]>
Commit | Line | Data |
---|---|---|
a246b010 CL |
1 | /* |
2 | * linux/net/sunrpc/xprtsock.c | |
3 | * | |
4 | * Client-side transport implementation for sockets. | |
5 | * | |
6 | * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com> | |
7 | * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com> | |
8 | * TCP NFS related read + write fixes | |
9 | * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> | |
10 | * | |
11 | * Rewrite of larges part of the code in order to stabilize TCP stuff. | |
12 | * Fix behaviour when socket buffer is full. | |
13 | * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> | |
55aa4f58 CL |
14 | * |
15 | * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> | |
a246b010 CL |
16 | */ |
17 | ||
18 | #include <linux/types.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/capability.h> | |
21 | #include <linux/sched.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/errno.h> | |
24 | #include <linux/socket.h> | |
25 | #include <linux/in.h> | |
26 | #include <linux/net.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/udp.h> | |
29 | #include <linux/tcp.h> | |
30 | #include <linux/sunrpc/clnt.h> | |
02107148 | 31 | #include <linux/sunrpc/sched.h> |
a246b010 CL |
32 | #include <linux/file.h> |
33 | ||
34 | #include <net/sock.h> | |
35 | #include <net/checksum.h> | |
36 | #include <net/udp.h> | |
37 | #include <net/tcp.h> | |
38 | ||
c556b754 CL |
39 | /* |
40 | * xprtsock tunables | |
41 | */ | |
42 | unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; | |
43 | unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE; | |
44 | ||
45 | unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; | |
46 | unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; | |
47 | ||
262965f5 CL |
48 | /* |
49 | * How many times to try sending a request on a socket before waiting | |
50 | * for the socket buffer to clear. | |
51 | */ | |
52 | #define XS_SENDMSG_RETRY (10U) | |
53 | ||
03bf4b70 CL |
54 | /* |
55 | * Time out for an RPC UDP socket connect. UDP socket connects are | |
56 | * synchronous, but we set a timeout anyway in case of resource | |
57 | * exhaustion on the local host. | |
58 | */ | |
59 | #define XS_UDP_CONN_TO (5U * HZ) | |
60 | ||
61 | /* | |
62 | * Wait duration for an RPC TCP connection to be established. Solaris | |
63 | * NFS over TCP uses 60 seconds, for example, which is in line with how | |
64 | * long a server takes to reboot. | |
65 | */ | |
66 | #define XS_TCP_CONN_TO (60U * HZ) | |
67 | ||
68 | /* | |
69 | * Wait duration for a reply from the RPC portmapper. | |
70 | */ | |
71 | #define XS_BIND_TO (60U * HZ) | |
72 | ||
73 | /* | |
74 | * Delay if a UDP socket connect error occurs. This is most likely some | |
75 | * kind of resource problem on the local host. | |
76 | */ | |
77 | #define XS_UDP_REEST_TO (2U * HZ) | |
78 | ||
79 | /* | |
80 | * The reestablish timeout allows clients to delay for a bit before attempting | |
81 | * to reconnect to a server that just dropped our connection. | |
82 | * | |
83 | * We implement an exponential backoff when trying to reestablish a TCP | |
84 | * transport connection with the server. Some servers like to drop a TCP | |
85 | * connection when they are overworked, so we start with a short timeout and | |
86 | * increase over time if the server is down or not responding. | |
87 | */ | |
88 | #define XS_TCP_INIT_REEST_TO (3U * HZ) | |
89 | #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ) | |
90 | ||
91 | /* | |
92 | * TCP idle timeout; client drops the transport socket if it is idle | |
93 | * for this long. Note that we also timeout UDP sockets to prevent | |
94 | * holding port numbers when there is no RPC traffic. | |
95 | */ | |
96 | #define XS_IDLE_DISC_TO (5U * 60 * HZ) | |
97 | ||
a246b010 CL |
98 | #ifdef RPC_DEBUG |
99 | # undef RPC_DEBUG_DATA | |
9903cd1c | 100 | # define RPCDBG_FACILITY RPCDBG_TRANS |
a246b010 CL |
101 | #endif |
102 | ||
a246b010 | 103 | #ifdef RPC_DEBUG_DATA |
9903cd1c | 104 | static void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 | 105 | { |
9903cd1c CL |
106 | u8 *buf = (u8 *) packet; |
107 | int j; | |
a246b010 CL |
108 | |
109 | dprintk("RPC: %s\n", msg); | |
110 | for (j = 0; j < count && j < 128; j += 4) { | |
111 | if (!(j & 31)) { | |
112 | if (j) | |
113 | dprintk("\n"); | |
114 | dprintk("0x%04x ", j); | |
115 | } | |
116 | dprintk("%02x%02x%02x%02x ", | |
117 | buf[j], buf[j+1], buf[j+2], buf[j+3]); | |
118 | } | |
119 | dprintk("\n"); | |
120 | } | |
121 | #else | |
9903cd1c | 122 | static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 CL |
123 | { |
124 | /* NOP */ | |
125 | } | |
126 | #endif | |
127 | ||
b4b5cc85 CL |
128 | #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) |
129 | ||
130 | static 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 | ||
150 | static 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 | ||
9903cd1c CL |
163 | /** |
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 | * | |
a246b010 | 171 | */ |
262965f5 | 172 | static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base) |
a246b010 CL |
173 | { |
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 | ||
262965f5 CL |
179 | if (unlikely(!sock)) |
180 | return -ENOTCONN; | |
181 | ||
182 | clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); | |
183 | ||
a246b010 CL |
184 | len = xdr->head[0].iov_len; |
185 | if (base < len || (addr != NULL && base == 0)) { | |
b4b5cc85 | 186 | err = xs_send_head(sock, addr, addrlen, xdr, base, len); |
a246b010 CL |
187 | if (ret == 0) |
188 | ret = err; | |
189 | else if (err > 0) | |
190 | ret += err; | |
b4b5cc85 | 191 | if (err != (len - base)) |
a246b010 CL |
192 | goto out; |
193 | base = 0; | |
194 | } else | |
195 | base -= len; | |
196 | ||
b4b5cc85 | 197 | if (unlikely(pglen == 0)) |
a246b010 | 198 | goto copy_tail; |
b4b5cc85 | 199 | if (unlikely(base >= pglen)) { |
a246b010 CL |
200 | base -= pglen; |
201 | goto copy_tail; | |
202 | } | |
203 | if (base || xdr->page_base) { | |
204 | pglen -= base; | |
9903cd1c | 205 | base += xdr->page_base; |
a246b010 CL |
206 | ppage += base >> PAGE_CACHE_SHIFT; |
207 | base &= ~PAGE_CACHE_MASK; | |
208 | } | |
209 | ||
210 | sendpage = sock->ops->sendpage ? : sock_no_sendpage; | |
211 | do { | |
b4b5cc85 | 212 | int flags = XS_SENDMSG_FLAGS; |
a246b010 CL |
213 | |
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); | |
236 | copy_tail: | |
237 | len = xdr->tail[0].iov_len; | |
238 | if (base < len) { | |
b4b5cc85 | 239 | err = xs_send_tail(sock, xdr, base, len); |
a246b010 CL |
240 | if (ret == 0) |
241 | ret = err; | |
242 | else if (err > 0) | |
243 | ret += err; | |
244 | } | |
245 | out: | |
246 | return ret; | |
247 | } | |
248 | ||
9903cd1c | 249 | /** |
262965f5 CL |
250 | * xs_nospace - place task on wait queue if transmit was incomplete |
251 | * @task: task to put to sleep | |
9903cd1c | 252 | * |
a246b010 | 253 | */ |
262965f5 | 254 | static void xs_nospace(struct rpc_task *task) |
a246b010 | 255 | { |
262965f5 CL |
256 | struct rpc_rqst *req = task->tk_rqstp; |
257 | struct rpc_xprt *xprt = req->rq_xprt; | |
a246b010 | 258 | |
262965f5 CL |
259 | 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 | */ | |
290 | static 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; | |
a246b010 | 296 | |
9903cd1c | 297 | xs_pktdump("packet data:", |
a246b010 CL |
298 | req->rq_svec->iov_base, |
299 | req->rq_svec->iov_len); | |
300 | ||
262965f5 CL |
301 | req->rq_xtime = jiffies; |
302 | status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr, | |
303 | sizeof(xprt->addr), xdr, req->rq_bytes_sent); | |
a246b010 | 304 | |
262965f5 CL |
305 | dprintk("RPC: xs_udp_send_request(%u) = %d\n", |
306 | xdr->len - req->rq_bytes_sent, status); | |
a246b010 | 307 | |
262965f5 CL |
308 | if (likely(status >= (int) req->rq_slen)) |
309 | return 0; | |
a246b010 | 310 | |
262965f5 CL |
311 | /* Still some bytes left; set up for a retry later. */ |
312 | if (status > 0) | |
313 | status = -EAGAIN; | |
a246b010 | 314 | |
262965f5 CL |
315 | switch (status) { |
316 | case -ENETUNREACH: | |
317 | case -EPIPE: | |
a246b010 CL |
318 | case -ECONNREFUSED: |
319 | /* When the server has died, an ICMP port unreachable message | |
9903cd1c | 320 | * prompts ECONNREFUSED. */ |
a246b010 | 321 | break; |
262965f5 CL |
322 | case -EAGAIN: |
323 | xs_nospace(task); | |
a246b010 CL |
324 | break; |
325 | default: | |
262965f5 CL |
326 | dprintk("RPC: sendmsg returned unrecognized error %d\n", |
327 | -status); | |
9903cd1c | 328 | break; |
a246b010 | 329 | } |
262965f5 CL |
330 | |
331 | return status; | |
a246b010 CL |
332 | } |
333 | ||
808012fb CL |
334 | static 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 | ||
9903cd1c | 341 | /** |
262965f5 | 342 | * xs_tcp_send_request - write an RPC request to a TCP socket |
9903cd1c CL |
343 | * @task: address of RPC task that manages the state of an RPC request |
344 | * | |
345 | * Return values: | |
262965f5 CL |
346 | * 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 | |
9903cd1c CL |
351 | * |
352 | * XXX: In the case of soft timeouts, should we eventually give up | |
262965f5 | 353 | * if sendmsg is not able to make progress? |
9903cd1c | 354 | */ |
262965f5 | 355 | static int xs_tcp_send_request(struct rpc_task *task) |
a246b010 CL |
356 | { |
357 | struct rpc_rqst *req = task->tk_rqstp; | |
358 | struct rpc_xprt *xprt = req->rq_xprt; | |
262965f5 | 359 | struct xdr_buf *xdr = &req->rq_snd_buf; |
a246b010 CL |
360 | int status, retry = 0; |
361 | ||
808012fb | 362 | xs_encode_tcp_record_marker(&req->rq_snd_buf); |
a246b010 | 363 | |
262965f5 CL |
364 | xs_pktdump("packet data:", |
365 | req->rq_svec->iov_base, | |
366 | req->rq_svec->iov_len); | |
a246b010 CL |
367 | |
368 | /* Continue transmitting the packet/record. We must be careful | |
369 | * to cope with writespace callbacks arriving _after_ we have | |
262965f5 | 370 | * called sendmsg(). */ |
a246b010 CL |
371 | while (1) { |
372 | req->rq_xtime = jiffies; | |
262965f5 CL |
373 | status = xs_sendpages(xprt->sock, NULL, 0, xdr, |
374 | req->rq_bytes_sent); | |
a246b010 | 375 | |
262965f5 CL |
376 | dprintk("RPC: xs_tcp_send_request(%u) = %d\n", |
377 | xdr->len - req->rq_bytes_sent, status); | |
a246b010 | 378 | |
262965f5 | 379 | if (unlikely(status < 0)) |
a246b010 | 380 | break; |
a246b010 | 381 | |
262965f5 CL |
382 | /* If we've sent the entire packet, immediately |
383 | * reset the count of bytes sent. */ | |
384 | req->rq_bytes_sent += status; | |
ef759a2e | 385 | task->tk_bytes_sent += status; |
262965f5 CL |
386 | if (likely(req->rq_bytes_sent >= req->rq_slen)) { |
387 | req->rq_bytes_sent = 0; | |
388 | return 0; | |
389 | } | |
a246b010 CL |
390 | |
391 | status = -EAGAIN; | |
262965f5 | 392 | if (retry++ > XS_SENDMSG_RETRY) |
a246b010 CL |
393 | break; |
394 | } | |
395 | ||
262965f5 CL |
396 | switch (status) { |
397 | case -EAGAIN: | |
398 | xs_nospace(task); | |
399 | break; | |
400 | case -ECONNREFUSED: | |
401 | case -ECONNRESET: | |
402 | case -ENOTCONN: | |
403 | case -EPIPE: | |
404 | status = -ENOTCONN; | |
405 | break; | |
406 | default: | |
407 | dprintk("RPC: sendmsg returned unrecognized error %d\n", | |
408 | -status); | |
43118c29 | 409 | xprt_disconnect(xprt); |
262965f5 | 410 | break; |
a246b010 | 411 | } |
262965f5 | 412 | |
a246b010 CL |
413 | return status; |
414 | } | |
415 | ||
9903cd1c CL |
416 | /** |
417 | * xs_close - close a socket | |
418 | * @xprt: transport | |
419 | * | |
3167e12c CL |
420 | * This is used when all requests are complete; ie, no DRC state remains |
421 | * on the server we want to save. | |
a246b010 | 422 | */ |
9903cd1c | 423 | static void xs_close(struct rpc_xprt *xprt) |
a246b010 | 424 | { |
9903cd1c CL |
425 | struct socket *sock = xprt->sock; |
426 | struct sock *sk = xprt->inet; | |
a246b010 CL |
427 | |
428 | if (!sk) | |
632e3bdc | 429 | goto clear_close_wait; |
a246b010 | 430 | |
9903cd1c CL |
431 | dprintk("RPC: xs_close xprt %p\n", xprt); |
432 | ||
a246b010 CL |
433 | write_lock_bh(&sk->sk_callback_lock); |
434 | xprt->inet = NULL; | |
435 | xprt->sock = NULL; | |
436 | ||
9903cd1c CL |
437 | sk->sk_user_data = NULL; |
438 | sk->sk_data_ready = xprt->old_data_ready; | |
a246b010 | 439 | sk->sk_state_change = xprt->old_state_change; |
9903cd1c | 440 | sk->sk_write_space = xprt->old_write_space; |
a246b010 CL |
441 | write_unlock_bh(&sk->sk_callback_lock); |
442 | ||
9903cd1c | 443 | sk->sk_no_check = 0; |
a246b010 CL |
444 | |
445 | sock_release(sock); | |
632e3bdc TM |
446 | clear_close_wait: |
447 | smp_mb__before_clear_bit(); | |
448 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
449 | smp_mb__after_clear_bit(); | |
a246b010 CL |
450 | } |
451 | ||
9903cd1c CL |
452 | /** |
453 | * xs_destroy - prepare to shutdown a transport | |
454 | * @xprt: doomed transport | |
455 | * | |
456 | */ | |
457 | static void xs_destroy(struct rpc_xprt *xprt) | |
a246b010 | 458 | { |
9903cd1c CL |
459 | dprintk("RPC: xs_destroy xprt %p\n", xprt); |
460 | ||
55aa4f58 | 461 | cancel_delayed_work(&xprt->connect_worker); |
a246b010 CL |
462 | flush_scheduled_work(); |
463 | ||
464 | xprt_disconnect(xprt); | |
9903cd1c | 465 | xs_close(xprt); |
a246b010 CL |
466 | kfree(xprt->slot); |
467 | } | |
468 | ||
9903cd1c CL |
469 | static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) |
470 | { | |
471 | return (struct rpc_xprt *) sk->sk_user_data; | |
472 | } | |
473 | ||
474 | /** | |
475 | * xs_udp_data_ready - "data ready" callback for UDP sockets | |
476 | * @sk: socket with data to read | |
477 | * @len: how much data to read | |
478 | * | |
a246b010 | 479 | */ |
9903cd1c | 480 | static void xs_udp_data_ready(struct sock *sk, int len) |
a246b010 | 481 | { |
9903cd1c CL |
482 | struct rpc_task *task; |
483 | struct rpc_xprt *xprt; | |
a246b010 | 484 | struct rpc_rqst *rovr; |
9903cd1c | 485 | struct sk_buff *skb; |
a246b010 CL |
486 | int err, repsize, copied; |
487 | u32 _xid, *xp; | |
488 | ||
489 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
490 | dprintk("RPC: xs_udp_data_ready...\n"); |
491 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 492 | goto out; |
a246b010 CL |
493 | |
494 | if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) | |
495 | goto out; | |
496 | ||
497 | if (xprt->shutdown) | |
498 | goto dropit; | |
499 | ||
500 | repsize = skb->len - sizeof(struct udphdr); | |
501 | if (repsize < 4) { | |
9903cd1c | 502 | dprintk("RPC: impossible RPC reply size %d!\n", repsize); |
a246b010 CL |
503 | goto dropit; |
504 | } | |
505 | ||
506 | /* Copy the XID from the skb... */ | |
507 | xp = skb_header_pointer(skb, sizeof(struct udphdr), | |
508 | sizeof(_xid), &_xid); | |
509 | if (xp == NULL) | |
510 | goto dropit; | |
511 | ||
512 | /* Look up and lock the request corresponding to the given XID */ | |
4a0f8c04 | 513 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
514 | rovr = xprt_lookup_rqst(xprt, *xp); |
515 | if (!rovr) | |
516 | goto out_unlock; | |
517 | task = rovr->rq_task; | |
518 | ||
a246b010 CL |
519 | if ((copied = rovr->rq_private_buf.buflen) > repsize) |
520 | copied = repsize; | |
521 | ||
522 | /* Suck it into the iovec, verify checksum if not done by hw. */ | |
523 | if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) | |
524 | goto out_unlock; | |
525 | ||
526 | /* Something worked... */ | |
527 | dst_confirm(skb->dst); | |
528 | ||
1570c1e4 CL |
529 | xprt_adjust_cwnd(task, copied); |
530 | xprt_update_rtt(task); | |
531 | xprt_complete_rqst(task, copied); | |
a246b010 CL |
532 | |
533 | out_unlock: | |
4a0f8c04 | 534 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
535 | dropit: |
536 | skb_free_datagram(sk, skb); | |
537 | out: | |
538 | read_unlock(&sk->sk_callback_lock); | |
539 | } | |
540 | ||
9903cd1c | 541 | static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len) |
a246b010 CL |
542 | { |
543 | if (len > desc->count) | |
544 | len = desc->count; | |
545 | if (skb_copy_bits(desc->skb, desc->offset, p, len)) { | |
546 | dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n", | |
547 | len, desc->count); | |
548 | return 0; | |
549 | } | |
550 | desc->offset += len; | |
551 | desc->count -= len; | |
552 | dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n", | |
553 | len, desc->count); | |
554 | return len; | |
555 | } | |
556 | ||
9903cd1c | 557 | static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
558 | { |
559 | size_t len, used; | |
560 | char *p; | |
561 | ||
562 | p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset; | |
563 | len = sizeof(xprt->tcp_recm) - xprt->tcp_offset; | |
9903cd1c | 564 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
565 | xprt->tcp_offset += used; |
566 | if (used != len) | |
567 | return; | |
808012fb | 568 | |
a246b010 | 569 | xprt->tcp_reclen = ntohl(xprt->tcp_recm); |
808012fb | 570 | if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) |
a246b010 CL |
571 | xprt->tcp_flags |= XPRT_LAST_FRAG; |
572 | else | |
573 | xprt->tcp_flags &= ~XPRT_LAST_FRAG; | |
808012fb CL |
574 | xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; |
575 | ||
a246b010 CL |
576 | xprt->tcp_flags &= ~XPRT_COPY_RECM; |
577 | xprt->tcp_offset = 0; | |
808012fb | 578 | |
a246b010 | 579 | /* Sanity check of the record length */ |
808012fb | 580 | if (unlikely(xprt->tcp_reclen < 4)) { |
9903cd1c | 581 | dprintk("RPC: invalid TCP record fragment length\n"); |
a246b010 | 582 | xprt_disconnect(xprt); |
9903cd1c | 583 | return; |
a246b010 CL |
584 | } |
585 | dprintk("RPC: reading TCP record fragment of length %d\n", | |
586 | xprt->tcp_reclen); | |
587 | } | |
588 | ||
9903cd1c | 589 | static void xs_tcp_check_recm(struct rpc_xprt *xprt) |
a246b010 CL |
590 | { |
591 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n", | |
592 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags); | |
593 | if (xprt->tcp_offset == xprt->tcp_reclen) { | |
594 | xprt->tcp_flags |= XPRT_COPY_RECM; | |
595 | xprt->tcp_offset = 0; | |
596 | if (xprt->tcp_flags & XPRT_LAST_FRAG) { | |
597 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
598 | xprt->tcp_flags |= XPRT_COPY_XID; | |
599 | xprt->tcp_copied = 0; | |
600 | } | |
601 | } | |
602 | } | |
603 | ||
9903cd1c | 604 | static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
605 | { |
606 | size_t len, used; | |
607 | char *p; | |
608 | ||
609 | len = sizeof(xprt->tcp_xid) - xprt->tcp_offset; | |
610 | dprintk("RPC: reading XID (%Zu bytes)\n", len); | |
611 | p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset; | |
9903cd1c | 612 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
613 | xprt->tcp_offset += used; |
614 | if (used != len) | |
615 | return; | |
616 | xprt->tcp_flags &= ~XPRT_COPY_XID; | |
617 | xprt->tcp_flags |= XPRT_COPY_DATA; | |
618 | xprt->tcp_copied = 4; | |
619 | dprintk("RPC: reading reply for XID %08x\n", | |
620 | ntohl(xprt->tcp_xid)); | |
9903cd1c | 621 | xs_tcp_check_recm(xprt); |
a246b010 CL |
622 | } |
623 | ||
9903cd1c | 624 | static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
625 | { |
626 | struct rpc_rqst *req; | |
627 | struct xdr_buf *rcvbuf; | |
628 | size_t len; | |
629 | ssize_t r; | |
630 | ||
631 | /* Find and lock the request corresponding to this xid */ | |
4a0f8c04 | 632 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
633 | req = xprt_lookup_rqst(xprt, xprt->tcp_xid); |
634 | if (!req) { | |
635 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
636 | dprintk("RPC: XID %08x request not found!\n", | |
637 | ntohl(xprt->tcp_xid)); | |
4a0f8c04 | 638 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
639 | return; |
640 | } | |
641 | ||
642 | rcvbuf = &req->rq_private_buf; | |
643 | len = desc->count; | |
644 | if (len > xprt->tcp_reclen - xprt->tcp_offset) { | |
645 | skb_reader_t my_desc; | |
646 | ||
647 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
648 | memcpy(&my_desc, desc, sizeof(my_desc)); | |
649 | my_desc.count = len; | |
650 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 651 | &my_desc, xs_tcp_copy_data); |
a246b010 CL |
652 | desc->count -= r; |
653 | desc->offset += r; | |
654 | } else | |
655 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 656 | desc, xs_tcp_copy_data); |
a246b010 CL |
657 | |
658 | if (r > 0) { | |
659 | xprt->tcp_copied += r; | |
660 | xprt->tcp_offset += r; | |
661 | } | |
662 | if (r != len) { | |
663 | /* Error when copying to the receive buffer, | |
664 | * usually because we weren't able to allocate | |
665 | * additional buffer pages. All we can do now | |
666 | * is turn off XPRT_COPY_DATA, so the request | |
667 | * will not receive any additional updates, | |
668 | * and time out. | |
669 | * Any remaining data from this record will | |
670 | * be discarded. | |
671 | */ | |
672 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
673 | dprintk("RPC: XID %08x truncated request\n", | |
674 | ntohl(xprt->tcp_xid)); | |
675 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
676 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
677 | goto out; | |
678 | } | |
679 | ||
680 | dprintk("RPC: XID %08x read %Zd bytes\n", | |
681 | ntohl(xprt->tcp_xid), r); | |
682 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
683 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
684 | ||
685 | if (xprt->tcp_copied == req->rq_private_buf.buflen) | |
686 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
687 | else if (xprt->tcp_offset == xprt->tcp_reclen) { | |
688 | if (xprt->tcp_flags & XPRT_LAST_FRAG) | |
689 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
690 | } | |
691 | ||
692 | out: | |
1570c1e4 CL |
693 | if (!(xprt->tcp_flags & XPRT_COPY_DATA)) |
694 | xprt_complete_rqst(req->rq_task, xprt->tcp_copied); | |
4a0f8c04 | 695 | spin_unlock(&xprt->transport_lock); |
9903cd1c | 696 | xs_tcp_check_recm(xprt); |
a246b010 CL |
697 | } |
698 | ||
9903cd1c | 699 | static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
700 | { |
701 | size_t len; | |
702 | ||
703 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
704 | if (len > desc->count) | |
705 | len = desc->count; | |
706 | desc->count -= len; | |
707 | desc->offset += len; | |
708 | xprt->tcp_offset += len; | |
709 | dprintk("RPC: discarded %Zu bytes\n", len); | |
9903cd1c | 710 | xs_tcp_check_recm(xprt); |
a246b010 CL |
711 | } |
712 | ||
9903cd1c | 713 | static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) |
a246b010 CL |
714 | { |
715 | struct rpc_xprt *xprt = rd_desc->arg.data; | |
716 | skb_reader_t desc = { | |
717 | .skb = skb, | |
718 | .offset = offset, | |
719 | .count = len, | |
720 | .csum = 0 | |
9903cd1c | 721 | }; |
a246b010 | 722 | |
9903cd1c | 723 | dprintk("RPC: xs_tcp_data_recv started\n"); |
a246b010 CL |
724 | do { |
725 | /* Read in a new fragment marker if necessary */ | |
726 | /* Can we ever really expect to get completely empty fragments? */ | |
727 | if (xprt->tcp_flags & XPRT_COPY_RECM) { | |
9903cd1c | 728 | xs_tcp_read_fraghdr(xprt, &desc); |
a246b010 CL |
729 | continue; |
730 | } | |
731 | /* Read in the xid if necessary */ | |
732 | if (xprt->tcp_flags & XPRT_COPY_XID) { | |
9903cd1c | 733 | xs_tcp_read_xid(xprt, &desc); |
a246b010 CL |
734 | continue; |
735 | } | |
736 | /* Read in the request data */ | |
737 | if (xprt->tcp_flags & XPRT_COPY_DATA) { | |
9903cd1c | 738 | xs_tcp_read_request(xprt, &desc); |
a246b010 CL |
739 | continue; |
740 | } | |
741 | /* Skip over any trailing bytes on short reads */ | |
9903cd1c | 742 | xs_tcp_read_discard(xprt, &desc); |
a246b010 | 743 | } while (desc.count); |
9903cd1c | 744 | dprintk("RPC: xs_tcp_data_recv done\n"); |
a246b010 CL |
745 | return len - desc.count; |
746 | } | |
747 | ||
9903cd1c CL |
748 | /** |
749 | * xs_tcp_data_ready - "data ready" callback for TCP sockets | |
750 | * @sk: socket with data to read | |
751 | * @bytes: how much data to read | |
752 | * | |
753 | */ | |
754 | static void xs_tcp_data_ready(struct sock *sk, int bytes) | |
a246b010 CL |
755 | { |
756 | struct rpc_xprt *xprt; | |
757 | read_descriptor_t rd_desc; | |
758 | ||
759 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
760 | dprintk("RPC: xs_tcp_data_ready...\n"); |
761 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 762 | goto out; |
a246b010 CL |
763 | if (xprt->shutdown) |
764 | goto out; | |
765 | ||
9903cd1c | 766 | /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ |
a246b010 CL |
767 | rd_desc.arg.data = xprt; |
768 | rd_desc.count = 65536; | |
9903cd1c | 769 | tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); |
a246b010 CL |
770 | out: |
771 | read_unlock(&sk->sk_callback_lock); | |
772 | } | |
773 | ||
9903cd1c CL |
774 | /** |
775 | * xs_tcp_state_change - callback to handle TCP socket state changes | |
776 | * @sk: socket whose state has changed | |
777 | * | |
778 | */ | |
779 | static void xs_tcp_state_change(struct sock *sk) | |
a246b010 | 780 | { |
9903cd1c | 781 | struct rpc_xprt *xprt; |
a246b010 CL |
782 | |
783 | read_lock(&sk->sk_callback_lock); | |
784 | if (!(xprt = xprt_from_sock(sk))) | |
785 | goto out; | |
9903cd1c | 786 | dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); |
a246b010 CL |
787 | dprintk("RPC: state %x conn %d dead %d zapped %d\n", |
788 | sk->sk_state, xprt_connected(xprt), | |
789 | sock_flag(sk, SOCK_DEAD), | |
790 | sock_flag(sk, SOCK_ZAPPED)); | |
791 | ||
792 | switch (sk->sk_state) { | |
793 | case TCP_ESTABLISHED: | |
4a0f8c04 | 794 | spin_lock_bh(&xprt->transport_lock); |
a246b010 CL |
795 | if (!xprt_test_and_set_connected(xprt)) { |
796 | /* Reset TCP record info */ | |
797 | xprt->tcp_offset = 0; | |
798 | xprt->tcp_reclen = 0; | |
799 | xprt->tcp_copied = 0; | |
800 | xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID; | |
03bf4b70 | 801 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; |
44fbac22 | 802 | xprt_wake_pending_tasks(xprt, 0); |
a246b010 | 803 | } |
4a0f8c04 | 804 | spin_unlock_bh(&xprt->transport_lock); |
a246b010 CL |
805 | break; |
806 | case TCP_SYN_SENT: | |
807 | case TCP_SYN_RECV: | |
808 | break; | |
632e3bdc TM |
809 | case TCP_CLOSE_WAIT: |
810 | /* Try to schedule an autoclose RPC calls */ | |
811 | set_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
812 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) | |
813 | schedule_work(&xprt->task_cleanup); | |
a246b010 CL |
814 | default: |
815 | xprt_disconnect(xprt); | |
a246b010 CL |
816 | } |
817 | out: | |
818 | read_unlock(&sk->sk_callback_lock); | |
819 | } | |
820 | ||
9903cd1c | 821 | /** |
c7b2cae8 CL |
822 | * xs_udp_write_space - callback invoked when socket buffer space |
823 | * becomes available | |
9903cd1c CL |
824 | * @sk: socket whose state has changed |
825 | * | |
a246b010 CL |
826 | * Called when more output buffer space is available for this socket. |
827 | * We try not to wake our writers until they can make "significant" | |
c7b2cae8 | 828 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg |
a246b010 CL |
829 | * with a bunch of small requests. |
830 | */ | |
c7b2cae8 | 831 | static void xs_udp_write_space(struct sock *sk) |
a246b010 | 832 | { |
a246b010 | 833 | read_lock(&sk->sk_callback_lock); |
a246b010 | 834 | |
c7b2cae8 CL |
835 | /* from net/core/sock.c:sock_def_write_space */ |
836 | if (sock_writeable(sk)) { | |
837 | struct socket *sock; | |
838 | struct rpc_xprt *xprt; | |
839 | ||
840 | if (unlikely(!(sock = sk->sk_socket))) | |
a246b010 | 841 | goto out; |
c7b2cae8 CL |
842 | if (unlikely(!(xprt = xprt_from_sock(sk)))) |
843 | goto out; | |
844 | if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) | |
a246b010 | 845 | goto out; |
c7b2cae8 CL |
846 | |
847 | xprt_write_space(xprt); | |
a246b010 CL |
848 | } |
849 | ||
c7b2cae8 CL |
850 | out: |
851 | read_unlock(&sk->sk_callback_lock); | |
852 | } | |
a246b010 | 853 | |
c7b2cae8 CL |
854 | /** |
855 | * xs_tcp_write_space - callback invoked when socket buffer space | |
856 | * becomes available | |
857 | * @sk: socket whose state has changed | |
858 | * | |
859 | * Called when more output buffer space is available for this socket. | |
860 | * We try not to wake our writers until they can make "significant" | |
861 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg | |
862 | * with a bunch of small requests. | |
863 | */ | |
864 | static void xs_tcp_write_space(struct sock *sk) | |
865 | { | |
866 | read_lock(&sk->sk_callback_lock); | |
867 | ||
868 | /* from net/core/stream.c:sk_stream_write_space */ | |
869 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { | |
870 | struct socket *sock; | |
871 | struct rpc_xprt *xprt; | |
872 | ||
873 | if (unlikely(!(sock = sk->sk_socket))) | |
874 | goto out; | |
875 | if (unlikely(!(xprt = xprt_from_sock(sk)))) | |
876 | goto out; | |
877 | if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) | |
878 | goto out; | |
879 | ||
880 | xprt_write_space(xprt); | |
881 | } | |
882 | ||
883 | out: | |
a246b010 CL |
884 | read_unlock(&sk->sk_callback_lock); |
885 | } | |
886 | ||
470056c2 | 887 | static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) |
a246b010 CL |
888 | { |
889 | struct sock *sk = xprt->inet; | |
890 | ||
a246b010 CL |
891 | if (xprt->rcvsize) { |
892 | sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
893 | sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2; | |
894 | } | |
895 | if (xprt->sndsize) { | |
896 | sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
897 | sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2; | |
898 | sk->sk_write_space(sk); | |
899 | } | |
900 | } | |
901 | ||
43118c29 | 902 | /** |
470056c2 | 903 | * xs_udp_set_buffer_size - set send and receive limits |
43118c29 | 904 | * @xprt: generic transport |
470056c2 CL |
905 | * @sndsize: requested size of send buffer, in bytes |
906 | * @rcvsize: requested size of receive buffer, in bytes | |
43118c29 | 907 | * |
470056c2 | 908 | * Set socket send and receive buffer size limits. |
43118c29 | 909 | */ |
470056c2 | 910 | static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) |
43118c29 | 911 | { |
470056c2 CL |
912 | xprt->sndsize = 0; |
913 | if (sndsize) | |
914 | xprt->sndsize = sndsize + 1024; | |
915 | xprt->rcvsize = 0; | |
916 | if (rcvsize) | |
917 | xprt->rcvsize = rcvsize + 1024; | |
918 | ||
919 | xs_udp_do_set_buffer_size(xprt); | |
43118c29 CL |
920 | } |
921 | ||
46c0ee8b CL |
922 | /** |
923 | * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport | |
924 | * @task: task that timed out | |
925 | * | |
926 | * Adjust the congestion window after a retransmit timeout has occurred. | |
927 | */ | |
928 | static void xs_udp_timer(struct rpc_task *task) | |
929 | { | |
930 | xprt_adjust_cwnd(task, -ETIMEDOUT); | |
931 | } | |
932 | ||
92200412 CL |
933 | /** |
934 | * xs_set_port - reset the port number in the remote endpoint address | |
935 | * @xprt: generic transport | |
936 | * @port: new port number | |
937 | * | |
938 | */ | |
939 | static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) | |
940 | { | |
941 | dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); | |
942 | xprt->addr.sin_port = htons(port); | |
943 | } | |
944 | ||
9903cd1c | 945 | static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock) |
a246b010 CL |
946 | { |
947 | struct sockaddr_in myaddr = { | |
948 | .sin_family = AF_INET, | |
949 | }; | |
529b33c6 CL |
950 | int err; |
951 | unsigned short port = xprt->port; | |
a246b010 | 952 | |
a246b010 CL |
953 | do { |
954 | myaddr.sin_port = htons(port); | |
955 | err = sock->ops->bind(sock, (struct sockaddr *) &myaddr, | |
956 | sizeof(myaddr)); | |
957 | if (err == 0) { | |
958 | xprt->port = port; | |
9903cd1c CL |
959 | dprintk("RPC: xs_bindresvport bound to port %u\n", |
960 | port); | |
a246b010 CL |
961 | return 0; |
962 | } | |
529b33c6 CL |
963 | if (port <= xprt_min_resvport) |
964 | port = xprt_max_resvport; | |
965 | else | |
966 | port--; | |
a246b010 CL |
967 | } while (err == -EADDRINUSE && port != xprt->port); |
968 | ||
9903cd1c | 969 | dprintk("RPC: can't bind to reserved port (%d).\n", -err); |
a246b010 CL |
970 | return err; |
971 | } | |
972 | ||
b0d93ad5 CL |
973 | /** |
974 | * xs_udp_connect_worker - set up a UDP socket | |
975 | * @args: RPC transport to connect | |
976 | * | |
977 | * Invoked by a work queue tasklet. | |
978 | */ | |
979 | static void xs_udp_connect_worker(void *args) | |
a246b010 | 980 | { |
b0d93ad5 CL |
981 | struct rpc_xprt *xprt = (struct rpc_xprt *) args; |
982 | struct socket *sock = xprt->sock; | |
983 | int err, status = -EIO; | |
9903cd1c | 984 | |
b0d93ad5 CL |
985 | if (xprt->shutdown || xprt->addr.sin_port == 0) |
986 | goto out; | |
9903cd1c | 987 | |
b0d93ad5 | 988 | dprintk("RPC: xs_udp_connect_worker for xprt %p\n", xprt); |
9903cd1c | 989 | |
b0d93ad5 CL |
990 | /* Start by resetting any existing state */ |
991 | xs_close(xprt); | |
9903cd1c | 992 | |
b0d93ad5 CL |
993 | if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) { |
994 | dprintk("RPC: can't create UDP transport socket (%d).\n", -err); | |
995 | goto out; | |
996 | } | |
9903cd1c | 997 | |
b0d93ad5 CL |
998 | if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) { |
999 | sock_release(sock); | |
1000 | goto out; | |
1001 | } | |
9903cd1c | 1002 | |
b0d93ad5 CL |
1003 | if (!xprt->inet) { |
1004 | struct sock *sk = sock->sk; | |
a246b010 | 1005 | |
b0d93ad5 | 1006 | write_lock_bh(&sk->sk_callback_lock); |
a246b010 | 1007 | |
b0d93ad5 CL |
1008 | sk->sk_user_data = xprt; |
1009 | xprt->old_data_ready = sk->sk_data_ready; | |
1010 | xprt->old_state_change = sk->sk_state_change; | |
1011 | xprt->old_write_space = sk->sk_write_space; | |
9903cd1c | 1012 | sk->sk_data_ready = xs_udp_data_ready; |
c7b2cae8 | 1013 | sk->sk_write_space = xs_udp_write_space; |
a246b010 | 1014 | sk->sk_no_check = UDP_CSUM_NORCV; |
b079fa7b | 1015 | sk->sk_allocation = GFP_ATOMIC; |
b0d93ad5 | 1016 | |
a246b010 | 1017 | xprt_set_connected(xprt); |
a246b010 | 1018 | |
b0d93ad5 CL |
1019 | /* Reset to new socket */ |
1020 | xprt->sock = sock; | |
1021 | xprt->inet = sk; | |
a246b010 | 1022 | |
b0d93ad5 CL |
1023 | write_unlock_bh(&sk->sk_callback_lock); |
1024 | } | |
470056c2 | 1025 | xs_udp_do_set_buffer_size(xprt); |
b0d93ad5 CL |
1026 | status = 0; |
1027 | out: | |
1028 | xprt_wake_pending_tasks(xprt, status); | |
1029 | xprt_clear_connecting(xprt); | |
a246b010 CL |
1030 | } |
1031 | ||
3167e12c CL |
1032 | /* |
1033 | * We need to preserve the port number so the reply cache on the server can | |
1034 | * find our cached RPC replies when we get around to reconnecting. | |
1035 | */ | |
1036 | static void xs_tcp_reuse_connection(struct rpc_xprt *xprt) | |
1037 | { | |
1038 | int result; | |
1039 | struct socket *sock = xprt->sock; | |
1040 | struct sockaddr any; | |
1041 | ||
1042 | dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt); | |
1043 | ||
1044 | /* | |
1045 | * Disconnect the transport socket by doing a connect operation | |
1046 | * with AF_UNSPEC. This should return immediately... | |
1047 | */ | |
1048 | memset(&any, 0, sizeof(any)); | |
1049 | any.sa_family = AF_UNSPEC; | |
1050 | result = sock->ops->connect(sock, &any, sizeof(any), 0); | |
1051 | if (result) | |
1052 | dprintk("RPC: AF_UNSPEC connect return code %d\n", | |
1053 | result); | |
1054 | } | |
1055 | ||
9903cd1c | 1056 | /** |
b0d93ad5 | 1057 | * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint |
9903cd1c CL |
1058 | * @args: RPC transport to connect |
1059 | * | |
1060 | * Invoked by a work queue tasklet. | |
a246b010 | 1061 | */ |
b0d93ad5 | 1062 | static void xs_tcp_connect_worker(void *args) |
a246b010 CL |
1063 | { |
1064 | struct rpc_xprt *xprt = (struct rpc_xprt *)args; | |
1065 | struct socket *sock = xprt->sock; | |
b0d93ad5 | 1066 | int err, status = -EIO; |
a246b010 CL |
1067 | |
1068 | if (xprt->shutdown || xprt->addr.sin_port == 0) | |
1069 | goto out; | |
1070 | ||
b0d93ad5 | 1071 | dprintk("RPC: xs_tcp_connect_worker for xprt %p\n", xprt); |
9903cd1c | 1072 | |
3167e12c CL |
1073 | if (!xprt->sock) { |
1074 | /* start from scratch */ | |
1075 | if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) { | |
1076 | dprintk("RPC: can't create TCP transport socket (%d).\n", -err); | |
1077 | goto out; | |
1078 | } | |
a246b010 | 1079 | |
3167e12c CL |
1080 | if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) { |
1081 | sock_release(sock); | |
1082 | goto out; | |
1083 | } | |
1084 | } else | |
1085 | /* "close" the socket, preserving the local port */ | |
1086 | xs_tcp_reuse_connection(xprt); | |
a246b010 | 1087 | |
b0d93ad5 CL |
1088 | if (!xprt->inet) { |
1089 | struct sock *sk = sock->sk; | |
1090 | ||
1091 | write_lock_bh(&sk->sk_callback_lock); | |
1092 | ||
1093 | sk->sk_user_data = xprt; | |
1094 | xprt->old_data_ready = sk->sk_data_ready; | |
1095 | xprt->old_state_change = sk->sk_state_change; | |
1096 | xprt->old_write_space = sk->sk_write_space; | |
1097 | sk->sk_data_ready = xs_tcp_data_ready; | |
1098 | sk->sk_state_change = xs_tcp_state_change; | |
1099 | sk->sk_write_space = xs_tcp_write_space; | |
b079fa7b | 1100 | sk->sk_allocation = GFP_ATOMIC; |
3167e12c CL |
1101 | |
1102 | /* socket options */ | |
1103 | sk->sk_userlocks |= SOCK_BINDPORT_LOCK; | |
1104 | sock_reset_flag(sk, SOCK_LINGER); | |
1105 | tcp_sk(sk)->linger2 = 0; | |
1106 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; | |
b0d93ad5 CL |
1107 | |
1108 | xprt_clear_connected(xprt); | |
1109 | ||
1110 | /* Reset to new socket */ | |
1111 | xprt->sock = sock; | |
1112 | xprt->inet = sk; | |
1113 | ||
1114 | write_unlock_bh(&sk->sk_callback_lock); | |
1115 | } | |
1116 | ||
1117 | /* Tell the socket layer to start connecting... */ | |
262ca07d CL |
1118 | xprt->stat.connect_count++; |
1119 | xprt->stat.connect_start = jiffies; | |
a246b010 CL |
1120 | status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr, |
1121 | sizeof(xprt->addr), O_NONBLOCK); | |
1122 | dprintk("RPC: %p connect status %d connected %d sock state %d\n", | |
1123 | xprt, -status, xprt_connected(xprt), sock->sk->sk_state); | |
1124 | if (status < 0) { | |
1125 | switch (status) { | |
1126 | case -EINPROGRESS: | |
1127 | case -EALREADY: | |
1128 | goto out_clear; | |
3167e12c CL |
1129 | case -ECONNREFUSED: |
1130 | case -ECONNRESET: | |
1131 | /* retry with existing socket, after a delay */ | |
1132 | break; | |
1133 | default: | |
1134 | /* get rid of existing socket, and retry */ | |
1135 | xs_close(xprt); | |
1136 | break; | |
a246b010 CL |
1137 | } |
1138 | } | |
1139 | out: | |
44fbac22 | 1140 | xprt_wake_pending_tasks(xprt, status); |
a246b010 | 1141 | out_clear: |
2226feb6 | 1142 | xprt_clear_connecting(xprt); |
a246b010 CL |
1143 | } |
1144 | ||
9903cd1c CL |
1145 | /** |
1146 | * xs_connect - connect a socket to a remote endpoint | |
1147 | * @task: address of RPC task that manages state of connect request | |
1148 | * | |
1149 | * TCP: If the remote end dropped the connection, delay reconnecting. | |
03bf4b70 CL |
1150 | * |
1151 | * UDP socket connects are synchronous, but we use a work queue anyway | |
1152 | * to guarantee that even unprivileged user processes can set up a | |
1153 | * socket on a privileged port. | |
1154 | * | |
1155 | * If a UDP socket connect fails, the delay behavior here prevents | |
1156 | * retry floods (hard mounts). | |
9903cd1c CL |
1157 | */ |
1158 | static void xs_connect(struct rpc_task *task) | |
a246b010 CL |
1159 | { |
1160 | struct rpc_xprt *xprt = task->tk_xprt; | |
1161 | ||
b0d93ad5 CL |
1162 | if (xprt_test_and_set_connecting(xprt)) |
1163 | return; | |
1164 | ||
1165 | if (xprt->sock != NULL) { | |
03bf4b70 CL |
1166 | dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n", |
1167 | xprt, xprt->reestablish_timeout / HZ); | |
b0d93ad5 | 1168 | schedule_delayed_work(&xprt->connect_worker, |
03bf4b70 CL |
1169 | xprt->reestablish_timeout); |
1170 | xprt->reestablish_timeout <<= 1; | |
1171 | if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) | |
1172 | xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; | |
b0d93ad5 CL |
1173 | } else { |
1174 | dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); | |
1175 | schedule_work(&xprt->connect_worker); | |
1176 | ||
1177 | /* flush_scheduled_work can sleep... */ | |
1178 | if (!RPC_IS_ASYNC(task)) | |
1179 | flush_scheduled_work(); | |
a246b010 CL |
1180 | } |
1181 | } | |
1182 | ||
262ca07d CL |
1183 | /** |
1184 | * xs_udp_print_stats - display UDP socket-specifc stats | |
1185 | * @xprt: rpc_xprt struct containing statistics | |
1186 | * @seq: output file | |
1187 | * | |
1188 | */ | |
1189 | static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1190 | { | |
1191 | seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n", | |
1192 | xprt->port, | |
1193 | xprt->stat.bind_count, | |
1194 | xprt->stat.sends, | |
1195 | xprt->stat.recvs, | |
1196 | xprt->stat.bad_xids, | |
1197 | xprt->stat.req_u, | |
1198 | xprt->stat.bklog_u); | |
1199 | } | |
1200 | ||
1201 | /** | |
1202 | * xs_tcp_print_stats - display TCP socket-specifc stats | |
1203 | * @xprt: rpc_xprt struct containing statistics | |
1204 | * @seq: output file | |
1205 | * | |
1206 | */ | |
1207 | static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1208 | { | |
1209 | long idle_time = 0; | |
1210 | ||
1211 | if (xprt_connected(xprt)) | |
1212 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
1213 | ||
1214 | seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n", | |
1215 | xprt->port, | |
1216 | xprt->stat.bind_count, | |
1217 | xprt->stat.connect_count, | |
1218 | xprt->stat.connect_time, | |
1219 | idle_time, | |
1220 | xprt->stat.sends, | |
1221 | xprt->stat.recvs, | |
1222 | xprt->stat.bad_xids, | |
1223 | xprt->stat.req_u, | |
1224 | xprt->stat.bklog_u); | |
1225 | } | |
1226 | ||
262965f5 | 1227 | static struct rpc_xprt_ops xs_udp_ops = { |
43118c29 | 1228 | .set_buffer_size = xs_udp_set_buffer_size, |
12a80469 | 1229 | .reserve_xprt = xprt_reserve_xprt_cong, |
49e9a890 | 1230 | .release_xprt = xprt_release_xprt_cong, |
92200412 | 1231 | .set_port = xs_set_port, |
262965f5 | 1232 | .connect = xs_connect, |
02107148 CL |
1233 | .buf_alloc = rpc_malloc, |
1234 | .buf_free = rpc_free, | |
262965f5 | 1235 | .send_request = xs_udp_send_request, |
fe3aca29 | 1236 | .set_retrans_timeout = xprt_set_retrans_timeout_rtt, |
46c0ee8b | 1237 | .timer = xs_udp_timer, |
a58dd398 | 1238 | .release_request = xprt_release_rqst_cong, |
262965f5 CL |
1239 | .close = xs_close, |
1240 | .destroy = xs_destroy, | |
262ca07d | 1241 | .print_stats = xs_udp_print_stats, |
262965f5 CL |
1242 | }; |
1243 | ||
1244 | static struct rpc_xprt_ops xs_tcp_ops = { | |
12a80469 | 1245 | .reserve_xprt = xprt_reserve_xprt, |
49e9a890 | 1246 | .release_xprt = xprt_release_xprt, |
92200412 | 1247 | .set_port = xs_set_port, |
9903cd1c | 1248 | .connect = xs_connect, |
02107148 CL |
1249 | .buf_alloc = rpc_malloc, |
1250 | .buf_free = rpc_free, | |
262965f5 | 1251 | .send_request = xs_tcp_send_request, |
fe3aca29 | 1252 | .set_retrans_timeout = xprt_set_retrans_timeout_def, |
9903cd1c CL |
1253 | .close = xs_close, |
1254 | .destroy = xs_destroy, | |
262ca07d | 1255 | .print_stats = xs_tcp_print_stats, |
a246b010 CL |
1256 | }; |
1257 | ||
9903cd1c CL |
1258 | /** |
1259 | * xs_setup_udp - Set up transport to use a UDP socket | |
1260 | * @xprt: transport to set up | |
1261 | * @to: timeout parameters | |
1262 | * | |
1263 | */ | |
a246b010 CL |
1264 | int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
1265 | { | |
1266 | size_t slot_table_size; | |
1267 | ||
1268 | dprintk("RPC: setting up udp-ipv4 transport...\n"); | |
1269 | ||
1270 | xprt->max_reqs = xprt_udp_slot_table_entries; | |
1271 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
1272 | xprt->slot = kmalloc(slot_table_size, GFP_KERNEL); | |
1273 | if (xprt->slot == NULL) | |
1274 | return -ENOMEM; | |
1275 | memset(xprt->slot, 0, slot_table_size); | |
1276 | ||
1277 | xprt->prot = IPPROTO_UDP; | |
529b33c6 | 1278 | xprt->port = xprt_max_resvport; |
808012fb | 1279 | xprt->tsh_size = 0; |
a246b010 CL |
1280 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; |
1281 | /* XXX: header size can vary due to auth type, IPv6, etc. */ | |
1282 | xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); | |
1283 | ||
b0d93ad5 | 1284 | INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt); |
03bf4b70 CL |
1285 | xprt->bind_timeout = XS_BIND_TO; |
1286 | xprt->connect_timeout = XS_UDP_CONN_TO; | |
1287 | xprt->reestablish_timeout = XS_UDP_REEST_TO; | |
1288 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 1289 | |
262965f5 | 1290 | xprt->ops = &xs_udp_ops; |
a246b010 CL |
1291 | |
1292 | if (to) | |
1293 | xprt->timeout = *to; | |
1294 | else | |
9903cd1c | 1295 | xprt_set_timeout(&xprt->timeout, 5, 5 * HZ); |
a246b010 CL |
1296 | |
1297 | return 0; | |
1298 | } | |
1299 | ||
9903cd1c CL |
1300 | /** |
1301 | * xs_setup_tcp - Set up transport to use a TCP socket | |
1302 | * @xprt: transport to set up | |
1303 | * @to: timeout parameters | |
1304 | * | |
1305 | */ | |
a246b010 CL |
1306 | int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
1307 | { | |
1308 | size_t slot_table_size; | |
1309 | ||
1310 | dprintk("RPC: setting up tcp-ipv4 transport...\n"); | |
1311 | ||
1312 | xprt->max_reqs = xprt_tcp_slot_table_entries; | |
1313 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
1314 | xprt->slot = kmalloc(slot_table_size, GFP_KERNEL); | |
1315 | if (xprt->slot == NULL) | |
1316 | return -ENOMEM; | |
1317 | memset(xprt->slot, 0, slot_table_size); | |
1318 | ||
1319 | xprt->prot = IPPROTO_TCP; | |
529b33c6 | 1320 | xprt->port = xprt_max_resvport; |
808012fb | 1321 | xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); |
a246b010 | 1322 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; |
808012fb | 1323 | xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; |
a246b010 | 1324 | |
b0d93ad5 | 1325 | INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt); |
03bf4b70 CL |
1326 | xprt->bind_timeout = XS_BIND_TO; |
1327 | xprt->connect_timeout = XS_TCP_CONN_TO; | |
1328 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
1329 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 1330 | |
262965f5 | 1331 | xprt->ops = &xs_tcp_ops; |
a246b010 CL |
1332 | |
1333 | if (to) | |
1334 | xprt->timeout = *to; | |
1335 | else | |
9903cd1c | 1336 | xprt_set_timeout(&xprt->timeout, 2, 60 * HZ); |
a246b010 CL |
1337 | |
1338 | return 0; | |
1339 | } |