sk;
1021
1022 write_unlock_bh(&sk->sk_callback_lock);
1023 }
Chuck Lever470056c2005-08-25 16:25:56 -07001024 xs_udp_do_set_buffer_size(xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001025 status = 0;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001026out:
1027 xprt_wake_pending_tasks(xprt, status);
1028 xprt_clear_connecting(xprt);
1029}
1030
Chuck Lever3167e122005-08-25 16:25:55 -07001031/*
1032 * We need to preserve the port number so the reply cache on the server can
1033 * find our cached RPC replies when we get around to reconnecting.
1034 */
1035static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1036{
1037 int result;
1038 struct socket *sock = xprt->sock;
1039 struct sockaddr any;
1040
1041 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1042
1043 /*
1044 * Disconnect the transport socket by doing a connect operation
1045 * with AF_UNSPEC. This should return immediately...
1046 */
1047 memset(&any, 0, sizeof(any));
1048 any.sa_family = AF_UNSPEC;
1049 result = sock->ops->connect(sock, &any, sizeof(any), 0);
1050 if (result)
1051 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1052 result);
1053}
1054
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001055/**
1056 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1057 * @args: RPC transport to connect
1058 *
1059 * Invoked by a work queue tasklet.
1060 */
1061static void xs_tcp_connect_worker(void *args)
1062{
1063 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1064 struct socket *sock = xprt->sock;
1065 int err, status = -EIO;
1066
1067 if (xprt->shutdown || xprt->addr.sin_port == 0)
Chuck Levera246b012005-08-11 16:25:23 -04001068 goto out;
1069
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001070 dprintk("RPC: xs_tcp_connect_worker for xprt %p\n", xprt);
1071
Chuck Lever3167e122005-08-25 16:25:55 -07001072 if (!xprt->sock) {
1073 /* start from scratch */
1074 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1075 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1076 goto out;
1077 }
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001078
Chuck Lever3167e122005-08-25 16:25:55 -07001079 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1080 sock_release(sock);
1081 goto out;
1082 }
1083 } else
1084 /* "close" the socket, preserving the local port */
1085 xs_tcp_reuse_connection(xprt);
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001086
1087 if (!xprt->inet) {
1088 struct sock *sk = sock->sk;
1089
1090 write_lock_bh(&sk->sk_callback_lock);
1091
1092 sk->sk_user_data = xprt;
1093 xprt->old_data_ready = sk->sk_data_ready;
1094 xprt->old_state_change = sk->sk_state_change;
1095 xprt->old_write_space = sk->sk_write_space;
1096 sk->sk_data_ready = xs_tcp_data_ready;
1097 sk->sk_state_change = xs_tcp_state_change;
1098 sk->sk_write_space = xs_tcp_write_space;
Trond Myklebustb079fa7b2005-12-13 16:13:52 -05001099 sk->sk_allocation = GFP_ATOMIC;
Chuck Lever3167e122005-08-25 16:25:55 -07001100
1101 /* socket options */
1102 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1103 sock_reset_flag(sk, SOCK_LINGER);
1104 tcp_sk(sk)->linger2 = 0;
1105 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001106
1107 xprt_clear_connected(xprt);
1108
1109 /* Reset to new socket */
1110 xprt->sock = sock;
1111 xprt->inet = sk;
1112
1113 write_unlock_bh(&sk->sk_callback_lock);
1114 }
1115
1116 /* Tell the socket layer to start connecting... */
Chuck Levera246b012005-08-11 16:25:23 -04001117 status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
1118 sizeof(xprt->addr), O_NONBLOCK);
1119 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1120 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1121 if (status < 0) {
1122 switch (status) {
1123 case -EINPROGRESS:
1124 case -EALREADY:
1125 goto out_clear;
Chuck Lever3167e122005-08-25 16:25:55 -07001126 case -ECONNREFUSED:
1127 case -ECONNRESET:
1128 /* retry with existing socket, after a delay */
1129 break;
1130 default:
1131 /* get rid of existing socket, and retry */
1132 xs_close(xprt);
1133 break;
Chuck Levera246b012005-08-11 16:25:23 -04001134 }
1135 }
1136out:
Chuck Lever44fbac22005-08-11 16:25:44 -04001137 xprt_wake_pending_tasks(xprt, status);
Chuck Levera246b012005-08-11 16:25:23 -04001138out_clear:
Chuck Lever2226feb2005-08-11 16:25:38 -04001139 xprt_clear_connecting(xprt);
Chuck Levera246b012005-08-11 16:25:23 -04001140}
1141
Chuck Lever9903cd12005-08-11 16:25:26 -04001142/**
1143 * xs_connect - connect a socket to a remote endpoint
1144 * @task: address of RPC task that manages state of connect request
1145 *
1146 * TCP: If the remote end dropped the connection, delay reconnecting.
Chuck Lever03bf4b72005-08-25 16:25:55 -07001147 *
1148 * UDP socket connects are synchronous, but we use a work queue anyway
1149 * to guarantee that even unprivileged user processes can set up a
1150 * socket on a privileged port.
1151 *
1152 * If a UDP socket connect fails, the delay behavior here prevents
1153 * retry floods (hard mounts).
Chuck Lever9903cd12005-08-11 16:25:26 -04001154 */
1155static void xs_connect(struct rpc_task *task)
Chuck Levera246b012005-08-11 16:25:23 -04001156{
1157 struct rpc_xprt *xprt = task->tk_xprt;
1158
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001159 if (xprt_test_and_set_connecting(xprt))
1160 return;
1161
1162 if (xprt->sock != NULL) {
Chuck Lever03bf4b72005-08-25 16:25:55 -07001163 dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n",
1164 xprt, xprt->reestablish_timeout / HZ);
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001165 schedule_delayed_work(&xprt->connect_worker,
Chuck Lever03bf4b72005-08-25 16:25:55 -07001166 xprt->reestablish_timeout);
1167 xprt->reestablish_timeout <<= 1;
1168 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1169 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001170 } else {
1171 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1172 schedule_work(&xprt->connect_worker);
1173
1174 /* flush_scheduled_work can sleep... */
1175 if (!RPC_IS_ASYNC(task))
1176 flush_scheduled_work();
Chuck Levera246b012005-08-11 16:25:23 -04001177 }
1178}
1179
Chuck Lever262965f2005-08-11 16:25:56 -04001180static struct rpc_xprt_ops xs_udp_ops = {
Chuck Lever43118c22005-08-25 16:25:49 -07001181 .set_buffer_size = xs_udp_set_buffer_size,
Chuck Lever12a80462005-08-25 16:25:51 -07001182 .reserve_xprt = xprt_reserve_xprt_cong,
Chuck Lever49e9a892005-08-25 16:25:51 -07001183 .release_xprt = xprt_release_xprt_cong,
Chuck Lever92200412006-01-03 09:55:51 +01001184 .set_port = xs_set_port,
Chuck Lever9903cd12005-08-11 16:25:26 -04001185 .connect = xs_connect,
Chuck Lever02107142006-01-03 09:55:49 +01001186 .buf_alloc = rpc_malloc,
1187 .buf_free = rpc_free,
Chuck Lever262965f2005-08-11 16:25:56 -04001188 .send_request = xs_udp_send_request,
Chuck Leverfe3aca22005-08-25 16:25:50 -07001189 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
Chuck Lever46c0ee82005-08-25 16:25:52 -07001190 .timer = xs_udp_timer,
Chuck Levera58dd392005-08-25 16:25:53 -07001191 .release_request = xprt_release_rqst_cong,
Chuck Lever262965f2005-08-11 16:25:56 -04001192 .close = xs_close,
1193 .destroy = xs_destroy,
1194};
1195
1196static struct rpc_xprt_ops xs_tcp_ops = {
Chuck Lever12a80462005-08-25 16:25:51 -07001197 .reserve_xprt = xprt_reserve_xprt,
Chuck Lever49e9a892005-08-25 16:25:51 -07001198 .release_xprt = xprt_release_xprt,
Chuck Lever92200412006-01-03 09:55:51 +01001199 .set_port = xs_set_port,
Chuck Lever262965f2005-08-11 16:25:56 -04001200 .connect = xs_connect,
Chuck Lever02107142006-01-03 09:55:49 +01001201 .buf_alloc = rpc_malloc,
1202 .buf_free = rpc_free,
Chuck Lever262965f2005-08-11 16:25:56 -04001203 .send_request = xs_tcp_send_request,
Chuck Leverfe3aca22005-08-25 16:25:50 -07001204 .set_retrans_timeout = xprt_set_retrans_timeout_def,
Chuck Lever9903cd12005-08-11 16:25:26 -04001205 .close = xs_close,
1206 .destroy = xs_destroy,
Chuck Levera246b012005-08-11 16:25:23 -04001207};
1208
Chuck Lever9903cd12005-08-11 16:25:26 -04001209/**
1210 * xs_setup_udp - Set up transport to use a UDP socket
1211 * @xprt: transport to set up
1212 * @to: timeout parameters
1213 *
1214 */
Chuck Levera246b012005-08-11 16:25:23 -04001215int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1216{
1217 size_t slot_table_size;
1218
1219 dprintk("RPC: setting up udp-ipv4 transport...\n");
1220
1221 xprt->max_reqs = xprt_udp_slot_table_entries;
1222 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1223 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1224 if (xprt->slot == NULL)
1225 return -ENOMEM;
1226 memset(xprt->slot, 0, slot_table_size);
1227
1228 xprt->prot = IPPROTO_UDP;
Chuck Lever529b33c2005-08-25 16:25:54 -07001229 xprt->port = xprt_max_resvport;
Chuck Lever808012f2005-08-25 16:25:49 -07001230 xprt->tsh_size = 0;
Chuck Levera246b012005-08-11 16:25:23 -04001231 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1232 /* XXX: header size can vary due to auth type, IPv6, etc. */
1233 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1234
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001235 INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
Chuck Lever03bf4b72005-08-25 16:25:55 -07001236 xprt->bind_timeout = XS_BIND_TO;
1237 xprt->connect_timeout = XS_UDP_CONN_TO;
1238 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1239 xprt->idle_timeout = XS_IDLE_DISC_TO;
Chuck Levera246b012005-08-11 16:25:23 -04001240
Chuck Lever262965f2005-08-11 16:25:56 -04001241 xprt->ops = &xs_udp_ops;
Chuck Levera246b012005-08-11 16:25:23 -04001242
1243 if (to)
1244 xprt->timeout = *to;
1245 else
Chuck Lever9903cd12005-08-11 16:25:26 -04001246 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
Chuck Levera246b012005-08-11 16:25:23 -04001247
1248 return 0;
1249}
1250
Chuck Lever9903cd12005-08-11 16:25:26 -04001251/**
1252 * xs_setup_tcp - Set up transport to use a TCP socket
1253 * @xprt: transport to set up
1254 * @to: timeout parameters
1255 *
1256 */
Chuck Levera246b012005-08-11 16:25:23 -04001257int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1258{
1259 size_t slot_table_size;
1260
1261 dprintk("RPC: setting up tcp-ipv4 transport...\n");
1262
1263 xprt->max_reqs = xprt_tcp_slot_table_entries;
1264 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1265 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1266 if (xprt->slot == NULL)
1267 return -ENOMEM;
1268 memset(xprt->slot, 0, slot_table_size);
1269
1270 xprt->prot = IPPROTO_TCP;
Chuck Lever529b33c2005-08-25 16:25:54 -07001271 xprt->port = xprt_max_resvport;
Chuck Lever808012f2005-08-25 16:25:49 -07001272 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
Chuck Levera246b012005-08-11 16:25:23 -04001273 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
Chuck Lever808012f2005-08-25 16:25:49 -07001274 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
Chuck Levera246b012005-08-11 16:25:23 -04001275
Chuck Leverb0d93ad2005-08-11 16:25:53 -04001276 INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
Chuck Lever03bf4b72005-08-25 16:25:55 -07001277 xprt->bind_timeout = XS_BIND_TO;
1278 xprt->connect_timeout = XS_TCP_CONN_TO;
1279 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1280 xprt->idle_timeout = XS_IDLE_DISC_TO;
Chuck Levera246b012005-08-11 16:25:23 -04001281
Chuck Lever262965f2005-08-11 16:25:56 -04001282 xprt->ops = &xs_tcp_ops;
Chuck Levera246b012005-08-11 16:25:23 -04001283
1284 if (to)
1285 xprt->timeout = *to;
1286 else
Chuck Lever9903cd12005-08-11 16:25:26 -04001287 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
Chuck Levera246b012005-08-11 16:25:23 -04001288
1289 return 0;
1290}