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 | ||
edb267a6 CL |
128 | static void xs_format_peer_addresses(struct rpc_xprt *xprt) |
129 | { | |
130 | struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr; | |
131 | char *buf; | |
132 | ||
133 | buf = kzalloc(20, GFP_KERNEL); | |
134 | if (buf) { | |
135 | snprintf(buf, 20, "%u.%u.%u.%u", | |
136 | NIPQUAD(addr->sin_addr.s_addr)); | |
137 | } | |
138 | xprt->address_strings[RPC_DISPLAY_ADDR] = buf; | |
139 | ||
140 | buf = kzalloc(8, GFP_KERNEL); | |
141 | if (buf) { | |
142 | snprintf(buf, 8, "%u", | |
143 | ntohs(addr->sin_port)); | |
144 | } | |
145 | xprt->address_strings[RPC_DISPLAY_PORT] = buf; | |
146 | ||
147 | if (xprt->prot == IPPROTO_UDP) | |
148 | xprt->address_strings[RPC_DISPLAY_PROTO] = "udp"; | |
149 | else | |
150 | xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp"; | |
151 | ||
152 | buf = kzalloc(48, GFP_KERNEL); | |
153 | if (buf) { | |
154 | snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s", | |
155 | NIPQUAD(addr->sin_addr.s_addr), | |
156 | ntohs(addr->sin_port), | |
157 | xprt->prot == IPPROTO_UDP ? "udp" : "tcp"); | |
158 | } | |
159 | xprt->address_strings[RPC_DISPLAY_ALL] = buf; | |
160 | } | |
161 | ||
162 | static void xs_free_peer_addresses(struct rpc_xprt *xprt) | |
163 | { | |
164 | kfree(xprt->address_strings[RPC_DISPLAY_ADDR]); | |
165 | kfree(xprt->address_strings[RPC_DISPLAY_PORT]); | |
166 | kfree(xprt->address_strings[RPC_DISPLAY_ALL]); | |
167 | } | |
168 | ||
b4b5cc85 CL |
169 | #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) |
170 | ||
171 | static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len) | |
172 | { | |
173 | struct kvec iov = { | |
174 | .iov_base = xdr->head[0].iov_base + base, | |
175 | .iov_len = len - base, | |
176 | }; | |
177 | struct msghdr msg = { | |
178 | .msg_name = addr, | |
179 | .msg_namelen = addrlen, | |
180 | .msg_flags = XS_SENDMSG_FLAGS, | |
181 | }; | |
182 | ||
183 | if (xdr->len > len) | |
184 | msg.msg_flags |= MSG_MORE; | |
185 | ||
186 | if (likely(iov.iov_len)) | |
187 | return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); | |
188 | return kernel_sendmsg(sock, &msg, NULL, 0, 0); | |
189 | } | |
190 | ||
191 | static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len) | |
192 | { | |
193 | struct kvec iov = { | |
194 | .iov_base = xdr->tail[0].iov_base + base, | |
195 | .iov_len = len - base, | |
196 | }; | |
197 | struct msghdr msg = { | |
198 | .msg_flags = XS_SENDMSG_FLAGS, | |
199 | }; | |
200 | ||
201 | return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); | |
202 | } | |
203 | ||
9903cd1c CL |
204 | /** |
205 | * xs_sendpages - write pages directly to a socket | |
206 | * @sock: socket to send on | |
207 | * @addr: UDP only -- address of destination | |
208 | * @addrlen: UDP only -- length of destination address | |
209 | * @xdr: buffer containing this request | |
210 | * @base: starting position in the buffer | |
211 | * | |
a246b010 | 212 | */ |
262965f5 | 213 | static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base) |
a246b010 CL |
214 | { |
215 | struct page **ppage = xdr->pages; | |
216 | unsigned int len, pglen = xdr->page_len; | |
217 | int err, ret = 0; | |
218 | ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int); | |
219 | ||
262965f5 CL |
220 | if (unlikely(!sock)) |
221 | return -ENOTCONN; | |
222 | ||
223 | clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); | |
224 | ||
a246b010 CL |
225 | len = xdr->head[0].iov_len; |
226 | if (base < len || (addr != NULL && base == 0)) { | |
b4b5cc85 | 227 | err = xs_send_head(sock, addr, addrlen, xdr, base, len); |
a246b010 CL |
228 | if (ret == 0) |
229 | ret = err; | |
230 | else if (err > 0) | |
231 | ret += err; | |
b4b5cc85 | 232 | if (err != (len - base)) |
a246b010 CL |
233 | goto out; |
234 | base = 0; | |
235 | } else | |
236 | base -= len; | |
237 | ||
b4b5cc85 | 238 | if (unlikely(pglen == 0)) |
a246b010 | 239 | goto copy_tail; |
b4b5cc85 | 240 | if (unlikely(base >= pglen)) { |
a246b010 CL |
241 | base -= pglen; |
242 | goto copy_tail; | |
243 | } | |
244 | if (base || xdr->page_base) { | |
245 | pglen -= base; | |
9903cd1c | 246 | base += xdr->page_base; |
a246b010 CL |
247 | ppage += base >> PAGE_CACHE_SHIFT; |
248 | base &= ~PAGE_CACHE_MASK; | |
249 | } | |
250 | ||
251 | sendpage = sock->ops->sendpage ? : sock_no_sendpage; | |
252 | do { | |
b4b5cc85 | 253 | int flags = XS_SENDMSG_FLAGS; |
a246b010 CL |
254 | |
255 | len = PAGE_CACHE_SIZE; | |
256 | if (base) | |
257 | len -= base; | |
258 | if (pglen < len) | |
259 | len = pglen; | |
260 | ||
261 | if (pglen != len || xdr->tail[0].iov_len != 0) | |
262 | flags |= MSG_MORE; | |
263 | ||
264 | /* Hmm... We might be dealing with highmem pages */ | |
265 | if (PageHighMem(*ppage)) | |
266 | sendpage = sock_no_sendpage; | |
267 | err = sendpage(sock, *ppage, base, len, flags); | |
268 | if (ret == 0) | |
269 | ret = err; | |
270 | else if (err > 0) | |
271 | ret += err; | |
272 | if (err != len) | |
273 | goto out; | |
274 | base = 0; | |
275 | ppage++; | |
276 | } while ((pglen -= len) != 0); | |
277 | copy_tail: | |
278 | len = xdr->tail[0].iov_len; | |
279 | if (base < len) { | |
b4b5cc85 | 280 | err = xs_send_tail(sock, xdr, base, len); |
a246b010 CL |
281 | if (ret == 0) |
282 | ret = err; | |
283 | else if (err > 0) | |
284 | ret += err; | |
285 | } | |
286 | out: | |
287 | return ret; | |
288 | } | |
289 | ||
9903cd1c | 290 | /** |
262965f5 CL |
291 | * xs_nospace - place task on wait queue if transmit was incomplete |
292 | * @task: task to put to sleep | |
9903cd1c | 293 | * |
a246b010 | 294 | */ |
262965f5 | 295 | static void xs_nospace(struct rpc_task *task) |
a246b010 | 296 | { |
262965f5 CL |
297 | struct rpc_rqst *req = task->tk_rqstp; |
298 | struct rpc_xprt *xprt = req->rq_xprt; | |
a246b010 | 299 | |
262965f5 CL |
300 | dprintk("RPC: %4d xmit incomplete (%u left of %u)\n", |
301 | task->tk_pid, req->rq_slen - req->rq_bytes_sent, | |
302 | req->rq_slen); | |
303 | ||
304 | if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) { | |
305 | /* Protect against races with write_space */ | |
306 | spin_lock_bh(&xprt->transport_lock); | |
307 | ||
308 | /* Don't race with disconnect */ | |
309 | if (!xprt_connected(xprt)) | |
310 | task->tk_status = -ENOTCONN; | |
311 | else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags)) | |
312 | xprt_wait_for_buffer_space(task); | |
313 | ||
314 | spin_unlock_bh(&xprt->transport_lock); | |
315 | } else | |
316 | /* Keep holding the socket if it is blocked */ | |
317 | rpc_delay(task, HZ>>4); | |
318 | } | |
319 | ||
320 | /** | |
321 | * xs_udp_send_request - write an RPC request to a UDP socket | |
322 | * @task: address of RPC task that manages the state of an RPC request | |
323 | * | |
324 | * Return values: | |
325 | * 0: The request has been sent | |
326 | * EAGAIN: The socket was blocked, please call again later to | |
327 | * complete the request | |
328 | * ENOTCONN: Caller needs to invoke connect logic then call again | |
329 | * other: Some other error occured, the request was not sent | |
330 | */ | |
331 | static int xs_udp_send_request(struct rpc_task *task) | |
332 | { | |
333 | struct rpc_rqst *req = task->tk_rqstp; | |
334 | struct rpc_xprt *xprt = req->rq_xprt; | |
335 | struct xdr_buf *xdr = &req->rq_snd_buf; | |
336 | int status; | |
a246b010 | 337 | |
9903cd1c | 338 | xs_pktdump("packet data:", |
a246b010 CL |
339 | req->rq_svec->iov_base, |
340 | req->rq_svec->iov_len); | |
341 | ||
262965f5 CL |
342 | req->rq_xtime = jiffies; |
343 | status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr, | |
344 | sizeof(xprt->addr), xdr, req->rq_bytes_sent); | |
a246b010 | 345 | |
262965f5 CL |
346 | dprintk("RPC: xs_udp_send_request(%u) = %d\n", |
347 | xdr->len - req->rq_bytes_sent, status); | |
a246b010 | 348 | |
262965f5 CL |
349 | if (likely(status >= (int) req->rq_slen)) |
350 | return 0; | |
a246b010 | 351 | |
262965f5 CL |
352 | /* Still some bytes left; set up for a retry later. */ |
353 | if (status > 0) | |
354 | status = -EAGAIN; | |
a246b010 | 355 | |
262965f5 CL |
356 | switch (status) { |
357 | case -ENETUNREACH: | |
358 | case -EPIPE: | |
a246b010 CL |
359 | case -ECONNREFUSED: |
360 | /* When the server has died, an ICMP port unreachable message | |
9903cd1c | 361 | * prompts ECONNREFUSED. */ |
a246b010 | 362 | break; |
262965f5 CL |
363 | case -EAGAIN: |
364 | xs_nospace(task); | |
a246b010 CL |
365 | break; |
366 | default: | |
262965f5 CL |
367 | dprintk("RPC: sendmsg returned unrecognized error %d\n", |
368 | -status); | |
9903cd1c | 369 | break; |
a246b010 | 370 | } |
262965f5 CL |
371 | |
372 | return status; | |
a246b010 CL |
373 | } |
374 | ||
808012fb CL |
375 | static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf) |
376 | { | |
377 | u32 reclen = buf->len - sizeof(rpc_fraghdr); | |
378 | rpc_fraghdr *base = buf->head[0].iov_base; | |
379 | *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen); | |
380 | } | |
381 | ||
9903cd1c | 382 | /** |
262965f5 | 383 | * xs_tcp_send_request - write an RPC request to a TCP socket |
9903cd1c CL |
384 | * @task: address of RPC task that manages the state of an RPC request |
385 | * | |
386 | * Return values: | |
262965f5 CL |
387 | * 0: The request has been sent |
388 | * EAGAIN: The socket was blocked, please call again later to | |
389 | * complete the request | |
390 | * ENOTCONN: Caller needs to invoke connect logic then call again | |
391 | * other: Some other error occured, the request was not sent | |
9903cd1c CL |
392 | * |
393 | * XXX: In the case of soft timeouts, should we eventually give up | |
262965f5 | 394 | * if sendmsg is not able to make progress? |
9903cd1c | 395 | */ |
262965f5 | 396 | static int xs_tcp_send_request(struct rpc_task *task) |
a246b010 CL |
397 | { |
398 | struct rpc_rqst *req = task->tk_rqstp; | |
399 | struct rpc_xprt *xprt = req->rq_xprt; | |
262965f5 | 400 | struct xdr_buf *xdr = &req->rq_snd_buf; |
a246b010 CL |
401 | int status, retry = 0; |
402 | ||
808012fb | 403 | xs_encode_tcp_record_marker(&req->rq_snd_buf); |
a246b010 | 404 | |
262965f5 CL |
405 | xs_pktdump("packet data:", |
406 | req->rq_svec->iov_base, | |
407 | req->rq_svec->iov_len); | |
a246b010 CL |
408 | |
409 | /* Continue transmitting the packet/record. We must be careful | |
410 | * to cope with writespace callbacks arriving _after_ we have | |
262965f5 | 411 | * called sendmsg(). */ |
a246b010 CL |
412 | while (1) { |
413 | req->rq_xtime = jiffies; | |
262965f5 CL |
414 | status = xs_sendpages(xprt->sock, NULL, 0, xdr, |
415 | req->rq_bytes_sent); | |
a246b010 | 416 | |
262965f5 CL |
417 | dprintk("RPC: xs_tcp_send_request(%u) = %d\n", |
418 | xdr->len - req->rq_bytes_sent, status); | |
a246b010 | 419 | |
262965f5 | 420 | if (unlikely(status < 0)) |
a246b010 | 421 | break; |
a246b010 | 422 | |
262965f5 CL |
423 | /* If we've sent the entire packet, immediately |
424 | * reset the count of bytes sent. */ | |
425 | req->rq_bytes_sent += status; | |
ef759a2e | 426 | task->tk_bytes_sent += status; |
262965f5 CL |
427 | if (likely(req->rq_bytes_sent >= req->rq_slen)) { |
428 | req->rq_bytes_sent = 0; | |
429 | return 0; | |
430 | } | |
a246b010 CL |
431 | |
432 | status = -EAGAIN; | |
262965f5 | 433 | if (retry++ > XS_SENDMSG_RETRY) |
a246b010 CL |
434 | break; |
435 | } | |
436 | ||
262965f5 CL |
437 | switch (status) { |
438 | case -EAGAIN: | |
439 | xs_nospace(task); | |
440 | break; | |
441 | case -ECONNREFUSED: | |
442 | case -ECONNRESET: | |
443 | case -ENOTCONN: | |
444 | case -EPIPE: | |
445 | status = -ENOTCONN; | |
446 | break; | |
447 | default: | |
448 | dprintk("RPC: sendmsg returned unrecognized error %d\n", | |
449 | -status); | |
43118c29 | 450 | xprt_disconnect(xprt); |
262965f5 | 451 | break; |
a246b010 | 452 | } |
262965f5 | 453 | |
a246b010 CL |
454 | return status; |
455 | } | |
456 | ||
e0ab53de TM |
457 | /** |
458 | * xs_tcp_release_xprt - clean up after a tcp transmission | |
459 | * @xprt: transport | |
460 | * @task: rpc task | |
461 | * | |
462 | * This cleans up if an error causes us to abort the transmission of a request. | |
463 | * In this case, the socket may need to be reset in order to avoid confusing | |
464 | * the server. | |
465 | */ | |
466 | static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) | |
467 | { | |
468 | struct rpc_rqst *req; | |
469 | ||
470 | if (task != xprt->snd_task) | |
471 | return; | |
472 | if (task == NULL) | |
473 | goto out_release; | |
474 | req = task->tk_rqstp; | |
475 | if (req->rq_bytes_sent == 0) | |
476 | goto out_release; | |
477 | if (req->rq_bytes_sent == req->rq_snd_buf.len) | |
478 | goto out_release; | |
479 | set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state); | |
480 | out_release: | |
481 | xprt_release_xprt(xprt, task); | |
482 | } | |
483 | ||
9903cd1c CL |
484 | /** |
485 | * xs_close - close a socket | |
486 | * @xprt: transport | |
487 | * | |
3167e12c CL |
488 | * This is used when all requests are complete; ie, no DRC state remains |
489 | * on the server we want to save. | |
a246b010 | 490 | */ |
9903cd1c | 491 | static void xs_close(struct rpc_xprt *xprt) |
a246b010 | 492 | { |
9903cd1c CL |
493 | struct socket *sock = xprt->sock; |
494 | struct sock *sk = xprt->inet; | |
a246b010 CL |
495 | |
496 | if (!sk) | |
632e3bdc | 497 | goto clear_close_wait; |
a246b010 | 498 | |
9903cd1c CL |
499 | dprintk("RPC: xs_close xprt %p\n", xprt); |
500 | ||
a246b010 CL |
501 | write_lock_bh(&sk->sk_callback_lock); |
502 | xprt->inet = NULL; | |
503 | xprt->sock = NULL; | |
504 | ||
9903cd1c CL |
505 | sk->sk_user_data = NULL; |
506 | sk->sk_data_ready = xprt->old_data_ready; | |
a246b010 | 507 | sk->sk_state_change = xprt->old_state_change; |
9903cd1c | 508 | sk->sk_write_space = xprt->old_write_space; |
a246b010 CL |
509 | write_unlock_bh(&sk->sk_callback_lock); |
510 | ||
9903cd1c | 511 | sk->sk_no_check = 0; |
a246b010 CL |
512 | |
513 | sock_release(sock); | |
632e3bdc TM |
514 | clear_close_wait: |
515 | smp_mb__before_clear_bit(); | |
516 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
517 | smp_mb__after_clear_bit(); | |
a246b010 CL |
518 | } |
519 | ||
9903cd1c CL |
520 | /** |
521 | * xs_destroy - prepare to shutdown a transport | |
522 | * @xprt: doomed transport | |
523 | * | |
524 | */ | |
525 | static void xs_destroy(struct rpc_xprt *xprt) | |
a246b010 | 526 | { |
9903cd1c CL |
527 | dprintk("RPC: xs_destroy xprt %p\n", xprt); |
528 | ||
55aa4f58 | 529 | cancel_delayed_work(&xprt->connect_worker); |
a246b010 CL |
530 | flush_scheduled_work(); |
531 | ||
532 | xprt_disconnect(xprt); | |
9903cd1c | 533 | xs_close(xprt); |
edb267a6 | 534 | xs_free_peer_addresses(xprt); |
a246b010 CL |
535 | kfree(xprt->slot); |
536 | } | |
537 | ||
9903cd1c CL |
538 | static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) |
539 | { | |
540 | return (struct rpc_xprt *) sk->sk_user_data; | |
541 | } | |
542 | ||
543 | /** | |
544 | * xs_udp_data_ready - "data ready" callback for UDP sockets | |
545 | * @sk: socket with data to read | |
546 | * @len: how much data to read | |
547 | * | |
a246b010 | 548 | */ |
9903cd1c | 549 | static void xs_udp_data_ready(struct sock *sk, int len) |
a246b010 | 550 | { |
9903cd1c CL |
551 | struct rpc_task *task; |
552 | struct rpc_xprt *xprt; | |
a246b010 | 553 | struct rpc_rqst *rovr; |
9903cd1c | 554 | struct sk_buff *skb; |
a246b010 CL |
555 | int err, repsize, copied; |
556 | u32 _xid, *xp; | |
557 | ||
558 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
559 | dprintk("RPC: xs_udp_data_ready...\n"); |
560 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 561 | goto out; |
a246b010 CL |
562 | |
563 | if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) | |
564 | goto out; | |
565 | ||
566 | if (xprt->shutdown) | |
567 | goto dropit; | |
568 | ||
569 | repsize = skb->len - sizeof(struct udphdr); | |
570 | if (repsize < 4) { | |
9903cd1c | 571 | dprintk("RPC: impossible RPC reply size %d!\n", repsize); |
a246b010 CL |
572 | goto dropit; |
573 | } | |
574 | ||
575 | /* Copy the XID from the skb... */ | |
576 | xp = skb_header_pointer(skb, sizeof(struct udphdr), | |
577 | sizeof(_xid), &_xid); | |
578 | if (xp == NULL) | |
579 | goto dropit; | |
580 | ||
581 | /* Look up and lock the request corresponding to the given XID */ | |
4a0f8c04 | 582 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
583 | rovr = xprt_lookup_rqst(xprt, *xp); |
584 | if (!rovr) | |
585 | goto out_unlock; | |
586 | task = rovr->rq_task; | |
587 | ||
a246b010 CL |
588 | if ((copied = rovr->rq_private_buf.buflen) > repsize) |
589 | copied = repsize; | |
590 | ||
591 | /* Suck it into the iovec, verify checksum if not done by hw. */ | |
592 | if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) | |
593 | goto out_unlock; | |
594 | ||
595 | /* Something worked... */ | |
596 | dst_confirm(skb->dst); | |
597 | ||
1570c1e4 CL |
598 | xprt_adjust_cwnd(task, copied); |
599 | xprt_update_rtt(task); | |
600 | xprt_complete_rqst(task, copied); | |
a246b010 CL |
601 | |
602 | out_unlock: | |
4a0f8c04 | 603 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
604 | dropit: |
605 | skb_free_datagram(sk, skb); | |
606 | out: | |
607 | read_unlock(&sk->sk_callback_lock); | |
608 | } | |
609 | ||
9903cd1c | 610 | static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len) |
a246b010 CL |
611 | { |
612 | if (len > desc->count) | |
613 | len = desc->count; | |
614 | if (skb_copy_bits(desc->skb, desc->offset, p, len)) { | |
615 | dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n", | |
616 | len, desc->count); | |
617 | return 0; | |
618 | } | |
619 | desc->offset += len; | |
620 | desc->count -= len; | |
621 | dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n", | |
622 | len, desc->count); | |
623 | return len; | |
624 | } | |
625 | ||
9903cd1c | 626 | static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
627 | { |
628 | size_t len, used; | |
629 | char *p; | |
630 | ||
631 | p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset; | |
632 | len = sizeof(xprt->tcp_recm) - xprt->tcp_offset; | |
9903cd1c | 633 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
634 | xprt->tcp_offset += used; |
635 | if (used != len) | |
636 | return; | |
808012fb | 637 | |
a246b010 | 638 | xprt->tcp_reclen = ntohl(xprt->tcp_recm); |
808012fb | 639 | if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) |
a246b010 CL |
640 | xprt->tcp_flags |= XPRT_LAST_FRAG; |
641 | else | |
642 | xprt->tcp_flags &= ~XPRT_LAST_FRAG; | |
808012fb CL |
643 | xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; |
644 | ||
a246b010 CL |
645 | xprt->tcp_flags &= ~XPRT_COPY_RECM; |
646 | xprt->tcp_offset = 0; | |
808012fb | 647 | |
a246b010 | 648 | /* Sanity check of the record length */ |
808012fb | 649 | if (unlikely(xprt->tcp_reclen < 4)) { |
9903cd1c | 650 | dprintk("RPC: invalid TCP record fragment length\n"); |
a246b010 | 651 | xprt_disconnect(xprt); |
9903cd1c | 652 | return; |
a246b010 CL |
653 | } |
654 | dprintk("RPC: reading TCP record fragment of length %d\n", | |
655 | xprt->tcp_reclen); | |
656 | } | |
657 | ||
9903cd1c | 658 | static void xs_tcp_check_recm(struct rpc_xprt *xprt) |
a246b010 CL |
659 | { |
660 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n", | |
661 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags); | |
662 | if (xprt->tcp_offset == xprt->tcp_reclen) { | |
663 | xprt->tcp_flags |= XPRT_COPY_RECM; | |
664 | xprt->tcp_offset = 0; | |
665 | if (xprt->tcp_flags & XPRT_LAST_FRAG) { | |
666 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
667 | xprt->tcp_flags |= XPRT_COPY_XID; | |
668 | xprt->tcp_copied = 0; | |
669 | } | |
670 | } | |
671 | } | |
672 | ||
9903cd1c | 673 | static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
674 | { |
675 | size_t len, used; | |
676 | char *p; | |
677 | ||
678 | len = sizeof(xprt->tcp_xid) - xprt->tcp_offset; | |
679 | dprintk("RPC: reading XID (%Zu bytes)\n", len); | |
680 | p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset; | |
9903cd1c | 681 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
682 | xprt->tcp_offset += used; |
683 | if (used != len) | |
684 | return; | |
685 | xprt->tcp_flags &= ~XPRT_COPY_XID; | |
686 | xprt->tcp_flags |= XPRT_COPY_DATA; | |
687 | xprt->tcp_copied = 4; | |
688 | dprintk("RPC: reading reply for XID %08x\n", | |
689 | ntohl(xprt->tcp_xid)); | |
9903cd1c | 690 | xs_tcp_check_recm(xprt); |
a246b010 CL |
691 | } |
692 | ||
9903cd1c | 693 | static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
694 | { |
695 | struct rpc_rqst *req; | |
696 | struct xdr_buf *rcvbuf; | |
697 | size_t len; | |
698 | ssize_t r; | |
699 | ||
700 | /* Find and lock the request corresponding to this xid */ | |
4a0f8c04 | 701 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
702 | req = xprt_lookup_rqst(xprt, xprt->tcp_xid); |
703 | if (!req) { | |
704 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
705 | dprintk("RPC: XID %08x request not found!\n", | |
706 | ntohl(xprt->tcp_xid)); | |
4a0f8c04 | 707 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
708 | return; |
709 | } | |
710 | ||
711 | rcvbuf = &req->rq_private_buf; | |
712 | len = desc->count; | |
713 | if (len > xprt->tcp_reclen - xprt->tcp_offset) { | |
714 | skb_reader_t my_desc; | |
715 | ||
716 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
717 | memcpy(&my_desc, desc, sizeof(my_desc)); | |
718 | my_desc.count = len; | |
719 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 720 | &my_desc, xs_tcp_copy_data); |
a246b010 CL |
721 | desc->count -= r; |
722 | desc->offset += r; | |
723 | } else | |
724 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 725 | desc, xs_tcp_copy_data); |
a246b010 CL |
726 | |
727 | if (r > 0) { | |
728 | xprt->tcp_copied += r; | |
729 | xprt->tcp_offset += r; | |
730 | } | |
731 | if (r != len) { | |
732 | /* Error when copying to the receive buffer, | |
733 | * usually because we weren't able to allocate | |
734 | * additional buffer pages. All we can do now | |
735 | * is turn off XPRT_COPY_DATA, so the request | |
736 | * will not receive any additional updates, | |
737 | * and time out. | |
738 | * Any remaining data from this record will | |
739 | * be discarded. | |
740 | */ | |
741 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
742 | dprintk("RPC: XID %08x truncated request\n", | |
743 | ntohl(xprt->tcp_xid)); | |
744 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
745 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
746 | goto out; | |
747 | } | |
748 | ||
749 | dprintk("RPC: XID %08x read %Zd bytes\n", | |
750 | ntohl(xprt->tcp_xid), r); | |
751 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
752 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
753 | ||
754 | if (xprt->tcp_copied == req->rq_private_buf.buflen) | |
755 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
756 | else if (xprt->tcp_offset == xprt->tcp_reclen) { | |
757 | if (xprt->tcp_flags & XPRT_LAST_FRAG) | |
758 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
759 | } | |
760 | ||
761 | out: | |
1570c1e4 CL |
762 | if (!(xprt->tcp_flags & XPRT_COPY_DATA)) |
763 | xprt_complete_rqst(req->rq_task, xprt->tcp_copied); | |
4a0f8c04 | 764 | spin_unlock(&xprt->transport_lock); |
9903cd1c | 765 | xs_tcp_check_recm(xprt); |
a246b010 CL |
766 | } |
767 | ||
9903cd1c | 768 | static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
769 | { |
770 | size_t len; | |
771 | ||
772 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
773 | if (len > desc->count) | |
774 | len = desc->count; | |
775 | desc->count -= len; | |
776 | desc->offset += len; | |
777 | xprt->tcp_offset += len; | |
778 | dprintk("RPC: discarded %Zu bytes\n", len); | |
9903cd1c | 779 | xs_tcp_check_recm(xprt); |
a246b010 CL |
780 | } |
781 | ||
9903cd1c | 782 | static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) |
a246b010 CL |
783 | { |
784 | struct rpc_xprt *xprt = rd_desc->arg.data; | |
785 | skb_reader_t desc = { | |
786 | .skb = skb, | |
787 | .offset = offset, | |
788 | .count = len, | |
789 | .csum = 0 | |
9903cd1c | 790 | }; |
a246b010 | 791 | |
9903cd1c | 792 | dprintk("RPC: xs_tcp_data_recv started\n"); |
a246b010 CL |
793 | do { |
794 | /* Read in a new fragment marker if necessary */ | |
795 | /* Can we ever really expect to get completely empty fragments? */ | |
796 | if (xprt->tcp_flags & XPRT_COPY_RECM) { | |
9903cd1c | 797 | xs_tcp_read_fraghdr(xprt, &desc); |
a246b010 CL |
798 | continue; |
799 | } | |
800 | /* Read in the xid if necessary */ | |
801 | if (xprt->tcp_flags & XPRT_COPY_XID) { | |
9903cd1c | 802 | xs_tcp_read_xid(xprt, &desc); |
a246b010 CL |
803 | continue; |
804 | } | |
805 | /* Read in the request data */ | |
806 | if (xprt->tcp_flags & XPRT_COPY_DATA) { | |
9903cd1c | 807 | xs_tcp_read_request(xprt, &desc); |
a246b010 CL |
808 | continue; |
809 | } | |
810 | /* Skip over any trailing bytes on short reads */ | |
9903cd1c | 811 | xs_tcp_read_discard(xprt, &desc); |
a246b010 | 812 | } while (desc.count); |
9903cd1c | 813 | dprintk("RPC: xs_tcp_data_recv done\n"); |
a246b010 CL |
814 | return len - desc.count; |
815 | } | |
816 | ||
9903cd1c CL |
817 | /** |
818 | * xs_tcp_data_ready - "data ready" callback for TCP sockets | |
819 | * @sk: socket with data to read | |
820 | * @bytes: how much data to read | |
821 | * | |
822 | */ | |
823 | static void xs_tcp_data_ready(struct sock *sk, int bytes) | |
a246b010 CL |
824 | { |
825 | struct rpc_xprt *xprt; | |
826 | read_descriptor_t rd_desc; | |
827 | ||
828 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
829 | dprintk("RPC: xs_tcp_data_ready...\n"); |
830 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 831 | goto out; |
a246b010 CL |
832 | if (xprt->shutdown) |
833 | goto out; | |
834 | ||
9903cd1c | 835 | /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ |
a246b010 CL |
836 | rd_desc.arg.data = xprt; |
837 | rd_desc.count = 65536; | |
9903cd1c | 838 | tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); |
a246b010 CL |
839 | out: |
840 | read_unlock(&sk->sk_callback_lock); | |
841 | } | |
842 | ||
9903cd1c CL |
843 | /** |
844 | * xs_tcp_state_change - callback to handle TCP socket state changes | |
845 | * @sk: socket whose state has changed | |
846 | * | |
847 | */ | |
848 | static void xs_tcp_state_change(struct sock *sk) | |
a246b010 | 849 | { |
9903cd1c | 850 | struct rpc_xprt *xprt; |
a246b010 CL |
851 | |
852 | read_lock(&sk->sk_callback_lock); | |
853 | if (!(xprt = xprt_from_sock(sk))) | |
854 | goto out; | |
9903cd1c | 855 | dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); |
a246b010 CL |
856 | dprintk("RPC: state %x conn %d dead %d zapped %d\n", |
857 | sk->sk_state, xprt_connected(xprt), | |
858 | sock_flag(sk, SOCK_DEAD), | |
859 | sock_flag(sk, SOCK_ZAPPED)); | |
860 | ||
861 | switch (sk->sk_state) { | |
862 | case TCP_ESTABLISHED: | |
4a0f8c04 | 863 | spin_lock_bh(&xprt->transport_lock); |
a246b010 CL |
864 | if (!xprt_test_and_set_connected(xprt)) { |
865 | /* Reset TCP record info */ | |
866 | xprt->tcp_offset = 0; | |
867 | xprt->tcp_reclen = 0; | |
868 | xprt->tcp_copied = 0; | |
869 | xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID; | |
03bf4b70 | 870 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; |
44fbac22 | 871 | xprt_wake_pending_tasks(xprt, 0); |
a246b010 | 872 | } |
4a0f8c04 | 873 | spin_unlock_bh(&xprt->transport_lock); |
a246b010 CL |
874 | break; |
875 | case TCP_SYN_SENT: | |
876 | case TCP_SYN_RECV: | |
877 | break; | |
632e3bdc TM |
878 | case TCP_CLOSE_WAIT: |
879 | /* Try to schedule an autoclose RPC calls */ | |
880 | set_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
881 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) | |
882 | schedule_work(&xprt->task_cleanup); | |
a246b010 CL |
883 | default: |
884 | xprt_disconnect(xprt); | |
a246b010 CL |
885 | } |
886 | out: | |
887 | read_unlock(&sk->sk_callback_lock); | |
888 | } | |
889 | ||
9903cd1c | 890 | /** |
c7b2cae8 CL |
891 | * xs_udp_write_space - callback invoked when socket buffer space |
892 | * becomes available | |
9903cd1c CL |
893 | * @sk: socket whose state has changed |
894 | * | |
a246b010 CL |
895 | * Called when more output buffer space is available for this socket. |
896 | * We try not to wake our writers until they can make "significant" | |
c7b2cae8 | 897 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg |
a246b010 CL |
898 | * with a bunch of small requests. |
899 | */ | |
c7b2cae8 | 900 | static void xs_udp_write_space(struct sock *sk) |
a246b010 | 901 | { |
a246b010 | 902 | read_lock(&sk->sk_callback_lock); |
a246b010 | 903 | |
c7b2cae8 CL |
904 | /* from net/core/sock.c:sock_def_write_space */ |
905 | if (sock_writeable(sk)) { | |
906 | struct socket *sock; | |
907 | struct rpc_xprt *xprt; | |
908 | ||
909 | if (unlikely(!(sock = sk->sk_socket))) | |
a246b010 | 910 | goto out; |
c7b2cae8 CL |
911 | if (unlikely(!(xprt = xprt_from_sock(sk)))) |
912 | goto out; | |
913 | if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) | |
a246b010 | 914 | goto out; |
c7b2cae8 CL |
915 | |
916 | xprt_write_space(xprt); | |
a246b010 CL |
917 | } |
918 | ||
c7b2cae8 CL |
919 | out: |
920 | read_unlock(&sk->sk_callback_lock); | |
921 | } | |
a246b010 | 922 | |
c7b2cae8 CL |
923 | /** |
924 | * xs_tcp_write_space - callback invoked when socket buffer space | |
925 | * becomes available | |
926 | * @sk: socket whose state has changed | |
927 | * | |
928 | * Called when more output buffer space is available for this socket. | |
929 | * We try not to wake our writers until they can make "significant" | |
930 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg | |
931 | * with a bunch of small requests. | |
932 | */ | |
933 | static void xs_tcp_write_space(struct sock *sk) | |
934 | { | |
935 | read_lock(&sk->sk_callback_lock); | |
936 | ||
937 | /* from net/core/stream.c:sk_stream_write_space */ | |
938 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { | |
939 | struct socket *sock; | |
940 | struct rpc_xprt *xprt; | |
941 | ||
942 | if (unlikely(!(sock = sk->sk_socket))) | |
943 | goto out; | |
944 | if (unlikely(!(xprt = xprt_from_sock(sk)))) | |
945 | goto out; | |
946 | if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) | |
947 | goto out; | |
948 | ||
949 | xprt_write_space(xprt); | |
950 | } | |
951 | ||
952 | out: | |
a246b010 CL |
953 | read_unlock(&sk->sk_callback_lock); |
954 | } | |
955 | ||
470056c2 | 956 | static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) |
a246b010 CL |
957 | { |
958 | struct sock *sk = xprt->inet; | |
959 | ||
a246b010 CL |
960 | if (xprt->rcvsize) { |
961 | sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
962 | sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2; | |
963 | } | |
964 | if (xprt->sndsize) { | |
965 | sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
966 | sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2; | |
967 | sk->sk_write_space(sk); | |
968 | } | |
969 | } | |
970 | ||
43118c29 | 971 | /** |
470056c2 | 972 | * xs_udp_set_buffer_size - set send and receive limits |
43118c29 | 973 | * @xprt: generic transport |
470056c2 CL |
974 | * @sndsize: requested size of send buffer, in bytes |
975 | * @rcvsize: requested size of receive buffer, in bytes | |
43118c29 | 976 | * |
470056c2 | 977 | * Set socket send and receive buffer size limits. |
43118c29 | 978 | */ |
470056c2 | 979 | static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) |
43118c29 | 980 | { |
470056c2 CL |
981 | xprt->sndsize = 0; |
982 | if (sndsize) | |
983 | xprt->sndsize = sndsize + 1024; | |
984 | xprt->rcvsize = 0; | |
985 | if (rcvsize) | |
986 | xprt->rcvsize = rcvsize + 1024; | |
987 | ||
988 | xs_udp_do_set_buffer_size(xprt); | |
43118c29 CL |
989 | } |
990 | ||
46c0ee8b CL |
991 | /** |
992 | * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport | |
993 | * @task: task that timed out | |
994 | * | |
995 | * Adjust the congestion window after a retransmit timeout has occurred. | |
996 | */ | |
997 | static void xs_udp_timer(struct rpc_task *task) | |
998 | { | |
999 | xprt_adjust_cwnd(task, -ETIMEDOUT); | |
1000 | } | |
1001 | ||
b85d8806 CL |
1002 | static unsigned short xs_get_random_port(void) |
1003 | { | |
1004 | unsigned short range = xprt_max_resvport - xprt_min_resvport; | |
1005 | unsigned short rand = (unsigned short) net_random() % range; | |
1006 | return rand + xprt_min_resvport; | |
1007 | } | |
1008 | ||
edb267a6 CL |
1009 | /** |
1010 | * xs_print_peer_address - format an IPv4 address for printing | |
1011 | * @xprt: generic transport | |
1012 | * @format: flags field indicating which parts of the address to render | |
1013 | */ | |
1014 | static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format) | |
1015 | { | |
1016 | if (xprt->address_strings[format] != NULL) | |
1017 | return xprt->address_strings[format]; | |
1018 | else | |
1019 | return "unprintable"; | |
1020 | } | |
1021 | ||
92200412 CL |
1022 | /** |
1023 | * xs_set_port - reset the port number in the remote endpoint address | |
1024 | * @xprt: generic transport | |
1025 | * @port: new port number | |
1026 | * | |
1027 | */ | |
1028 | static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) | |
1029 | { | |
1030 | dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); | |
1031 | xprt->addr.sin_port = htons(port); | |
1032 | } | |
1033 | ||
9903cd1c | 1034 | static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock) |
a246b010 CL |
1035 | { |
1036 | struct sockaddr_in myaddr = { | |
1037 | .sin_family = AF_INET, | |
1038 | }; | |
529b33c6 CL |
1039 | int err; |
1040 | unsigned short port = xprt->port; | |
a246b010 | 1041 | |
a246b010 CL |
1042 | do { |
1043 | myaddr.sin_port = htons(port); | |
1044 | err = sock->ops->bind(sock, (struct sockaddr *) &myaddr, | |
1045 | sizeof(myaddr)); | |
1046 | if (err == 0) { | |
1047 | xprt->port = port; | |
9903cd1c CL |
1048 | dprintk("RPC: xs_bindresvport bound to port %u\n", |
1049 | port); | |
a246b010 CL |
1050 | return 0; |
1051 | } | |
529b33c6 CL |
1052 | if (port <= xprt_min_resvport) |
1053 | port = xprt_max_resvport; | |
1054 | else | |
1055 | port--; | |
a246b010 CL |
1056 | } while (err == -EADDRINUSE && port != xprt->port); |
1057 | ||
9903cd1c | 1058 | dprintk("RPC: can't bind to reserved port (%d).\n", -err); |
a246b010 CL |
1059 | return err; |
1060 | } | |
1061 | ||
b0d93ad5 CL |
1062 | /** |
1063 | * xs_udp_connect_worker - set up a UDP socket | |
1064 | * @args: RPC transport to connect | |
1065 | * | |
1066 | * Invoked by a work queue tasklet. | |
1067 | */ | |
1068 | static void xs_udp_connect_worker(void *args) | |
a246b010 | 1069 | { |
b0d93ad5 CL |
1070 | struct rpc_xprt *xprt = (struct rpc_xprt *) args; |
1071 | struct socket *sock = xprt->sock; | |
1072 | int err, status = -EIO; | |
9903cd1c | 1073 | |
ec739ef0 | 1074 | if (xprt->shutdown || !xprt_bound(xprt)) |
b0d93ad5 | 1075 | goto out; |
9903cd1c | 1076 | |
b0d93ad5 CL |
1077 | /* Start by resetting any existing state */ |
1078 | xs_close(xprt); | |
9903cd1c | 1079 | |
b0d93ad5 CL |
1080 | if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) { |
1081 | dprintk("RPC: can't create UDP transport socket (%d).\n", -err); | |
1082 | goto out; | |
1083 | } | |
9903cd1c | 1084 | |
b0d93ad5 CL |
1085 | if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) { |
1086 | sock_release(sock); | |
1087 | goto out; | |
1088 | } | |
9903cd1c | 1089 | |
edb267a6 CL |
1090 | dprintk("RPC: worker connecting xprt %p to address: %s\n", |
1091 | xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL)); | |
1092 | ||
b0d93ad5 CL |
1093 | if (!xprt->inet) { |
1094 | struct sock *sk = sock->sk; | |
a246b010 | 1095 | |
b0d93ad5 | 1096 | write_lock_bh(&sk->sk_callback_lock); |
a246b010 | 1097 | |
b0d93ad5 CL |
1098 | sk->sk_user_data = xprt; |
1099 | xprt->old_data_ready = sk->sk_data_ready; | |
1100 | xprt->old_state_change = sk->sk_state_change; | |
1101 | xprt->old_write_space = sk->sk_write_space; | |
9903cd1c | 1102 | sk->sk_data_ready = xs_udp_data_ready; |
c7b2cae8 | 1103 | sk->sk_write_space = xs_udp_write_space; |
a246b010 | 1104 | sk->sk_no_check = UDP_CSUM_NORCV; |
b079fa7b | 1105 | sk->sk_allocation = GFP_ATOMIC; |
b0d93ad5 | 1106 | |
a246b010 | 1107 | xprt_set_connected(xprt); |
a246b010 | 1108 | |
b0d93ad5 CL |
1109 | /* Reset to new socket */ |
1110 | xprt->sock = sock; | |
1111 | xprt->inet = sk; | |
a246b010 | 1112 | |
b0d93ad5 CL |
1113 | write_unlock_bh(&sk->sk_callback_lock); |
1114 | } | |
470056c2 | 1115 | xs_udp_do_set_buffer_size(xprt); |
b0d93ad5 CL |
1116 | status = 0; |
1117 | out: | |
1118 | xprt_wake_pending_tasks(xprt, status); | |
1119 | xprt_clear_connecting(xprt); | |
a246b010 CL |
1120 | } |
1121 | ||
3167e12c CL |
1122 | /* |
1123 | * We need to preserve the port number so the reply cache on the server can | |
1124 | * find our cached RPC replies when we get around to reconnecting. | |
1125 | */ | |
1126 | static void xs_tcp_reuse_connection(struct rpc_xprt *xprt) | |
1127 | { | |
1128 | int result; | |
1129 | struct socket *sock = xprt->sock; | |
1130 | struct sockaddr any; | |
1131 | ||
1132 | dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt); | |
1133 | ||
1134 | /* | |
1135 | * Disconnect the transport socket by doing a connect operation | |
1136 | * with AF_UNSPEC. This should return immediately... | |
1137 | */ | |
1138 | memset(&any, 0, sizeof(any)); | |
1139 | any.sa_family = AF_UNSPEC; | |
1140 | result = sock->ops->connect(sock, &any, sizeof(any), 0); | |
1141 | if (result) | |
1142 | dprintk("RPC: AF_UNSPEC connect return code %d\n", | |
1143 | result); | |
1144 | } | |
1145 | ||
9903cd1c | 1146 | /** |
b0d93ad5 | 1147 | * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint |
9903cd1c CL |
1148 | * @args: RPC transport to connect |
1149 | * | |
1150 | * Invoked by a work queue tasklet. | |
a246b010 | 1151 | */ |
b0d93ad5 | 1152 | static void xs_tcp_connect_worker(void *args) |
a246b010 CL |
1153 | { |
1154 | struct rpc_xprt *xprt = (struct rpc_xprt *)args; | |
1155 | struct socket *sock = xprt->sock; | |
b0d93ad5 | 1156 | int err, status = -EIO; |
a246b010 | 1157 | |
ec739ef0 | 1158 | if (xprt->shutdown || !xprt_bound(xprt)) |
a246b010 CL |
1159 | goto out; |
1160 | ||
3167e12c CL |
1161 | if (!xprt->sock) { |
1162 | /* start from scratch */ | |
1163 | if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) { | |
1164 | dprintk("RPC: can't create TCP transport socket (%d).\n", -err); | |
1165 | goto out; | |
1166 | } | |
a246b010 | 1167 | |
3167e12c CL |
1168 | if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) { |
1169 | sock_release(sock); | |
1170 | goto out; | |
1171 | } | |
1172 | } else | |
1173 | /* "close" the socket, preserving the local port */ | |
1174 | xs_tcp_reuse_connection(xprt); | |
a246b010 | 1175 | |
edb267a6 CL |
1176 | dprintk("RPC: worker connecting xprt %p to address: %s\n", |
1177 | xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL)); | |
1178 | ||
b0d93ad5 CL |
1179 | if (!xprt->inet) { |
1180 | struct sock *sk = sock->sk; | |
1181 | ||
1182 | write_lock_bh(&sk->sk_callback_lock); | |
1183 | ||
1184 | sk->sk_user_data = xprt; | |
1185 | xprt->old_data_ready = sk->sk_data_ready; | |
1186 | xprt->old_state_change = sk->sk_state_change; | |
1187 | xprt->old_write_space = sk->sk_write_space; | |
1188 | sk->sk_data_ready = xs_tcp_data_ready; | |
1189 | sk->sk_state_change = xs_tcp_state_change; | |
1190 | sk->sk_write_space = xs_tcp_write_space; | |
b079fa7b | 1191 | sk->sk_allocation = GFP_ATOMIC; |
3167e12c CL |
1192 | |
1193 | /* socket options */ | |
1194 | sk->sk_userlocks |= SOCK_BINDPORT_LOCK; | |
1195 | sock_reset_flag(sk, SOCK_LINGER); | |
1196 | tcp_sk(sk)->linger2 = 0; | |
1197 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; | |
b0d93ad5 CL |
1198 | |
1199 | xprt_clear_connected(xprt); | |
1200 | ||
1201 | /* Reset to new socket */ | |
1202 | xprt->sock = sock; | |
1203 | xprt->inet = sk; | |
1204 | ||
1205 | write_unlock_bh(&sk->sk_callback_lock); | |
1206 | } | |
1207 | ||
1208 | /* Tell the socket layer to start connecting... */ | |
262ca07d CL |
1209 | xprt->stat.connect_count++; |
1210 | xprt->stat.connect_start = jiffies; | |
a246b010 CL |
1211 | status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr, |
1212 | sizeof(xprt->addr), O_NONBLOCK); | |
1213 | dprintk("RPC: %p connect status %d connected %d sock state %d\n", | |
1214 | xprt, -status, xprt_connected(xprt), sock->sk->sk_state); | |
1215 | if (status < 0) { | |
1216 | switch (status) { | |
1217 | case -EINPROGRESS: | |
1218 | case -EALREADY: | |
1219 | goto out_clear; | |
3167e12c CL |
1220 | case -ECONNREFUSED: |
1221 | case -ECONNRESET: | |
1222 | /* retry with existing socket, after a delay */ | |
1223 | break; | |
1224 | default: | |
1225 | /* get rid of existing socket, and retry */ | |
1226 | xs_close(xprt); | |
1227 | break; | |
a246b010 CL |
1228 | } |
1229 | } | |
1230 | out: | |
44fbac22 | 1231 | xprt_wake_pending_tasks(xprt, status); |
a246b010 | 1232 | out_clear: |
2226feb6 | 1233 | xprt_clear_connecting(xprt); |
a246b010 CL |
1234 | } |
1235 | ||
9903cd1c CL |
1236 | /** |
1237 | * xs_connect - connect a socket to a remote endpoint | |
1238 | * @task: address of RPC task that manages state of connect request | |
1239 | * | |
1240 | * TCP: If the remote end dropped the connection, delay reconnecting. | |
03bf4b70 CL |
1241 | * |
1242 | * UDP socket connects are synchronous, but we use a work queue anyway | |
1243 | * to guarantee that even unprivileged user processes can set up a | |
1244 | * socket on a privileged port. | |
1245 | * | |
1246 | * If a UDP socket connect fails, the delay behavior here prevents | |
1247 | * retry floods (hard mounts). | |
9903cd1c CL |
1248 | */ |
1249 | static void xs_connect(struct rpc_task *task) | |
a246b010 CL |
1250 | { |
1251 | struct rpc_xprt *xprt = task->tk_xprt; | |
1252 | ||
b0d93ad5 CL |
1253 | if (xprt_test_and_set_connecting(xprt)) |
1254 | return; | |
1255 | ||
1256 | if (xprt->sock != NULL) { | |
03bf4b70 CL |
1257 | dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n", |
1258 | xprt, xprt->reestablish_timeout / HZ); | |
b0d93ad5 | 1259 | schedule_delayed_work(&xprt->connect_worker, |
03bf4b70 CL |
1260 | xprt->reestablish_timeout); |
1261 | xprt->reestablish_timeout <<= 1; | |
1262 | if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) | |
1263 | xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; | |
b0d93ad5 CL |
1264 | } else { |
1265 | dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); | |
1266 | schedule_work(&xprt->connect_worker); | |
1267 | ||
1268 | /* flush_scheduled_work can sleep... */ | |
1269 | if (!RPC_IS_ASYNC(task)) | |
1270 | flush_scheduled_work(); | |
a246b010 CL |
1271 | } |
1272 | } | |
1273 | ||
262ca07d CL |
1274 | /** |
1275 | * xs_udp_print_stats - display UDP socket-specifc stats | |
1276 | * @xprt: rpc_xprt struct containing statistics | |
1277 | * @seq: output file | |
1278 | * | |
1279 | */ | |
1280 | static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1281 | { | |
1282 | seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n", | |
1283 | xprt->port, | |
1284 | xprt->stat.bind_count, | |
1285 | xprt->stat.sends, | |
1286 | xprt->stat.recvs, | |
1287 | xprt->stat.bad_xids, | |
1288 | xprt->stat.req_u, | |
1289 | xprt->stat.bklog_u); | |
1290 | } | |
1291 | ||
1292 | /** | |
1293 | * xs_tcp_print_stats - display TCP socket-specifc stats | |
1294 | * @xprt: rpc_xprt struct containing statistics | |
1295 | * @seq: output file | |
1296 | * | |
1297 | */ | |
1298 | static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1299 | { | |
1300 | long idle_time = 0; | |
1301 | ||
1302 | if (xprt_connected(xprt)) | |
1303 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
1304 | ||
1305 | seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n", | |
1306 | xprt->port, | |
1307 | xprt->stat.bind_count, | |
1308 | xprt->stat.connect_count, | |
1309 | xprt->stat.connect_time, | |
1310 | idle_time, | |
1311 | xprt->stat.sends, | |
1312 | xprt->stat.recvs, | |
1313 | xprt->stat.bad_xids, | |
1314 | xprt->stat.req_u, | |
1315 | xprt->stat.bklog_u); | |
1316 | } | |
1317 | ||
262965f5 | 1318 | static struct rpc_xprt_ops xs_udp_ops = { |
43118c29 | 1319 | .set_buffer_size = xs_udp_set_buffer_size, |
edb267a6 | 1320 | .print_addr = xs_print_peer_address, |
12a80469 | 1321 | .reserve_xprt = xprt_reserve_xprt_cong, |
49e9a890 | 1322 | .release_xprt = xprt_release_xprt_cong, |
bbf7c1dd | 1323 | .rpcbind = rpc_getport, |
92200412 | 1324 | .set_port = xs_set_port, |
262965f5 | 1325 | .connect = xs_connect, |
02107148 CL |
1326 | .buf_alloc = rpc_malloc, |
1327 | .buf_free = rpc_free, | |
262965f5 | 1328 | .send_request = xs_udp_send_request, |
fe3aca29 | 1329 | .set_retrans_timeout = xprt_set_retrans_timeout_rtt, |
46c0ee8b | 1330 | .timer = xs_udp_timer, |
a58dd398 | 1331 | .release_request = xprt_release_rqst_cong, |
262965f5 CL |
1332 | .close = xs_close, |
1333 | .destroy = xs_destroy, | |
262ca07d | 1334 | .print_stats = xs_udp_print_stats, |
262965f5 CL |
1335 | }; |
1336 | ||
1337 | static struct rpc_xprt_ops xs_tcp_ops = { | |
edb267a6 | 1338 | .print_addr = xs_print_peer_address, |
12a80469 | 1339 | .reserve_xprt = xprt_reserve_xprt, |
e0ab53de | 1340 | .release_xprt = xs_tcp_release_xprt, |
bbf7c1dd | 1341 | .rpcbind = rpc_getport, |
92200412 | 1342 | .set_port = xs_set_port, |
9903cd1c | 1343 | .connect = xs_connect, |
02107148 CL |
1344 | .buf_alloc = rpc_malloc, |
1345 | .buf_free = rpc_free, | |
262965f5 | 1346 | .send_request = xs_tcp_send_request, |
fe3aca29 | 1347 | .set_retrans_timeout = xprt_set_retrans_timeout_def, |
9903cd1c CL |
1348 | .close = xs_close, |
1349 | .destroy = xs_destroy, | |
262ca07d | 1350 | .print_stats = xs_tcp_print_stats, |
a246b010 CL |
1351 | }; |
1352 | ||
9903cd1c CL |
1353 | /** |
1354 | * xs_setup_udp - Set up transport to use a UDP socket | |
1355 | * @xprt: transport to set up | |
1356 | * @to: timeout parameters | |
1357 | * | |
1358 | */ | |
a246b010 CL |
1359 | int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
1360 | { | |
1361 | size_t slot_table_size; | |
1362 | ||
a246b010 CL |
1363 | xprt->max_reqs = xprt_udp_slot_table_entries; |
1364 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
0da974f4 | 1365 | xprt->slot = kzalloc(slot_table_size, GFP_KERNEL); |
a246b010 CL |
1366 | if (xprt->slot == NULL) |
1367 | return -ENOMEM; | |
a246b010 | 1368 | |
ec739ef0 CL |
1369 | if (ntohs(xprt->addr.sin_port) != 0) |
1370 | xprt_set_bound(xprt); | |
b85d8806 | 1371 | xprt->port = xs_get_random_port(); |
ec739ef0 CL |
1372 | |
1373 | xprt->prot = IPPROTO_UDP; | |
808012fb | 1374 | xprt->tsh_size = 0; |
a246b010 CL |
1375 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; |
1376 | /* XXX: header size can vary due to auth type, IPv6, etc. */ | |
1377 | xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); | |
1378 | ||
b0d93ad5 | 1379 | INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt); |
03bf4b70 CL |
1380 | xprt->bind_timeout = XS_BIND_TO; |
1381 | xprt->connect_timeout = XS_UDP_CONN_TO; | |
1382 | xprt->reestablish_timeout = XS_UDP_REEST_TO; | |
1383 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 1384 | |
262965f5 | 1385 | xprt->ops = &xs_udp_ops; |
a246b010 CL |
1386 | |
1387 | if (to) | |
1388 | xprt->timeout = *to; | |
1389 | else | |
9903cd1c | 1390 | xprt_set_timeout(&xprt->timeout, 5, 5 * HZ); |
a246b010 | 1391 | |
edb267a6 CL |
1392 | xs_format_peer_addresses(xprt); |
1393 | dprintk("RPC: set up transport to address %s\n", | |
1394 | xs_print_peer_address(xprt, RPC_DISPLAY_ALL)); | |
1395 | ||
a246b010 CL |
1396 | return 0; |
1397 | } | |
1398 | ||
9903cd1c CL |
1399 | /** |
1400 | * xs_setup_tcp - Set up transport to use a TCP socket | |
1401 | * @xprt: transport to set up | |
1402 | * @to: timeout parameters | |
1403 | * | |
1404 | */ | |
a246b010 CL |
1405 | int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
1406 | { | |
1407 | size_t slot_table_size; | |
1408 | ||
a246b010 CL |
1409 | xprt->max_reqs = xprt_tcp_slot_table_entries; |
1410 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
0da974f4 | 1411 | xprt->slot = kzalloc(slot_table_size, GFP_KERNEL); |
a246b010 CL |
1412 | if (xprt->slot == NULL) |
1413 | return -ENOMEM; | |
a246b010 | 1414 | |
ec739ef0 CL |
1415 | if (ntohs(xprt->addr.sin_port) != 0) |
1416 | xprt_set_bound(xprt); | |
b85d8806 | 1417 | xprt->port = xs_get_random_port(); |
ec739ef0 CL |
1418 | |
1419 | xprt->prot = IPPROTO_TCP; | |
808012fb | 1420 | xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); |
a246b010 | 1421 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; |
808012fb | 1422 | xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; |
a246b010 | 1423 | |
b0d93ad5 | 1424 | INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt); |
03bf4b70 CL |
1425 | xprt->bind_timeout = XS_BIND_TO; |
1426 | xprt->connect_timeout = XS_TCP_CONN_TO; | |
1427 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
1428 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 1429 | |
262965f5 | 1430 | xprt->ops = &xs_tcp_ops; |
a246b010 CL |
1431 | |
1432 | if (to) | |
1433 | xprt->timeout = *to; | |
1434 | else | |
9903cd1c | 1435 | xprt_set_timeout(&xprt->timeout, 2, 60 * HZ); |
a246b010 | 1436 | |
edb267a6 CL |
1437 | xs_format_peer_addresses(xprt); |
1438 | dprintk("RPC: set up transport to address %s\n", | |
1439 | xs_print_peer_address(xprt, RPC_DISPLAY_ALL)); | |
1440 | ||
a246b010 CL |
1441 | return 0; |
1442 | } |