]>
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> | |
14 | */ | |
15 | ||
16 | #include <linux/types.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/capability.h> | |
19 | #include <linux/sched.h> | |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/socket.h> | |
23 | #include <linux/in.h> | |
24 | #include <linux/net.h> | |
25 | #include <linux/mm.h> | |
26 | #include <linux/udp.h> | |
27 | #include <linux/tcp.h> | |
28 | #include <linux/sunrpc/clnt.h> | |
29 | #include <linux/file.h> | |
30 | ||
31 | #include <net/sock.h> | |
32 | #include <net/checksum.h> | |
33 | #include <net/udp.h> | |
34 | #include <net/tcp.h> | |
35 | ||
9903cd1c CL |
36 | /* |
37 | * Maximum port number to use when requesting a reserved port. | |
38 | */ | |
39 | #define XS_MAX_RESVPORT (800U) | |
40 | ||
a246b010 CL |
41 | #ifdef RPC_DEBUG |
42 | # undef RPC_DEBUG_DATA | |
9903cd1c | 43 | # define RPCDBG_FACILITY RPCDBG_TRANS |
a246b010 CL |
44 | #endif |
45 | ||
a246b010 | 46 | #ifdef RPC_DEBUG_DATA |
9903cd1c | 47 | static void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 | 48 | { |
9903cd1c CL |
49 | u8 *buf = (u8 *) packet; |
50 | int j; | |
a246b010 CL |
51 | |
52 | dprintk("RPC: %s\n", msg); | |
53 | for (j = 0; j < count && j < 128; j += 4) { | |
54 | if (!(j & 31)) { | |
55 | if (j) | |
56 | dprintk("\n"); | |
57 | dprintk("0x%04x ", j); | |
58 | } | |
59 | dprintk("%02x%02x%02x%02x ", | |
60 | buf[j], buf[j+1], buf[j+2], buf[j+3]); | |
61 | } | |
62 | dprintk("\n"); | |
63 | } | |
64 | #else | |
9903cd1c | 65 | static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 CL |
66 | { |
67 | /* NOP */ | |
68 | } | |
69 | #endif | |
70 | ||
b4b5cc85 CL |
71 | #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) |
72 | ||
73 | static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len) | |
74 | { | |
75 | struct kvec iov = { | |
76 | .iov_base = xdr->head[0].iov_base + base, | |
77 | .iov_len = len - base, | |
78 | }; | |
79 | struct msghdr msg = { | |
80 | .msg_name = addr, | |
81 | .msg_namelen = addrlen, | |
82 | .msg_flags = XS_SENDMSG_FLAGS, | |
83 | }; | |
84 | ||
85 | if (xdr->len > len) | |
86 | msg.msg_flags |= MSG_MORE; | |
87 | ||
88 | if (likely(iov.iov_len)) | |
89 | return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); | |
90 | return kernel_sendmsg(sock, &msg, NULL, 0, 0); | |
91 | } | |
92 | ||
93 | static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len) | |
94 | { | |
95 | struct kvec iov = { | |
96 | .iov_base = xdr->tail[0].iov_base + base, | |
97 | .iov_len = len - base, | |
98 | }; | |
99 | struct msghdr msg = { | |
100 | .msg_flags = XS_SENDMSG_FLAGS, | |
101 | }; | |
102 | ||
103 | return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); | |
104 | } | |
105 | ||
9903cd1c CL |
106 | /** |
107 | * xs_sendpages - write pages directly to a socket | |
108 | * @sock: socket to send on | |
109 | * @addr: UDP only -- address of destination | |
110 | * @addrlen: UDP only -- length of destination address | |
111 | * @xdr: buffer containing this request | |
112 | * @base: starting position in the buffer | |
113 | * | |
a246b010 | 114 | */ |
b4b5cc85 | 115 | static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base) |
a246b010 CL |
116 | { |
117 | struct page **ppage = xdr->pages; | |
118 | unsigned int len, pglen = xdr->page_len; | |
119 | int err, ret = 0; | |
120 | ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int); | |
121 | ||
122 | len = xdr->head[0].iov_len; | |
123 | if (base < len || (addr != NULL && base == 0)) { | |
b4b5cc85 | 124 | err = xs_send_head(sock, addr, addrlen, xdr, base, len); |
a246b010 CL |
125 | if (ret == 0) |
126 | ret = err; | |
127 | else if (err > 0) | |
128 | ret += err; | |
b4b5cc85 | 129 | if (err != (len - base)) |
a246b010 CL |
130 | goto out; |
131 | base = 0; | |
132 | } else | |
133 | base -= len; | |
134 | ||
b4b5cc85 | 135 | if (unlikely(pglen == 0)) |
a246b010 | 136 | goto copy_tail; |
b4b5cc85 | 137 | if (unlikely(base >= pglen)) { |
a246b010 CL |
138 | base -= pglen; |
139 | goto copy_tail; | |
140 | } | |
141 | if (base || xdr->page_base) { | |
142 | pglen -= base; | |
9903cd1c | 143 | base += xdr->page_base; |
a246b010 CL |
144 | ppage += base >> PAGE_CACHE_SHIFT; |
145 | base &= ~PAGE_CACHE_MASK; | |
146 | } | |
147 | ||
148 | sendpage = sock->ops->sendpage ? : sock_no_sendpage; | |
149 | do { | |
b4b5cc85 | 150 | int flags = XS_SENDMSG_FLAGS; |
a246b010 CL |
151 | |
152 | len = PAGE_CACHE_SIZE; | |
153 | if (base) | |
154 | len -= base; | |
155 | if (pglen < len) | |
156 | len = pglen; | |
157 | ||
158 | if (pglen != len || xdr->tail[0].iov_len != 0) | |
159 | flags |= MSG_MORE; | |
160 | ||
161 | /* Hmm... We might be dealing with highmem pages */ | |
162 | if (PageHighMem(*ppage)) | |
163 | sendpage = sock_no_sendpage; | |
164 | err = sendpage(sock, *ppage, base, len, flags); | |
165 | if (ret == 0) | |
166 | ret = err; | |
167 | else if (err > 0) | |
168 | ret += err; | |
169 | if (err != len) | |
170 | goto out; | |
171 | base = 0; | |
172 | ppage++; | |
173 | } while ((pglen -= len) != 0); | |
174 | copy_tail: | |
175 | len = xdr->tail[0].iov_len; | |
176 | if (base < len) { | |
b4b5cc85 | 177 | err = xs_send_tail(sock, xdr, base, len); |
a246b010 CL |
178 | if (ret == 0) |
179 | ret = err; | |
180 | else if (err > 0) | |
181 | ret += err; | |
182 | } | |
183 | out: | |
184 | return ret; | |
185 | } | |
186 | ||
9903cd1c CL |
187 | /** |
188 | * xs_sendmsg - write an RPC request to a socket | |
189 | * @xprt: generic transport | |
190 | * @req: the RPC request to write | |
191 | * | |
a246b010 | 192 | */ |
9903cd1c | 193 | static int xs_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req) |
a246b010 | 194 | { |
9903cd1c CL |
195 | struct socket *sock = xprt->sock; |
196 | struct xdr_buf *xdr = &req->rq_snd_buf; | |
a246b010 CL |
197 | struct sockaddr *addr = NULL; |
198 | int addrlen = 0; | |
9903cd1c CL |
199 | unsigned int skip; |
200 | int result; | |
a246b010 CL |
201 | |
202 | if (!sock) | |
203 | return -ENOTCONN; | |
204 | ||
9903cd1c | 205 | xs_pktdump("packet data:", |
a246b010 CL |
206 | req->rq_svec->iov_base, |
207 | req->rq_svec->iov_len); | |
208 | ||
209 | /* For UDP, we need to provide an address */ | |
210 | if (!xprt->stream) { | |
211 | addr = (struct sockaddr *) &xprt->addr; | |
212 | addrlen = sizeof(xprt->addr); | |
213 | } | |
9903cd1c | 214 | /* Don't repeat bytes */ |
a246b010 CL |
215 | skip = req->rq_bytes_sent; |
216 | ||
217 | clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); | |
b4b5cc85 | 218 | result = xs_sendpages(sock, addr, addrlen, xdr, skip); |
a246b010 | 219 | |
9903cd1c | 220 | dprintk("RPC: xs_sendmsg(%d) = %d\n", xdr->len - skip, result); |
a246b010 CL |
221 | |
222 | if (result >= 0) | |
223 | return result; | |
224 | ||
225 | switch (result) { | |
226 | case -ECONNREFUSED: | |
227 | /* When the server has died, an ICMP port unreachable message | |
9903cd1c | 228 | * prompts ECONNREFUSED. */ |
a246b010 CL |
229 | case -EAGAIN: |
230 | break; | |
231 | case -ECONNRESET: | |
232 | case -ENOTCONN: | |
233 | case -EPIPE: | |
234 | /* connection broken */ | |
235 | if (xprt->stream) | |
236 | result = -ENOTCONN; | |
237 | break; | |
238 | default: | |
9903cd1c | 239 | break; |
a246b010 CL |
240 | } |
241 | return result; | |
242 | } | |
243 | ||
9903cd1c CL |
244 | /** |
245 | * xs_send_request - write an RPC request to a socket | |
246 | * @task: address of RPC task that manages the state of an RPC request | |
247 | * | |
248 | * Return values: | |
249 | * 0: The request has been sent | |
250 | * EAGAIN: The socket was blocked, please call again later to | |
251 | * complete the request | |
252 | * other: Some other error occured, the request was not sent | |
253 | * | |
254 | * XXX: In the case of soft timeouts, should we eventually give up | |
255 | * if the socket is not able to make progress? | |
256 | */ | |
257 | static int xs_send_request(struct rpc_task *task) | |
a246b010 CL |
258 | { |
259 | struct rpc_rqst *req = task->tk_rqstp; | |
260 | struct rpc_xprt *xprt = req->rq_xprt; | |
261 | int status, retry = 0; | |
262 | ||
263 | /* set up everything as needed. */ | |
264 | /* Write the record marker */ | |
265 | if (xprt->stream) { | |
9903cd1c | 266 | u32 *marker = req->rq_svec[0].iov_base; |
a246b010 CL |
267 | |
268 | *marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker))); | |
269 | } | |
270 | ||
271 | /* Continue transmitting the packet/record. We must be careful | |
272 | * to cope with writespace callbacks arriving _after_ we have | |
9903cd1c | 273 | * called sendmsg(). |
a246b010 CL |
274 | */ |
275 | while (1) { | |
276 | req->rq_xtime = jiffies; | |
9903cd1c | 277 | status = xs_sendmsg(xprt, req); |
a246b010 CL |
278 | |
279 | if (status < 0) | |
280 | break; | |
281 | ||
282 | if (xprt->stream) { | |
283 | req->rq_bytes_sent += status; | |
284 | ||
285 | /* If we've sent the entire packet, immediately | |
286 | * reset the count of bytes sent. */ | |
287 | if (req->rq_bytes_sent >= req->rq_slen) { | |
288 | req->rq_bytes_sent = 0; | |
289 | return 0; | |
290 | } | |
291 | } else { | |
292 | if (status >= req->rq_slen) | |
293 | return 0; | |
294 | status = -EAGAIN; | |
295 | break; | |
296 | } | |
297 | ||
298 | dprintk("RPC: %4d xmit incomplete (%d left of %d)\n", | |
299 | task->tk_pid, req->rq_slen - req->rq_bytes_sent, | |
300 | req->rq_slen); | |
301 | ||
302 | status = -EAGAIN; | |
303 | if (retry++ > 50) | |
304 | break; | |
305 | } | |
306 | ||
307 | if (status == -EAGAIN) { | |
308 | if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) { | |
9903cd1c | 309 | /* Protect against races with xs_write_space */ |
a246b010 CL |
310 | spin_lock_bh(&xprt->sock_lock); |
311 | /* Don't race with disconnect */ | |
312 | if (!xprt_connected(xprt)) | |
313 | task->tk_status = -ENOTCONN; | |
314 | else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags)) { | |
315 | task->tk_timeout = req->rq_timeout; | |
316 | rpc_sleep_on(&xprt->pending, task, NULL, NULL); | |
317 | } | |
318 | spin_unlock_bh(&xprt->sock_lock); | |
319 | return status; | |
320 | } | |
321 | /* Keep holding the socket if it is blocked */ | |
322 | rpc_delay(task, HZ>>4); | |
323 | } | |
324 | return status; | |
325 | } | |
326 | ||
9903cd1c CL |
327 | /** |
328 | * xs_close - close a socket | |
329 | * @xprt: transport | |
330 | * | |
a246b010 | 331 | */ |
9903cd1c | 332 | static void xs_close(struct rpc_xprt *xprt) |
a246b010 | 333 | { |
9903cd1c CL |
334 | struct socket *sock = xprt->sock; |
335 | struct sock *sk = xprt->inet; | |
a246b010 CL |
336 | |
337 | if (!sk) | |
338 | return; | |
339 | ||
9903cd1c CL |
340 | dprintk("RPC: xs_close xprt %p\n", xprt); |
341 | ||
a246b010 CL |
342 | write_lock_bh(&sk->sk_callback_lock); |
343 | xprt->inet = NULL; | |
344 | xprt->sock = NULL; | |
345 | ||
9903cd1c CL |
346 | sk->sk_user_data = NULL; |
347 | sk->sk_data_ready = xprt->old_data_ready; | |
a246b010 | 348 | sk->sk_state_change = xprt->old_state_change; |
9903cd1c | 349 | sk->sk_write_space = xprt->old_write_space; |
a246b010 CL |
350 | write_unlock_bh(&sk->sk_callback_lock); |
351 | ||
9903cd1c | 352 | sk->sk_no_check = 0; |
a246b010 CL |
353 | |
354 | sock_release(sock); | |
355 | } | |
356 | ||
9903cd1c CL |
357 | /** |
358 | * xs_destroy - prepare to shutdown a transport | |
359 | * @xprt: doomed transport | |
360 | * | |
361 | */ | |
362 | static void xs_destroy(struct rpc_xprt *xprt) | |
a246b010 | 363 | { |
9903cd1c CL |
364 | dprintk("RPC: xs_destroy xprt %p\n", xprt); |
365 | ||
a246b010 CL |
366 | cancel_delayed_work(&xprt->sock_connect); |
367 | flush_scheduled_work(); | |
368 | ||
369 | xprt_disconnect(xprt); | |
9903cd1c | 370 | xs_close(xprt); |
a246b010 CL |
371 | kfree(xprt->slot); |
372 | } | |
373 | ||
9903cd1c CL |
374 | static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) |
375 | { | |
376 | return (struct rpc_xprt *) sk->sk_user_data; | |
377 | } | |
378 | ||
379 | /** | |
380 | * xs_udp_data_ready - "data ready" callback for UDP sockets | |
381 | * @sk: socket with data to read | |
382 | * @len: how much data to read | |
383 | * | |
a246b010 | 384 | */ |
9903cd1c | 385 | static void xs_udp_data_ready(struct sock *sk, int len) |
a246b010 | 386 | { |
9903cd1c CL |
387 | struct rpc_task *task; |
388 | struct rpc_xprt *xprt; | |
a246b010 | 389 | struct rpc_rqst *rovr; |
9903cd1c | 390 | struct sk_buff *skb; |
a246b010 CL |
391 | int err, repsize, copied; |
392 | u32 _xid, *xp; | |
393 | ||
394 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
395 | dprintk("RPC: xs_udp_data_ready...\n"); |
396 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 397 | goto out; |
a246b010 CL |
398 | |
399 | if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) | |
400 | goto out; | |
401 | ||
402 | if (xprt->shutdown) | |
403 | goto dropit; | |
404 | ||
405 | repsize = skb->len - sizeof(struct udphdr); | |
406 | if (repsize < 4) { | |
9903cd1c | 407 | dprintk("RPC: impossible RPC reply size %d!\n", repsize); |
a246b010 CL |
408 | goto dropit; |
409 | } | |
410 | ||
411 | /* Copy the XID from the skb... */ | |
412 | xp = skb_header_pointer(skb, sizeof(struct udphdr), | |
413 | sizeof(_xid), &_xid); | |
414 | if (xp == NULL) | |
415 | goto dropit; | |
416 | ||
417 | /* Look up and lock the request corresponding to the given XID */ | |
418 | spin_lock(&xprt->sock_lock); | |
419 | rovr = xprt_lookup_rqst(xprt, *xp); | |
420 | if (!rovr) | |
421 | goto out_unlock; | |
422 | task = rovr->rq_task; | |
423 | ||
424 | dprintk("RPC: %4d received reply\n", task->tk_pid); | |
425 | ||
426 | if ((copied = rovr->rq_private_buf.buflen) > repsize) | |
427 | copied = repsize; | |
428 | ||
429 | /* Suck it into the iovec, verify checksum if not done by hw. */ | |
430 | if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) | |
431 | goto out_unlock; | |
432 | ||
433 | /* Something worked... */ | |
434 | dst_confirm(skb->dst); | |
435 | ||
436 | xprt_complete_rqst(xprt, rovr, copied); | |
437 | ||
438 | out_unlock: | |
439 | spin_unlock(&xprt->sock_lock); | |
440 | dropit: | |
441 | skb_free_datagram(sk, skb); | |
442 | out: | |
443 | read_unlock(&sk->sk_callback_lock); | |
444 | } | |
445 | ||
9903cd1c | 446 | static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len) |
a246b010 CL |
447 | { |
448 | if (len > desc->count) | |
449 | len = desc->count; | |
450 | if (skb_copy_bits(desc->skb, desc->offset, p, len)) { | |
451 | dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n", | |
452 | len, desc->count); | |
453 | return 0; | |
454 | } | |
455 | desc->offset += len; | |
456 | desc->count -= len; | |
457 | dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n", | |
458 | len, desc->count); | |
459 | return len; | |
460 | } | |
461 | ||
9903cd1c | 462 | static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
463 | { |
464 | size_t len, used; | |
465 | char *p; | |
466 | ||
467 | p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset; | |
468 | len = sizeof(xprt->tcp_recm) - xprt->tcp_offset; | |
9903cd1c | 469 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
470 | xprt->tcp_offset += used; |
471 | if (used != len) | |
472 | return; | |
473 | xprt->tcp_reclen = ntohl(xprt->tcp_recm); | |
474 | if (xprt->tcp_reclen & 0x80000000) | |
475 | xprt->tcp_flags |= XPRT_LAST_FRAG; | |
476 | else | |
477 | xprt->tcp_flags &= ~XPRT_LAST_FRAG; | |
478 | xprt->tcp_reclen &= 0x7fffffff; | |
479 | xprt->tcp_flags &= ~XPRT_COPY_RECM; | |
480 | xprt->tcp_offset = 0; | |
481 | /* Sanity check of the record length */ | |
482 | if (xprt->tcp_reclen < 4) { | |
9903cd1c | 483 | dprintk("RPC: invalid TCP record fragment length\n"); |
a246b010 | 484 | xprt_disconnect(xprt); |
9903cd1c | 485 | return; |
a246b010 CL |
486 | } |
487 | dprintk("RPC: reading TCP record fragment of length %d\n", | |
488 | xprt->tcp_reclen); | |
489 | } | |
490 | ||
9903cd1c | 491 | static void xs_tcp_check_recm(struct rpc_xprt *xprt) |
a246b010 CL |
492 | { |
493 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n", | |
494 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags); | |
495 | if (xprt->tcp_offset == xprt->tcp_reclen) { | |
496 | xprt->tcp_flags |= XPRT_COPY_RECM; | |
497 | xprt->tcp_offset = 0; | |
498 | if (xprt->tcp_flags & XPRT_LAST_FRAG) { | |
499 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
500 | xprt->tcp_flags |= XPRT_COPY_XID; | |
501 | xprt->tcp_copied = 0; | |
502 | } | |
503 | } | |
504 | } | |
505 | ||
9903cd1c | 506 | static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
507 | { |
508 | size_t len, used; | |
509 | char *p; | |
510 | ||
511 | len = sizeof(xprt->tcp_xid) - xprt->tcp_offset; | |
512 | dprintk("RPC: reading XID (%Zu bytes)\n", len); | |
513 | p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset; | |
9903cd1c | 514 | used = xs_tcp_copy_data(desc, p, len); |
a246b010 CL |
515 | xprt->tcp_offset += used; |
516 | if (used != len) | |
517 | return; | |
518 | xprt->tcp_flags &= ~XPRT_COPY_XID; | |
519 | xprt->tcp_flags |= XPRT_COPY_DATA; | |
520 | xprt->tcp_copied = 4; | |
521 | dprintk("RPC: reading reply for XID %08x\n", | |
522 | ntohl(xprt->tcp_xid)); | |
9903cd1c | 523 | xs_tcp_check_recm(xprt); |
a246b010 CL |
524 | } |
525 | ||
9903cd1c | 526 | static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
527 | { |
528 | struct rpc_rqst *req; | |
529 | struct xdr_buf *rcvbuf; | |
530 | size_t len; | |
531 | ssize_t r; | |
532 | ||
533 | /* Find and lock the request corresponding to this xid */ | |
534 | spin_lock(&xprt->sock_lock); | |
535 | req = xprt_lookup_rqst(xprt, xprt->tcp_xid); | |
536 | if (!req) { | |
537 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
538 | dprintk("RPC: XID %08x request not found!\n", | |
539 | ntohl(xprt->tcp_xid)); | |
540 | spin_unlock(&xprt->sock_lock); | |
541 | return; | |
542 | } | |
543 | ||
544 | rcvbuf = &req->rq_private_buf; | |
545 | len = desc->count; | |
546 | if (len > xprt->tcp_reclen - xprt->tcp_offset) { | |
547 | skb_reader_t my_desc; | |
548 | ||
549 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
550 | memcpy(&my_desc, desc, sizeof(my_desc)); | |
551 | my_desc.count = len; | |
552 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 553 | &my_desc, xs_tcp_copy_data); |
a246b010 CL |
554 | desc->count -= r; |
555 | desc->offset += r; | |
556 | } else | |
557 | r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied, | |
9903cd1c | 558 | desc, xs_tcp_copy_data); |
a246b010 CL |
559 | |
560 | if (r > 0) { | |
561 | xprt->tcp_copied += r; | |
562 | xprt->tcp_offset += r; | |
563 | } | |
564 | if (r != len) { | |
565 | /* Error when copying to the receive buffer, | |
566 | * usually because we weren't able to allocate | |
567 | * additional buffer pages. All we can do now | |
568 | * is turn off XPRT_COPY_DATA, so the request | |
569 | * will not receive any additional updates, | |
570 | * and time out. | |
571 | * Any remaining data from this record will | |
572 | * be discarded. | |
573 | */ | |
574 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
575 | dprintk("RPC: XID %08x truncated request\n", | |
576 | ntohl(xprt->tcp_xid)); | |
577 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
578 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
579 | goto out; | |
580 | } | |
581 | ||
582 | dprintk("RPC: XID %08x read %Zd bytes\n", | |
583 | ntohl(xprt->tcp_xid), r); | |
584 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n", | |
585 | xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen); | |
586 | ||
587 | if (xprt->tcp_copied == req->rq_private_buf.buflen) | |
588 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
589 | else if (xprt->tcp_offset == xprt->tcp_reclen) { | |
590 | if (xprt->tcp_flags & XPRT_LAST_FRAG) | |
591 | xprt->tcp_flags &= ~XPRT_COPY_DATA; | |
592 | } | |
593 | ||
594 | out: | |
595 | if (!(xprt->tcp_flags & XPRT_COPY_DATA)) { | |
596 | dprintk("RPC: %4d received reply complete\n", | |
597 | req->rq_task->tk_pid); | |
598 | xprt_complete_rqst(xprt, req, xprt->tcp_copied); | |
599 | } | |
600 | spin_unlock(&xprt->sock_lock); | |
9903cd1c | 601 | xs_tcp_check_recm(xprt); |
a246b010 CL |
602 | } |
603 | ||
9903cd1c | 604 | static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc) |
a246b010 CL |
605 | { |
606 | size_t len; | |
607 | ||
608 | len = xprt->tcp_reclen - xprt->tcp_offset; | |
609 | if (len > desc->count) | |
610 | len = desc->count; | |
611 | desc->count -= len; | |
612 | desc->offset += len; | |
613 | xprt->tcp_offset += len; | |
614 | dprintk("RPC: discarded %Zu bytes\n", len); | |
9903cd1c | 615 | xs_tcp_check_recm(xprt); |
a246b010 CL |
616 | } |
617 | ||
9903cd1c | 618 | static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) |
a246b010 CL |
619 | { |
620 | struct rpc_xprt *xprt = rd_desc->arg.data; | |
621 | skb_reader_t desc = { | |
622 | .skb = skb, | |
623 | .offset = offset, | |
624 | .count = len, | |
625 | .csum = 0 | |
9903cd1c | 626 | }; |
a246b010 | 627 | |
9903cd1c | 628 | dprintk("RPC: xs_tcp_data_recv started\n"); |
a246b010 CL |
629 | do { |
630 | /* Read in a new fragment marker if necessary */ | |
631 | /* Can we ever really expect to get completely empty fragments? */ | |
632 | if (xprt->tcp_flags & XPRT_COPY_RECM) { | |
9903cd1c | 633 | xs_tcp_read_fraghdr(xprt, &desc); |
a246b010 CL |
634 | continue; |
635 | } | |
636 | /* Read in the xid if necessary */ | |
637 | if (xprt->tcp_flags & XPRT_COPY_XID) { | |
9903cd1c | 638 | xs_tcp_read_xid(xprt, &desc); |
a246b010 CL |
639 | continue; |
640 | } | |
641 | /* Read in the request data */ | |
642 | if (xprt->tcp_flags & XPRT_COPY_DATA) { | |
9903cd1c | 643 | xs_tcp_read_request(xprt, &desc); |
a246b010 CL |
644 | continue; |
645 | } | |
646 | /* Skip over any trailing bytes on short reads */ | |
9903cd1c | 647 | xs_tcp_read_discard(xprt, &desc); |
a246b010 | 648 | } while (desc.count); |
9903cd1c | 649 | dprintk("RPC: xs_tcp_data_recv done\n"); |
a246b010 CL |
650 | return len - desc.count; |
651 | } | |
652 | ||
9903cd1c CL |
653 | /** |
654 | * xs_tcp_data_ready - "data ready" callback for TCP sockets | |
655 | * @sk: socket with data to read | |
656 | * @bytes: how much data to read | |
657 | * | |
658 | */ | |
659 | static void xs_tcp_data_ready(struct sock *sk, int bytes) | |
a246b010 CL |
660 | { |
661 | struct rpc_xprt *xprt; | |
662 | read_descriptor_t rd_desc; | |
663 | ||
664 | read_lock(&sk->sk_callback_lock); | |
9903cd1c CL |
665 | dprintk("RPC: xs_tcp_data_ready...\n"); |
666 | if (!(xprt = xprt_from_sock(sk))) | |
a246b010 | 667 | goto out; |
a246b010 CL |
668 | if (xprt->shutdown) |
669 | goto out; | |
670 | ||
9903cd1c | 671 | /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ |
a246b010 CL |
672 | rd_desc.arg.data = xprt; |
673 | rd_desc.count = 65536; | |
9903cd1c | 674 | tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); |
a246b010 CL |
675 | out: |
676 | read_unlock(&sk->sk_callback_lock); | |
677 | } | |
678 | ||
9903cd1c CL |
679 | /** |
680 | * xs_tcp_state_change - callback to handle TCP socket state changes | |
681 | * @sk: socket whose state has changed | |
682 | * | |
683 | */ | |
684 | static void xs_tcp_state_change(struct sock *sk) | |
a246b010 | 685 | { |
9903cd1c | 686 | struct rpc_xprt *xprt; |
a246b010 CL |
687 | |
688 | read_lock(&sk->sk_callback_lock); | |
689 | if (!(xprt = xprt_from_sock(sk))) | |
690 | goto out; | |
9903cd1c | 691 | dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); |
a246b010 CL |
692 | dprintk("RPC: state %x conn %d dead %d zapped %d\n", |
693 | sk->sk_state, xprt_connected(xprt), | |
694 | sock_flag(sk, SOCK_DEAD), | |
695 | sock_flag(sk, SOCK_ZAPPED)); | |
696 | ||
697 | switch (sk->sk_state) { | |
698 | case TCP_ESTABLISHED: | |
699 | spin_lock_bh(&xprt->sock_lock); | |
700 | if (!xprt_test_and_set_connected(xprt)) { | |
701 | /* Reset TCP record info */ | |
702 | xprt->tcp_offset = 0; | |
703 | xprt->tcp_reclen = 0; | |
704 | xprt->tcp_copied = 0; | |
705 | xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID; | |
706 | rpc_wake_up(&xprt->pending); | |
707 | } | |
708 | spin_unlock_bh(&xprt->sock_lock); | |
709 | break; | |
710 | case TCP_SYN_SENT: | |
711 | case TCP_SYN_RECV: | |
712 | break; | |
713 | default: | |
714 | xprt_disconnect(xprt); | |
715 | break; | |
716 | } | |
717 | out: | |
718 | read_unlock(&sk->sk_callback_lock); | |
719 | } | |
720 | ||
9903cd1c CL |
721 | /** |
722 | * xs_write_space - callback invoked when socket buffer space becomes | |
723 | * available | |
724 | * @sk: socket whose state has changed | |
725 | * | |
a246b010 CL |
726 | * Called when more output buffer space is available for this socket. |
727 | * We try not to wake our writers until they can make "significant" | |
728 | * progress, otherwise we'll waste resources thrashing sock_sendmsg | |
729 | * with a bunch of small requests. | |
730 | */ | |
9903cd1c | 731 | static void xs_write_space(struct sock *sk) |
a246b010 | 732 | { |
9903cd1c CL |
733 | struct rpc_xprt *xprt; |
734 | struct socket *sock; | |
a246b010 CL |
735 | |
736 | read_lock(&sk->sk_callback_lock); | |
737 | if (!(xprt = xprt_from_sock(sk)) || !(sock = sk->sk_socket)) | |
738 | goto out; | |
739 | if (xprt->shutdown) | |
740 | goto out; | |
741 | ||
742 | /* Wait until we have enough socket memory */ | |
743 | if (xprt->stream) { | |
744 | /* from net/core/stream.c:sk_stream_write_space */ | |
745 | if (sk_stream_wspace(sk) < sk_stream_min_wspace(sk)) | |
746 | goto out; | |
747 | } else { | |
748 | /* from net/core/sock.c:sock_def_write_space */ | |
749 | if (!sock_writeable(sk)) | |
750 | goto out; | |
751 | } | |
752 | ||
753 | if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)) | |
754 | goto out; | |
755 | ||
756 | spin_lock_bh(&xprt->sock_lock); | |
757 | if (xprt->snd_task) | |
758 | rpc_wake_up_task(xprt->snd_task); | |
759 | spin_unlock_bh(&xprt->sock_lock); | |
760 | out: | |
761 | read_unlock(&sk->sk_callback_lock); | |
762 | } | |
763 | ||
9903cd1c CL |
764 | /** |
765 | * xs_set_buffer_size - set send and receive limits | |
766 | * @xprt: generic transport | |
767 | * | |
768 | * Set socket send and receive limits based on the | |
769 | * sndsize and rcvsize fields in the generic transport | |
770 | * structure. This applies only to UDP sockets. | |
a246b010 | 771 | */ |
9903cd1c | 772 | static void xs_set_buffer_size(struct rpc_xprt *xprt) |
a246b010 CL |
773 | { |
774 | struct sock *sk = xprt->inet; | |
775 | ||
776 | if (xprt->stream) | |
777 | return; | |
778 | if (xprt->rcvsize) { | |
779 | sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
780 | sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2; | |
781 | } | |
782 | if (xprt->sndsize) { | |
783 | sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
784 | sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2; | |
785 | sk->sk_write_space(sk); | |
786 | } | |
787 | } | |
788 | ||
9903cd1c | 789 | static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock) |
a246b010 CL |
790 | { |
791 | struct sockaddr_in myaddr = { | |
792 | .sin_family = AF_INET, | |
793 | }; | |
9903cd1c | 794 | int err, port; |
a246b010 CL |
795 | |
796 | /* Were we already bound to a given port? Try to reuse it */ | |
797 | port = xprt->port; | |
798 | do { | |
799 | myaddr.sin_port = htons(port); | |
800 | err = sock->ops->bind(sock, (struct sockaddr *) &myaddr, | |
801 | sizeof(myaddr)); | |
802 | if (err == 0) { | |
803 | xprt->port = port; | |
9903cd1c CL |
804 | dprintk("RPC: xs_bindresvport bound to port %u\n", |
805 | port); | |
a246b010 CL |
806 | return 0; |
807 | } | |
808 | if (--port == 0) | |
9903cd1c | 809 | port = XS_MAX_RESVPORT; |
a246b010 CL |
810 | } while (err == -EADDRINUSE && port != xprt->port); |
811 | ||
9903cd1c | 812 | dprintk("RPC: can't bind to reserved port (%d).\n", -err); |
a246b010 CL |
813 | return err; |
814 | } | |
815 | ||
9903cd1c | 816 | static struct socket *xs_create(struct rpc_xprt *xprt, int proto, int resvport) |
a246b010 | 817 | { |
9903cd1c CL |
818 | struct socket *sock; |
819 | int type, err; | |
820 | ||
821 | dprintk("RPC: xs_create(%s %d)\n", | |
822 | (proto == IPPROTO_UDP)? "udp" : "tcp", proto); | |
823 | ||
824 | type = (proto == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
825 | ||
826 | if ((err = sock_create_kern(PF_INET, type, proto, &sock)) < 0) { | |
827 | dprintk("RPC: can't create socket (%d).\n", -err); | |
828 | return NULL; | |
829 | } | |
830 | ||
831 | /* If the caller has the capability, bind to a reserved port */ | |
832 | if (resvport && xs_bindresvport(xprt, sock) < 0) | |
833 | goto failed; | |
834 | ||
835 | return sock; | |
836 | ||
837 | failed: | |
838 | sock_release(sock); | |
839 | return NULL; | |
840 | } | |
841 | ||
842 | static void xs_bind(struct rpc_xprt *xprt, struct socket *sock) | |
843 | { | |
844 | struct sock *sk = sock->sk; | |
a246b010 CL |
845 | |
846 | if (xprt->inet) | |
847 | return; | |
848 | ||
849 | write_lock_bh(&sk->sk_callback_lock); | |
850 | sk->sk_user_data = xprt; | |
851 | xprt->old_data_ready = sk->sk_data_ready; | |
852 | xprt->old_state_change = sk->sk_state_change; | |
853 | xprt->old_write_space = sk->sk_write_space; | |
854 | if (xprt->prot == IPPROTO_UDP) { | |
9903cd1c | 855 | sk->sk_data_ready = xs_udp_data_ready; |
a246b010 CL |
856 | sk->sk_no_check = UDP_CSUM_NORCV; |
857 | xprt_set_connected(xprt); | |
858 | } else { | |
859 | tcp_sk(sk)->nonagle = 1; /* disable Nagle's algorithm */ | |
9903cd1c CL |
860 | sk->sk_data_ready = xs_tcp_data_ready; |
861 | sk->sk_state_change = xs_tcp_state_change; | |
a246b010 CL |
862 | xprt_clear_connected(xprt); |
863 | } | |
9903cd1c | 864 | sk->sk_write_space = xs_write_space; |
a246b010 CL |
865 | |
866 | /* Reset to new socket */ | |
867 | xprt->sock = sock; | |
868 | xprt->inet = sk; | |
869 | write_unlock_bh(&sk->sk_callback_lock); | |
870 | ||
871 | return; | |
872 | } | |
873 | ||
9903cd1c CL |
874 | /** |
875 | * xs_connect_worker - try to connect a socket to a remote endpoint | |
876 | * @args: RPC transport to connect | |
877 | * | |
878 | * Invoked by a work queue tasklet. | |
a246b010 | 879 | */ |
9903cd1c | 880 | static void xs_connect_worker(void *args) |
a246b010 CL |
881 | { |
882 | struct rpc_xprt *xprt = (struct rpc_xprt *)args; | |
883 | struct socket *sock = xprt->sock; | |
884 | int status = -EIO; | |
885 | ||
886 | if (xprt->shutdown || xprt->addr.sin_port == 0) | |
887 | goto out; | |
888 | ||
9903cd1c CL |
889 | dprintk("RPC: xs_connect_worker xprt %p\n", xprt); |
890 | ||
a246b010 CL |
891 | /* |
892 | * Start by resetting any existing state | |
893 | */ | |
9903cd1c CL |
894 | xs_close(xprt); |
895 | sock = xs_create(xprt, xprt->prot, xprt->resvport); | |
a246b010 CL |
896 | if (sock == NULL) { |
897 | /* couldn't create socket or bind to reserved port; | |
898 | * this is likely a permanent error, so cause an abort */ | |
899 | goto out; | |
900 | } | |
9903cd1c CL |
901 | xs_bind(xprt, sock); |
902 | xs_set_buffer_size(xprt); | |
a246b010 CL |
903 | |
904 | status = 0; | |
905 | if (!xprt->stream) | |
906 | goto out; | |
907 | ||
908 | /* | |
909 | * Tell the socket layer to start connecting... | |
910 | */ | |
911 | status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr, | |
912 | sizeof(xprt->addr), O_NONBLOCK); | |
913 | dprintk("RPC: %p connect status %d connected %d sock state %d\n", | |
914 | xprt, -status, xprt_connected(xprt), sock->sk->sk_state); | |
915 | if (status < 0) { | |
916 | switch (status) { | |
917 | case -EINPROGRESS: | |
918 | case -EALREADY: | |
919 | goto out_clear; | |
920 | } | |
921 | } | |
922 | out: | |
923 | if (status < 0) | |
924 | rpc_wake_up_status(&xprt->pending, status); | |
925 | else | |
926 | rpc_wake_up(&xprt->pending); | |
927 | out_clear: | |
928 | smp_mb__before_clear_bit(); | |
929 | clear_bit(XPRT_CONNECTING, &xprt->sockstate); | |
930 | smp_mb__after_clear_bit(); | |
931 | } | |
932 | ||
9903cd1c CL |
933 | /** |
934 | * xs_connect - connect a socket to a remote endpoint | |
935 | * @task: address of RPC task that manages state of connect request | |
936 | * | |
937 | * TCP: If the remote end dropped the connection, delay reconnecting. | |
938 | */ | |
939 | static void xs_connect(struct rpc_task *task) | |
a246b010 CL |
940 | { |
941 | struct rpc_xprt *xprt = task->tk_xprt; | |
942 | ||
943 | if (!test_and_set_bit(XPRT_CONNECTING, &xprt->sockstate)) { | |
9903cd1c CL |
944 | if (xprt->sock != NULL) { |
945 | dprintk("RPC: xs_connect delayed xprt %p\n", xprt); | |
a246b010 CL |
946 | schedule_delayed_work(&xprt->sock_connect, |
947 | RPC_REESTABLISH_TIMEOUT); | |
9903cd1c CL |
948 | } else { |
949 | dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); | |
a246b010 CL |
950 | schedule_work(&xprt->sock_connect); |
951 | /* flush_scheduled_work can sleep... */ | |
952 | if (!RPC_IS_ASYNC(task)) | |
953 | flush_scheduled_work(); | |
954 | } | |
955 | } | |
956 | } | |
957 | ||
9903cd1c CL |
958 | static struct rpc_xprt_ops xs_ops = { |
959 | .set_buffer_size = xs_set_buffer_size, | |
960 | .connect = xs_connect, | |
961 | .send_request = xs_send_request, | |
962 | .close = xs_close, | |
963 | .destroy = xs_destroy, | |
a246b010 CL |
964 | }; |
965 | ||
966 | extern unsigned int xprt_udp_slot_table_entries; | |
967 | extern unsigned int xprt_tcp_slot_table_entries; | |
968 | ||
9903cd1c CL |
969 | /** |
970 | * xs_setup_udp - Set up transport to use a UDP socket | |
971 | * @xprt: transport to set up | |
972 | * @to: timeout parameters | |
973 | * | |
974 | */ | |
a246b010 CL |
975 | int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
976 | { | |
977 | size_t slot_table_size; | |
978 | ||
979 | dprintk("RPC: setting up udp-ipv4 transport...\n"); | |
980 | ||
981 | xprt->max_reqs = xprt_udp_slot_table_entries; | |
982 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
983 | xprt->slot = kmalloc(slot_table_size, GFP_KERNEL); | |
984 | if (xprt->slot == NULL) | |
985 | return -ENOMEM; | |
986 | memset(xprt->slot, 0, slot_table_size); | |
987 | ||
988 | xprt->prot = IPPROTO_UDP; | |
9903cd1c | 989 | xprt->port = XS_MAX_RESVPORT; |
a246b010 CL |
990 | xprt->stream = 0; |
991 | xprt->nocong = 0; | |
992 | xprt->cwnd = RPC_INITCWND; | |
993 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; | |
994 | /* XXX: header size can vary due to auth type, IPv6, etc. */ | |
995 | xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); | |
996 | ||
9903cd1c | 997 | INIT_WORK(&xprt->sock_connect, xs_connect_worker, xprt); |
a246b010 | 998 | |
9903cd1c | 999 | xprt->ops = &xs_ops; |
a246b010 CL |
1000 | |
1001 | if (to) | |
1002 | xprt->timeout = *to; | |
1003 | else | |
9903cd1c | 1004 | xprt_set_timeout(&xprt->timeout, 5, 5 * HZ); |
a246b010 CL |
1005 | |
1006 | return 0; | |
1007 | } | |
1008 | ||
9903cd1c CL |
1009 | /** |
1010 | * xs_setup_tcp - Set up transport to use a TCP socket | |
1011 | * @xprt: transport to set up | |
1012 | * @to: timeout parameters | |
1013 | * | |
1014 | */ | |
a246b010 CL |
1015 | int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to) |
1016 | { | |
1017 | size_t slot_table_size; | |
1018 | ||
1019 | dprintk("RPC: setting up tcp-ipv4 transport...\n"); | |
1020 | ||
1021 | xprt->max_reqs = xprt_tcp_slot_table_entries; | |
1022 | slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]); | |
1023 | xprt->slot = kmalloc(slot_table_size, GFP_KERNEL); | |
1024 | if (xprt->slot == NULL) | |
1025 | return -ENOMEM; | |
1026 | memset(xprt->slot, 0, slot_table_size); | |
1027 | ||
1028 | xprt->prot = IPPROTO_TCP; | |
9903cd1c | 1029 | xprt->port = XS_MAX_RESVPORT; |
a246b010 CL |
1030 | xprt->stream = 1; |
1031 | xprt->nocong = 1; | |
1032 | xprt->cwnd = RPC_MAXCWND(xprt); | |
1033 | xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0; | |
1034 | xprt->max_payload = (1U << 31) - 1; | |
1035 | ||
9903cd1c | 1036 | INIT_WORK(&xprt->sock_connect, xs_connect_worker, xprt); |
a246b010 | 1037 | |
9903cd1c | 1038 | xprt->ops = &xs_ops; |
a246b010 CL |
1039 | |
1040 | if (to) | |
1041 | xprt->timeout = *to; | |
1042 | else | |
9903cd1c | 1043 | xprt_set_timeout(&xprt->timeout, 2, 60 * HZ); |
a246b010 CL |
1044 | |
1045 | return 0; | |
1046 | } |