Commit | Line | Data |
---|---|---|
a246b010 CL |
1 | /* |
2 | * linux/net/sunrpc/xprtsock.c | |
3 | * | |
4 | * Client-side transport implementation for sockets. | |
5 | * | |
113aa838 AC |
6 | * TCP callback races fixes (C) 1998 Red Hat |
7 | * TCP send fixes (C) 1998 Red Hat | |
a246b010 CL |
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> | |
8f9d5b1a CL |
16 | * |
17 | * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. | |
18 | * <gilles.quillard@bull.net> | |
a246b010 CL |
19 | */ |
20 | ||
21 | #include <linux/types.h> | |
22 | #include <linux/slab.h> | |
bc25571e | 23 | #include <linux/module.h> |
a246b010 | 24 | #include <linux/capability.h> |
a246b010 CL |
25 | #include <linux/pagemap.h> |
26 | #include <linux/errno.h> | |
27 | #include <linux/socket.h> | |
28 | #include <linux/in.h> | |
29 | #include <linux/net.h> | |
30 | #include <linux/mm.h> | |
31 | #include <linux/udp.h> | |
32 | #include <linux/tcp.h> | |
33 | #include <linux/sunrpc/clnt.h> | |
02107148 | 34 | #include <linux/sunrpc/sched.h> |
49c36fcc | 35 | #include <linux/sunrpc/xprtsock.h> |
a246b010 CL |
36 | #include <linux/file.h> |
37 | ||
38 | #include <net/sock.h> | |
39 | #include <net/checksum.h> | |
40 | #include <net/udp.h> | |
41 | #include <net/tcp.h> | |
42 | ||
c556b754 CL |
43 | /* |
44 | * xprtsock tunables | |
45 | */ | |
46 | unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; | |
47 | unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE; | |
48 | ||
49 | unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; | |
50 | unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; | |
51 | ||
fbf76683 CL |
52 | /* |
53 | * We can register our own files under /proc/sys/sunrpc by | |
54 | * calling register_sysctl_table() again. The files in that | |
55 | * directory become the union of all files registered there. | |
56 | * | |
57 | * We simply need to make sure that we don't collide with | |
58 | * someone else's file names! | |
59 | */ | |
60 | ||
61 | #ifdef RPC_DEBUG | |
62 | ||
63 | static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; | |
64 | static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; | |
65 | static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; | |
66 | static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; | |
67 | ||
68 | static struct ctl_table_header *sunrpc_table_header; | |
69 | ||
70 | /* | |
71 | * FIXME: changing the UDP slot table size should also resize the UDP | |
72 | * socket buffers for existing UDP transports | |
73 | */ | |
74 | static ctl_table xs_tunables_table[] = { | |
75 | { | |
76 | .ctl_name = CTL_SLOTTABLE_UDP, | |
77 | .procname = "udp_slot_table_entries", | |
78 | .data = &xprt_udp_slot_table_entries, | |
79 | .maxlen = sizeof(unsigned int), | |
80 | .mode = 0644, | |
81 | .proc_handler = &proc_dointvec_minmax, | |
82 | .strategy = &sysctl_intvec, | |
83 | .extra1 = &min_slot_table_size, | |
84 | .extra2 = &max_slot_table_size | |
85 | }, | |
86 | { | |
87 | .ctl_name = CTL_SLOTTABLE_TCP, | |
88 | .procname = "tcp_slot_table_entries", | |
89 | .data = &xprt_tcp_slot_table_entries, | |
90 | .maxlen = sizeof(unsigned int), | |
91 | .mode = 0644, | |
92 | .proc_handler = &proc_dointvec_minmax, | |
93 | .strategy = &sysctl_intvec, | |
94 | .extra1 = &min_slot_table_size, | |
95 | .extra2 = &max_slot_table_size | |
96 | }, | |
97 | { | |
98 | .ctl_name = CTL_MIN_RESVPORT, | |
99 | .procname = "min_resvport", | |
100 | .data = &xprt_min_resvport, | |
101 | .maxlen = sizeof(unsigned int), | |
102 | .mode = 0644, | |
103 | .proc_handler = &proc_dointvec_minmax, | |
104 | .strategy = &sysctl_intvec, | |
105 | .extra1 = &xprt_min_resvport_limit, | |
106 | .extra2 = &xprt_max_resvport_limit | |
107 | }, | |
108 | { | |
109 | .ctl_name = CTL_MAX_RESVPORT, | |
110 | .procname = "max_resvport", | |
111 | .data = &xprt_max_resvport, | |
112 | .maxlen = sizeof(unsigned int), | |
113 | .mode = 0644, | |
114 | .proc_handler = &proc_dointvec_minmax, | |
115 | .strategy = &sysctl_intvec, | |
116 | .extra1 = &xprt_min_resvport_limit, | |
117 | .extra2 = &xprt_max_resvport_limit | |
118 | }, | |
119 | { | |
120 | .ctl_name = 0, | |
121 | }, | |
122 | }; | |
123 | ||
124 | static ctl_table sunrpc_table[] = { | |
125 | { | |
126 | .ctl_name = CTL_SUNRPC, | |
127 | .procname = "sunrpc", | |
128 | .mode = 0555, | |
129 | .child = xs_tunables_table | |
130 | }, | |
131 | { | |
132 | .ctl_name = 0, | |
133 | }, | |
134 | }; | |
135 | ||
136 | #endif | |
137 | ||
03bf4b70 CL |
138 | /* |
139 | * Time out for an RPC UDP socket connect. UDP socket connects are | |
140 | * synchronous, but we set a timeout anyway in case of resource | |
141 | * exhaustion on the local host. | |
142 | */ | |
143 | #define XS_UDP_CONN_TO (5U * HZ) | |
144 | ||
145 | /* | |
146 | * Wait duration for an RPC TCP connection to be established. Solaris | |
147 | * NFS over TCP uses 60 seconds, for example, which is in line with how | |
148 | * long a server takes to reboot. | |
149 | */ | |
150 | #define XS_TCP_CONN_TO (60U * HZ) | |
151 | ||
152 | /* | |
153 | * Wait duration for a reply from the RPC portmapper. | |
154 | */ | |
155 | #define XS_BIND_TO (60U * HZ) | |
156 | ||
157 | /* | |
158 | * Delay if a UDP socket connect error occurs. This is most likely some | |
159 | * kind of resource problem on the local host. | |
160 | */ | |
161 | #define XS_UDP_REEST_TO (2U * HZ) | |
162 | ||
163 | /* | |
164 | * The reestablish timeout allows clients to delay for a bit before attempting | |
165 | * to reconnect to a server that just dropped our connection. | |
166 | * | |
167 | * We implement an exponential backoff when trying to reestablish a TCP | |
168 | * transport connection with the server. Some servers like to drop a TCP | |
169 | * connection when they are overworked, so we start with a short timeout and | |
170 | * increase over time if the server is down or not responding. | |
171 | */ | |
172 | #define XS_TCP_INIT_REEST_TO (3U * HZ) | |
173 | #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ) | |
174 | ||
175 | /* | |
176 | * TCP idle timeout; client drops the transport socket if it is idle | |
177 | * for this long. Note that we also timeout UDP sockets to prevent | |
178 | * holding port numbers when there is no RPC traffic. | |
179 | */ | |
180 | #define XS_IDLE_DISC_TO (5U * 60 * HZ) | |
181 | ||
a246b010 CL |
182 | #ifdef RPC_DEBUG |
183 | # undef RPC_DEBUG_DATA | |
9903cd1c | 184 | # define RPCDBG_FACILITY RPCDBG_TRANS |
a246b010 CL |
185 | #endif |
186 | ||
a246b010 | 187 | #ifdef RPC_DEBUG_DATA |
9903cd1c | 188 | static void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 | 189 | { |
9903cd1c CL |
190 | u8 *buf = (u8 *) packet; |
191 | int j; | |
a246b010 | 192 | |
46121cf7 | 193 | dprintk("RPC: %s\n", msg); |
a246b010 CL |
194 | for (j = 0; j < count && j < 128; j += 4) { |
195 | if (!(j & 31)) { | |
196 | if (j) | |
197 | dprintk("\n"); | |
198 | dprintk("0x%04x ", j); | |
199 | } | |
200 | dprintk("%02x%02x%02x%02x ", | |
201 | buf[j], buf[j+1], buf[j+2], buf[j+3]); | |
202 | } | |
203 | dprintk("\n"); | |
204 | } | |
205 | #else | |
9903cd1c | 206 | static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 CL |
207 | { |
208 | /* NOP */ | |
209 | } | |
210 | #endif | |
211 | ||
ffc2e518 CL |
212 | struct sock_xprt { |
213 | struct rpc_xprt xprt; | |
ee0ac0c2 CL |
214 | |
215 | /* | |
216 | * Network layer | |
217 | */ | |
218 | struct socket * sock; | |
219 | struct sock * inet; | |
51971139 CL |
220 | |
221 | /* | |
222 | * State of TCP reply receive | |
223 | */ | |
224 | __be32 tcp_fraghdr, | |
225 | tcp_xid; | |
226 | ||
227 | u32 tcp_offset, | |
228 | tcp_reclen; | |
229 | ||
230 | unsigned long tcp_copied, | |
231 | tcp_flags; | |
c8475461 CL |
232 | |
233 | /* | |
234 | * Connection of transports | |
235 | */ | |
34161db6 | 236 | struct delayed_work connect_worker; |
d3bc9a1d | 237 | struct sockaddr_storage addr; |
c8475461 | 238 | unsigned short port; |
7c6e066e CL |
239 | |
240 | /* | |
241 | * UDP socket buffer size parameters | |
242 | */ | |
243 | size_t rcvsize, | |
244 | sndsize; | |
314dfd79 CL |
245 | |
246 | /* | |
247 | * Saved socket callback addresses | |
248 | */ | |
249 | void (*old_data_ready)(struct sock *, int); | |
250 | void (*old_state_change)(struct sock *); | |
251 | void (*old_write_space)(struct sock *); | |
2a9e1cfa | 252 | void (*old_error_report)(struct sock *); |
ffc2e518 CL |
253 | }; |
254 | ||
e136d092 CL |
255 | /* |
256 | * TCP receive state flags | |
257 | */ | |
258 | #define TCP_RCV_LAST_FRAG (1UL << 0) | |
259 | #define TCP_RCV_COPY_FRAGHDR (1UL << 1) | |
260 | #define TCP_RCV_COPY_XID (1UL << 2) | |
261 | #define TCP_RCV_COPY_DATA (1UL << 3) | |
262 | ||
95392c59 CL |
263 | static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) |
264 | { | |
265 | return (struct sockaddr *) &xprt->addr; | |
266 | } | |
267 | ||
268 | static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) | |
edb267a6 | 269 | { |
95392c59 CL |
270 | return (struct sockaddr_in *) &xprt->addr; |
271 | } | |
272 | ||
273 | static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) | |
274 | { | |
275 | return (struct sockaddr_in6 *) &xprt->addr; | |
276 | } | |
277 | ||
b454ae90 CL |
278 | static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt, |
279 | const char *protocol, | |
280 | const char *netid) | |
edb267a6 | 281 | { |
95392c59 | 282 | struct sockaddr_in *addr = xs_addr_in(xprt); |
edb267a6 CL |
283 | char *buf; |
284 | ||
285 | buf = kzalloc(20, GFP_KERNEL); | |
286 | if (buf) { | |
e0db4a78 | 287 | snprintf(buf, 20, "%pI4", &addr->sin_addr.s_addr); |
edb267a6 CL |
288 | } |
289 | xprt->address_strings[RPC_DISPLAY_ADDR] = buf; | |
290 | ||
291 | buf = kzalloc(8, GFP_KERNEL); | |
292 | if (buf) { | |
293 | snprintf(buf, 8, "%u", | |
294 | ntohs(addr->sin_port)); | |
295 | } | |
296 | xprt->address_strings[RPC_DISPLAY_PORT] = buf; | |
297 | ||
b454ae90 | 298 | xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; |
edb267a6 CL |
299 | |
300 | buf = kzalloc(48, GFP_KERNEL); | |
301 | if (buf) { | |
21454aaa HH |
302 | snprintf(buf, 48, "addr=%pI4 port=%u proto=%s", |
303 | &addr->sin_addr.s_addr, | |
edb267a6 | 304 | ntohs(addr->sin_port), |
b454ae90 | 305 | protocol); |
edb267a6 CL |
306 | } |
307 | xprt->address_strings[RPC_DISPLAY_ALL] = buf; | |
fbfe3cc6 CL |
308 | |
309 | buf = kzalloc(10, GFP_KERNEL); | |
310 | if (buf) { | |
311 | snprintf(buf, 10, "%02x%02x%02x%02x", | |
312 | NIPQUAD(addr->sin_addr.s_addr)); | |
313 | } | |
314 | xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf; | |
315 | ||
316 | buf = kzalloc(8, GFP_KERNEL); | |
317 | if (buf) { | |
318 | snprintf(buf, 8, "%4hx", | |
319 | ntohs(addr->sin_port)); | |
320 | } | |
321 | xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf; | |
756805e7 CL |
322 | |
323 | buf = kzalloc(30, GFP_KERNEL); | |
324 | if (buf) { | |
21454aaa HH |
325 | snprintf(buf, 30, "%pI4.%u.%u", |
326 | &addr->sin_addr.s_addr, | |
756805e7 CL |
327 | ntohs(addr->sin_port) >> 8, |
328 | ntohs(addr->sin_port) & 0xff); | |
329 | } | |
330 | xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf; | |
4417c8c4 | 331 | |
b454ae90 | 332 | xprt->address_strings[RPC_DISPLAY_NETID] = netid; |
edb267a6 CL |
333 | } |
334 | ||
b454ae90 CL |
335 | static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt, |
336 | const char *protocol, | |
337 | const char *netid) | |
4b6473fb | 338 | { |
95392c59 | 339 | struct sockaddr_in6 *addr = xs_addr_in6(xprt); |
4b6473fb CL |
340 | char *buf; |
341 | ||
342 | buf = kzalloc(40, GFP_KERNEL); | |
343 | if (buf) { | |
5b095d98 | 344 | snprintf(buf, 40, "%pI6",&addr->sin6_addr); |
4b6473fb CL |
345 | } |
346 | xprt->address_strings[RPC_DISPLAY_ADDR] = buf; | |
347 | ||
348 | buf = kzalloc(8, GFP_KERNEL); | |
349 | if (buf) { | |
350 | snprintf(buf, 8, "%u", | |
351 | ntohs(addr->sin6_port)); | |
352 | } | |
353 | xprt->address_strings[RPC_DISPLAY_PORT] = buf; | |
354 | ||
b454ae90 | 355 | xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; |
4b6473fb CL |
356 | |
357 | buf = kzalloc(64, GFP_KERNEL); | |
358 | if (buf) { | |
5b095d98 | 359 | snprintf(buf, 64, "addr=%pI6 port=%u proto=%s", |
fdb46ee7 | 360 | &addr->sin6_addr, |
4b6473fb | 361 | ntohs(addr->sin6_port), |
b454ae90 | 362 | protocol); |
4b6473fb CL |
363 | } |
364 | xprt->address_strings[RPC_DISPLAY_ALL] = buf; | |
365 | ||
366 | buf = kzalloc(36, GFP_KERNEL); | |
b071195d | 367 | if (buf) |
4b7a4274 | 368 | snprintf(buf, 36, "%pi6", &addr->sin6_addr); |
b071195d | 369 | |
4b6473fb CL |
370 | xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf; |
371 | ||
372 | buf = kzalloc(8, GFP_KERNEL); | |
373 | if (buf) { | |
374 | snprintf(buf, 8, "%4hx", | |
375 | ntohs(addr->sin6_port)); | |
376 | } | |
377 | xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf; | |
756805e7 CL |
378 | |
379 | buf = kzalloc(50, GFP_KERNEL); | |
380 | if (buf) { | |
5b095d98 | 381 | snprintf(buf, 50, "%pI6.%u.%u", |
fdb46ee7 HH |
382 | &addr->sin6_addr, |
383 | ntohs(addr->sin6_port) >> 8, | |
384 | ntohs(addr->sin6_port) & 0xff); | |
756805e7 CL |
385 | } |
386 | xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf; | |
4417c8c4 | 387 | |
b454ae90 | 388 | xprt->address_strings[RPC_DISPLAY_NETID] = netid; |
edb267a6 CL |
389 | } |
390 | ||
391 | static void xs_free_peer_addresses(struct rpc_xprt *xprt) | |
392 | { | |
33e01dc7 CL |
393 | unsigned int i; |
394 | ||
395 | for (i = 0; i < RPC_DISPLAY_MAX; i++) | |
396 | switch (i) { | |
397 | case RPC_DISPLAY_PROTO: | |
398 | case RPC_DISPLAY_NETID: | |
399 | continue; | |
400 | default: | |
401 | kfree(xprt->address_strings[i]); | |
402 | } | |
edb267a6 CL |
403 | } |
404 | ||
b4b5cc85 CL |
405 | #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) |
406 | ||
24c5684b | 407 | static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) |
b4b5cc85 | 408 | { |
b4b5cc85 CL |
409 | struct msghdr msg = { |
410 | .msg_name = addr, | |
411 | .msg_namelen = addrlen, | |
24c5684b TM |
412 | .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), |
413 | }; | |
414 | struct kvec iov = { | |
415 | .iov_base = vec->iov_base + base, | |
416 | .iov_len = vec->iov_len - base, | |
b4b5cc85 CL |
417 | }; |
418 | ||
24c5684b | 419 | if (iov.iov_len != 0) |
b4b5cc85 CL |
420 | return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); |
421 | return kernel_sendmsg(sock, &msg, NULL, 0, 0); | |
422 | } | |
423 | ||
24c5684b | 424 | static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more) |
b4b5cc85 | 425 | { |
24c5684b TM |
426 | struct page **ppage; |
427 | unsigned int remainder; | |
428 | int err, sent = 0; | |
429 | ||
430 | remainder = xdr->page_len - base; | |
431 | base += xdr->page_base; | |
432 | ppage = xdr->pages + (base >> PAGE_SHIFT); | |
433 | base &= ~PAGE_MASK; | |
434 | for(;;) { | |
435 | unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); | |
436 | int flags = XS_SENDMSG_FLAGS; | |
b4b5cc85 | 437 | |
24c5684b TM |
438 | remainder -= len; |
439 | if (remainder != 0 || more) | |
440 | flags |= MSG_MORE; | |
441 | err = sock->ops->sendpage(sock, *ppage, base, len, flags); | |
442 | if (remainder == 0 || err != len) | |
443 | break; | |
444 | sent += err; | |
445 | ppage++; | |
446 | base = 0; | |
447 | } | |
448 | if (sent == 0) | |
449 | return err; | |
450 | if (err > 0) | |
451 | sent += err; | |
452 | return sent; | |
b4b5cc85 CL |
453 | } |
454 | ||
9903cd1c CL |
455 | /** |
456 | * xs_sendpages - write pages directly to a socket | |
457 | * @sock: socket to send on | |
458 | * @addr: UDP only -- address of destination | |
459 | * @addrlen: UDP only -- length of destination address | |
460 | * @xdr: buffer containing this request | |
461 | * @base: starting position in the buffer | |
462 | * | |
a246b010 | 463 | */ |
24c5684b | 464 | static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base) |
a246b010 | 465 | { |
24c5684b TM |
466 | unsigned int remainder = xdr->len - base; |
467 | int err, sent = 0; | |
a246b010 | 468 | |
262965f5 | 469 | if (unlikely(!sock)) |
fba91afb | 470 | return -ENOTSOCK; |
262965f5 CL |
471 | |
472 | clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); | |
24c5684b TM |
473 | if (base != 0) { |
474 | addr = NULL; | |
475 | addrlen = 0; | |
476 | } | |
262965f5 | 477 | |
24c5684b TM |
478 | if (base < xdr->head[0].iov_len || addr != NULL) { |
479 | unsigned int len = xdr->head[0].iov_len - base; | |
480 | remainder -= len; | |
481 | err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); | |
482 | if (remainder == 0 || err != len) | |
a246b010 | 483 | goto out; |
24c5684b | 484 | sent += err; |
a246b010 CL |
485 | base = 0; |
486 | } else | |
24c5684b | 487 | base -= xdr->head[0].iov_len; |
a246b010 | 488 | |
24c5684b TM |
489 | if (base < xdr->page_len) { |
490 | unsigned int len = xdr->page_len - base; | |
491 | remainder -= len; | |
492 | err = xs_send_pagedata(sock, xdr, base, remainder != 0); | |
493 | if (remainder == 0 || err != len) | |
a246b010 | 494 | goto out; |
24c5684b | 495 | sent += err; |
a246b010 | 496 | base = 0; |
24c5684b TM |
497 | } else |
498 | base -= xdr->page_len; | |
499 | ||
500 | if (base >= xdr->tail[0].iov_len) | |
501 | return sent; | |
502 | err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); | |
a246b010 | 503 | out: |
24c5684b TM |
504 | if (sent == 0) |
505 | return err; | |
506 | if (err > 0) | |
507 | sent += err; | |
508 | return sent; | |
a246b010 CL |
509 | } |
510 | ||
b6ddf64f TM |
511 | static void xs_nospace_callback(struct rpc_task *task) |
512 | { | |
513 | struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt); | |
514 | ||
515 | transport->inet->sk_write_pending--; | |
516 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); | |
517 | } | |
518 | ||
9903cd1c | 519 | /** |
262965f5 CL |
520 | * xs_nospace - place task on wait queue if transmit was incomplete |
521 | * @task: task to put to sleep | |
9903cd1c | 522 | * |
a246b010 | 523 | */ |
262965f5 | 524 | static void xs_nospace(struct rpc_task *task) |
a246b010 | 525 | { |
262965f5 CL |
526 | struct rpc_rqst *req = task->tk_rqstp; |
527 | struct rpc_xprt *xprt = req->rq_xprt; | |
ee0ac0c2 | 528 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 | 529 | |
46121cf7 | 530 | dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", |
262965f5 CL |
531 | task->tk_pid, req->rq_slen - req->rq_bytes_sent, |
532 | req->rq_slen); | |
533 | ||
b6ddf64f TM |
534 | /* Protect against races with write_space */ |
535 | spin_lock_bh(&xprt->transport_lock); | |
536 | ||
537 | /* Don't race with disconnect */ | |
538 | if (xprt_connected(xprt)) { | |
539 | if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) { | |
540 | /* | |
541 | * Notify TCP that we're limited by the application | |
542 | * window size | |
543 | */ | |
544 | set_bit(SOCK_NOSPACE, &transport->sock->flags); | |
545 | transport->inet->sk_write_pending++; | |
546 | /* ...and wait for more buffer space */ | |
547 | xprt_wait_for_buffer_space(task, xs_nospace_callback); | |
548 | } | |
549 | } else { | |
550 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); | |
551 | task->tk_status = -ENOTCONN; | |
552 | } | |
262965f5 | 553 | |
b6ddf64f | 554 | spin_unlock_bh(&xprt->transport_lock); |
262965f5 CL |
555 | } |
556 | ||
557 | /** | |
558 | * xs_udp_send_request - write an RPC request to a UDP socket | |
559 | * @task: address of RPC task that manages the state of an RPC request | |
560 | * | |
561 | * Return values: | |
562 | * 0: The request has been sent | |
563 | * EAGAIN: The socket was blocked, please call again later to | |
564 | * complete the request | |
565 | * ENOTCONN: Caller needs to invoke connect logic then call again | |
566 | * other: Some other error occured, the request was not sent | |
567 | */ | |
568 | static int xs_udp_send_request(struct rpc_task *task) | |
569 | { | |
570 | struct rpc_rqst *req = task->tk_rqstp; | |
571 | struct rpc_xprt *xprt = req->rq_xprt; | |
ee0ac0c2 | 572 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
262965f5 CL |
573 | struct xdr_buf *xdr = &req->rq_snd_buf; |
574 | int status; | |
a246b010 | 575 | |
9903cd1c | 576 | xs_pktdump("packet data:", |
a246b010 CL |
577 | req->rq_svec->iov_base, |
578 | req->rq_svec->iov_len); | |
579 | ||
01d37c42 TM |
580 | if (!xprt_bound(xprt)) |
581 | return -ENOTCONN; | |
ee0ac0c2 | 582 | status = xs_sendpages(transport->sock, |
95392c59 | 583 | xs_addr(xprt), |
ee0ac0c2 CL |
584 | xprt->addrlen, xdr, |
585 | req->rq_bytes_sent); | |
a246b010 | 586 | |
46121cf7 | 587 | dprintk("RPC: xs_udp_send_request(%u) = %d\n", |
262965f5 | 588 | xdr->len - req->rq_bytes_sent, status); |
a246b010 | 589 | |
2199700f TM |
590 | if (status >= 0) { |
591 | task->tk_bytes_sent += status; | |
592 | if (status >= req->rq_slen) | |
593 | return 0; | |
594 | /* Still some bytes left; set up for a retry later. */ | |
262965f5 | 595 | status = -EAGAIN; |
2199700f | 596 | } |
a246b010 | 597 | |
262965f5 | 598 | switch (status) { |
fba91afb TM |
599 | case -ENOTSOCK: |
600 | status = -ENOTCONN; | |
601 | /* Should we call xs_close() here? */ | |
602 | break; | |
b6ddf64f TM |
603 | case -EAGAIN: |
604 | xs_nospace(task); | |
605 | break; | |
262965f5 CL |
606 | case -ENETUNREACH: |
607 | case -EPIPE: | |
a246b010 CL |
608 | case -ECONNREFUSED: |
609 | /* When the server has died, an ICMP port unreachable message | |
9903cd1c | 610 | * prompts ECONNREFUSED. */ |
b6ddf64f | 611 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); |
a246b010 CL |
612 | break; |
613 | default: | |
b6ddf64f | 614 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); |
46121cf7 | 615 | dprintk("RPC: sendmsg returned unrecognized error %d\n", |
262965f5 | 616 | -status); |
a246b010 | 617 | } |
262965f5 CL |
618 | |
619 | return status; | |
a246b010 CL |
620 | } |
621 | ||
e06799f9 TM |
622 | /** |
623 | * xs_tcp_shutdown - gracefully shut down a TCP socket | |
624 | * @xprt: transport | |
625 | * | |
626 | * Initiates a graceful shutdown of the TCP socket by calling the | |
627 | * equivalent of shutdown(SHUT_WR); | |
628 | */ | |
629 | static void xs_tcp_shutdown(struct rpc_xprt *xprt) | |
630 | { | |
631 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); | |
632 | struct socket *sock = transport->sock; | |
633 | ||
634 | if (sock != NULL) | |
635 | kernel_sock_shutdown(sock, SHUT_WR); | |
636 | } | |
637 | ||
808012fb CL |
638 | static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf) |
639 | { | |
640 | u32 reclen = buf->len - sizeof(rpc_fraghdr); | |
641 | rpc_fraghdr *base = buf->head[0].iov_base; | |
642 | *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen); | |
643 | } | |
644 | ||
9903cd1c | 645 | /** |
262965f5 | 646 | * xs_tcp_send_request - write an RPC request to a TCP socket |
9903cd1c CL |
647 | * @task: address of RPC task that manages the state of an RPC request |
648 | * | |
649 | * Return values: | |
262965f5 CL |
650 | * 0: The request has been sent |
651 | * EAGAIN: The socket was blocked, please call again later to | |
652 | * complete the request | |
653 | * ENOTCONN: Caller needs to invoke connect logic then call again | |
654 | * other: Some other error occured, the request was not sent | |
9903cd1c CL |
655 | * |
656 | * XXX: In the case of soft timeouts, should we eventually give up | |
262965f5 | 657 | * if sendmsg is not able to make progress? |
9903cd1c | 658 | */ |
262965f5 | 659 | static int xs_tcp_send_request(struct rpc_task *task) |
a246b010 CL |
660 | { |
661 | struct rpc_rqst *req = task->tk_rqstp; | |
662 | struct rpc_xprt *xprt = req->rq_xprt; | |
ee0ac0c2 | 663 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
262965f5 | 664 | struct xdr_buf *xdr = &req->rq_snd_buf; |
b595bb15 | 665 | int status; |
a246b010 | 666 | |
808012fb | 667 | xs_encode_tcp_record_marker(&req->rq_snd_buf); |
a246b010 | 668 | |
262965f5 CL |
669 | xs_pktdump("packet data:", |
670 | req->rq_svec->iov_base, | |
671 | req->rq_svec->iov_len); | |
a246b010 CL |
672 | |
673 | /* Continue transmitting the packet/record. We must be careful | |
674 | * to cope with writespace callbacks arriving _after_ we have | |
262965f5 | 675 | * called sendmsg(). */ |
a246b010 | 676 | while (1) { |
ee0ac0c2 CL |
677 | status = xs_sendpages(transport->sock, |
678 | NULL, 0, xdr, req->rq_bytes_sent); | |
a246b010 | 679 | |
46121cf7 | 680 | dprintk("RPC: xs_tcp_send_request(%u) = %d\n", |
262965f5 | 681 | xdr->len - req->rq_bytes_sent, status); |
a246b010 | 682 | |
262965f5 | 683 | if (unlikely(status < 0)) |
a246b010 | 684 | break; |
a246b010 | 685 | |
262965f5 CL |
686 | /* If we've sent the entire packet, immediately |
687 | * reset the count of bytes sent. */ | |
688 | req->rq_bytes_sent += status; | |
ef759a2e | 689 | task->tk_bytes_sent += status; |
262965f5 CL |
690 | if (likely(req->rq_bytes_sent >= req->rq_slen)) { |
691 | req->rq_bytes_sent = 0; | |
692 | return 0; | |
693 | } | |
a246b010 | 694 | |
06b4b681 TM |
695 | if (status != 0) |
696 | continue; | |
a246b010 | 697 | status = -EAGAIN; |
06b4b681 | 698 | break; |
a246b010 CL |
699 | } |
700 | ||
262965f5 | 701 | switch (status) { |
fba91afb TM |
702 | case -ENOTSOCK: |
703 | status = -ENOTCONN; | |
704 | /* Should we call xs_close() here? */ | |
705 | break; | |
262965f5 CL |
706 | case -EAGAIN: |
707 | xs_nospace(task); | |
708 | break; | |
262965f5 | 709 | case -ECONNRESET: |
2a9e1cfa TM |
710 | xs_tcp_shutdown(xprt); |
711 | case -ECONNREFUSED: | |
262965f5 CL |
712 | case -ENOTCONN: |
713 | case -EPIPE: | |
714 | status = -ENOTCONN; | |
b6ddf64f | 715 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); |
262965f5 CL |
716 | break; |
717 | default: | |
46121cf7 | 718 | dprintk("RPC: sendmsg returned unrecognized error %d\n", |
262965f5 | 719 | -status); |
b6ddf64f | 720 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); |
e06799f9 | 721 | xs_tcp_shutdown(xprt); |
a246b010 | 722 | } |
262965f5 | 723 | |
a246b010 CL |
724 | return status; |
725 | } | |
726 | ||
e0ab53de TM |
727 | /** |
728 | * xs_tcp_release_xprt - clean up after a tcp transmission | |
729 | * @xprt: transport | |
730 | * @task: rpc task | |
731 | * | |
732 | * This cleans up if an error causes us to abort the transmission of a request. | |
733 | * In this case, the socket may need to be reset in order to avoid confusing | |
734 | * the server. | |
735 | */ | |
736 | static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) | |
737 | { | |
738 | struct rpc_rqst *req; | |
739 | ||
740 | if (task != xprt->snd_task) | |
741 | return; | |
742 | if (task == NULL) | |
743 | goto out_release; | |
744 | req = task->tk_rqstp; | |
745 | if (req->rq_bytes_sent == 0) | |
746 | goto out_release; | |
747 | if (req->rq_bytes_sent == req->rq_snd_buf.len) | |
748 | goto out_release; | |
749 | set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state); | |
750 | out_release: | |
751 | xprt_release_xprt(xprt, task); | |
752 | } | |
753 | ||
2a9e1cfa TM |
754 | static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) |
755 | { | |
756 | transport->old_data_ready = sk->sk_data_ready; | |
757 | transport->old_state_change = sk->sk_state_change; | |
758 | transport->old_write_space = sk->sk_write_space; | |
759 | transport->old_error_report = sk->sk_error_report; | |
760 | } | |
761 | ||
762 | static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) | |
763 | { | |
764 | sk->sk_data_ready = transport->old_data_ready; | |
765 | sk->sk_state_change = transport->old_state_change; | |
766 | sk->sk_write_space = transport->old_write_space; | |
767 | sk->sk_error_report = transport->old_error_report; | |
768 | } | |
769 | ||
9903cd1c CL |
770 | /** |
771 | * xs_close - close a socket | |
772 | * @xprt: transport | |
773 | * | |
3167e12c CL |
774 | * This is used when all requests are complete; ie, no DRC state remains |
775 | * on the server we want to save. | |
a246b010 | 776 | */ |
9903cd1c | 777 | static void xs_close(struct rpc_xprt *xprt) |
a246b010 | 778 | { |
ee0ac0c2 CL |
779 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
780 | struct socket *sock = transport->sock; | |
781 | struct sock *sk = transport->inet; | |
a246b010 CL |
782 | |
783 | if (!sk) | |
632e3bdc | 784 | goto clear_close_wait; |
a246b010 | 785 | |
46121cf7 | 786 | dprintk("RPC: xs_close xprt %p\n", xprt); |
9903cd1c | 787 | |
a246b010 | 788 | write_lock_bh(&sk->sk_callback_lock); |
ee0ac0c2 CL |
789 | transport->inet = NULL; |
790 | transport->sock = NULL; | |
a246b010 | 791 | |
9903cd1c | 792 | sk->sk_user_data = NULL; |
2a9e1cfa TM |
793 | |
794 | xs_restore_old_callbacks(transport, sk); | |
a246b010 CL |
795 | write_unlock_bh(&sk->sk_callback_lock); |
796 | ||
9903cd1c | 797 | sk->sk_no_check = 0; |
a246b010 CL |
798 | |
799 | sock_release(sock); | |
632e3bdc TM |
800 | clear_close_wait: |
801 | smp_mb__before_clear_bit(); | |
802 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
3b948ae5 | 803 | clear_bit(XPRT_CLOSING, &xprt->state); |
632e3bdc | 804 | smp_mb__after_clear_bit(); |
62da3b24 | 805 | xprt_disconnect_done(xprt); |
a246b010 CL |
806 | } |
807 | ||
9903cd1c CL |
808 | /** |
809 | * xs_destroy - prepare to shutdown a transport | |
810 | * @xprt: doomed transport | |
811 | * | |
812 | */ | |
813 | static void xs_destroy(struct rpc_xprt *xprt) | |
a246b010 | 814 | { |
c8475461 CL |
815 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
816 | ||
46121cf7 | 817 | dprintk("RPC: xs_destroy xprt %p\n", xprt); |
9903cd1c | 818 | |
c1384c9c | 819 | cancel_rearming_delayed_work(&transport->connect_worker); |
a246b010 | 820 | |
9903cd1c | 821 | xs_close(xprt); |
edb267a6 | 822 | xs_free_peer_addresses(xprt); |
a246b010 | 823 | kfree(xprt->slot); |
c8541ecd | 824 | kfree(xprt); |
bc25571e | 825 | module_put(THIS_MODULE); |
a246b010 CL |
826 | } |
827 | ||
9903cd1c CL |
828 | static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) |
829 | { | |
830 | return (struct rpc_xprt *) sk->sk_user_data; | |
831 | } | |
832 | ||
833 | /** | |
834 | * xs_udp_data_ready - "data ready" callback for UDP sockets | |
835 | * @sk: socket with data to read | |
836 | * @len: how much data to read | |
837 | * | |
a246b010 | 838 | */ |
9903cd1c | 839 | static void xs_udp_data_ready(struct sock *sk, int len) |
a246b010 | 840 | { |
9903cd1c CL |
841 | struct rpc_task *task; |
842 | struct rpc_xprt *xprt; | |
a246b010 | 843 | struct rpc_rqst *rovr; |
9903cd1c | 844 | struct sk_buff *skb; |
a246b010 | 845 | int err, repsize, copied; |
d8ed029d AD |
846 | u32 _xid; |
847 | __be32 *xp; | |
a246b010 CL |
848 | |
849 | read_lock(&sk->sk_callback_lock); | |
46121cf7 | 850 | dprintk("RPC: xs_udp_data_ready...\n"); |
9903cd1c | 851 | if (!(xprt = xprt_from_sock(sk))) |
a246b010 | 852 | goto out; |
a246b010 CL |
853 | |
854 | if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) | |
855 | goto out; | |
856 | ||
857 | if (xprt->shutdown) | |
858 | goto dropit; | |
859 | ||
860 | repsize = skb->len - sizeof(struct udphdr); | |
861 | if (repsize < 4) { | |
46121cf7 | 862 | dprintk("RPC: impossible RPC reply size %d!\n", repsize); |
a246b010 CL |
863 | goto dropit; |
864 | } | |
865 | ||
866 | /* Copy the XID from the skb... */ | |
867 | xp = skb_header_pointer(skb, sizeof(struct udphdr), | |
868 | sizeof(_xid), &_xid); | |
869 | if (xp == NULL) | |
870 | goto dropit; | |
871 | ||
872 | /* Look up and lock the request corresponding to the given XID */ | |
4a0f8c04 | 873 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
874 | rovr = xprt_lookup_rqst(xprt, *xp); |
875 | if (!rovr) | |
876 | goto out_unlock; | |
877 | task = rovr->rq_task; | |
878 | ||
a246b010 CL |
879 | if ((copied = rovr->rq_private_buf.buflen) > repsize) |
880 | copied = repsize; | |
881 | ||
882 | /* Suck it into the iovec, verify checksum if not done by hw. */ | |
1781f7f5 HX |
883 | if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { |
884 | UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); | |
a246b010 | 885 | goto out_unlock; |
1781f7f5 HX |
886 | } |
887 | ||
888 | UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); | |
a246b010 CL |
889 | |
890 | /* Something worked... */ | |
891 | dst_confirm(skb->dst); | |
892 | ||
1570c1e4 CL |
893 | xprt_adjust_cwnd(task, copied); |
894 | xprt_update_rtt(task); | |
895 | xprt_complete_rqst(task, copied); | |
a246b010 CL |
896 | |
897 | out_unlock: | |
4a0f8c04 | 898 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
899 | dropit: |
900 | skb_free_datagram(sk, skb); | |
901 | out: | |
902 | read_unlock(&sk->sk_callback_lock); | |
903 | } | |
904 | ||
dd456471 | 905 | static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) |
a246b010 | 906 | { |
51971139 | 907 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 CL |
908 | size_t len, used; |
909 | char *p; | |
910 | ||
51971139 CL |
911 | p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; |
912 | len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; | |
9d292316 | 913 | used = xdr_skb_read_bits(desc, p, len); |
51971139 | 914 | transport->tcp_offset += used; |
a246b010 CL |
915 | if (used != len) |
916 | return; | |
808012fb | 917 | |
51971139 CL |
918 | transport->tcp_reclen = ntohl(transport->tcp_fraghdr); |
919 | if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) | |
e136d092 | 920 | transport->tcp_flags |= TCP_RCV_LAST_FRAG; |
a246b010 | 921 | else |
e136d092 | 922 | transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; |
51971139 | 923 | transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; |
808012fb | 924 | |
e136d092 | 925 | transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; |
51971139 | 926 | transport->tcp_offset = 0; |
808012fb | 927 | |
a246b010 | 928 | /* Sanity check of the record length */ |
51971139 | 929 | if (unlikely(transport->tcp_reclen < 4)) { |
46121cf7 | 930 | dprintk("RPC: invalid TCP record fragment length\n"); |
3ebb067d | 931 | xprt_force_disconnect(xprt); |
9903cd1c | 932 | return; |
a246b010 | 933 | } |
46121cf7 | 934 | dprintk("RPC: reading TCP record fragment of length %d\n", |
51971139 | 935 | transport->tcp_reclen); |
a246b010 CL |
936 | } |
937 | ||
51971139 | 938 | static void xs_tcp_check_fraghdr(struct sock_xprt *transport) |
a246b010 | 939 | { |
51971139 | 940 | if (transport->tcp_offset == transport->tcp_reclen) { |
e136d092 | 941 | transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; |
51971139 | 942 | transport->tcp_offset = 0; |
e136d092 CL |
943 | if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { |
944 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; | |
945 | transport->tcp_flags |= TCP_RCV_COPY_XID; | |
51971139 | 946 | transport->tcp_copied = 0; |
a246b010 CL |
947 | } |
948 | } | |
949 | } | |
950 | ||
dd456471 | 951 | static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) |
a246b010 CL |
952 | { |
953 | size_t len, used; | |
954 | char *p; | |
955 | ||
51971139 | 956 | len = sizeof(transport->tcp_xid) - transport->tcp_offset; |
46121cf7 | 957 | dprintk("RPC: reading XID (%Zu bytes)\n", len); |
51971139 | 958 | p = ((char *) &transport->tcp_xid) + transport->tcp_offset; |
9d292316 | 959 | used = xdr_skb_read_bits(desc, p, len); |
51971139 | 960 | transport->tcp_offset += used; |
a246b010 CL |
961 | if (used != len) |
962 | return; | |
e136d092 CL |
963 | transport->tcp_flags &= ~TCP_RCV_COPY_XID; |
964 | transport->tcp_flags |= TCP_RCV_COPY_DATA; | |
51971139 | 965 | transport->tcp_copied = 4; |
46121cf7 | 966 | dprintk("RPC: reading reply for XID %08x\n", |
51971139 CL |
967 | ntohl(transport->tcp_xid)); |
968 | xs_tcp_check_fraghdr(transport); | |
a246b010 CL |
969 | } |
970 | ||
dd456471 | 971 | static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) |
a246b010 | 972 | { |
51971139 | 973 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 CL |
974 | struct rpc_rqst *req; |
975 | struct xdr_buf *rcvbuf; | |
976 | size_t len; | |
977 | ssize_t r; | |
978 | ||
979 | /* Find and lock the request corresponding to this xid */ | |
4a0f8c04 | 980 | spin_lock(&xprt->transport_lock); |
51971139 | 981 | req = xprt_lookup_rqst(xprt, transport->tcp_xid); |
a246b010 | 982 | if (!req) { |
e136d092 | 983 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; |
46121cf7 | 984 | dprintk("RPC: XID %08x request not found!\n", |
51971139 | 985 | ntohl(transport->tcp_xid)); |
4a0f8c04 | 986 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
987 | return; |
988 | } | |
989 | ||
990 | rcvbuf = &req->rq_private_buf; | |
991 | len = desc->count; | |
51971139 | 992 | if (len > transport->tcp_reclen - transport->tcp_offset) { |
dd456471 | 993 | struct xdr_skb_reader my_desc; |
a246b010 | 994 | |
51971139 | 995 | len = transport->tcp_reclen - transport->tcp_offset; |
a246b010 CL |
996 | memcpy(&my_desc, desc, sizeof(my_desc)); |
997 | my_desc.count = len; | |
51971139 | 998 | r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, |
9d292316 | 999 | &my_desc, xdr_skb_read_bits); |
a246b010 CL |
1000 | desc->count -= r; |
1001 | desc->offset += r; | |
1002 | } else | |
51971139 | 1003 | r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, |
9d292316 | 1004 | desc, xdr_skb_read_bits); |
a246b010 CL |
1005 | |
1006 | if (r > 0) { | |
51971139 CL |
1007 | transport->tcp_copied += r; |
1008 | transport->tcp_offset += r; | |
a246b010 CL |
1009 | } |
1010 | if (r != len) { | |
1011 | /* Error when copying to the receive buffer, | |
1012 | * usually because we weren't able to allocate | |
1013 | * additional buffer pages. All we can do now | |
e136d092 | 1014 | * is turn off TCP_RCV_COPY_DATA, so the request |
a246b010 CL |
1015 | * will not receive any additional updates, |
1016 | * and time out. | |
1017 | * Any remaining data from this record will | |
1018 | * be discarded. | |
1019 | */ | |
e136d092 | 1020 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; |
46121cf7 | 1021 | dprintk("RPC: XID %08x truncated request\n", |
51971139 | 1022 | ntohl(transport->tcp_xid)); |
46121cf7 CL |
1023 | dprintk("RPC: xprt = %p, tcp_copied = %lu, " |
1024 | "tcp_offset = %u, tcp_reclen = %u\n", | |
1025 | xprt, transport->tcp_copied, | |
1026 | transport->tcp_offset, transport->tcp_reclen); | |
a246b010 CL |
1027 | goto out; |
1028 | } | |
1029 | ||
46121cf7 | 1030 | dprintk("RPC: XID %08x read %Zd bytes\n", |
51971139 | 1031 | ntohl(transport->tcp_xid), r); |
46121cf7 CL |
1032 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " |
1033 | "tcp_reclen = %u\n", xprt, transport->tcp_copied, | |
1034 | transport->tcp_offset, transport->tcp_reclen); | |
51971139 CL |
1035 | |
1036 | if (transport->tcp_copied == req->rq_private_buf.buflen) | |
e136d092 | 1037 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; |
51971139 | 1038 | else if (transport->tcp_offset == transport->tcp_reclen) { |
e136d092 CL |
1039 | if (transport->tcp_flags & TCP_RCV_LAST_FRAG) |
1040 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; | |
a246b010 CL |
1041 | } |
1042 | ||
1043 | out: | |
e136d092 | 1044 | if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) |
51971139 | 1045 | xprt_complete_rqst(req->rq_task, transport->tcp_copied); |
4a0f8c04 | 1046 | spin_unlock(&xprt->transport_lock); |
51971139 | 1047 | xs_tcp_check_fraghdr(transport); |
a246b010 CL |
1048 | } |
1049 | ||
dd456471 | 1050 | static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) |
a246b010 CL |
1051 | { |
1052 | size_t len; | |
1053 | ||
51971139 | 1054 | len = transport->tcp_reclen - transport->tcp_offset; |
a246b010 CL |
1055 | if (len > desc->count) |
1056 | len = desc->count; | |
1057 | desc->count -= len; | |
1058 | desc->offset += len; | |
51971139 | 1059 | transport->tcp_offset += len; |
46121cf7 | 1060 | dprintk("RPC: discarded %Zu bytes\n", len); |
51971139 | 1061 | xs_tcp_check_fraghdr(transport); |
a246b010 CL |
1062 | } |
1063 | ||
9903cd1c | 1064 | static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) |
a246b010 CL |
1065 | { |
1066 | struct rpc_xprt *xprt = rd_desc->arg.data; | |
51971139 | 1067 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
dd456471 | 1068 | struct xdr_skb_reader desc = { |
a246b010 CL |
1069 | .skb = skb, |
1070 | .offset = offset, | |
1071 | .count = len, | |
9903cd1c | 1072 | }; |
a246b010 | 1073 | |
46121cf7 | 1074 | dprintk("RPC: xs_tcp_data_recv started\n"); |
a246b010 CL |
1075 | do { |
1076 | /* Read in a new fragment marker if necessary */ | |
1077 | /* Can we ever really expect to get completely empty fragments? */ | |
e136d092 | 1078 | if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { |
9903cd1c | 1079 | xs_tcp_read_fraghdr(xprt, &desc); |
a246b010 CL |
1080 | continue; |
1081 | } | |
1082 | /* Read in the xid if necessary */ | |
e136d092 | 1083 | if (transport->tcp_flags & TCP_RCV_COPY_XID) { |
51971139 | 1084 | xs_tcp_read_xid(transport, &desc); |
a246b010 CL |
1085 | continue; |
1086 | } | |
1087 | /* Read in the request data */ | |
e136d092 | 1088 | if (transport->tcp_flags & TCP_RCV_COPY_DATA) { |
9903cd1c | 1089 | xs_tcp_read_request(xprt, &desc); |
a246b010 CL |
1090 | continue; |
1091 | } | |
1092 | /* Skip over any trailing bytes on short reads */ | |
51971139 | 1093 | xs_tcp_read_discard(transport, &desc); |
a246b010 | 1094 | } while (desc.count); |
46121cf7 | 1095 | dprintk("RPC: xs_tcp_data_recv done\n"); |
a246b010 CL |
1096 | return len - desc.count; |
1097 | } | |
1098 | ||
9903cd1c CL |
1099 | /** |
1100 | * xs_tcp_data_ready - "data ready" callback for TCP sockets | |
1101 | * @sk: socket with data to read | |
1102 | * @bytes: how much data to read | |
1103 | * | |
1104 | */ | |
1105 | static void xs_tcp_data_ready(struct sock *sk, int bytes) | |
a246b010 CL |
1106 | { |
1107 | struct rpc_xprt *xprt; | |
1108 | read_descriptor_t rd_desc; | |
ff2d7db8 | 1109 | int read; |
a246b010 | 1110 | |
46121cf7 CL |
1111 | dprintk("RPC: xs_tcp_data_ready...\n"); |
1112 | ||
a246b010 | 1113 | read_lock(&sk->sk_callback_lock); |
9903cd1c | 1114 | if (!(xprt = xprt_from_sock(sk))) |
a246b010 | 1115 | goto out; |
a246b010 CL |
1116 | if (xprt->shutdown) |
1117 | goto out; | |
1118 | ||
9903cd1c | 1119 | /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ |
a246b010 | 1120 | rd_desc.arg.data = xprt; |
ff2d7db8 TM |
1121 | do { |
1122 | rd_desc.count = 65536; | |
1123 | read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); | |
1124 | } while (read > 0); | |
a246b010 CL |
1125 | out: |
1126 | read_unlock(&sk->sk_callback_lock); | |
1127 | } | |
1128 | ||
9903cd1c CL |
1129 | /** |
1130 | * xs_tcp_state_change - callback to handle TCP socket state changes | |
1131 | * @sk: socket whose state has changed | |
1132 | * | |
1133 | */ | |
1134 | static void xs_tcp_state_change(struct sock *sk) | |
a246b010 | 1135 | { |
9903cd1c | 1136 | struct rpc_xprt *xprt; |
a246b010 CL |
1137 | |
1138 | read_lock(&sk->sk_callback_lock); | |
1139 | if (!(xprt = xprt_from_sock(sk))) | |
1140 | goto out; | |
46121cf7 CL |
1141 | dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); |
1142 | dprintk("RPC: state %x conn %d dead %d zapped %d\n", | |
1143 | sk->sk_state, xprt_connected(xprt), | |
1144 | sock_flag(sk, SOCK_DEAD), | |
1145 | sock_flag(sk, SOCK_ZAPPED)); | |
a246b010 CL |
1146 | |
1147 | switch (sk->sk_state) { | |
1148 | case TCP_ESTABLISHED: | |
4a0f8c04 | 1149 | spin_lock_bh(&xprt->transport_lock); |
a246b010 | 1150 | if (!xprt_test_and_set_connected(xprt)) { |
51971139 CL |
1151 | struct sock_xprt *transport = container_of(xprt, |
1152 | struct sock_xprt, xprt); | |
1153 | ||
a246b010 | 1154 | /* Reset TCP record info */ |
51971139 CL |
1155 | transport->tcp_offset = 0; |
1156 | transport->tcp_reclen = 0; | |
1157 | transport->tcp_copied = 0; | |
e136d092 CL |
1158 | transport->tcp_flags = |
1159 | TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; | |
51971139 | 1160 | |
44fbac22 | 1161 | xprt_wake_pending_tasks(xprt, 0); |
a246b010 | 1162 | } |
4a0f8c04 | 1163 | spin_unlock_bh(&xprt->transport_lock); |
a246b010 | 1164 | break; |
3b948ae5 TM |
1165 | case TCP_FIN_WAIT1: |
1166 | /* The client initiated a shutdown of the socket */ | |
7c1d71cf | 1167 | xprt->connect_cookie++; |
663b8858 | 1168 | xprt->reestablish_timeout = 0; |
3b948ae5 TM |
1169 | set_bit(XPRT_CLOSING, &xprt->state); |
1170 | smp_mb__before_clear_bit(); | |
1171 | clear_bit(XPRT_CONNECTED, &xprt->state); | |
ef803670 | 1172 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); |
3b948ae5 | 1173 | smp_mb__after_clear_bit(); |
a246b010 | 1174 | break; |
632e3bdc | 1175 | case TCP_CLOSE_WAIT: |
3b948ae5 TM |
1176 | /* The server initiated a shutdown of the socket */ |
1177 | set_bit(XPRT_CLOSING, &xprt->state); | |
66af1e55 | 1178 | xprt_force_disconnect(xprt); |
663b8858 | 1179 | case TCP_SYN_SENT: |
7c1d71cf | 1180 | xprt->connect_cookie++; |
663b8858 TM |
1181 | case TCP_CLOSING: |
1182 | /* | |
1183 | * If the server closed down the connection, make sure that | |
1184 | * we back off before reconnecting | |
1185 | */ | |
1186 | if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) | |
1187 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
3b948ae5 TM |
1188 | break; |
1189 | case TCP_LAST_ACK: | |
1190 | smp_mb__before_clear_bit(); | |
1191 | clear_bit(XPRT_CONNECTED, &xprt->state); | |
1192 | smp_mb__after_clear_bit(); | |
1193 | break; | |
1194 | case TCP_CLOSE: | |
1195 | smp_mb__before_clear_bit(); | |
ef803670 | 1196 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); |
3b948ae5 TM |
1197 | clear_bit(XPRT_CLOSING, &xprt->state); |
1198 | smp_mb__after_clear_bit(); | |
1199 | /* Mark transport as closed and wake up all pending tasks */ | |
62da3b24 | 1200 | xprt_disconnect_done(xprt); |
a246b010 CL |
1201 | } |
1202 | out: | |
1203 | read_unlock(&sk->sk_callback_lock); | |
1204 | } | |
1205 | ||
2a9e1cfa TM |
1206 | /** |
1207 | * xs_tcp_error_report - callback mainly for catching RST events | |
1208 | * @sk: socket | |
1209 | */ | |
1210 | static void xs_tcp_error_report(struct sock *sk) | |
1211 | { | |
1212 | struct rpc_xprt *xprt; | |
1213 | ||
1214 | read_lock(&sk->sk_callback_lock); | |
1215 | if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED) | |
1216 | goto out; | |
1217 | if (!(xprt = xprt_from_sock(sk))) | |
1218 | goto out; | |
1219 | dprintk("RPC: %s client %p...\n" | |
1220 | "RPC: error %d\n", | |
1221 | __func__, xprt, sk->sk_err); | |
1222 | ||
1223 | xprt_force_disconnect(xprt); | |
1224 | out: | |
1225 | read_unlock(&sk->sk_callback_lock); | |
1226 | } | |
1227 | ||
9903cd1c | 1228 | /** |
c7b2cae8 CL |
1229 | * xs_udp_write_space - callback invoked when socket buffer space |
1230 | * becomes available | |
9903cd1c CL |
1231 | * @sk: socket whose state has changed |
1232 | * | |
a246b010 CL |
1233 | * Called when more output buffer space is available for this socket. |
1234 | * We try not to wake our writers until they can make "significant" | |
c7b2cae8 | 1235 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg |
a246b010 CL |
1236 | * with a bunch of small requests. |
1237 | */ | |
c7b2cae8 | 1238 | static void xs_udp_write_space(struct sock *sk) |
a246b010 | 1239 | { |
a246b010 | 1240 | read_lock(&sk->sk_callback_lock); |
a246b010 | 1241 | |
c7b2cae8 CL |
1242 | /* from net/core/sock.c:sock_def_write_space */ |
1243 | if (sock_writeable(sk)) { | |
1244 | struct socket *sock; | |
1245 | struct rpc_xprt *xprt; | |
1246 | ||
1247 | if (unlikely(!(sock = sk->sk_socket))) | |
a246b010 | 1248 | goto out; |
b6ddf64f TM |
1249 | clear_bit(SOCK_NOSPACE, &sock->flags); |
1250 | ||
c7b2cae8 CL |
1251 | if (unlikely(!(xprt = xprt_from_sock(sk)))) |
1252 | goto out; | |
b6ddf64f | 1253 | if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) |
a246b010 | 1254 | goto out; |
c7b2cae8 CL |
1255 | |
1256 | xprt_write_space(xprt); | |
a246b010 CL |
1257 | } |
1258 | ||
c7b2cae8 CL |
1259 | out: |
1260 | read_unlock(&sk->sk_callback_lock); | |
1261 | } | |
a246b010 | 1262 | |
c7b2cae8 CL |
1263 | /** |
1264 | * xs_tcp_write_space - callback invoked when socket buffer space | |
1265 | * becomes available | |
1266 | * @sk: socket whose state has changed | |
1267 | * | |
1268 | * Called when more output buffer space is available for this socket. | |
1269 | * We try not to wake our writers until they can make "significant" | |
1270 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg | |
1271 | * with a bunch of small requests. | |
1272 | */ | |
1273 | static void xs_tcp_write_space(struct sock *sk) | |
1274 | { | |
1275 | read_lock(&sk->sk_callback_lock); | |
1276 | ||
1277 | /* from net/core/stream.c:sk_stream_write_space */ | |
1278 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { | |
1279 | struct socket *sock; | |
1280 | struct rpc_xprt *xprt; | |
1281 | ||
1282 | if (unlikely(!(sock = sk->sk_socket))) | |
1283 | goto out; | |
b6ddf64f TM |
1284 | clear_bit(SOCK_NOSPACE, &sock->flags); |
1285 | ||
c7b2cae8 CL |
1286 | if (unlikely(!(xprt = xprt_from_sock(sk)))) |
1287 | goto out; | |
b6ddf64f | 1288 | if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) |
c7b2cae8 CL |
1289 | goto out; |
1290 | ||
1291 | xprt_write_space(xprt); | |
1292 | } | |
1293 | ||
1294 | out: | |
a246b010 CL |
1295 | read_unlock(&sk->sk_callback_lock); |
1296 | } | |
1297 | ||
470056c2 | 1298 | static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) |
a246b010 | 1299 | { |
ee0ac0c2 CL |
1300 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
1301 | struct sock *sk = transport->inet; | |
a246b010 | 1302 | |
7c6e066e | 1303 | if (transport->rcvsize) { |
a246b010 | 1304 | sk->sk_userlocks |= SOCK_RCVBUF_LOCK; |
7c6e066e | 1305 | sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; |
a246b010 | 1306 | } |
7c6e066e | 1307 | if (transport->sndsize) { |
a246b010 | 1308 | sk->sk_userlocks |= SOCK_SNDBUF_LOCK; |
7c6e066e | 1309 | sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; |
a246b010 CL |
1310 | sk->sk_write_space(sk); |
1311 | } | |
1312 | } | |
1313 | ||
43118c29 | 1314 | /** |
470056c2 | 1315 | * xs_udp_set_buffer_size - set send and receive limits |
43118c29 | 1316 | * @xprt: generic transport |
470056c2 CL |
1317 | * @sndsize: requested size of send buffer, in bytes |
1318 | * @rcvsize: requested size of receive buffer, in bytes | |
43118c29 | 1319 | * |
470056c2 | 1320 | * Set socket send and receive buffer size limits. |
43118c29 | 1321 | */ |
470056c2 | 1322 | static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) |
43118c29 | 1323 | { |
7c6e066e CL |
1324 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
1325 | ||
1326 | transport->sndsize = 0; | |
470056c2 | 1327 | if (sndsize) |
7c6e066e CL |
1328 | transport->sndsize = sndsize + 1024; |
1329 | transport->rcvsize = 0; | |
470056c2 | 1330 | if (rcvsize) |
7c6e066e | 1331 | transport->rcvsize = rcvsize + 1024; |
470056c2 CL |
1332 | |
1333 | xs_udp_do_set_buffer_size(xprt); | |
43118c29 CL |
1334 | } |
1335 | ||
46c0ee8b CL |
1336 | /** |
1337 | * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport | |
1338 | * @task: task that timed out | |
1339 | * | |
1340 | * Adjust the congestion window after a retransmit timeout has occurred. | |
1341 | */ | |
1342 | static void xs_udp_timer(struct rpc_task *task) | |
1343 | { | |
1344 | xprt_adjust_cwnd(task, -ETIMEDOUT); | |
1345 | } | |
1346 | ||
b85d8806 CL |
1347 | static unsigned short xs_get_random_port(void) |
1348 | { | |
1349 | unsigned short range = xprt_max_resvport - xprt_min_resvport; | |
1350 | unsigned short rand = (unsigned short) net_random() % range; | |
1351 | return rand + xprt_min_resvport; | |
1352 | } | |
1353 | ||
92200412 CL |
1354 | /** |
1355 | * xs_set_port - reset the port number in the remote endpoint address | |
1356 | * @xprt: generic transport | |
1357 | * @port: new port number | |
1358 | * | |
1359 | */ | |
1360 | static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) | |
1361 | { | |
95392c59 | 1362 | struct sockaddr *addr = xs_addr(xprt); |
c4efcb1d | 1363 | |
46121cf7 | 1364 | dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); |
c4efcb1d | 1365 | |
20612005 CL |
1366 | switch (addr->sa_family) { |
1367 | case AF_INET: | |
1368 | ((struct sockaddr_in *)addr)->sin_port = htons(port); | |
1369 | break; | |
1370 | case AF_INET6: | |
1371 | ((struct sockaddr_in6 *)addr)->sin6_port = htons(port); | |
1372 | break; | |
1373 | default: | |
1374 | BUG(); | |
1375 | } | |
92200412 CL |
1376 | } |
1377 | ||
67a391d7 TM |
1378 | static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock) |
1379 | { | |
1380 | unsigned short port = transport->port; | |
1381 | ||
1382 | if (port == 0 && transport->xprt.resvport) | |
1383 | port = xs_get_random_port(); | |
1384 | return port; | |
1385 | } | |
1386 | ||
1387 | static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port) | |
1388 | { | |
1389 | if (transport->port != 0) | |
1390 | transport->port = 0; | |
1391 | if (!transport->xprt.resvport) | |
1392 | return 0; | |
1393 | if (port <= xprt_min_resvport || port > xprt_max_resvport) | |
1394 | return xprt_max_resvport; | |
1395 | return --port; | |
1396 | } | |
1397 | ||
7dc753f0 | 1398 | static int xs_bind4(struct sock_xprt *transport, struct socket *sock) |
a246b010 CL |
1399 | { |
1400 | struct sockaddr_in myaddr = { | |
1401 | .sin_family = AF_INET, | |
1402 | }; | |
d3bc9a1d | 1403 | struct sockaddr_in *sa; |
67a391d7 TM |
1404 | int err, nloop = 0; |
1405 | unsigned short port = xs_get_srcport(transport, sock); | |
1406 | unsigned short last; | |
a246b010 | 1407 | |
d3bc9a1d FM |
1408 | sa = (struct sockaddr_in *)&transport->addr; |
1409 | myaddr.sin_addr = sa->sin_addr; | |
a246b010 CL |
1410 | do { |
1411 | myaddr.sin_port = htons(port); | |
e6242e92 | 1412 | err = kernel_bind(sock, (struct sockaddr *) &myaddr, |
a246b010 | 1413 | sizeof(myaddr)); |
67a391d7 | 1414 | if (port == 0) |
d3bc9a1d | 1415 | break; |
a246b010 | 1416 | if (err == 0) { |
c8475461 | 1417 | transport->port = port; |
d3bc9a1d | 1418 | break; |
a246b010 | 1419 | } |
67a391d7 TM |
1420 | last = port; |
1421 | port = xs_next_srcport(transport, sock, port); | |
1422 | if (port > last) | |
1423 | nloop++; | |
1424 | } while (err == -EADDRINUSE && nloop != 2); | |
21454aaa HH |
1425 | dprintk("RPC: %s %pI4:%u: %s (%d)\n", |
1426 | __func__, &myaddr.sin_addr, | |
7dc753f0 | 1427 | port, err ? "failed" : "ok", err); |
a246b010 CL |
1428 | return err; |
1429 | } | |
1430 | ||
90058d37 CL |
1431 | static int xs_bind6(struct sock_xprt *transport, struct socket *sock) |
1432 | { | |
1433 | struct sockaddr_in6 myaddr = { | |
1434 | .sin6_family = AF_INET6, | |
1435 | }; | |
1436 | struct sockaddr_in6 *sa; | |
67a391d7 TM |
1437 | int err, nloop = 0; |
1438 | unsigned short port = xs_get_srcport(transport, sock); | |
1439 | unsigned short last; | |
90058d37 | 1440 | |
90058d37 CL |
1441 | sa = (struct sockaddr_in6 *)&transport->addr; |
1442 | myaddr.sin6_addr = sa->sin6_addr; | |
1443 | do { | |
1444 | myaddr.sin6_port = htons(port); | |
1445 | err = kernel_bind(sock, (struct sockaddr *) &myaddr, | |
1446 | sizeof(myaddr)); | |
67a391d7 | 1447 | if (port == 0) |
90058d37 CL |
1448 | break; |
1449 | if (err == 0) { | |
1450 | transport->port = port; | |
1451 | break; | |
1452 | } | |
67a391d7 TM |
1453 | last = port; |
1454 | port = xs_next_srcport(transport, sock, port); | |
1455 | if (port > last) | |
1456 | nloop++; | |
1457 | } while (err == -EADDRINUSE && nloop != 2); | |
5b095d98 | 1458 | dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n", |
fdb46ee7 | 1459 | &myaddr.sin6_addr, port, err ? "failed" : "ok", err); |
a246b010 CL |
1460 | return err; |
1461 | } | |
1462 | ||
ed07536e PZ |
1463 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1464 | static struct lock_class_key xs_key[2]; | |
1465 | static struct lock_class_key xs_slock_key[2]; | |
1466 | ||
8945ee5e | 1467 | static inline void xs_reclassify_socket4(struct socket *sock) |
ed07536e PZ |
1468 | { |
1469 | struct sock *sk = sock->sk; | |
8945ee5e | 1470 | |
02b3d346 | 1471 | BUG_ON(sock_owned_by_user(sk)); |
8945ee5e CL |
1472 | sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", |
1473 | &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); | |
1474 | } | |
ed07536e | 1475 | |
8945ee5e CL |
1476 | static inline void xs_reclassify_socket6(struct socket *sock) |
1477 | { | |
1478 | struct sock *sk = sock->sk; | |
ed07536e | 1479 | |
f4921aff | 1480 | BUG_ON(sock_owned_by_user(sk)); |
8945ee5e CL |
1481 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", |
1482 | &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); | |
ed07536e PZ |
1483 | } |
1484 | #else | |
8945ee5e CL |
1485 | static inline void xs_reclassify_socket4(struct socket *sock) |
1486 | { | |
1487 | } | |
1488 | ||
1489 | static inline void xs_reclassify_socket6(struct socket *sock) | |
ed07536e PZ |
1490 | { |
1491 | } | |
1492 | #endif | |
1493 | ||
16be2d20 CL |
1494 | static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) |
1495 | { | |
1496 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); | |
1497 | ||
1498 | if (!transport->inet) { | |
1499 | struct sock *sk = sock->sk; | |
1500 | ||
1501 | write_lock_bh(&sk->sk_callback_lock); | |
1502 | ||
2a9e1cfa TM |
1503 | xs_save_old_callbacks(transport, sk); |
1504 | ||
16be2d20 | 1505 | sk->sk_user_data = xprt; |
16be2d20 CL |
1506 | sk->sk_data_ready = xs_udp_data_ready; |
1507 | sk->sk_write_space = xs_udp_write_space; | |
1508 | sk->sk_no_check = UDP_CSUM_NORCV; | |
1509 | sk->sk_allocation = GFP_ATOMIC; | |
1510 | ||
1511 | xprt_set_connected(xprt); | |
1512 | ||
1513 | /* Reset to new socket */ | |
1514 | transport->sock = sock; | |
1515 | transport->inet = sk; | |
1516 | ||
1517 | write_unlock_bh(&sk->sk_callback_lock); | |
1518 | } | |
1519 | xs_udp_do_set_buffer_size(xprt); | |
1520 | } | |
1521 | ||
b0d93ad5 | 1522 | /** |
9c3d72de | 1523 | * xs_udp_connect_worker4 - set up a UDP socket |
65f27f38 | 1524 | * @work: RPC transport to connect |
b0d93ad5 CL |
1525 | * |
1526 | * Invoked by a work queue tasklet. | |
1527 | */ | |
9c3d72de | 1528 | static void xs_udp_connect_worker4(struct work_struct *work) |
a246b010 | 1529 | { |
34161db6 TM |
1530 | struct sock_xprt *transport = |
1531 | container_of(work, struct sock_xprt, connect_worker.work); | |
c8475461 | 1532 | struct rpc_xprt *xprt = &transport->xprt; |
ee0ac0c2 | 1533 | struct socket *sock = transport->sock; |
b0d93ad5 | 1534 | int err, status = -EIO; |
9903cd1c | 1535 | |
01d37c42 | 1536 | if (xprt->shutdown) |
b0d93ad5 | 1537 | goto out; |
9903cd1c | 1538 | |
b0d93ad5 CL |
1539 | /* Start by resetting any existing state */ |
1540 | xs_close(xprt); | |
9903cd1c | 1541 | |
b0d93ad5 | 1542 | if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) { |
46121cf7 | 1543 | dprintk("RPC: can't create UDP transport socket (%d).\n", -err); |
b0d93ad5 CL |
1544 | goto out; |
1545 | } | |
8945ee5e | 1546 | xs_reclassify_socket4(sock); |
9903cd1c | 1547 | |
7dc753f0 | 1548 | if (xs_bind4(transport, sock)) { |
b0d93ad5 CL |
1549 | sock_release(sock); |
1550 | goto out; | |
1551 | } | |
9903cd1c | 1552 | |
46121cf7 | 1553 | dprintk("RPC: worker connecting xprt %p to address: %s\n", |
7559c7a2 | 1554 | xprt, xprt->address_strings[RPC_DISPLAY_ALL]); |
edb267a6 | 1555 | |
16be2d20 | 1556 | xs_udp_finish_connecting(xprt, sock); |
b0d93ad5 CL |
1557 | status = 0; |
1558 | out: | |
1559 | xprt_wake_pending_tasks(xprt, status); | |
1560 | xprt_clear_connecting(xprt); | |
a246b010 | 1561 | } |
a246b010 | 1562 | |
68e220bd CL |
1563 | /** |
1564 | * xs_udp_connect_worker6 - set up a UDP socket | |
1565 | * @work: RPC transport to connect | |
1566 | * | |
1567 | * Invoked by a work queue tasklet. | |
1568 | */ | |
1569 | static void xs_udp_connect_worker6(struct work_struct *work) | |
1570 | { | |
1571 | struct sock_xprt *transport = | |
1572 | container_of(work, struct sock_xprt, connect_worker.work); | |
1573 | struct rpc_xprt *xprt = &transport->xprt; | |
1574 | struct socket *sock = transport->sock; | |
1575 | int err, status = -EIO; | |
a246b010 | 1576 | |
01d37c42 | 1577 | if (xprt->shutdown) |
68e220bd | 1578 | goto out; |
b0d93ad5 | 1579 | |
68e220bd CL |
1580 | /* Start by resetting any existing state */ |
1581 | xs_close(xprt); | |
a246b010 | 1582 | |
68e220bd CL |
1583 | if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) { |
1584 | dprintk("RPC: can't create UDP transport socket (%d).\n", -err); | |
1585 | goto out; | |
1586 | } | |
8945ee5e | 1587 | xs_reclassify_socket6(sock); |
a246b010 | 1588 | |
68e220bd CL |
1589 | if (xs_bind6(transport, sock) < 0) { |
1590 | sock_release(sock); | |
1591 | goto out; | |
b0d93ad5 | 1592 | } |
68e220bd CL |
1593 | |
1594 | dprintk("RPC: worker connecting xprt %p to address: %s\n", | |
1595 | xprt, xprt->address_strings[RPC_DISPLAY_ALL]); | |
1596 | ||
1597 | xs_udp_finish_connecting(xprt, sock); | |
b0d93ad5 CL |
1598 | status = 0; |
1599 | out: | |
1600 | xprt_wake_pending_tasks(xprt, status); | |
1601 | xprt_clear_connecting(xprt); | |
a246b010 CL |
1602 | } |
1603 | ||
3167e12c CL |
1604 | /* |
1605 | * We need to preserve the port number so the reply cache on the server can | |
1606 | * find our cached RPC replies when we get around to reconnecting. | |
1607 | */ | |
1608 | static void xs_tcp_reuse_connection(struct rpc_xprt *xprt) | |
1609 | { | |
1610 | int result; | |
ee0ac0c2 | 1611 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
3167e12c CL |
1612 | struct sockaddr any; |
1613 | ||
46121cf7 | 1614 | dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt); |
3167e12c CL |
1615 | |
1616 | /* | |
1617 | * Disconnect the transport socket by doing a connect operation | |
1618 | * with AF_UNSPEC. This should return immediately... | |
1619 | */ | |
1620 | memset(&any, 0, sizeof(any)); | |
1621 | any.sa_family = AF_UNSPEC; | |
ee0ac0c2 | 1622 | result = kernel_connect(transport->sock, &any, sizeof(any), 0); |
3167e12c | 1623 | if (result) |
46121cf7 | 1624 | dprintk("RPC: AF_UNSPEC connect return code %d\n", |
3167e12c CL |
1625 | result); |
1626 | } | |
1627 | ||
16be2d20 | 1628 | static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) |
a246b010 | 1629 | { |
16be2d20 | 1630 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
edb267a6 | 1631 | |
ee0ac0c2 | 1632 | if (!transport->inet) { |
b0d93ad5 CL |
1633 | struct sock *sk = sock->sk; |
1634 | ||
1635 | write_lock_bh(&sk->sk_callback_lock); | |
1636 | ||
2a9e1cfa TM |
1637 | xs_save_old_callbacks(transport, sk); |
1638 | ||
b0d93ad5 | 1639 | sk->sk_user_data = xprt; |
b0d93ad5 CL |
1640 | sk->sk_data_ready = xs_tcp_data_ready; |
1641 | sk->sk_state_change = xs_tcp_state_change; | |
1642 | sk->sk_write_space = xs_tcp_write_space; | |
2a9e1cfa | 1643 | sk->sk_error_report = xs_tcp_error_report; |
b079fa7b | 1644 | sk->sk_allocation = GFP_ATOMIC; |
3167e12c CL |
1645 | |
1646 | /* socket options */ | |
1647 | sk->sk_userlocks |= SOCK_BINDPORT_LOCK; | |
1648 | sock_reset_flag(sk, SOCK_LINGER); | |
1649 | tcp_sk(sk)->linger2 = 0; | |
1650 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; | |
b0d93ad5 CL |
1651 | |
1652 | xprt_clear_connected(xprt); | |
1653 | ||
1654 | /* Reset to new socket */ | |
ee0ac0c2 CL |
1655 | transport->sock = sock; |
1656 | transport->inet = sk; | |
b0d93ad5 CL |
1657 | |
1658 | write_unlock_bh(&sk->sk_callback_lock); | |
1659 | } | |
1660 | ||
01d37c42 TM |
1661 | if (!xprt_bound(xprt)) |
1662 | return -ENOTCONN; | |
1663 | ||
b0d93ad5 | 1664 | /* Tell the socket layer to start connecting... */ |
262ca07d CL |
1665 | xprt->stat.connect_count++; |
1666 | xprt->stat.connect_start = jiffies; | |
95392c59 | 1667 | return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); |
16be2d20 CL |
1668 | } |
1669 | ||
9903cd1c | 1670 | /** |
9c3d72de | 1671 | * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint |
65f27f38 | 1672 | * @work: RPC transport to connect |
9903cd1c CL |
1673 | * |
1674 | * Invoked by a work queue tasklet. | |
a246b010 | 1675 | */ |
9c3d72de | 1676 | static void xs_tcp_connect_worker4(struct work_struct *work) |
a246b010 | 1677 | { |
34161db6 TM |
1678 | struct sock_xprt *transport = |
1679 | container_of(work, struct sock_xprt, connect_worker.work); | |
c8475461 | 1680 | struct rpc_xprt *xprt = &transport->xprt; |
ee0ac0c2 | 1681 | struct socket *sock = transport->sock; |
b0d93ad5 | 1682 | int err, status = -EIO; |
a246b010 | 1683 | |
01d37c42 | 1684 | if (xprt->shutdown) |
a246b010 CL |
1685 | goto out; |
1686 | ||
ee0ac0c2 | 1687 | if (!sock) { |
3167e12c CL |
1688 | /* start from scratch */ |
1689 | if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) { | |
16be2d20 | 1690 | dprintk("RPC: can't create TCP transport socket (%d).\n", -err); |
3167e12c CL |
1691 | goto out; |
1692 | } | |
8945ee5e | 1693 | xs_reclassify_socket4(sock); |
a246b010 | 1694 | |
16be2d20 | 1695 | if (xs_bind4(transport, sock) < 0) { |
3167e12c CL |
1696 | sock_release(sock); |
1697 | goto out; | |
1698 | } | |
1699 | } else | |
1700 | /* "close" the socket, preserving the local port */ | |
1701 | xs_tcp_reuse_connection(xprt); | |
a246b010 | 1702 | |
46121cf7 | 1703 | dprintk("RPC: worker connecting xprt %p to address: %s\n", |
7559c7a2 | 1704 | xprt, xprt->address_strings[RPC_DISPLAY_ALL]); |
edb267a6 | 1705 | |
16be2d20 | 1706 | status = xs_tcp_finish_connecting(xprt, sock); |
46121cf7 CL |
1707 | dprintk("RPC: %p connect status %d connected %d sock state %d\n", |
1708 | xprt, -status, xprt_connected(xprt), | |
1709 | sock->sk->sk_state); | |
a246b010 CL |
1710 | if (status < 0) { |
1711 | switch (status) { | |
1712 | case -EINPROGRESS: | |
1713 | case -EALREADY: | |
1714 | goto out_clear; | |
3167e12c CL |
1715 | case -ECONNREFUSED: |
1716 | case -ECONNRESET: | |
1717 | /* retry with existing socket, after a delay */ | |
1718 | break; | |
1719 | default: | |
1720 | /* get rid of existing socket, and retry */ | |
e06799f9 | 1721 | xs_tcp_shutdown(xprt); |
a246b010 CL |
1722 | } |
1723 | } | |
1724 | out: | |
44fbac22 | 1725 | xprt_wake_pending_tasks(xprt, status); |
68e220bd CL |
1726 | out_clear: |
1727 | xprt_clear_connecting(xprt); | |
1728 | } | |
b0d93ad5 | 1729 | |
68e220bd CL |
1730 | /** |
1731 | * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint | |
1732 | * @work: RPC transport to connect | |
1733 | * | |
1734 | * Invoked by a work queue tasklet. | |
1735 | */ | |
1736 | static void xs_tcp_connect_worker6(struct work_struct *work) | |
1737 | { | |
1738 | struct sock_xprt *transport = | |
1739 | container_of(work, struct sock_xprt, connect_worker.work); | |
1740 | struct rpc_xprt *xprt = &transport->xprt; | |
1741 | struct socket *sock = transport->sock; | |
1742 | int err, status = -EIO; | |
3167e12c | 1743 | |
01d37c42 | 1744 | if (xprt->shutdown) |
68e220bd | 1745 | goto out; |
b0d93ad5 | 1746 | |
68e220bd CL |
1747 | if (!sock) { |
1748 | /* start from scratch */ | |
1749 | if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) { | |
1750 | dprintk("RPC: can't create TCP transport socket (%d).\n", -err); | |
1751 | goto out; | |
1752 | } | |
8945ee5e | 1753 | xs_reclassify_socket6(sock); |
b0d93ad5 | 1754 | |
68e220bd CL |
1755 | if (xs_bind6(transport, sock) < 0) { |
1756 | sock_release(sock); | |
1757 | goto out; | |
1758 | } | |
1759 | } else | |
1760 | /* "close" the socket, preserving the local port */ | |
1761 | xs_tcp_reuse_connection(xprt); | |
b0d93ad5 | 1762 | |
68e220bd CL |
1763 | dprintk("RPC: worker connecting xprt %p to address: %s\n", |
1764 | xprt, xprt->address_strings[RPC_DISPLAY_ALL]); | |
b0d93ad5 | 1765 | |
68e220bd | 1766 | status = xs_tcp_finish_connecting(xprt, sock); |
46121cf7 | 1767 | dprintk("RPC: %p connect status %d connected %d sock state %d\n", |
68e220bd | 1768 | xprt, -status, xprt_connected(xprt), sock->sk->sk_state); |
a246b010 CL |
1769 | if (status < 0) { |
1770 | switch (status) { | |
1771 | case -EINPROGRESS: | |
1772 | case -EALREADY: | |
1773 | goto out_clear; | |
3167e12c CL |
1774 | case -ECONNREFUSED: |
1775 | case -ECONNRESET: | |
1776 | /* retry with existing socket, after a delay */ | |
1777 | break; | |
1778 | default: | |
1779 | /* get rid of existing socket, and retry */ | |
e06799f9 | 1780 | xs_tcp_shutdown(xprt); |
a246b010 CL |
1781 | } |
1782 | } | |
1783 | out: | |
44fbac22 | 1784 | xprt_wake_pending_tasks(xprt, status); |
a246b010 | 1785 | out_clear: |
2226feb6 | 1786 | xprt_clear_connecting(xprt); |
a246b010 CL |
1787 | } |
1788 | ||
9903cd1c CL |
1789 | /** |
1790 | * xs_connect - connect a socket to a remote endpoint | |
1791 | * @task: address of RPC task that manages state of connect request | |
1792 | * | |
1793 | * TCP: If the remote end dropped the connection, delay reconnecting. | |
03bf4b70 CL |
1794 | * |
1795 | * UDP socket connects are synchronous, but we use a work queue anyway | |
1796 | * to guarantee that even unprivileged user processes can set up a | |
1797 | * socket on a privileged port. | |
1798 | * | |
1799 | * If a UDP socket connect fails, the delay behavior here prevents | |
1800 | * retry floods (hard mounts). | |
9903cd1c CL |
1801 | */ |
1802 | static void xs_connect(struct rpc_task *task) | |
a246b010 CL |
1803 | { |
1804 | struct rpc_xprt *xprt = task->tk_xprt; | |
ee0ac0c2 | 1805 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 | 1806 | |
b0d93ad5 CL |
1807 | if (xprt_test_and_set_connecting(xprt)) |
1808 | return; | |
1809 | ||
ee0ac0c2 | 1810 | if (transport->sock != NULL) { |
46121cf7 CL |
1811 | dprintk("RPC: xs_connect delayed xprt %p for %lu " |
1812 | "seconds\n", | |
03bf4b70 | 1813 | xprt, xprt->reestablish_timeout / HZ); |
c1384c9c TM |
1814 | queue_delayed_work(rpciod_workqueue, |
1815 | &transport->connect_worker, | |
1816 | xprt->reestablish_timeout); | |
03bf4b70 CL |
1817 | xprt->reestablish_timeout <<= 1; |
1818 | if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) | |
1819 | xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; | |
b0d93ad5 | 1820 | } else { |
46121cf7 | 1821 | dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); |
c1384c9c TM |
1822 | queue_delayed_work(rpciod_workqueue, |
1823 | &transport->connect_worker, 0); | |
a246b010 CL |
1824 | } |
1825 | } | |
1826 | ||
e06799f9 TM |
1827 | static void xs_tcp_connect(struct rpc_task *task) |
1828 | { | |
1829 | struct rpc_xprt *xprt = task->tk_xprt; | |
1830 | ||
1831 | /* Initiate graceful shutdown of the socket if not already done */ | |
1832 | if (test_bit(XPRT_CONNECTED, &xprt->state)) | |
1833 | xs_tcp_shutdown(xprt); | |
1834 | /* Exit if we need to wait for socket shutdown to complete */ | |
1835 | if (test_bit(XPRT_CLOSING, &xprt->state)) | |
1836 | return; | |
1837 | xs_connect(task); | |
1838 | } | |
1839 | ||
262ca07d CL |
1840 | /** |
1841 | * xs_udp_print_stats - display UDP socket-specifc stats | |
1842 | * @xprt: rpc_xprt struct containing statistics | |
1843 | * @seq: output file | |
1844 | * | |
1845 | */ | |
1846 | static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1847 | { | |
c8475461 CL |
1848 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
1849 | ||
262ca07d | 1850 | seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n", |
c8475461 | 1851 | transport->port, |
262ca07d CL |
1852 | xprt->stat.bind_count, |
1853 | xprt->stat.sends, | |
1854 | xprt->stat.recvs, | |
1855 | xprt->stat.bad_xids, | |
1856 | xprt->stat.req_u, | |
1857 | xprt->stat.bklog_u); | |
1858 | } | |
1859 | ||
1860 | /** | |
1861 | * xs_tcp_print_stats - display TCP socket-specifc stats | |
1862 | * @xprt: rpc_xprt struct containing statistics | |
1863 | * @seq: output file | |
1864 | * | |
1865 | */ | |
1866 | static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
1867 | { | |
c8475461 | 1868 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
262ca07d CL |
1869 | long idle_time = 0; |
1870 | ||
1871 | if (xprt_connected(xprt)) | |
1872 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
1873 | ||
1874 | seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n", | |
c8475461 | 1875 | transport->port, |
262ca07d CL |
1876 | xprt->stat.bind_count, |
1877 | xprt->stat.connect_count, | |
1878 | xprt->stat.connect_time, | |
1879 | idle_time, | |
1880 | xprt->stat.sends, | |
1881 | xprt->stat.recvs, | |
1882 | xprt->stat.bad_xids, | |
1883 | xprt->stat.req_u, | |
1884 | xprt->stat.bklog_u); | |
1885 | } | |
1886 | ||
262965f5 | 1887 | static struct rpc_xprt_ops xs_udp_ops = { |
43118c29 | 1888 | .set_buffer_size = xs_udp_set_buffer_size, |
12a80469 | 1889 | .reserve_xprt = xprt_reserve_xprt_cong, |
49e9a890 | 1890 | .release_xprt = xprt_release_xprt_cong, |
45160d62 | 1891 | .rpcbind = rpcb_getport_async, |
92200412 | 1892 | .set_port = xs_set_port, |
262965f5 | 1893 | .connect = xs_connect, |
02107148 CL |
1894 | .buf_alloc = rpc_malloc, |
1895 | .buf_free = rpc_free, | |
262965f5 | 1896 | .send_request = xs_udp_send_request, |
fe3aca29 | 1897 | .set_retrans_timeout = xprt_set_retrans_timeout_rtt, |
46c0ee8b | 1898 | .timer = xs_udp_timer, |
a58dd398 | 1899 | .release_request = xprt_release_rqst_cong, |
262965f5 CL |
1900 | .close = xs_close, |
1901 | .destroy = xs_destroy, | |
262ca07d | 1902 | .print_stats = xs_udp_print_stats, |
262965f5 CL |
1903 | }; |
1904 | ||
1905 | static struct rpc_xprt_ops xs_tcp_ops = { | |
12a80469 | 1906 | .reserve_xprt = xprt_reserve_xprt, |
e0ab53de | 1907 | .release_xprt = xs_tcp_release_xprt, |
45160d62 | 1908 | .rpcbind = rpcb_getport_async, |
92200412 | 1909 | .set_port = xs_set_port, |
e06799f9 | 1910 | .connect = xs_tcp_connect, |
02107148 CL |
1911 | .buf_alloc = rpc_malloc, |
1912 | .buf_free = rpc_free, | |
262965f5 | 1913 | .send_request = xs_tcp_send_request, |
fe3aca29 | 1914 | .set_retrans_timeout = xprt_set_retrans_timeout_def, |
e06799f9 | 1915 | .close = xs_tcp_shutdown, |
9903cd1c | 1916 | .destroy = xs_destroy, |
262ca07d | 1917 | .print_stats = xs_tcp_print_stats, |
a246b010 CL |
1918 | }; |
1919 | ||
3c341b0b | 1920 | static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, |
bc25571e | 1921 | unsigned int slot_table_size) |
c8541ecd CL |
1922 | { |
1923 | struct rpc_xprt *xprt; | |
ffc2e518 | 1924 | struct sock_xprt *new; |
c8541ecd | 1925 | |
96802a09 | 1926 | if (args->addrlen > sizeof(xprt->addr)) { |
46121cf7 | 1927 | dprintk("RPC: xs_setup_xprt: address too large\n"); |
c8541ecd CL |
1928 | return ERR_PTR(-EBADF); |
1929 | } | |
1930 | ||
ffc2e518 CL |
1931 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1932 | if (new == NULL) { | |
46121cf7 CL |
1933 | dprintk("RPC: xs_setup_xprt: couldn't allocate " |
1934 | "rpc_xprt\n"); | |
c8541ecd CL |
1935 | return ERR_PTR(-ENOMEM); |
1936 | } | |
ffc2e518 | 1937 | xprt = &new->xprt; |
c8541ecd CL |
1938 | |
1939 | xprt->max_reqs = slot_table_size; | |
1940 | xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL); | |
1941 | if (xprt->slot == NULL) { | |
1942 | kfree(xprt); | |
46121cf7 CL |
1943 | dprintk("RPC: xs_setup_xprt: couldn't allocate slot " |
1944 | "table\n"); | |
c8541ecd CL |
1945 | return ERR_PTR(-ENOMEM); |
1946 | } | |
1947 | ||
96802a09 FM |
1948 | memcpy(&xprt->addr, args->dstaddr, args->addrlen); |
1949 | xprt->addrlen = args->addrlen; | |
d3bc9a1d FM |
1950 | if (args->srcaddr) |
1951 | memcpy(&new->addr, args->srcaddr, args->addrlen); | |
c8541ecd CL |
1952 | |
1953 | return xprt; | |
1954 | } | |
1955 | ||
2881ae74 TM |
1956 | static const struct rpc_timeout xs_udp_default_timeout = { |
1957 | .to_initval = 5 * HZ, | |
1958 | .to_maxval = 30 * HZ, | |
1959 | .to_increment = 5 * HZ, | |
1960 | .to_retries = 5, | |
1961 | }; | |
1962 | ||
9903cd1c CL |
1963 | /** |
1964 | * xs_setup_udp - Set up transport to use a UDP socket | |
96802a09 | 1965 | * @args: rpc transport creation arguments |
9903cd1c CL |
1966 | * |
1967 | */ | |
483066d6 | 1968 | static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) |
a246b010 | 1969 | { |
8f9d5b1a | 1970 | struct sockaddr *addr = args->dstaddr; |
c8541ecd | 1971 | struct rpc_xprt *xprt; |
c8475461 | 1972 | struct sock_xprt *transport; |
a246b010 | 1973 | |
96802a09 | 1974 | xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries); |
c8541ecd CL |
1975 | if (IS_ERR(xprt)) |
1976 | return xprt; | |
c8475461 | 1977 | transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 | 1978 | |
ec739ef0 | 1979 | xprt->prot = IPPROTO_UDP; |
808012fb | 1980 | xprt->tsh_size = 0; |
a246b010 CL |
1981 | /* XXX: header size can vary due to auth type, IPv6, etc. */ |
1982 | xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); | |
1983 | ||
03bf4b70 CL |
1984 | xprt->bind_timeout = XS_BIND_TO; |
1985 | xprt->connect_timeout = XS_UDP_CONN_TO; | |
1986 | xprt->reestablish_timeout = XS_UDP_REEST_TO; | |
1987 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 1988 | |
262965f5 | 1989 | xprt->ops = &xs_udp_ops; |
a246b010 | 1990 | |
ba7392bb | 1991 | xprt->timeout = &xs_udp_default_timeout; |
a246b010 | 1992 | |
8f9d5b1a CL |
1993 | switch (addr->sa_family) { |
1994 | case AF_INET: | |
1995 | if (((struct sockaddr_in *)addr)->sin_port != htons(0)) | |
1996 | xprt_set_bound(xprt); | |
1997 | ||
1998 | INIT_DELAYED_WORK(&transport->connect_worker, | |
1999 | xs_udp_connect_worker4); | |
b454ae90 | 2000 | xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); |
8f9d5b1a CL |
2001 | break; |
2002 | case AF_INET6: | |
2003 | if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) | |
2004 | xprt_set_bound(xprt); | |
2005 | ||
2006 | INIT_DELAYED_WORK(&transport->connect_worker, | |
2007 | xs_udp_connect_worker6); | |
b454ae90 | 2008 | xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); |
8f9d5b1a CL |
2009 | break; |
2010 | default: | |
2011 | kfree(xprt); | |
2012 | return ERR_PTR(-EAFNOSUPPORT); | |
2013 | } | |
2014 | ||
46121cf7 | 2015 | dprintk("RPC: set up transport to address %s\n", |
7559c7a2 | 2016 | xprt->address_strings[RPC_DISPLAY_ALL]); |
edb267a6 | 2017 | |
bc25571e TT |
2018 | if (try_module_get(THIS_MODULE)) |
2019 | return xprt; | |
2020 | ||
2021 | kfree(xprt->slot); | |
2022 | kfree(xprt); | |
2023 | return ERR_PTR(-EINVAL); | |
a246b010 CL |
2024 | } |
2025 | ||
2881ae74 TM |
2026 | static const struct rpc_timeout xs_tcp_default_timeout = { |
2027 | .to_initval = 60 * HZ, | |
2028 | .to_maxval = 60 * HZ, | |
2029 | .to_retries = 2, | |
2030 | }; | |
2031 | ||
9903cd1c CL |
2032 | /** |
2033 | * xs_setup_tcp - Set up transport to use a TCP socket | |
96802a09 | 2034 | * @args: rpc transport creation arguments |
9903cd1c CL |
2035 | * |
2036 | */ | |
483066d6 | 2037 | static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) |
a246b010 | 2038 | { |
8f9d5b1a | 2039 | struct sockaddr *addr = args->dstaddr; |
c8541ecd | 2040 | struct rpc_xprt *xprt; |
c8475461 | 2041 | struct sock_xprt *transport; |
a246b010 | 2042 | |
96802a09 | 2043 | xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries); |
c8541ecd CL |
2044 | if (IS_ERR(xprt)) |
2045 | return xprt; | |
c8475461 | 2046 | transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 | 2047 | |
ec739ef0 | 2048 | xprt->prot = IPPROTO_TCP; |
808012fb | 2049 | xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); |
808012fb | 2050 | xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; |
a246b010 | 2051 | |
03bf4b70 CL |
2052 | xprt->bind_timeout = XS_BIND_TO; |
2053 | xprt->connect_timeout = XS_TCP_CONN_TO; | |
2054 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
2055 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 2056 | |
262965f5 | 2057 | xprt->ops = &xs_tcp_ops; |
ba7392bb | 2058 | xprt->timeout = &xs_tcp_default_timeout; |
a246b010 | 2059 | |
8f9d5b1a CL |
2060 | switch (addr->sa_family) { |
2061 | case AF_INET: | |
2062 | if (((struct sockaddr_in *)addr)->sin_port != htons(0)) | |
2063 | xprt_set_bound(xprt); | |
2064 | ||
2065 | INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4); | |
b454ae90 | 2066 | xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); |
8f9d5b1a CL |
2067 | break; |
2068 | case AF_INET6: | |
2069 | if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) | |
2070 | xprt_set_bound(xprt); | |
2071 | ||
2072 | INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6); | |
b454ae90 | 2073 | xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); |
8f9d5b1a CL |
2074 | break; |
2075 | default: | |
2076 | kfree(xprt); | |
2077 | return ERR_PTR(-EAFNOSUPPORT); | |
2078 | } | |
2079 | ||
46121cf7 | 2080 | dprintk("RPC: set up transport to address %s\n", |
7559c7a2 | 2081 | xprt->address_strings[RPC_DISPLAY_ALL]); |
edb267a6 | 2082 | |
bc25571e TT |
2083 | if (try_module_get(THIS_MODULE)) |
2084 | return xprt; | |
2085 | ||
2086 | kfree(xprt->slot); | |
2087 | kfree(xprt); | |
2088 | return ERR_PTR(-EINVAL); | |
a246b010 | 2089 | } |
282b32e1 | 2090 | |
bc25571e TT |
2091 | static struct xprt_class xs_udp_transport = { |
2092 | .list = LIST_HEAD_INIT(xs_udp_transport.list), | |
2093 | .name = "udp", | |
2094 | .owner = THIS_MODULE, | |
4fa016eb | 2095 | .ident = IPPROTO_UDP, |
bc25571e TT |
2096 | .setup = xs_setup_udp, |
2097 | }; | |
2098 | ||
2099 | static struct xprt_class xs_tcp_transport = { | |
2100 | .list = LIST_HEAD_INIT(xs_tcp_transport.list), | |
2101 | .name = "tcp", | |
2102 | .owner = THIS_MODULE, | |
4fa016eb | 2103 | .ident = IPPROTO_TCP, |
bc25571e TT |
2104 | .setup = xs_setup_tcp, |
2105 | }; | |
2106 | ||
282b32e1 | 2107 | /** |
bc25571e | 2108 | * init_socket_xprt - set up xprtsock's sysctls, register with RPC client |
282b32e1 CL |
2109 | * |
2110 | */ | |
2111 | int init_socket_xprt(void) | |
2112 | { | |
fbf76683 | 2113 | #ifdef RPC_DEBUG |
2b1bec5f | 2114 | if (!sunrpc_table_header) |
0b4d4147 | 2115 | sunrpc_table_header = register_sysctl_table(sunrpc_table); |
fbf76683 CL |
2116 | #endif |
2117 | ||
bc25571e TT |
2118 | xprt_register_transport(&xs_udp_transport); |
2119 | xprt_register_transport(&xs_tcp_transport); | |
2120 | ||
282b32e1 CL |
2121 | return 0; |
2122 | } | |
2123 | ||
2124 | /** | |
bc25571e | 2125 | * cleanup_socket_xprt - remove xprtsock's sysctls, unregister |
282b32e1 CL |
2126 | * |
2127 | */ | |
2128 | void cleanup_socket_xprt(void) | |
2129 | { | |
fbf76683 CL |
2130 | #ifdef RPC_DEBUG |
2131 | if (sunrpc_table_header) { | |
2132 | unregister_sysctl_table(sunrpc_table_header); | |
2133 | sunrpc_table_header = NULL; | |
2134 | } | |
2135 | #endif | |
bc25571e TT |
2136 | |
2137 | xprt_unregister_transport(&xs_udp_transport); | |
2138 | xprt_unregister_transport(&xs_tcp_transport); | |
282b32e1 | 2139 | } |