]>
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 | ||
e0ab53de TM |
416 | /** |
417 | * xs_tcp_release_xprt - clean up after a tcp transmission | |
418 | * @xprt: transport | |
419 | * @task: rpc task | |
420 | * | |
421 | * This cleans up if an error causes us to abort the transmission of a request. | |
422 | * In this case, the socket may need to be reset in order to avoid confusing | |
423 | * the server. | |
424 | */ | |
425 | static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) | |
426 | { | |
427 | struct rpc_rqst *req; | |
428 | ||
429 | if (task != xprt->snd_task) | |
430 | return; | |
431 | if (task == NULL) | |
432 | goto out_release; | |
433 | req = task->tk_rqstp; | |
434 | if (req->rq_bytes_sent == 0) | |
435 | goto out_release; | |
436 | if (req->rq_bytes_sent == req->rq_snd_buf.len) | |
437 | goto out_release; | |
438 | set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state); | |
439 | out_release: | |
440 | xprt_release_xprt(xprt, task); | |
441 | } | |
442 | ||
9903cd1c CL |
443 | /** |
444 | * xs_close - close a socket | |
445 | * @xprt: transport | |
446 | * | |
3167e12c CL |
447 | * This is used when all requests are complete; ie, no DRC state remains |
448 | * on the server we want to save. | |
a246b010 | 449 | */ |
9903cd1c | 450 | static void xs_close(struct rpc_xprt *xprt) |
a246b010 | 451 | { |
9903cd1c CL |
452 | struct socket *sock = xprt->sock; |
453 | struct sock *sk = xprt->inet; | |
a246b010 CL |
454 | |
455 | if (!sk) | |
632e3bdc | 456 | goto clear_close_wait; |
a246b010 | 457 | |
9903cd1c CL |
458 | dprintk("RPC: xs_close xprt %p\n", xprt); |
459 | ||
a246b010 CL |
460 | write_lock_bh(&sk->sk_callback_lock); |
461 | xprt->inet = NULL; | |
462 | xprt->sock = NULL; | |
463 | ||
9903cd1c CL |
464 | sk->sk_user_data = NULL; |
465 | sk->sk_data_ready = xprt->old_data_ready; | |
a246b010 | 466 | sk->sk_state_change = xprt->old_state_change; |
9903cd1c | 467 | sk->sk_write_space = xprt->old_write_space; |
a246b010 CL |
468 | write_unlock_bh(&sk->sk_callback_lock); |
469 | ||
9903cd1c | 470 | sk->sk_no_check = 0; |
a246b010 CL |
471 | |
472 | sock_release(sock); | |
632e3bdc TM |
473 | clear_close_wait: |
474 | smp_mb__before_clear_bit(); | |
475 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
476 | smp_mb__after_clear_bit(); | |
a246b010 CL |
477 | } |
478 | ||
9903cd1c CL |
479 | /** |
480 | * xs_destroy - prepare to shutdown a transport | |
481 | * @xprt: doomed transport | |
482 | * | |
483 | */ | |
484 | static void xs_destroy(struct rpc_xprt *xprt) | |
a246b010 | 485 | { |
9903cd1c CL |
486 | dprintk("RPC: xs_destroy xprt %p\n", xprt); |
487 | ||
55aa4f58 | 488 | cancel_delayed_work(&xprt->connect_worker); |
a246b010 CL |
489 | flush_scheduled_work(); |
490 | ||
491 | xprt_disconnect(xprt); | |
9903cd1c | 492 | xs_close(xprt); |
a246b010 CL |
493 | kfree(xprt->slot); |
494 | } | |
495 | ||
9903cd1c CL |
496 | static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) |
497 | { | |
498 | return (struct rpc_xprt *) sk->sk_user_data; | |
499 | } | |
500 | ||
501 | /** | |
502 | * xs_udp_data_ready - "data ready" callback for UDP sockets | |
503 | * @sk: socket with data to read | |
504 | * @len: how much data to read | |
505 | * | |
a246b010 | 506 | */ |
9903cd1c | 507 | static void xs_udp_data_ready(struct sock *sk, int len) |
a246b010 | 508 | { |
9903cd1c CL |
509 | struct rpc_task *task; |
510 | struct rpc_xprt *xprt; | |
a246b010 | 511 | struct rpc_rqst *rovr; |
9903cd1c | 512 | struct sk_buff *skb; |
a246b010 CL |
513 | int err, repsize, copied; |
514 | u32 _xid, *xp; | |
515 | ||
516 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
517 | dprintk("RPC: xs_udp_data_ready...\n"); |
518 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 519 | goto out; |
a246b010 CL |
520 | |
521 | if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) | |
522 | goto out; | |
523 | ||
524 | if (xprt->shutdown) | |
525 | goto dropit; | |
526 | ||
527 | repsize = skb->len - sizeof(struct udphdr); | |
528 | if (repsize < 4) { | |
9903cd1c | 529 | dprintk("RPC: impossible RPC reply size %d!\n", repsize); |
a246b010 CL |
530 | goto dropit; |
531 | } | |
532 | ||
533 | /* Copy the XID from the skb... */ | |
534 | xp = skb_header_pointer(skb, sizeof(struct udphdr), | |
535 | sizeof(_xid), &_xid); | |
536 | if (xp == NULL) | |
537 | goto dropit; | |
538 | ||
539 | /* Look up and lock the request corresponding to the given XID */ | |
4a0f8c04 | 540 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
541 | rovr = xprt_lookup_rqst(xprt, *xp); |
542 | if (!rovr) | |
543 | goto out_unlock; | |
544 | task = rovr->rq_task; | |
545 | ||
a246b010 CL |
546 | if ((copied = rovr->rq_private_buf.buflen) > repsize) |
547 | copied = repsize; | |
548 | ||
549 | /* Suck it into the iovec, verify checksum if not done by hw. */ | |
550 | if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) | |
551 | goto out_unlock; | |
552 | ||
553 | /* Something worked... */ | |
554 | dst_confirm(skb->dst); | |
555 | ||
1570c1e4 CL |
556 | xprt_adjust_cwnd(task, copied); |
557 | xprt_update_rtt(task); | |
558 | xprt_complete_rqst(task, copied); | |
a246b010 CL |
559 | |
560 | out_unlock: | |
4a0f8c04 | 561 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
562 | dropit: |
563 | skb_free_datagram(sk, skb); | |
564 | out: | |
565 | read_unlock(&sk->sk_callback_lock); | |
566 | } | |
567 | ||
9903cd1c | 568 | static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len) |
a246b010 CL |
569 | { |
570 | if (len > desc->count) | |
571 | len = desc->count; | |
572 | if (skb_copy_bits(desc->skb, desc->offset, p, len)) { | |
573 | dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n", | |
574 | len, desc->count); | |
575 | return 0; | |
576 | } | |
577 | desc->offset += len; | |
578 | desc->count -= len; | |
579 | dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n", | |
580 | len, desc->count); | |
581 | return len; | |
582 | } | |
583 | ||
9903cd1c | 584 | static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
585 | { |
586 | size_t len, used; | |
587 | char *p; | |
588 | ||
589 | p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset; | |
590 | len = sizeof(xprt->tcp_recm) - xprt->tcp_offset; | |
9903cd1c | 591 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
592 | xprt->tcp_offset += used; |
593 | if (used != len) | |
594 | return; | |
808012fb | 595 | |
a246b010 | 596 | xprt->tcp_reclen = ntohl(xprt->tcp_recm); |
808012fb | 597 | if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) |
a246b010 CL |
598 | xprt->tcp_flags |= XPRT_LAST_FRAG; |
599 | else | |
600 | xprt->tcp_flags &= ~XPRT_LAST_FRAG; | |
808012fb CL |
601 | xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; |
602 | ||
a246b010 CL |
603 | xprt->tcp_flags &= ~XPRT_COPY_RECM; |
604 | xprt->tcp_offset = 0; | |
808012fb | 605 | |
a246b010 | 606 | /* Sanity check of the record length */ |
808012fb | 607 | if (unlikely(xprt->tcp_reclen < 4)) { |
9903cd1c | 608 | dprintk("RPC: invalid TCP record fragment length\n"); |
a246b010 | 609 | xprt_disconnect(xprt); |
9903cd1c | 610 | return; |
a246b010 CL |
611 | } |
612 | dprintk("RPC: reading TCP record fragment of length %d\n", | |
613 | xprt->tcp_reclen); | |
614 | } | |
615 | ||
9903cd1c | 616 | static void xs_tcp_check_recm(struct rpc_xprt *xprt) |
a246b010 CL |
617 | { |
618 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n", | |
619 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags); | |
620 | if (xprt->tcp_offset == xprt->tcp_reclen) { | |
621 | xprt->tcp_flags |= XPRT_COPY_RECM; | |
622 | xprt->tcp_offset = 0; | |
623 | if (xprt->tcp_flags & XPRT_LAST_FRAG) { | |
624 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
625 | xprt->tcp_flags |= XPRT_COPY_XID; | |
626 | xprt->tcp_copied = 0; | |
627 | } | |
628 | } | |
629 | } | |
630 | ||
9903cd1c | 631 | static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
632 | { |
633 | size_t len, used; | |
634 | char *p; | |
635 | ||
636 | len = sizeof(xprt->tcp_xid) - xprt->tcp_offset; | |
637 | dprintk("RPC: reading XID (%Zu bytes)\n", len); | |
638 | p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset; | |
9903cd1c | 639 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
640 | xprt->tcp_offset += used; |
641 | if (used != len) | |
642 | return; | |
643 | xprt->tcp_flags &= ~XPRT_COPY_XID; | |
644 | xprt->tcp_flags |= XPRT_COPY_DATA; | |
645 | xprt->tcp_copied = 4; | |
646 | dprintk("RPC: reading reply for XID %08x\n", | |
647 | ntohl(xprt->tcp_xid)); | |
9903cd1c | 648 | xs_tcp_check_recm(xprt); |
a246b010 CL |
649 | } |
650 | ||
9903cd1c | 651 | static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
652 | { |
653 | struct rpc_rqst *req; | |
654 | struct xdr_buf *rcvbuf; | |
655 | size_t len; | |
656 | ssize_t r; | |
657 | ||
658 | /* Find and lock the request corresponding to this xid */ | |
4a0f8c04 | 659 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
660 | req = xprt_lookup_rqst(xprt, xprt->tcp_xid); |
661 | if (!req) { | |
662 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
663 | dprintk("RPC: XID %08x request not found!\n", | |
664 | ntohl(xprt->tcp_xid)); | |
4a0f8c04 | 665 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
666 | return; |
667 | } | |
668 | ||
669 | rcvbuf = &req->rq_private_buf; | |
670 | len = desc->count; | |
671 | if (len > xprt->tcp_reclen - xprt->tcp_offset) { | |
672 | skb_reader_t my_desc; | |
673 | ||
674 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
675 | memcpy(&my_desc, desc, sizeof(my_desc)); | |
676 | my_desc.count = len; | |
677 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 678 | &my_desc, xs_tcp_copy_data); |
a246b010 CL |
679 | desc->count -= r; |
680 | desc->offset += r; | |
681 | } else | |
682 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 683 | desc, xs_tcp_copy_data); |
a246b010 CL |
684 | |
685 | if (r > 0) { | |
686 | xprt->tcp_copied += r; | |
687 | xprt->tcp_offset += r; | |
688 | } | |
689 | if (r != len) { | |
690 | /* Error when copying to the receive buffer, | |
691 | * usually because we weren't able to allocate | |
692 | * additional buffer pages. All we can do now | |
693 | * is turn off XPRT_COPY_DATA, so the request | |
694 | * will not receive any additional updates, | |
695 | * and time out. | |
696 | * Any remaining data from this record will | |
697 | * be discarded. | |
698 | */ | |
699 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
700 | dprintk("RPC: XID %08x truncated request\n", | |
701 | ntohl(xprt->tcp_xid)); | |
702 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
703 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
704 | goto out; | |
705 | } | |
706 | ||
707 | dprintk("RPC: XID %08x read %Zd bytes\n", | |
708 | ntohl(xprt->tcp_xid), r); | |
709 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
710 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
711 | ||
712 | if (xprt->tcp_copied == req->rq_private_buf.buflen) | |
713 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
714 | else if (xprt->tcp_offset == xprt->tcp_reclen) { | |
715 | if (xprt->tcp_flags & XPRT_LAST_FRAG) | |
716 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
717 | } | |
718 | ||
719 | out: | |
1570c1e4 CL |
720 | if (!(xprt->tcp_flags & XPRT_COPY_DATA)) |
721 | xprt_complete_rqst(req->rq_task, xprt->tcp_copied); | |
4a0f8c04 | 722 | spin_unlock(&xprt->transport_lock); |
9903cd1c | 723 | xs_tcp_check_recm(xprt); |
a246b010 CL |
724 | } |
725 | ||
9903cd1c | 726 | static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
727 | { |
728 | size_t len; | |
729 | ||
730 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
731 | if (len > desc->count) | |
732 | len = desc->count; | |
733 | desc->count -= len; | |
734 | desc->offset += len; | |
735 | xprt->tcp_offset += len; | |
736 | dprintk("RPC: discarded %Zu bytes\n", len); | |
9903cd1c | 737 | xs_tcp_check_recm(xprt); |
a246b010 CL |
738 | } |
739 | ||
9903cd1c | 740 | static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) |
a246b010 CL |
741 | { |
742 | struct rpc_xprt *xprt = rd_desc->arg.data; | |
743 | skb_reader_t desc = { | |
744 | .skb = skb, | |
745 | .offset = offset, | |
746 | .count = len, | |
747 | .csum = 0 | |
9903cd1c | 748 | }; |
a246b010 | 749 | |
9903cd1c | 750 | dprintk("RPC: xs_tcp_data_recv started\n"); |
a246b010 CL |
751 | do { |
752 | /* Read in a new fragment marker if necessary */ | |
753 | /* Can we ever really expect to get completely empty fragments? */ | |
754 | if (xprt->tcp_flags & XPRT_COPY_RECM) { | |
9903cd1c | 755 | xs_tcp_read_fraghdr(xprt, &desc); |
a246b010 CL |
756 | continue; |
757 | } | |
758 | /* Read in the xid if necessary */ | |
759 | if (xprt->tcp_flags & XPRT_COPY_XID) { | |
9903cd1c | 760 | xs_tcp_read_xid(xprt, &desc); |
a246b010 CL |
761 | continue; |
762 | } | |
763 | /* Read in the request data */ | |
764 | if (xprt->tcp_flags & XPRT_COPY_DATA) { | |
9903cd1c | 765 | xs_tcp_read_request(xprt, &desc); |
a246b010 CL |
766 | continue; |
767 | } | |
768 | /* Skip over any trailing bytes on short reads */ | |
9903cd1c | 769 | xs_tcp_read_discard(xprt, &desc); |
a246b010 | 770 | } while (desc.count); |
9903cd1c | 771 | dprintk("RPC: xs_tcp_data_recv done\n"); |
a246b010 CL |
772 | return len - desc.count; |
773 | } | |
774 | ||
9903cd1c CL |
775 | /** |
776 | * xs_tcp_data_ready - "data ready" callback for TCP sockets | |
777 | * @sk: socket with data to read | |
778 | * @bytes: how much data to read | |
779 | * | |
780 | */ | |
781 | static void xs_tcp_data_ready(struct sock *sk, int bytes) | |
a246b010 CL |
782 | { |
783 | struct rpc_xprt *xprt; | |
784 | read_descriptor_t rd_desc; | |
785 | ||
786 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
787 | dprintk("RPC: xs_tcp_data_ready...\n"); |
788 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 789 | goto out; |
a246b010 CL |
790 | if (xprt->shutdown) |
791 | goto out; | |
792 | ||
9903cd1c | 793 | /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ |
a246b010 CL |
794 | rd_desc.arg.data = xprt; |
795 | rd_desc.count = 65536; | |
9903cd1c | 796 | tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); |
a246b010 CL |
797 | out: |
798 | read_unlock(&sk->sk_callback_lock); | |
799 | } | |
800 | ||
9903cd1c CL |
801 | /** |
802 | * xs_tcp_state_change - callback to handle TCP socket state changes | |
803 | * @sk: socket whose state has changed | |
804 | * | |
805 | */ | |
806 | static void xs_tcp_state_change(struct sock *sk) | |
a246b010 | 807 | { |
9903cd1c | 808 | struct rpc_xprt *xprt; |
a246b010 CL |
809 | |
810 | read_lock(&sk->sk_callback_lock); | |
811 | if (!(xprt = xprt_from_sock(sk))) | |
812 | goto out; | |
9903cd1c | 813 | dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); |
a246b010 CL |
814 | dprintk("RPC: state %x conn %d dead %d zapped %d\n", |
815 | sk->sk_state, xprt_connected(xprt), | |
816 | sock_flag(sk, SOCK_DEAD), | |
817 | sock_flag(sk, SOCK_ZAPPED)); | |
818 | ||
819 | switch (sk->sk_state) { | |
820 | case TCP_ESTABLISHED: | |
4a0f8c04 | 821 | spin_lock_bh(&xprt->transport_lock); |
a246b010 CL |
822 | if (!xprt_test_and_set_connected(xprt)) { |
823 | /* Reset TCP record info */ | |
824 | xprt->tcp_offset = 0; | |
825 | xprt->tcp_reclen = 0; | |
826 | xprt->tcp_copied = 0; | |
827 | xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID; | |
03bf4b70 | 828 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; |
44fbac22 | 829 | xprt_wake_pending_tasks(xprt, 0); |
a246b010 | 830 | } |
4a0f8c04 | 831 | spin_unlock_bh(&xprt->transport_lock); |
a246b010 CL |
832 | break; |
833 | case TCP_SYN_SENT: | |
834 | case TCP_SYN_RECV: | |
835 | break; | |
632e3bdc TM |
836 | case TCP_CLOSE_WAIT: |
837 | /* Try to schedule an autoclose RPC calls */ | |
838 | set_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
839 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) | |
840 | schedule_work(&xprt->task_cleanup); | |
a246b010 CL |
841 | default: |
842 | xprt_disconnect(xprt); | |
a246b010 CL |
843 | } |
844 | out: | |
845 | read_unlock(&sk->sk_callback_lock); | |
846 | } | |
847 | ||
9903cd1c | 848 | /** |
c7b2cae8 CL |
849 | * xs_udp_write_space - callback invoked when socket buffer space |
850 | * becomes available | |
9903cd1c CL |
851 | * @sk: socket whose state has changed |
852 | * | |
a246b010 CL |
853 | * Called when more output buffer space is available for this socket. |
854 | * We try not to wake our writers until they can make "significant" | |
c7b2cae8 | 855 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg |
a246b010 CL |
856 | * with a bunch of small requests. |
857 | */ | |
c7b2cae8 | 858 | static void xs_udp_write_space(struct sock *sk) |
a246b010 | 859 | { |
a246b010 | 860 | read_lock(&sk->sk_callback_lock); |
a246b010 | 861 | |
c7b2cae8 CL |
862 | /* from net/core/sock.c:sock_def_write_space */ |
863 | if (sock_writeable(sk)) { | |
864 | struct socket *sock; | |
865 | struct rpc_xprt *xprt; | |
866 | ||
867 | if (unlikely(!(sock = sk->sk_socket))) | |
a246b010 | 868 | goto out; |
c7b2cae8 CL |
869 | if (unlikely(!(xprt = xprt_from_sock(sk)))) |
870 | goto out; | |
871 | if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) | |
a246b010 | 872 | goto out; |
c7b2cae8 CL |
873 | |
874 | xprt_write_space(xprt); | |
a246b010 CL |
875 | } |
876 | ||
c7b2cae8 CL |
877 | out: |
878 | read_unlock(&sk->sk_callback_lock); | |
879 | } | |
a246b010 | 880 | |
c7b2cae8 CL |
881 | /** |
882 | * xs_tcp_write_space - callback invoked when socket buffer space | |
883 | * becomes available | |
884 | * @sk: socket whose state has changed | |
885 | * | |
886 | * Called when more output buffer space is available for this socket. | |
887 | * We try not to wake our writers until they can make "significant" | |
888 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg | |
889 | * with a bunch of small requests. | |
890 | */ | |
891 | static void xs_tcp_write_space(struct sock *sk) | |
892 | { | |
893 | read_lock(&sk->sk_callback_lock); | |
894 | ||
895 | /* from net/core/stream.c:sk_stream_write_space */ | |
896 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { | |
897 | struct socket *sock; | |
898 | struct rpc_xprt *xprt; | |
899 | ||
900 | if (unlikely(!(sock = sk->sk_socket))) | |
901 | goto out; | |
902 | if (unlikely(!(xprt = xprt_from_sock(sk)))) | |
903 | goto out; | |
904 | if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) | |
905 | goto out; | |
906 | ||
907 | xprt_write_space(xprt); | |
908 | } | |
909 | ||
910 | out: | |
a246b010 CL |
911 | read_unlock(&sk->sk_callback_lock); |
912 | } | |
913 | ||
470056c2 | 914 | static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) |
a246b010 CL |
915 | { |
916 | struct sock *sk = xprt->inet; | |
917 | ||
a246b010 CL |
918 | if (xprt->rcvsize) { |
919 | sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
920 | sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2; | |
921 | } | |
922 | if (xprt->sndsize) { | |
923 | sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
924 | sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2; | |
925 | sk->sk_write_space(sk); | |
926 | } | |
927 | } | |
928 | ||
43118c29 | 929 | /** |
470056c2 | 930 | * xs_udp_set_buffer_size - set send and receive limits |
43118c29 | 931 | * @xprt: generic transport |
470056c2 CL |
932 | * @sndsize: requested size of send buffer, in bytes |
933 | * @rcvsize: requested size of receive buffer, in bytes | |
43118c29 | 934 | * |
470056c2 | 935 | * Set socket send and receive buffer size limits. |
43118c29 | 936 | */ |
470056c2 | 937 | static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) |
43118c29 | 938 | { |
470056c2 CL |
939 | xprt->sndsize = 0; |
940 | if (sndsize) | |
941 | xprt->sndsize = sndsize + 1024; | |
942 | xprt->rcvsize = 0; | |
943 | if (rcvsize) | |
944 | xprt->rcvsize = rcvsize + 1024; | |
945 | ||
946 | xs_udp_do_set_buffer_size(xprt); | |
43118c29 CL |
947 | } |
948 | ||
46c0ee8b CL |
949 | /** |
950 | * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport | |
951 | * @task: task that timed out | |
952 | * | |
953 | * Adjust the congestion window after a retransmit timeout has occurred. | |
954 | */ | |
955 | static void xs_udp_timer(struct rpc_task *task) | |
956 | { | |
957 | xprt_adjust_cwnd(task, -ETIMEDOUT); | |
958 | } | |
959 | ||
b85d8806 CL |
960 | static unsigned short xs_get_random_port(void) |
961 | { | |
962 | unsigned short range = xprt_max_resvport - xprt_min_resvport; | |
963 | unsigned short rand = (unsigned short) net_random() % range; | |
964 | return rand + xprt_min_resvport; | |
965 | } | |
966 | ||
92200412 CL |
967 | /** |
968 | * xs_set_port - reset the port number in the remote endpoint address | |
969 | * @xprt: generic transport | |
970 | * @port: new port number | |
971 | * | |
972 | */ | |
973 | static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) | |
974 | { | |
975 | dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); | |
976 | xprt->addr.sin_port = htons(port); | |
977 | } | |
978 | ||
9903cd1c | 979 | static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock) |
a246b010 CL |
980 | { |
981 | struct sockaddr_in myaddr = { | |
982 | .sin_family = AF_INET, | |
983 | }; | |
529b33c6 CL |
984 | int err; |
985 | unsigned short port = xprt->port; | |
a246b010 | 986 | |
a246b010 CL |
987 | do { |
988 | myaddr.sin_port = htons(port); | |
989 | err = sock->ops->bind(sock, (struct sockaddr *) &myaddr, | |
990 | sizeof(myaddr)); | |
991 | if (err == 0) { | |
992 | xprt->port = port; | |
9903cd1c CL |
993 | dprintk("RPC: xs_bindresvport bound to port %u\n", |
994 | port); | |
a246b010 CL |
995 | return 0; |
996 | } | |
529b33c6 CL |
997 | if (port <= xprt_min_resvport) |
998 | port = xprt_max_resvport; | |
999 | else | |
1000 | port--; | |
a246b010 CL |
1001 | } while (err == -EADDRINUSE && port != xprt->port); |
1002 | ||
9903cd1c | 1003 | dprintk("RPC: can't bind to reserved port (%d).\n", -err); |
a246b010 CL |
1004 | return err; |
1005 | } | |
1006 | ||
b0d93ad5 CL |
1007 | /** |
1008 | * xs_udp_connect_worker - set up a UDP socket | |
1009 | * @args: RPC transport to connect | |
1010 | * | |
1011 | * Invoked by a work queue tasklet. | |
1012 | */ | |
1013 | static void xs_udp_connect_worker(void *args) | |
a246b010 | 1014 | { |
b0d93ad5 CL |
1015 | struct rpc_xprt *xprt = (struct rpc_xprt *) args; |
1016 | struct socket *sock = xprt->sock; | |
1017 | int err, status = -EIO; | |
9903cd1c | 1018 | |
b0d93ad5 CL |
1019 | if (xprt->shutdown || xprt->addr.sin_port == 0) |
1020 | goto out; | |
9903cd1c | 1021 | |
b0d93ad5 | 1022 | dprintk("RPC: xs_udp_connect_worker for xprt %p\n", xprt); |
9903cd1c | 1023 | |
b0d93ad5 CL |
1024 | /* Start by resetting any existing state */ |
1025 | xs_close(xprt); | |
9903cd1c | 1026 | |
b0d93ad5 CL |
1027 | if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) { |
1028 | dprintk("RPC: can't create UDP transport socket (%d).\n", -err); | |
1029 | goto out; | |
1030 | } | |
9903cd1c | 1031 | |
b0d93ad5 CL |
1032 | if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) { |
1033 | sock_release(sock); | |
1034 | goto out; | |
1035 | } | |
9903cd1c | 1036 | |
b0d93ad5 CL |
1037 | if (!xprt->inet) { |
1038 | struct sock *sk = sock->sk; | |
a246b010 | 1039 | |
b0d93ad5 | 1040 | write_lock_bh(&sk->sk_callback_lock); |
a246b010 | 1041 | |
b0d93ad5 CL |
1042 | sk->sk_user_data = xprt; |
1043 | xprt->old_data_ready = sk->sk_data_ready; | |
1044 | xprt->old_state_change = sk->sk_state_change; | |
1045 | xprt->old_write_space = sk->sk_write_space; | |
9903cd1c | 1046 | sk->sk_data_ready = xs_udp_data_ready; |
c7b2cae8 | 1047 | sk->sk_write_space = xs_udp_write_space; |
a246b010 | 1048 | sk->sk_no_check = UDP_CSUM_NORCV; |
b079fa7b | 1049 | sk->sk_allocation = GFP_ATOMIC; |
b0d93ad5 | 1050 | |
a246b010 | 1051 | xprt_set_connected(xprt); |
a246b010 | 1052 | |
b0d93ad5 CL |
1053 | /* Reset to new socket */ |
1054 | xprt->sock = sock; | |
1055 | xprt->inet = sk; | |
a246b010 | 1056 | |
b0d93ad5 CL |
1057 | write_unlock_bh(&sk->sk_callback_lock); |
1058 | } | |
470056c2 | 1059 | xs_udp_do_set_buffer_size(xprt); |
b0d93ad5 CL |
1060 | status = 0; |
1061 | out: | |
1062 | xprt_wake_pending_tasks(xprt, status); | |
1063 | xprt_clear_connecting(xprt); | |
a246b010 CL |
1064 | } |
1065 | ||
3167e12c CL |
1066 | /* |
1067 | * We need to preserve the port number so the reply cache on the server can | |
1068 | * find our cached RPC replies when we get around to reconnecting. | |
1069 | */ | |
1070 | static void xs_tcp_reuse_connection(struct rpc_xprt *xprt) | |
1071 | { | |
1072 | int result; | |
1073 | struct socket *sock = xprt->sock; | |
1074 | struct sockaddr any; | |
1075 | ||
1076 | dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt); | |
1077 | ||
1078 | /* | |
1079 | * Disconnect the transport socket by doing a connect operation | |
1080 | * with AF_UNSPEC. This should return immediately... | |
1081 | */ | |
1082 | memset(&any, 0, sizeof(any)); | |
1083 | any.sa_family = AF_UNSPEC; | |
1084 | result = sock->ops->connect(sock, &any, sizeof(any), 0); | |
1085 | if (result) | |
1086 | dprintk("RPC: AF_UNSPEC connect return code %d\n", | |
1087 | result); | |
1088 | } | |
1089 | ||
9903cd1c | 1090 | /** |
b0d93ad5 | 1091 | * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint |
9903cd1c CL |
1092 | * @args: RPC transport to connect |
1093 | * | |
1094 | * Invoked by a work queue tasklet. | |
a246b010 | 1095 | */ |
b0d93ad5 | 1096 | static void xs_tcp_connect_worker(void *args) |
a246b010 CL |
1097 | { |
1098 | struct rpc_xprt *xprt = (struct rpc_xprt *)args; | |
1099 | struct socket *sock = xprt->sock; | |
b0d93ad5 | 1100 | int err, status = -EIO; |
a246b010 CL |
1101 | |
1102 | if (xprt->shutdown || xprt->addr.sin_port == 0) | |
1103 | goto out; | |
1104 | ||
b0d93ad5 | 1105 | dprintk("RPC: xs_tcp_connect_worker for xprt %p\n", xprt); |
9903cd1c | 1106 | |
3167e12c CL |
1107 | if (!xprt->sock) { |
1108 | /* start from scratch */ | |
1109 | if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) { | |
1110 | dprintk("RPC: can't create TCP transport socket (%d).\n", -err); | |
1111 | goto out; | |
1112 | } | |
a246b010 | 1113 | |
3167e12c CL |
1114 | if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) { |
1115 | sock_release(sock); | |
1116 | goto out; | |
1117 | } | |
1118 | } else | |
1119 | /* "close" the socket, preserving the local port */ | |
1120 | xs_tcp_reuse_connection(xprt); | |
a246b010 | 1121 | |
b0d93ad5 CL |
1122 | if (!xprt->inet) { |
1123 | struct sock *sk = sock->sk; | |
1124 | ||
1125 | write_lock_bh(&sk->sk_callback_lock); | |
1126 | ||
1127 | sk->sk_user_data = xprt; | |
1128 | xprt->old_data_ready = sk->sk_data_ready; | |
1129 | xprt->old_state_change = sk->sk_state_change; | |
1130 | xprt->old_write_space = sk->sk_write_space; | |
1131 | sk->sk_data_ready = xs_tcp_data_ready; | |
1132 | sk->sk_state_change = xs_tcp_state_change; | |
1133 | sk->sk_write_space = xs_tcp_write_space; | |
b079fa7b | 1134 | sk->sk_allocation = GFP_ATOMIC; |
3167e12c CL |
1135 | |
1136 | /* socket options */ | |
1137 | sk->sk_userlocks |= SOCK_BINDPORT_LOCK; | |
1138 | sock_reset_flag(sk, SOCK_LINGER); | |
1139 | tcp_sk(sk)->linger2 = 0; | |
1140 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; | |
b0d93ad5 CL |
1141 | |
1142 | xprt_clear_connected(xprt); | |
1143 | ||
1144 | /* Reset to new socket */ | |
1145 | xprt->sock = sock; | |
1146 | xprt->inet = sk; | |
1147 | ||
1148 | write_unlock_bh(&sk->sk_callback_lock); | |
1149 | } | |
1150 | ||
1151 | /* Tell the socket layer to start connecting... */ | |
262ca07d CL |
1152 | xprt->stat.connect_count++; |
1153 | xprt->stat.connect_start = jiffies; | |
a246b010 CL |
1154 | status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr, |
1155 | sizeof(xprt->addr), O_NONBLOCK); | |
1156 | dprintk("RPC: %p connect status %d connected %d sock state %d\n", | |
1157 | xprt, -status, xprt_connected(xprt), sock->sk->sk_state); | |
1158 | if (status < 0) { | |
1159 | switch (status) { | |
1160 | case -EINPROGRESS: | |
1161 | case -EALREADY: | |
1162 | goto out_clear; | |
3167e12c CL |
1163 | case -ECONNREFUSED: |
1164 | case -ECONNRESET: | |
1165 | /* retry with existing socket, after a delay */ | |
1166 | break; | |
1167 | default: | |
1168 | /* get rid of existing socket, and retry */ | |
1169 | xs_close(xprt); | |
1170 | break; | |
a246b010 CL |
1171 | } |
1172 | } | |
1173 | out: | |
44fbac22 | 1174 | xprt_wake_pending_tasks(xprt, status); |
a246b010 | 1175 | out_clear: |
2226feb6 | 1176 | xprt_clear_connecting(xprt); |
a246b010 CL |
1177 | } |
1178 | ||
9903cd1c CL |
1179 | /** |
1180 | * xs_connect - connect a socket to a remote endpoint | |
1181 | * @task: address of RPC task that manages state of connect request | |
1182 | * | |
1183 | * TCP: If the remote end dropped the connection, delay reconnecting. | |
03bf4b70 CL |
1184 | * |
1185 | * UDP socket connects are synchronous, but we use a work queue anyway | |
1186 | * to guarantee that even unprivileged user processes can set up a | |
1187 | * socket on a privileged port. | |
1188 | * | |
1189 | * If a UDP socket connect fails, the delay behavior here prevents | |
1190 | * retry floods (hard mounts). | |
9903cd1c CL |
1191 | */ |
1192 | static void xs_connect(struct rpc_task *task) | |
a246b010 CL |
1193 | { |
1194 | struct rpc_xprt *xprt = task->tk_xprt; | |
1195 | ||
b0d93ad5 CL |
1196 | if (xprt_test_and_set_connecting(xprt)) |
1197 | return; | |
1198 | ||
1199 | if (xprt->sock != NULL) { | |
03bf4b70 CL |
1200 | dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n", |
1201 | xprt, xprt->reestablish_timeout / HZ); | |
b0d93ad5 | 1202 | schedule_delayed_work(&xprt->connect_worker, |
03bf4b70 CL |
1203 | xprt->reestablish_timeout); |
1204 | xprt->reestablish_timeout <<= 1; | |
1205 | if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) | |
1206 | xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; | |
b0d93ad5 CL |
1207 | } else { |
1208 | dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); | |
1209 | schedule_work(&xprt->connect_worker); | |
1210 | ||
1211 | /* flush_scheduled_work can sleep... */ | |
1212 | if (!RPC_IS_ASYNC(task)) | |
1213 | flush_scheduled_work(); | |
a246b010 CL |
1214 | } |
1215 | } | |
1216 | ||
262ca07d CL |
1217 | /** |
1218 | * xs_udp_print_stats - display UDP socket-specifc stats | |
1219 | * @xprt: rpc_xprt struct containing statistics | |
1220 | * @seq: output file | |
1221 | * | |
1222 | */ | |
1223 | static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1224 | { | |
1225 | seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n", | |
1226 | xprt->port, | |
1227 | xprt->stat.bind_count, | |
1228 | xprt->stat.sends, | |
1229 | xprt->stat.recvs, | |
1230 | xprt->stat.bad_xids, | |
1231 | xprt->stat.req_u, | |
1232 | xprt->stat.bklog_u); | |
1233 | } | |
1234 | ||
1235 | /** | |
1236 | * xs_tcp_print_stats - display TCP socket-specifc stats | |
1237 | * @xprt: rpc_xprt struct containing statistics | |
1238 | * @seq: output file | |
1239 | * | |
1240 | */ | |
1241 | static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1242 | { | |
1243 | long idle_time = 0; | |
1244 | ||
1245 | if (xprt_connected(xprt)) | |
1246 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
1247 | ||
1248 | seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n", | |
1249 | xprt->port, | |
1250 | xprt->stat.bind_count, | |
1251 | xprt->stat.connect_count, | |
1252 | xprt->stat.connect_time, | |
1253 | idle_time, | |
1254 | xprt->stat.sends, | |
1255 | xprt->stat.recvs, | |
1256 | xprt->stat.bad_xids, | |
1257 | xprt->stat.req_u, | |
1258 | xprt->stat.bklog_u); | |
1259 | } | |
1260 | ||
262965f5 | 1261 | static struct rpc_xprt_ops xs_udp_ops = { |
43118c29 | 1262 | .set_buffer_size = xs_udp_set_buffer_size, |
12a80469 | 1263 | .reserve_xprt = xprt_reserve_xprt_cong, |
49e9a890 | 1264 | .release_xprt = xprt_release_xprt_cong, |
92200412 | 1265 | .set_port = xs_set_port, |
262965f5 | 1266 | .connect = xs_connect, |
02107148 CL |
1267 | .buf_alloc = rpc_malloc, |
1268 | .buf_free = rpc_free, | |
262965f5 | 1269 | .send_request = xs_udp_send_request, |
fe3aca29 | 1270 | .set_retrans_timeout = xprt_set_retrans_timeout_rtt, |
46c0ee8b | 1271 | .timer = xs_udp_timer, |
a58dd398 | 1272 | .release_request = xprt_release_rqst_cong, |
262965f5 CL |
1273 | .close = xs_close, |
1274 | .destroy = xs_destroy, | |
262ca07d | 1275 | .print_stats = xs_udp_print_stats, |
262965f5 CL |
1276 | }; |
1277 | ||
1278 | static struct rpc_xprt_ops xs_tcp_ops = { | |
12a80469 | 1279 | .reserve_xprt = xprt_reserve_xprt, |
e0ab53de | 1280 | .release_xprt = xs_tcp_release_xprt, |
92200412 | 1281 | .set_port = xs_set_port, |
9903cd1c | 1282 | .connect = xs_connect, |
02107148 CL |
1283 | .buf_alloc = rpc_malloc, |
1284 | .buf_free = rpc_free, | |
262965f5 | 1285 | .send_request = xs_tcp_send_request, |
fe3aca29 | 1286 | .set_retrans_timeout = xprt_set_retrans_timeout_def, |
9903cd1c CL |
1287 | .close = xs_close, |
1288 | .destroy = xs_destroy, | |
262ca07d | 1289 | .print_stats = xs_tcp_print_stats, |
a246b010 CL |
1290 | }; |
1291 | ||
9903cd1c CL |
1292 | /** |
1293 | * xs_setup_udp - Set up transport to use a UDP socket | |
1294 | * @xprt: transport to set up | |
1295 | * @to: timeout parameters | |
1296 | * | |
1297 | */ | |
a246b010 CL |
1298 | int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
1299 | { | |
1300 | size_t slot_table_size; | |
1301 | ||
1302 | dprintk("RPC: setting up udp-ipv4 transport...\n"); | |
1303 | ||
1304 | xprt->max_reqs = xprt_udp_slot_table_entries; | |
1305 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
0da974f4 | 1306 | xprt->slot = kzalloc(slot_table_size, GFP_KERNEL); |
a246b010 CL |
1307 | if (xprt->slot == NULL) |
1308 | return -ENOMEM; | |
a246b010 CL |
1309 | |
1310 | xprt->prot = IPPROTO_UDP; | |
b85d8806 | 1311 | xprt->port = xs_get_random_port(); |
808012fb | 1312 | xprt->tsh_size = 0; |
a246b010 CL |
1313 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; |
1314 | /* XXX: header size can vary due to auth type, IPv6, etc. */ | |
1315 | xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); | |
1316 | ||
b0d93ad5 | 1317 | INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt); |
03bf4b70 CL |
1318 | xprt->bind_timeout = XS_BIND_TO; |
1319 | xprt->connect_timeout = XS_UDP_CONN_TO; | |
1320 | xprt->reestablish_timeout = XS_UDP_REEST_TO; | |
1321 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 1322 | |
262965f5 | 1323 | xprt->ops = &xs_udp_ops; |
a246b010 CL |
1324 | |
1325 | if (to) | |
1326 | xprt->timeout = *to; | |
1327 | else | |
9903cd1c | 1328 | xprt_set_timeout(&xprt->timeout, 5, 5 * HZ); |
a246b010 CL |
1329 | |
1330 | return 0; | |
1331 | } | |
1332 | ||
9903cd1c CL |
1333 | /** |
1334 | * xs_setup_tcp - Set up transport to use a TCP socket | |
1335 | * @xprt: transport to set up | |
1336 | * @to: timeout parameters | |
1337 | * | |
1338 | */ | |
a246b010 CL |
1339 | int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
1340 | { | |
1341 | size_t slot_table_size; | |
1342 | ||
1343 | dprintk("RPC: setting up tcp-ipv4 transport...\n"); | |
1344 | ||
1345 | xprt->max_reqs = xprt_tcp_slot_table_entries; | |
1346 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
0da974f4 | 1347 | xprt->slot = kzalloc(slot_table_size, GFP_KERNEL); |
a246b010 CL |
1348 | if (xprt->slot == NULL) |
1349 | return -ENOMEM; | |
a246b010 CL |
1350 | |
1351 | xprt->prot = IPPROTO_TCP; | |
b85d8806 | 1352 | xprt->port = xs_get_random_port(); |
808012fb | 1353 | xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); |
a246b010 | 1354 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; |
808012fb | 1355 | xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; |
a246b010 | 1356 | |
b0d93ad5 | 1357 | INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt); |
03bf4b70 CL |
1358 | xprt->bind_timeout = XS_BIND_TO; |
1359 | xprt->connect_timeout = XS_TCP_CONN_TO; | |
1360 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
1361 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 1362 | |
262965f5 | 1363 | xprt->ops = &xs_tcp_ops; |
a246b010 CL |
1364 | |
1365 | if (to) | |
1366 | xprt->timeout = *to; | |
1367 | else | |
9903cd1c | 1368 | xprt_set_timeout(&xprt->timeout, 2, 60 * HZ); |
a246b010 CL |
1369 | |
1370 | return 0; | |
1371 | } |