]>
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> | |
176e21ee | 22 | #include <linux/string.h> |
a246b010 | 23 | #include <linux/slab.h> |
bc25571e | 24 | #include <linux/module.h> |
a246b010 | 25 | #include <linux/capability.h> |
a246b010 CL |
26 | #include <linux/pagemap.h> |
27 | #include <linux/errno.h> | |
28 | #include <linux/socket.h> | |
29 | #include <linux/in.h> | |
30 | #include <linux/net.h> | |
31 | #include <linux/mm.h> | |
176e21ee | 32 | #include <linux/un.h> |
a246b010 CL |
33 | #include <linux/udp.h> |
34 | #include <linux/tcp.h> | |
35 | #include <linux/sunrpc/clnt.h> | |
5976687a | 36 | #include <linux/sunrpc/addr.h> |
02107148 | 37 | #include <linux/sunrpc/sched.h> |
4cfc7e60 | 38 | #include <linux/sunrpc/svcsock.h> |
49c36fcc | 39 | #include <linux/sunrpc/xprtsock.h> |
a246b010 | 40 | #include <linux/file.h> |
9e00abc3 | 41 | #ifdef CONFIG_SUNRPC_BACKCHANNEL |
44b98efd RL |
42 | #include <linux/sunrpc/bc_xprt.h> |
43 | #endif | |
a246b010 CL |
44 | |
45 | #include <net/sock.h> | |
46 | #include <net/checksum.h> | |
47 | #include <net/udp.h> | |
48 | #include <net/tcp.h> | |
49 | ||
40b5ea0c TM |
50 | #include <trace/events/sunrpc.h> |
51 | ||
4cfc7e60 | 52 | #include "sunrpc.h" |
176e21ee CL |
53 | |
54 | static void xs_close(struct rpc_xprt *xprt); | |
55 | ||
c556b754 CL |
56 | /* |
57 | * xprtsock tunables | |
58 | */ | |
09acfea5 TM |
59 | static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; |
60 | static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; | |
61 | static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; | |
c556b754 | 62 | |
09acfea5 TM |
63 | static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; |
64 | static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; | |
c556b754 | 65 | |
7d1e8255 | 66 | #define XS_TCP_LINGER_TO (15U * HZ) |
25fe6142 | 67 | static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; |
7d1e8255 | 68 | |
fbf76683 CL |
69 | /* |
70 | * We can register our own files under /proc/sys/sunrpc by | |
71 | * calling register_sysctl_table() again. The files in that | |
72 | * directory become the union of all files registered there. | |
73 | * | |
74 | * We simply need to make sure that we don't collide with | |
75 | * someone else's file names! | |
76 | */ | |
77 | ||
78 | #ifdef RPC_DEBUG | |
79 | ||
80 | static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; | |
81 | static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; | |
d9ba131d | 82 | static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; |
fbf76683 CL |
83 | static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; |
84 | static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; | |
85 | ||
86 | static struct ctl_table_header *sunrpc_table_header; | |
87 | ||
88 | /* | |
89 | * FIXME: changing the UDP slot table size should also resize the UDP | |
90 | * socket buffers for existing UDP transports | |
91 | */ | |
fe2c6338 | 92 | static struct ctl_table xs_tunables_table[] = { |
fbf76683 | 93 | { |
fbf76683 CL |
94 | .procname = "udp_slot_table_entries", |
95 | .data = &xprt_udp_slot_table_entries, | |
96 | .maxlen = sizeof(unsigned int), | |
97 | .mode = 0644, | |
6d456111 | 98 | .proc_handler = proc_dointvec_minmax, |
fbf76683 CL |
99 | .extra1 = &min_slot_table_size, |
100 | .extra2 = &max_slot_table_size | |
101 | }, | |
102 | { | |
fbf76683 CL |
103 | .procname = "tcp_slot_table_entries", |
104 | .data = &xprt_tcp_slot_table_entries, | |
105 | .maxlen = sizeof(unsigned int), | |
106 | .mode = 0644, | |
6d456111 | 107 | .proc_handler = proc_dointvec_minmax, |
fbf76683 CL |
108 | .extra1 = &min_slot_table_size, |
109 | .extra2 = &max_slot_table_size | |
110 | }, | |
d9ba131d TM |
111 | { |
112 | .procname = "tcp_max_slot_table_entries", | |
113 | .data = &xprt_max_tcp_slot_table_entries, | |
114 | .maxlen = sizeof(unsigned int), | |
115 | .mode = 0644, | |
116 | .proc_handler = proc_dointvec_minmax, | |
117 | .extra1 = &min_slot_table_size, | |
118 | .extra2 = &max_tcp_slot_table_limit | |
119 | }, | |
fbf76683 | 120 | { |
fbf76683 CL |
121 | .procname = "min_resvport", |
122 | .data = &xprt_min_resvport, | |
123 | .maxlen = sizeof(unsigned int), | |
124 | .mode = 0644, | |
6d456111 | 125 | .proc_handler = proc_dointvec_minmax, |
fbf76683 CL |
126 | .extra1 = &xprt_min_resvport_limit, |
127 | .extra2 = &xprt_max_resvport_limit | |
128 | }, | |
129 | { | |
fbf76683 CL |
130 | .procname = "max_resvport", |
131 | .data = &xprt_max_resvport, | |
132 | .maxlen = sizeof(unsigned int), | |
133 | .mode = 0644, | |
6d456111 | 134 | .proc_handler = proc_dointvec_minmax, |
fbf76683 CL |
135 | .extra1 = &xprt_min_resvport_limit, |
136 | .extra2 = &xprt_max_resvport_limit | |
137 | }, | |
25fe6142 TM |
138 | { |
139 | .procname = "tcp_fin_timeout", | |
140 | .data = &xs_tcp_fin_timeout, | |
141 | .maxlen = sizeof(xs_tcp_fin_timeout), | |
142 | .mode = 0644, | |
6d456111 | 143 | .proc_handler = proc_dointvec_jiffies, |
fbf76683 | 144 | }, |
f8572d8f | 145 | { }, |
fbf76683 CL |
146 | }; |
147 | ||
fe2c6338 | 148 | static struct ctl_table sunrpc_table[] = { |
fbf76683 | 149 | { |
fbf76683 CL |
150 | .procname = "sunrpc", |
151 | .mode = 0555, | |
152 | .child = xs_tunables_table | |
153 | }, | |
f8572d8f | 154 | { }, |
fbf76683 CL |
155 | }; |
156 | ||
157 | #endif | |
158 | ||
03bf4b70 CL |
159 | /* |
160 | * Wait duration for a reply from the RPC portmapper. | |
161 | */ | |
162 | #define XS_BIND_TO (60U * HZ) | |
163 | ||
164 | /* | |
165 | * Delay if a UDP socket connect error occurs. This is most likely some | |
166 | * kind of resource problem on the local host. | |
167 | */ | |
168 | #define XS_UDP_REEST_TO (2U * HZ) | |
169 | ||
170 | /* | |
171 | * The reestablish timeout allows clients to delay for a bit before attempting | |
172 | * to reconnect to a server that just dropped our connection. | |
173 | * | |
174 | * We implement an exponential backoff when trying to reestablish a TCP | |
175 | * transport connection with the server. Some servers like to drop a TCP | |
176 | * connection when they are overworked, so we start with a short timeout and | |
177 | * increase over time if the server is down or not responding. | |
178 | */ | |
179 | #define XS_TCP_INIT_REEST_TO (3U * HZ) | |
180 | #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ) | |
181 | ||
182 | /* | |
183 | * TCP idle timeout; client drops the transport socket if it is idle | |
184 | * for this long. Note that we also timeout UDP sockets to prevent | |
185 | * holding port numbers when there is no RPC traffic. | |
186 | */ | |
187 | #define XS_IDLE_DISC_TO (5U * 60 * HZ) | |
188 | ||
a246b010 CL |
189 | #ifdef RPC_DEBUG |
190 | # undef RPC_DEBUG_DATA | |
9903cd1c | 191 | # define RPCDBG_FACILITY RPCDBG_TRANS |
a246b010 CL |
192 | #endif |
193 | ||
a246b010 | 194 | #ifdef RPC_DEBUG_DATA |
9903cd1c | 195 | static void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 | 196 | { |
9903cd1c CL |
197 | u8 *buf = (u8 *) packet; |
198 | int j; | |
a246b010 | 199 | |
46121cf7 | 200 | dprintk("RPC: %s\n", msg); |
a246b010 CL |
201 | for (j = 0; j < count && j < 128; j += 4) { |
202 | if (!(j & 31)) { | |
203 | if (j) | |
204 | dprintk("\n"); | |
205 | dprintk("0x%04x ", j); | |
206 | } | |
207 | dprintk("%02x%02x%02x%02x ", | |
208 | buf[j], buf[j+1], buf[j+2], buf[j+3]); | |
209 | } | |
210 | dprintk("\n"); | |
211 | } | |
212 | #else | |
9903cd1c | 213 | static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) |
a246b010 CL |
214 | { |
215 | /* NOP */ | |
216 | } | |
217 | #endif | |
218 | ||
ffc2e518 CL |
219 | struct sock_xprt { |
220 | struct rpc_xprt xprt; | |
ee0ac0c2 CL |
221 | |
222 | /* | |
223 | * Network layer | |
224 | */ | |
225 | struct socket * sock; | |
226 | struct sock * inet; | |
51971139 CL |
227 | |
228 | /* | |
229 | * State of TCP reply receive | |
230 | */ | |
231 | __be32 tcp_fraghdr, | |
b76ce561 TM |
232 | tcp_xid, |
233 | tcp_calldir; | |
51971139 CL |
234 | |
235 | u32 tcp_offset, | |
236 | tcp_reclen; | |
237 | ||
238 | unsigned long tcp_copied, | |
239 | tcp_flags; | |
c8475461 CL |
240 | |
241 | /* | |
242 | * Connection of transports | |
243 | */ | |
34161db6 | 244 | struct delayed_work connect_worker; |
fbfffbd5 CL |
245 | struct sockaddr_storage srcaddr; |
246 | unsigned short srcport; | |
7c6e066e CL |
247 | |
248 | /* | |
249 | * UDP socket buffer size parameters | |
250 | */ | |
251 | size_t rcvsize, | |
252 | sndsize; | |
314dfd79 CL |
253 | |
254 | /* | |
255 | * Saved socket callback addresses | |
256 | */ | |
676d2369 | 257 | void (*old_data_ready)(struct sock *); |
314dfd79 CL |
258 | void (*old_state_change)(struct sock *); |
259 | void (*old_write_space)(struct sock *); | |
2118071d | 260 | void (*old_error_report)(struct sock *); |
ffc2e518 CL |
261 | }; |
262 | ||
e136d092 CL |
263 | /* |
264 | * TCP receive state flags | |
265 | */ | |
266 | #define TCP_RCV_LAST_FRAG (1UL << 0) | |
267 | #define TCP_RCV_COPY_FRAGHDR (1UL << 1) | |
268 | #define TCP_RCV_COPY_XID (1UL << 2) | |
269 | #define TCP_RCV_COPY_DATA (1UL << 3) | |
f4a2e418 RL |
270 | #define TCP_RCV_READ_CALLDIR (1UL << 4) |
271 | #define TCP_RCV_COPY_CALLDIR (1UL << 5) | |
18dca02a RL |
272 | |
273 | /* | |
274 | * TCP RPC flags | |
275 | */ | |
f4a2e418 | 276 | #define TCP_RPC_REPLY (1UL << 6) |
e136d092 | 277 | |
2118071d TM |
278 | static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) |
279 | { | |
280 | return (struct rpc_xprt *) sk->sk_user_data; | |
281 | } | |
282 | ||
95392c59 CL |
283 | static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) |
284 | { | |
285 | return (struct sockaddr *) &xprt->addr; | |
286 | } | |
287 | ||
176e21ee CL |
288 | static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) |
289 | { | |
290 | return (struct sockaddr_un *) &xprt->addr; | |
291 | } | |
292 | ||
95392c59 | 293 | static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) |
edb267a6 | 294 | { |
95392c59 CL |
295 | return (struct sockaddr_in *) &xprt->addr; |
296 | } | |
297 | ||
298 | static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) | |
299 | { | |
300 | return (struct sockaddr_in6 *) &xprt->addr; | |
301 | } | |
302 | ||
c877b849 | 303 | static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) |
edb267a6 | 304 | { |
c877b849 | 305 | struct sockaddr *sap = xs_addr(xprt); |
9dc3b095 CL |
306 | struct sockaddr_in6 *sin6; |
307 | struct sockaddr_in *sin; | |
176e21ee | 308 | struct sockaddr_un *sun; |
c877b849 | 309 | char buf[128]; |
edb267a6 | 310 | |
9dc3b095 | 311 | switch (sap->sa_family) { |
176e21ee CL |
312 | case AF_LOCAL: |
313 | sun = xs_addr_un(xprt); | |
314 | strlcpy(buf, sun->sun_path, sizeof(buf)); | |
315 | xprt->address_strings[RPC_DISPLAY_ADDR] = | |
316 | kstrdup(buf, GFP_KERNEL); | |
317 | break; | |
9dc3b095 | 318 | case AF_INET: |
176e21ee CL |
319 | (void)rpc_ntop(sap, buf, sizeof(buf)); |
320 | xprt->address_strings[RPC_DISPLAY_ADDR] = | |
321 | kstrdup(buf, GFP_KERNEL); | |
9dc3b095 | 322 | sin = xs_addr_in(xprt); |
fc0b5791 | 323 | snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); |
9dc3b095 CL |
324 | break; |
325 | case AF_INET6: | |
176e21ee CL |
326 | (void)rpc_ntop(sap, buf, sizeof(buf)); |
327 | xprt->address_strings[RPC_DISPLAY_ADDR] = | |
328 | kstrdup(buf, GFP_KERNEL); | |
9dc3b095 | 329 | sin6 = xs_addr_in6(xprt); |
fc0b5791 | 330 | snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); |
9dc3b095 CL |
331 | break; |
332 | default: | |
333 | BUG(); | |
756805e7 | 334 | } |
176e21ee | 335 | |
9dc3b095 | 336 | xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); |
edb267a6 CL |
337 | } |
338 | ||
9dc3b095 | 339 | static void xs_format_common_peer_ports(struct rpc_xprt *xprt) |
4b6473fb | 340 | { |
9dc3b095 CL |
341 | struct sockaddr *sap = xs_addr(xprt); |
342 | char buf[128]; | |
4b6473fb | 343 | |
81160e66 | 344 | snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); |
c877b849 | 345 | xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); |
4b6473fb | 346 | |
81160e66 | 347 | snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); |
c877b849 CL |
348 | xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); |
349 | } | |
4b6473fb | 350 | |
9dc3b095 CL |
351 | static void xs_format_peer_addresses(struct rpc_xprt *xprt, |
352 | const char *protocol, | |
353 | const char *netid) | |
c877b849 | 354 | { |
b454ae90 | 355 | xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; |
b454ae90 | 356 | xprt->address_strings[RPC_DISPLAY_NETID] = netid; |
c877b849 | 357 | xs_format_common_peer_addresses(xprt); |
9dc3b095 | 358 | xs_format_common_peer_ports(xprt); |
edb267a6 | 359 | } |
4b6473fb | 360 | |
9dc3b095 | 361 | static void xs_update_peer_port(struct rpc_xprt *xprt) |
4b6473fb | 362 | { |
9dc3b095 CL |
363 | kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); |
364 | kfree(xprt->address_strings[RPC_DISPLAY_PORT]); | |
4417c8c4 | 365 | |
9dc3b095 | 366 | xs_format_common_peer_ports(xprt); |
edb267a6 CL |
367 | } |
368 | ||
369 | static void xs_free_peer_addresses(struct rpc_xprt *xprt) | |
370 | { | |
33e01dc7 CL |
371 | unsigned int i; |
372 | ||
373 | for (i = 0; i < RPC_DISPLAY_MAX; i++) | |
374 | switch (i) { | |
375 | case RPC_DISPLAY_PROTO: | |
376 | case RPC_DISPLAY_NETID: | |
377 | continue; | |
378 | default: | |
379 | kfree(xprt->address_strings[i]); | |
380 | } | |
edb267a6 CL |
381 | } |
382 | ||
b4b5cc85 CL |
383 | #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) |
384 | ||
24c5684b | 385 | static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) |
b4b5cc85 | 386 | { |
b4b5cc85 CL |
387 | struct msghdr msg = { |
388 | .msg_name = addr, | |
389 | .msg_namelen = addrlen, | |
24c5684b TM |
390 | .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), |
391 | }; | |
392 | struct kvec iov = { | |
393 | .iov_base = vec->iov_base + base, | |
394 | .iov_len = vec->iov_len - base, | |
b4b5cc85 CL |
395 | }; |
396 | ||
24c5684b | 397 | if (iov.iov_len != 0) |
b4b5cc85 CL |
398 | return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); |
399 | return kernel_sendmsg(sock, &msg, NULL, 0, 0); | |
400 | } | |
401 | ||
f279cd00 | 402 | static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p) |
b4b5cc85 | 403 | { |
a6b31d18 TM |
404 | ssize_t (*do_sendpage)(struct socket *sock, struct page *page, |
405 | int offset, size_t size, int flags); | |
24c5684b TM |
406 | struct page **ppage; |
407 | unsigned int remainder; | |
f279cd00 | 408 | int err; |
24c5684b TM |
409 | |
410 | remainder = xdr->page_len - base; | |
411 | base += xdr->page_base; | |
412 | ppage = xdr->pages + (base >> PAGE_SHIFT); | |
413 | base &= ~PAGE_MASK; | |
a6b31d18 TM |
414 | do_sendpage = sock->ops->sendpage; |
415 | if (!zerocopy) | |
416 | do_sendpage = sock_no_sendpage; | |
24c5684b TM |
417 | for(;;) { |
418 | unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); | |
419 | int flags = XS_SENDMSG_FLAGS; | |
b4b5cc85 | 420 | |
24c5684b TM |
421 | remainder -= len; |
422 | if (remainder != 0 || more) | |
423 | flags |= MSG_MORE; | |
a6b31d18 | 424 | err = do_sendpage(sock, *ppage, base, len, flags); |
24c5684b TM |
425 | if (remainder == 0 || err != len) |
426 | break; | |
f279cd00 | 427 | *sent_p += err; |
24c5684b TM |
428 | ppage++; |
429 | base = 0; | |
430 | } | |
f279cd00 JB |
431 | if (err > 0) { |
432 | *sent_p += err; | |
433 | err = 0; | |
434 | } | |
435 | return err; | |
b4b5cc85 CL |
436 | } |
437 | ||
9903cd1c CL |
438 | /** |
439 | * xs_sendpages - write pages directly to a socket | |
440 | * @sock: socket to send on | |
441 | * @addr: UDP only -- address of destination | |
442 | * @addrlen: UDP only -- length of destination address | |
443 | * @xdr: buffer containing this request | |
444 | * @base: starting position in the buffer | |
a6b31d18 | 445 | * @zerocopy: true if it is safe to use sendpage() |
f279cd00 | 446 | * @sent_p: return the total number of bytes successfully queued for sending |
9903cd1c | 447 | * |
a246b010 | 448 | */ |
f279cd00 | 449 | static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p) |
a246b010 | 450 | { |
24c5684b | 451 | unsigned int remainder = xdr->len - base; |
f279cd00 JB |
452 | int err = 0; |
453 | int sent = 0; | |
a246b010 | 454 | |
262965f5 | 455 | if (unlikely(!sock)) |
fba91afb | 456 | return -ENOTSOCK; |
262965f5 CL |
457 | |
458 | clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); | |
24c5684b TM |
459 | if (base != 0) { |
460 | addr = NULL; | |
461 | addrlen = 0; | |
462 | } | |
262965f5 | 463 | |
24c5684b TM |
464 | if (base < xdr->head[0].iov_len || addr != NULL) { |
465 | unsigned int len = xdr->head[0].iov_len - base; | |
466 | remainder -= len; | |
467 | err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); | |
468 | if (remainder == 0 || err != len) | |
a246b010 | 469 | goto out; |
f279cd00 | 470 | *sent_p += err; |
a246b010 CL |
471 | base = 0; |
472 | } else | |
24c5684b | 473 | base -= xdr->head[0].iov_len; |
a246b010 | 474 | |
24c5684b TM |
475 | if (base < xdr->page_len) { |
476 | unsigned int len = xdr->page_len - base; | |
477 | remainder -= len; | |
f279cd00 JB |
478 | err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent); |
479 | *sent_p += sent; | |
480 | if (remainder == 0 || sent != len) | |
a246b010 CL |
481 | goto out; |
482 | base = 0; | |
24c5684b TM |
483 | } else |
484 | base -= xdr->page_len; | |
485 | ||
486 | if (base >= xdr->tail[0].iov_len) | |
f279cd00 | 487 | return 0; |
24c5684b | 488 | err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); |
a246b010 | 489 | out: |
f279cd00 JB |
490 | if (err > 0) { |
491 | *sent_p += err; | |
492 | err = 0; | |
493 | } | |
494 | return err; | |
a246b010 CL |
495 | } |
496 | ||
b6ddf64f TM |
497 | static void xs_nospace_callback(struct rpc_task *task) |
498 | { | |
499 | struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt); | |
500 | ||
501 | transport->inet->sk_write_pending--; | |
502 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); | |
503 | } | |
504 | ||
9903cd1c | 505 | /** |
262965f5 CL |
506 | * xs_nospace - place task on wait queue if transmit was incomplete |
507 | * @task: task to put to sleep | |
9903cd1c | 508 | * |
a246b010 | 509 | */ |
5e3771ce | 510 | static int xs_nospace(struct rpc_task *task) |
a246b010 | 511 | { |
262965f5 CL |
512 | struct rpc_rqst *req = task->tk_rqstp; |
513 | struct rpc_xprt *xprt = req->rq_xprt; | |
ee0ac0c2 | 514 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
06ea0bfe | 515 | struct sock *sk = transport->inet; |
24ca9a84 | 516 | int ret = -EAGAIN; |
a246b010 | 517 | |
46121cf7 | 518 | dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", |
262965f5 CL |
519 | task->tk_pid, req->rq_slen - req->rq_bytes_sent, |
520 | req->rq_slen); | |
521 | ||
b6ddf64f TM |
522 | /* Protect against races with write_space */ |
523 | spin_lock_bh(&xprt->transport_lock); | |
524 | ||
525 | /* Don't race with disconnect */ | |
526 | if (xprt_connected(xprt)) { | |
527 | if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) { | |
528 | /* | |
529 | * Notify TCP that we're limited by the application | |
530 | * window size | |
531 | */ | |
532 | set_bit(SOCK_NOSPACE, &transport->sock->flags); | |
06ea0bfe | 533 | sk->sk_write_pending++; |
b6ddf64f TM |
534 | /* ...and wait for more buffer space */ |
535 | xprt_wait_for_buffer_space(task, xs_nospace_callback); | |
536 | } | |
537 | } else { | |
538 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); | |
5e3771ce | 539 | ret = -ENOTCONN; |
b6ddf64f | 540 | } |
262965f5 | 541 | |
b6ddf64f | 542 | spin_unlock_bh(&xprt->transport_lock); |
06ea0bfe TM |
543 | |
544 | /* Race breaker in case memory is freed before above code is called */ | |
545 | sk->sk_write_space(sk); | |
5e3771ce | 546 | return ret; |
262965f5 CL |
547 | } |
548 | ||
61677eee CL |
549 | /* |
550 | * Construct a stream transport record marker in @buf. | |
551 | */ | |
552 | static inline void xs_encode_stream_record_marker(struct xdr_buf *buf) | |
553 | { | |
554 | u32 reclen = buf->len - sizeof(rpc_fraghdr); | |
555 | rpc_fraghdr *base = buf->head[0].iov_base; | |
556 | *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen); | |
557 | } | |
558 | ||
176e21ee CL |
559 | /** |
560 | * xs_local_send_request - write an RPC request to an AF_LOCAL socket | |
561 | * @task: RPC task that manages the state of an RPC request | |
562 | * | |
563 | * Return values: | |
564 | * 0: The request has been sent | |
565 | * EAGAIN: The socket was blocked, please call again later to | |
566 | * complete the request | |
567 | * ENOTCONN: Caller needs to invoke connect logic then call again | |
568 | * other: Some other error occured, the request was not sent | |
569 | */ | |
570 | static int xs_local_send_request(struct rpc_task *task) | |
571 | { | |
572 | struct rpc_rqst *req = task->tk_rqstp; | |
573 | struct rpc_xprt *xprt = req->rq_xprt; | |
574 | struct sock_xprt *transport = | |
575 | container_of(xprt, struct sock_xprt, xprt); | |
576 | struct xdr_buf *xdr = &req->rq_snd_buf; | |
577 | int status; | |
f279cd00 | 578 | int sent = 0; |
176e21ee CL |
579 | |
580 | xs_encode_stream_record_marker(&req->rq_snd_buf); | |
581 | ||
582 | xs_pktdump("packet data:", | |
583 | req->rq_svec->iov_base, req->rq_svec->iov_len); | |
584 | ||
f279cd00 JB |
585 | status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent, |
586 | true, &sent); | |
176e21ee CL |
587 | dprintk("RPC: %s(%u) = %d\n", |
588 | __func__, xdr->len - req->rq_bytes_sent, status); | |
f279cd00 JB |
589 | if (likely(sent > 0) || status == 0) { |
590 | req->rq_bytes_sent += sent; | |
591 | req->rq_xmit_bytes_sent += sent; | |
176e21ee CL |
592 | if (likely(req->rq_bytes_sent >= req->rq_slen)) { |
593 | req->rq_bytes_sent = 0; | |
594 | return 0; | |
595 | } | |
596 | status = -EAGAIN; | |
597 | } | |
598 | ||
599 | switch (status) { | |
3601c4a9 | 600 | case -ENOBUFS: |
176e21ee CL |
601 | case -EAGAIN: |
602 | status = xs_nospace(task); | |
603 | break; | |
604 | default: | |
605 | dprintk("RPC: sendmsg returned unrecognized error %d\n", | |
606 | -status); | |
607 | case -EPIPE: | |
608 | xs_close(xprt); | |
609 | status = -ENOTCONN; | |
610 | } | |
611 | ||
612 | return status; | |
613 | } | |
614 | ||
262965f5 CL |
615 | /** |
616 | * xs_udp_send_request - write an RPC request to a UDP socket | |
617 | * @task: address of RPC task that manages the state of an RPC request | |
618 | * | |
619 | * Return values: | |
620 | * 0: The request has been sent | |
621 | * EAGAIN: The socket was blocked, please call again later to | |
622 | * complete the request | |
623 | * ENOTCONN: Caller needs to invoke connect logic then call again | |
25985edc | 624 | * other: Some other error occurred, the request was not sent |
262965f5 CL |
625 | */ |
626 | static int xs_udp_send_request(struct rpc_task *task) | |
627 | { | |
628 | struct rpc_rqst *req = task->tk_rqstp; | |
629 | struct rpc_xprt *xprt = req->rq_xprt; | |
ee0ac0c2 | 630 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
262965f5 | 631 | struct xdr_buf *xdr = &req->rq_snd_buf; |
f279cd00 | 632 | int sent = 0; |
262965f5 | 633 | int status; |
a246b010 | 634 | |
9903cd1c | 635 | xs_pktdump("packet data:", |
a246b010 CL |
636 | req->rq_svec->iov_base, |
637 | req->rq_svec->iov_len); | |
638 | ||
01d37c42 TM |
639 | if (!xprt_bound(xprt)) |
640 | return -ENOTCONN; | |
f279cd00 JB |
641 | status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen, |
642 | xdr, req->rq_bytes_sent, true, &sent); | |
a246b010 | 643 | |
46121cf7 | 644 | dprintk("RPC: xs_udp_send_request(%u) = %d\n", |
262965f5 | 645 | xdr->len - req->rq_bytes_sent, status); |
a246b010 | 646 | |
f279cd00 JB |
647 | if (sent > 0 || status == 0) { |
648 | req->rq_xmit_bytes_sent += sent; | |
649 | if (sent >= req->rq_slen) | |
2199700f TM |
650 | return 0; |
651 | /* Still some bytes left; set up for a retry later. */ | |
262965f5 | 652 | status = -EAGAIN; |
2199700f | 653 | } |
a246b010 | 654 | |
262965f5 | 655 | switch (status) { |
fba91afb TM |
656 | case -ENOTSOCK: |
657 | status = -ENOTCONN; | |
658 | /* Should we call xs_close() here? */ | |
659 | break; | |
b6ddf64f | 660 | case -EAGAIN: |
5e3771ce | 661 | status = xs_nospace(task); |
b6ddf64f | 662 | break; |
c8485e4d TM |
663 | default: |
664 | dprintk("RPC: sendmsg returned unrecognized error %d\n", | |
665 | -status); | |
262965f5 | 666 | case -ENETUNREACH: |
3601c4a9 | 667 | case -ENOBUFS: |
262965f5 | 668 | case -EPIPE: |
a246b010 CL |
669 | case -ECONNREFUSED: |
670 | /* When the server has died, an ICMP port unreachable message | |
9903cd1c | 671 | * prompts ECONNREFUSED. */ |
b6ddf64f | 672 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); |
a246b010 | 673 | } |
5fe46e9d | 674 | |
262965f5 | 675 | return status; |
a246b010 CL |
676 | } |
677 | ||
e06799f9 TM |
678 | /** |
679 | * xs_tcp_shutdown - gracefully shut down a TCP socket | |
680 | * @xprt: transport | |
681 | * | |
682 | * Initiates a graceful shutdown of the TCP socket by calling the | |
683 | * equivalent of shutdown(SHUT_WR); | |
684 | */ | |
685 | static void xs_tcp_shutdown(struct rpc_xprt *xprt) | |
686 | { | |
687 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); | |
688 | struct socket *sock = transport->sock; | |
689 | ||
40b5ea0c | 690 | if (sock != NULL) { |
e06799f9 | 691 | kernel_sock_shutdown(sock, SHUT_WR); |
40b5ea0c TM |
692 | trace_rpc_socket_shutdown(xprt, sock); |
693 | } | |
e06799f9 TM |
694 | } |
695 | ||
9903cd1c | 696 | /** |
262965f5 | 697 | * xs_tcp_send_request - write an RPC request to a TCP socket |
9903cd1c CL |
698 | * @task: address of RPC task that manages the state of an RPC request |
699 | * | |
700 | * Return values: | |
262965f5 CL |
701 | * 0: The request has been sent |
702 | * EAGAIN: The socket was blocked, please call again later to | |
703 | * complete the request | |
704 | * ENOTCONN: Caller needs to invoke connect logic then call again | |
25985edc | 705 | * other: Some other error occurred, the request was not sent |
9903cd1c CL |
706 | * |
707 | * XXX: In the case of soft timeouts, should we eventually give up | |
262965f5 | 708 | * if sendmsg is not able to make progress? |
9903cd1c | 709 | */ |
262965f5 | 710 | static int xs_tcp_send_request(struct rpc_task *task) |
a246b010 CL |
711 | { |
712 | struct rpc_rqst *req = task->tk_rqstp; | |
713 | struct rpc_xprt *xprt = req->rq_xprt; | |
ee0ac0c2 | 714 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
262965f5 | 715 | struct xdr_buf *xdr = &req->rq_snd_buf; |
a6b31d18 | 716 | bool zerocopy = true; |
b595bb15 | 717 | int status; |
f279cd00 | 718 | int sent; |
a246b010 | 719 | |
61677eee | 720 | xs_encode_stream_record_marker(&req->rq_snd_buf); |
a246b010 | 721 | |
262965f5 CL |
722 | xs_pktdump("packet data:", |
723 | req->rq_svec->iov_base, | |
724 | req->rq_svec->iov_len); | |
a6b31d18 TM |
725 | /* Don't use zero copy if this is a resend. If the RPC call |
726 | * completes while the socket holds a reference to the pages, | |
727 | * then we may end up resending corrupted data. | |
728 | */ | |
729 | if (task->tk_flags & RPC_TASK_SENT) | |
730 | zerocopy = false; | |
a246b010 CL |
731 | |
732 | /* Continue transmitting the packet/record. We must be careful | |
733 | * to cope with writespace callbacks arriving _after_ we have | |
262965f5 | 734 | * called sendmsg(). */ |
a246b010 | 735 | while (1) { |
f279cd00 JB |
736 | sent = 0; |
737 | status = xs_sendpages(transport->sock, NULL, 0, xdr, | |
738 | req->rq_bytes_sent, zerocopy, &sent); | |
a246b010 | 739 | |
46121cf7 | 740 | dprintk("RPC: xs_tcp_send_request(%u) = %d\n", |
262965f5 | 741 | xdr->len - req->rq_bytes_sent, status); |
a246b010 | 742 | |
f279cd00 | 743 | if (unlikely(sent == 0 && status < 0)) |
a246b010 | 744 | break; |
a246b010 | 745 | |
262965f5 CL |
746 | /* If we've sent the entire packet, immediately |
747 | * reset the count of bytes sent. */ | |
f279cd00 JB |
748 | req->rq_bytes_sent += sent; |
749 | req->rq_xmit_bytes_sent += sent; | |
262965f5 CL |
750 | if (likely(req->rq_bytes_sent >= req->rq_slen)) { |
751 | req->rq_bytes_sent = 0; | |
752 | return 0; | |
753 | } | |
a246b010 | 754 | |
f279cd00 | 755 | if (sent != 0) |
06b4b681 | 756 | continue; |
a246b010 | 757 | status = -EAGAIN; |
06b4b681 | 758 | break; |
a246b010 CL |
759 | } |
760 | ||
262965f5 | 761 | switch (status) { |
fba91afb TM |
762 | case -ENOTSOCK: |
763 | status = -ENOTCONN; | |
764 | /* Should we call xs_close() here? */ | |
765 | break; | |
3601c4a9 | 766 | case -ENOBUFS: |
262965f5 | 767 | case -EAGAIN: |
5e3771ce | 768 | status = xs_nospace(task); |
262965f5 | 769 | break; |
c8485e4d TM |
770 | default: |
771 | dprintk("RPC: sendmsg returned unrecognized error %d\n", | |
772 | -status); | |
262965f5 | 773 | case -ECONNRESET: |
2a9e1cfa TM |
774 | xs_tcp_shutdown(xprt); |
775 | case -ECONNREFUSED: | |
262965f5 | 776 | case -ENOTCONN: |
b9d2bb2e | 777 | case -EPIPE: |
b6ddf64f | 778 | clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); |
a246b010 | 779 | } |
5fe46e9d | 780 | |
a246b010 CL |
781 | return status; |
782 | } | |
783 | ||
e0ab53de TM |
784 | /** |
785 | * xs_tcp_release_xprt - clean up after a tcp transmission | |
786 | * @xprt: transport | |
787 | * @task: rpc task | |
788 | * | |
789 | * This cleans up if an error causes us to abort the transmission of a request. | |
790 | * In this case, the socket may need to be reset in order to avoid confusing | |
791 | * the server. | |
792 | */ | |
793 | static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) | |
794 | { | |
795 | struct rpc_rqst *req; | |
796 | ||
797 | if (task != xprt->snd_task) | |
798 | return; | |
799 | if (task == NULL) | |
800 | goto out_release; | |
801 | req = task->tk_rqstp; | |
43cedbf0 TM |
802 | if (req == NULL) |
803 | goto out_release; | |
e0ab53de TM |
804 | if (req->rq_bytes_sent == 0) |
805 | goto out_release; | |
806 | if (req->rq_bytes_sent == req->rq_snd_buf.len) | |
807 | goto out_release; | |
a4f0835c | 808 | set_bit(XPRT_CLOSE_WAIT, &xprt->state); |
e0ab53de TM |
809 | out_release: |
810 | xprt_release_xprt(xprt, task); | |
811 | } | |
812 | ||
2a9e1cfa TM |
813 | static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) |
814 | { | |
815 | transport->old_data_ready = sk->sk_data_ready; | |
816 | transport->old_state_change = sk->sk_state_change; | |
817 | transport->old_write_space = sk->sk_write_space; | |
2118071d | 818 | transport->old_error_report = sk->sk_error_report; |
2a9e1cfa TM |
819 | } |
820 | ||
821 | static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) | |
822 | { | |
823 | sk->sk_data_ready = transport->old_data_ready; | |
824 | sk->sk_state_change = transport->old_state_change; | |
825 | sk->sk_write_space = transport->old_write_space; | |
2118071d TM |
826 | sk->sk_error_report = transport->old_error_report; |
827 | } | |
828 | ||
829 | /** | |
830 | * xs_error_report - callback to handle TCP socket state errors | |
831 | * @sk: socket | |
832 | * | |
833 | * Note: we don't call sock_error() since there may be a rpc_task | |
834 | * using the socket, and so we don't want to clear sk->sk_err. | |
835 | */ | |
836 | static void xs_error_report(struct sock *sk) | |
837 | { | |
838 | struct rpc_xprt *xprt; | |
839 | int err; | |
840 | ||
841 | read_lock_bh(&sk->sk_callback_lock); | |
842 | if (!(xprt = xprt_from_sock(sk))) | |
843 | goto out; | |
844 | ||
845 | err = -sk->sk_err; | |
846 | if (err == 0) | |
847 | goto out; | |
848 | dprintk("RPC: xs_error_report client %p, error=%d...\n", | |
849 | xprt, -err); | |
e8353c76 | 850 | trace_rpc_socket_error(xprt, sk->sk_socket, err); |
a743419f BC |
851 | if (test_bit(XPRT_CONNECTION_REUSE, &xprt->state)) |
852 | goto out; | |
2118071d TM |
853 | xprt_wake_pending_tasks(xprt, err); |
854 | out: | |
855 | read_unlock_bh(&sk->sk_callback_lock); | |
2a9e1cfa TM |
856 | } |
857 | ||
fe315e76 | 858 | static void xs_reset_transport(struct sock_xprt *transport) |
a246b010 | 859 | { |
ee0ac0c2 CL |
860 | struct socket *sock = transport->sock; |
861 | struct sock *sk = transport->inet; | |
a246b010 | 862 | |
fe315e76 CL |
863 | if (sk == NULL) |
864 | return; | |
9903cd1c | 865 | |
246408dc TM |
866 | transport->srcport = 0; |
867 | ||
a246b010 | 868 | write_lock_bh(&sk->sk_callback_lock); |
ee0ac0c2 CL |
869 | transport->inet = NULL; |
870 | transport->sock = NULL; | |
a246b010 | 871 | |
9903cd1c | 872 | sk->sk_user_data = NULL; |
2a9e1cfa TM |
873 | |
874 | xs_restore_old_callbacks(transport, sk); | |
a246b010 CL |
875 | write_unlock_bh(&sk->sk_callback_lock); |
876 | ||
40b5ea0c | 877 | trace_rpc_socket_close(&transport->xprt, sock); |
a246b010 | 878 | sock_release(sock); |
fe315e76 CL |
879 | } |
880 | ||
881 | /** | |
882 | * xs_close - close a socket | |
883 | * @xprt: transport | |
884 | * | |
885 | * This is used when all requests are complete; ie, no DRC state remains | |
886 | * on the server we want to save. | |
f75e6745 TM |
887 | * |
888 | * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with | |
889 | * xs_reset_transport() zeroing the socket from underneath a writer. | |
fe315e76 CL |
890 | */ |
891 | static void xs_close(struct rpc_xprt *xprt) | |
892 | { | |
893 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); | |
894 | ||
895 | dprintk("RPC: xs_close xprt %p\n", xprt); | |
896 | ||
93dc41bd N |
897 | cancel_delayed_work_sync(&transport->connect_worker); |
898 | ||
fe315e76 | 899 | xs_reset_transport(transport); |
61d0a8e6 | 900 | xprt->reestablish_timeout = 0; |
fe315e76 | 901 | |
4e857c58 | 902 | smp_mb__before_atomic(); |
7d1e8255 | 903 | clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); |
632e3bdc | 904 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); |
3b948ae5 | 905 | clear_bit(XPRT_CLOSING, &xprt->state); |
4e857c58 | 906 | smp_mb__after_atomic(); |
62da3b24 | 907 | xprt_disconnect_done(xprt); |
a246b010 CL |
908 | } |
909 | ||
f75e6745 TM |
910 | static void xs_tcp_close(struct rpc_xprt *xprt) |
911 | { | |
912 | if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state)) | |
913 | xs_close(xprt); | |
914 | else | |
915 | xs_tcp_shutdown(xprt); | |
916 | } | |
917 | ||
315f3812 KM |
918 | static void xs_xprt_free(struct rpc_xprt *xprt) |
919 | { | |
920 | xs_free_peer_addresses(xprt); | |
921 | xprt_free(xprt); | |
922 | } | |
923 | ||
9903cd1c CL |
924 | /** |
925 | * xs_destroy - prepare to shutdown a transport | |
926 | * @xprt: doomed transport | |
927 | * | |
928 | */ | |
929 | static void xs_destroy(struct rpc_xprt *xprt) | |
a246b010 | 930 | { |
46121cf7 | 931 | dprintk("RPC: xs_destroy xprt %p\n", xprt); |
9903cd1c | 932 | |
a1311d87 | 933 | xs_close(xprt); |
315f3812 | 934 | xs_xprt_free(xprt); |
a1311d87 | 935 | module_put(THIS_MODULE); |
a246b010 CL |
936 | } |
937 | ||
176e21ee CL |
938 | static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb) |
939 | { | |
940 | struct xdr_skb_reader desc = { | |
941 | .skb = skb, | |
942 | .offset = sizeof(rpc_fraghdr), | |
943 | .count = skb->len - sizeof(rpc_fraghdr), | |
944 | }; | |
945 | ||
946 | if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0) | |
947 | return -1; | |
948 | if (desc.count) | |
949 | return -1; | |
950 | return 0; | |
951 | } | |
952 | ||
953 | /** | |
954 | * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets | |
955 | * @sk: socket with data to read | |
956 | * @len: how much data to read | |
957 | * | |
958 | * Currently this assumes we can read the whole reply in a single gulp. | |
959 | */ | |
676d2369 | 960 | static void xs_local_data_ready(struct sock *sk) |
176e21ee CL |
961 | { |
962 | struct rpc_task *task; | |
963 | struct rpc_xprt *xprt; | |
964 | struct rpc_rqst *rovr; | |
965 | struct sk_buff *skb; | |
966 | int err, repsize, copied; | |
967 | u32 _xid; | |
968 | __be32 *xp; | |
969 | ||
970 | read_lock_bh(&sk->sk_callback_lock); | |
971 | dprintk("RPC: %s...\n", __func__); | |
972 | xprt = xprt_from_sock(sk); | |
973 | if (xprt == NULL) | |
974 | goto out; | |
975 | ||
976 | skb = skb_recv_datagram(sk, 0, 1, &err); | |
977 | if (skb == NULL) | |
978 | goto out; | |
979 | ||
176e21ee CL |
980 | repsize = skb->len - sizeof(rpc_fraghdr); |
981 | if (repsize < 4) { | |
982 | dprintk("RPC: impossible RPC reply size %d\n", repsize); | |
983 | goto dropit; | |
984 | } | |
985 | ||
986 | /* Copy the XID from the skb... */ | |
987 | xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid); | |
988 | if (xp == NULL) | |
989 | goto dropit; | |
990 | ||
991 | /* Look up and lock the request corresponding to the given XID */ | |
992 | spin_lock(&xprt->transport_lock); | |
993 | rovr = xprt_lookup_rqst(xprt, *xp); | |
994 | if (!rovr) | |
995 | goto out_unlock; | |
996 | task = rovr->rq_task; | |
997 | ||
998 | copied = rovr->rq_private_buf.buflen; | |
999 | if (copied > repsize) | |
1000 | copied = repsize; | |
1001 | ||
1002 | if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) { | |
1003 | dprintk("RPC: sk_buff copy failed\n"); | |
1004 | goto out_unlock; | |
1005 | } | |
1006 | ||
1007 | xprt_complete_rqst(task, copied); | |
1008 | ||
1009 | out_unlock: | |
1010 | spin_unlock(&xprt->transport_lock); | |
1011 | dropit: | |
1012 | skb_free_datagram(sk, skb); | |
1013 | out: | |
1014 | read_unlock_bh(&sk->sk_callback_lock); | |
1015 | } | |
1016 | ||
9903cd1c CL |
1017 | /** |
1018 | * xs_udp_data_ready - "data ready" callback for UDP sockets | |
1019 | * @sk: socket with data to read | |
1020 | * @len: how much data to read | |
1021 | * | |
a246b010 | 1022 | */ |
676d2369 | 1023 | static void xs_udp_data_ready(struct sock *sk) |
a246b010 | 1024 | { |
9903cd1c CL |
1025 | struct rpc_task *task; |
1026 | struct rpc_xprt *xprt; | |
a246b010 | 1027 | struct rpc_rqst *rovr; |
9903cd1c | 1028 | struct sk_buff *skb; |
a246b010 | 1029 | int err, repsize, copied; |
d8ed029d AD |
1030 | u32 _xid; |
1031 | __be32 *xp; | |
a246b010 | 1032 | |
f064af1e | 1033 | read_lock_bh(&sk->sk_callback_lock); |
46121cf7 | 1034 | dprintk("RPC: xs_udp_data_ready...\n"); |
9903cd1c | 1035 | if (!(xprt = xprt_from_sock(sk))) |
a246b010 | 1036 | goto out; |
a246b010 CL |
1037 | |
1038 | if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) | |
1039 | goto out; | |
1040 | ||
a246b010 CL |
1041 | repsize = skb->len - sizeof(struct udphdr); |
1042 | if (repsize < 4) { | |
46121cf7 | 1043 | dprintk("RPC: impossible RPC reply size %d!\n", repsize); |
a246b010 CL |
1044 | goto dropit; |
1045 | } | |
1046 | ||
1047 | /* Copy the XID from the skb... */ | |
1048 | xp = skb_header_pointer(skb, sizeof(struct udphdr), | |
1049 | sizeof(_xid), &_xid); | |
1050 | if (xp == NULL) | |
1051 | goto dropit; | |
1052 | ||
1053 | /* Look up and lock the request corresponding to the given XID */ | |
4a0f8c04 | 1054 | spin_lock(&xprt->transport_lock); |
a246b010 CL |
1055 | rovr = xprt_lookup_rqst(xprt, *xp); |
1056 | if (!rovr) | |
1057 | goto out_unlock; | |
1058 | task = rovr->rq_task; | |
1059 | ||
a246b010 CL |
1060 | if ((copied = rovr->rq_private_buf.buflen) > repsize) |
1061 | copied = repsize; | |
1062 | ||
1063 | /* Suck it into the iovec, verify checksum if not done by hw. */ | |
1781f7f5 HX |
1064 | if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { |
1065 | UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); | |
a246b010 | 1066 | goto out_unlock; |
1781f7f5 HX |
1067 | } |
1068 | ||
1069 | UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); | |
a246b010 | 1070 | |
6a24dfb6 | 1071 | xprt_adjust_cwnd(xprt, task, copied); |
1570c1e4 | 1072 | xprt_complete_rqst(task, copied); |
a246b010 CL |
1073 | |
1074 | out_unlock: | |
4a0f8c04 | 1075 | spin_unlock(&xprt->transport_lock); |
a246b010 CL |
1076 | dropit: |
1077 | skb_free_datagram(sk, skb); | |
1078 | out: | |
f064af1e | 1079 | read_unlock_bh(&sk->sk_callback_lock); |
a246b010 CL |
1080 | } |
1081 | ||
a519fc7a TM |
1082 | /* |
1083 | * Helper function to force a TCP close if the server is sending | |
1084 | * junk and/or it has put us in CLOSE_WAIT | |
1085 | */ | |
1086 | static void xs_tcp_force_close(struct rpc_xprt *xprt) | |
1087 | { | |
1088 | set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); | |
1089 | xprt_force_disconnect(xprt); | |
1090 | } | |
1091 | ||
dd456471 | 1092 | static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) |
a246b010 | 1093 | { |
51971139 | 1094 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 CL |
1095 | size_t len, used; |
1096 | char *p; | |
1097 | ||
51971139 CL |
1098 | p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; |
1099 | len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; | |
9d292316 | 1100 | used = xdr_skb_read_bits(desc, p, len); |
51971139 | 1101 | transport->tcp_offset += used; |
a246b010 CL |
1102 | if (used != len) |
1103 | return; | |
808012fb | 1104 | |
51971139 CL |
1105 | transport->tcp_reclen = ntohl(transport->tcp_fraghdr); |
1106 | if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) | |
e136d092 | 1107 | transport->tcp_flags |= TCP_RCV_LAST_FRAG; |
a246b010 | 1108 | else |
e136d092 | 1109 | transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; |
51971139 | 1110 | transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; |
808012fb | 1111 | |
e136d092 | 1112 | transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; |
51971139 | 1113 | transport->tcp_offset = 0; |
808012fb | 1114 | |
a246b010 | 1115 | /* Sanity check of the record length */ |
18dca02a | 1116 | if (unlikely(transport->tcp_reclen < 8)) { |
46121cf7 | 1117 | dprintk("RPC: invalid TCP record fragment length\n"); |
a519fc7a | 1118 | xs_tcp_force_close(xprt); |
9903cd1c | 1119 | return; |
a246b010 | 1120 | } |
46121cf7 | 1121 | dprintk("RPC: reading TCP record fragment of length %d\n", |
51971139 | 1122 | transport->tcp_reclen); |
a246b010 CL |
1123 | } |
1124 | ||
51971139 | 1125 | static void xs_tcp_check_fraghdr(struct sock_xprt *transport) |
a246b010 | 1126 | { |
51971139 | 1127 | if (transport->tcp_offset == transport->tcp_reclen) { |
e136d092 | 1128 | transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; |
51971139 | 1129 | transport->tcp_offset = 0; |
e136d092 CL |
1130 | if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { |
1131 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; | |
1132 | transport->tcp_flags |= TCP_RCV_COPY_XID; | |
51971139 | 1133 | transport->tcp_copied = 0; |
a246b010 CL |
1134 | } |
1135 | } | |
1136 | } | |
1137 | ||
dd456471 | 1138 | static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) |
a246b010 CL |
1139 | { |
1140 | size_t len, used; | |
1141 | char *p; | |
1142 | ||
51971139 | 1143 | len = sizeof(transport->tcp_xid) - transport->tcp_offset; |
46121cf7 | 1144 | dprintk("RPC: reading XID (%Zu bytes)\n", len); |
51971139 | 1145 | p = ((char *) &transport->tcp_xid) + transport->tcp_offset; |
9d292316 | 1146 | used = xdr_skb_read_bits(desc, p, len); |
51971139 | 1147 | transport->tcp_offset += used; |
a246b010 CL |
1148 | if (used != len) |
1149 | return; | |
e136d092 | 1150 | transport->tcp_flags &= ~TCP_RCV_COPY_XID; |
f4a2e418 | 1151 | transport->tcp_flags |= TCP_RCV_READ_CALLDIR; |
51971139 | 1152 | transport->tcp_copied = 4; |
18dca02a RL |
1153 | dprintk("RPC: reading %s XID %08x\n", |
1154 | (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" | |
1155 | : "request with", | |
51971139 CL |
1156 | ntohl(transport->tcp_xid)); |
1157 | xs_tcp_check_fraghdr(transport); | |
a246b010 CL |
1158 | } |
1159 | ||
18dca02a RL |
1160 | static inline void xs_tcp_read_calldir(struct sock_xprt *transport, |
1161 | struct xdr_skb_reader *desc) | |
a246b010 | 1162 | { |
18dca02a RL |
1163 | size_t len, used; |
1164 | u32 offset; | |
b76ce561 | 1165 | char *p; |
18dca02a RL |
1166 | |
1167 | /* | |
1168 | * We want transport->tcp_offset to be 8 at the end of this routine | |
1169 | * (4 bytes for the xid and 4 bytes for the call/reply flag). | |
1170 | * When this function is called for the first time, | |
1171 | * transport->tcp_offset is 4 (after having already read the xid). | |
1172 | */ | |
1173 | offset = transport->tcp_offset - sizeof(transport->tcp_xid); | |
b76ce561 | 1174 | len = sizeof(transport->tcp_calldir) - offset; |
18dca02a | 1175 | dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len); |
b76ce561 TM |
1176 | p = ((char *) &transport->tcp_calldir) + offset; |
1177 | used = xdr_skb_read_bits(desc, p, len); | |
18dca02a RL |
1178 | transport->tcp_offset += used; |
1179 | if (used != len) | |
1180 | return; | |
f4a2e418 | 1181 | transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; |
f4a2e418 RL |
1182 | /* |
1183 | * We don't yet have the XDR buffer, so we will write the calldir | |
1184 | * out after we get the buffer from the 'struct rpc_rqst' | |
1185 | */ | |
b76ce561 TM |
1186 | switch (ntohl(transport->tcp_calldir)) { |
1187 | case RPC_REPLY: | |
1188 | transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; | |
1189 | transport->tcp_flags |= TCP_RCV_COPY_DATA; | |
18dca02a | 1190 | transport->tcp_flags |= TCP_RPC_REPLY; |
b76ce561 TM |
1191 | break; |
1192 | case RPC_CALL: | |
1193 | transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; | |
1194 | transport->tcp_flags |= TCP_RCV_COPY_DATA; | |
18dca02a | 1195 | transport->tcp_flags &= ~TCP_RPC_REPLY; |
b76ce561 TM |
1196 | break; |
1197 | default: | |
1198 | dprintk("RPC: invalid request message type\n"); | |
a519fc7a | 1199 | xs_tcp_force_close(&transport->xprt); |
b76ce561 | 1200 | } |
18dca02a RL |
1201 | xs_tcp_check_fraghdr(transport); |
1202 | } | |
1203 | ||
44b98efd RL |
1204 | static inline void xs_tcp_read_common(struct rpc_xprt *xprt, |
1205 | struct xdr_skb_reader *desc, | |
1206 | struct rpc_rqst *req) | |
a246b010 | 1207 | { |
44b98efd RL |
1208 | struct sock_xprt *transport = |
1209 | container_of(xprt, struct sock_xprt, xprt); | |
a246b010 CL |
1210 | struct xdr_buf *rcvbuf; |
1211 | size_t len; | |
1212 | ssize_t r; | |
1213 | ||
a246b010 | 1214 | rcvbuf = &req->rq_private_buf; |
f4a2e418 RL |
1215 | |
1216 | if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { | |
1217 | /* | |
1218 | * Save the RPC direction in the XDR buffer | |
1219 | */ | |
f4a2e418 | 1220 | memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, |
b76ce561 TM |
1221 | &transport->tcp_calldir, |
1222 | sizeof(transport->tcp_calldir)); | |
1223 | transport->tcp_copied += sizeof(transport->tcp_calldir); | |
f4a2e418 | 1224 | transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; |
a246b010 CL |
1225 | } |
1226 | ||
a246b010 | 1227 | len = desc->count; |
51971139 | 1228 | if (len > transport->tcp_reclen - transport->tcp_offset) { |
dd456471 | 1229 | struct xdr_skb_reader my_desc; |
a246b010 | 1230 | |
51971139 | 1231 | len = transport->tcp_reclen - transport->tcp_offset; |
a246b010 CL |
1232 | memcpy(&my_desc, desc, sizeof(my_desc)); |
1233 | my_desc.count = len; | |
51971139 | 1234 | r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, |
9d292316 | 1235 | &my_desc, xdr_skb_read_bits); |
a246b010 CL |
1236 | desc->count -= r; |
1237 | desc->offset += r; | |
1238 | } else | |
51971139 | 1239 | r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, |
9d292316 | 1240 | desc, xdr_skb_read_bits); |
a246b010 CL |
1241 | |
1242 | if (r > 0) { | |
51971139 CL |
1243 | transport->tcp_copied += r; |
1244 | transport->tcp_offset += r; | |
a246b010 CL |
1245 | } |
1246 | if (r != len) { | |
1247 | /* Error when copying to the receive buffer, | |
1248 | * usually because we weren't able to allocate | |
1249 | * additional buffer pages. All we can do now | |
e136d092 | 1250 | * is turn off TCP_RCV_COPY_DATA, so the request |
a246b010 CL |
1251 | * will not receive any additional updates, |
1252 | * and time out. | |
1253 | * Any remaining data from this record will | |
1254 | * be discarded. | |
1255 | */ | |
e136d092 | 1256 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; |
46121cf7 | 1257 | dprintk("RPC: XID %08x truncated request\n", |
51971139 | 1258 | ntohl(transport->tcp_xid)); |
46121cf7 CL |
1259 | dprintk("RPC: xprt = %p, tcp_copied = %lu, " |
1260 | "tcp_offset = %u, tcp_reclen = %u\n", | |
1261 | xprt, transport->tcp_copied, | |
1262 | transport->tcp_offset, transport->tcp_reclen); | |
44b98efd | 1263 | return; |
a246b010 CL |
1264 | } |
1265 | ||
46121cf7 | 1266 | dprintk("RPC: XID %08x read %Zd bytes\n", |
51971139 | 1267 | ntohl(transport->tcp_xid), r); |
46121cf7 CL |
1268 | dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " |
1269 | "tcp_reclen = %u\n", xprt, transport->tcp_copied, | |
1270 | transport->tcp_offset, transport->tcp_reclen); | |
51971139 CL |
1271 | |
1272 | if (transport->tcp_copied == req->rq_private_buf.buflen) | |
e136d092 | 1273 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; |
51971139 | 1274 | else if (transport->tcp_offset == transport->tcp_reclen) { |
e136d092 CL |
1275 | if (transport->tcp_flags & TCP_RCV_LAST_FRAG) |
1276 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; | |
a246b010 | 1277 | } |
44b98efd RL |
1278 | } |
1279 | ||
1280 | /* | |
1281 | * Finds the request corresponding to the RPC xid and invokes the common | |
1282 | * tcp read code to read the data. | |
1283 | */ | |
1284 | static inline int xs_tcp_read_reply(struct rpc_xprt *xprt, | |
1285 | struct xdr_skb_reader *desc) | |
1286 | { | |
1287 | struct sock_xprt *transport = | |
1288 | container_of(xprt, struct sock_xprt, xprt); | |
1289 | struct rpc_rqst *req; | |
1290 | ||
1291 | dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid)); | |
1292 | ||
1293 | /* Find and lock the request corresponding to this xid */ | |
1294 | spin_lock(&xprt->transport_lock); | |
1295 | req = xprt_lookup_rqst(xprt, transport->tcp_xid); | |
1296 | if (!req) { | |
1297 | dprintk("RPC: XID %08x request not found!\n", | |
1298 | ntohl(transport->tcp_xid)); | |
1299 | spin_unlock(&xprt->transport_lock); | |
1300 | return -1; | |
1301 | } | |
1302 | ||
1303 | xs_tcp_read_common(xprt, desc, req); | |
1304 | ||
e136d092 | 1305 | if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) |
51971139 | 1306 | xprt_complete_rqst(req->rq_task, transport->tcp_copied); |
44b98efd | 1307 | |
4a0f8c04 | 1308 | spin_unlock(&xprt->transport_lock); |
44b98efd RL |
1309 | return 0; |
1310 | } | |
1311 | ||
9e00abc3 | 1312 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
44b98efd RL |
1313 | /* |
1314 | * Obtains an rpc_rqst previously allocated and invokes the common | |
1315 | * tcp read code to read the data. The result is placed in the callback | |
1316 | * queue. | |
1317 | * If we're unable to obtain the rpc_rqst we schedule the closing of the | |
1318 | * connection and return -1. | |
1319 | */ | |
2ea24497 | 1320 | static int xs_tcp_read_callback(struct rpc_xprt *xprt, |
44b98efd RL |
1321 | struct xdr_skb_reader *desc) |
1322 | { | |
1323 | struct sock_xprt *transport = | |
1324 | container_of(xprt, struct sock_xprt, xprt); | |
1325 | struct rpc_rqst *req; | |
1326 | ||
2ea24497 TM |
1327 | /* Look up and lock the request corresponding to the given XID */ |
1328 | spin_lock(&xprt->transport_lock); | |
1329 | req = xprt_lookup_bc_request(xprt, transport->tcp_xid); | |
44b98efd | 1330 | if (req == NULL) { |
2ea24497 | 1331 | spin_unlock(&xprt->transport_lock); |
44b98efd RL |
1332 | printk(KERN_WARNING "Callback slot table overflowed\n"); |
1333 | xprt_force_disconnect(xprt); | |
1334 | return -1; | |
1335 | } | |
1336 | ||
44b98efd RL |
1337 | dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid)); |
1338 | xs_tcp_read_common(xprt, desc, req); | |
1339 | ||
2ea24497 TM |
1340 | if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) |
1341 | xprt_complete_bc_request(req, transport->tcp_copied); | |
1342 | spin_unlock(&xprt->transport_lock); | |
44b98efd RL |
1343 | |
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, | |
1348 | struct xdr_skb_reader *desc) | |
1349 | { | |
1350 | struct sock_xprt *transport = | |
1351 | container_of(xprt, struct sock_xprt, xprt); | |
1352 | ||
1353 | return (transport->tcp_flags & TCP_RPC_REPLY) ? | |
1354 | xs_tcp_read_reply(xprt, desc) : | |
1355 | xs_tcp_read_callback(xprt, desc); | |
1356 | } | |
1357 | #else | |
1358 | static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, | |
1359 | struct xdr_skb_reader *desc) | |
1360 | { | |
1361 | return xs_tcp_read_reply(xprt, desc); | |
1362 | } | |
9e00abc3 | 1363 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
44b98efd RL |
1364 | |
1365 | /* | |
1366 | * Read data off the transport. This can be either an RPC_CALL or an | |
1367 | * RPC_REPLY. Relay the processing to helper functions. | |
1368 | */ | |
1369 | static void xs_tcp_read_data(struct rpc_xprt *xprt, | |
1370 | struct xdr_skb_reader *desc) | |
1371 | { | |
1372 | struct sock_xprt *transport = | |
1373 | container_of(xprt, struct sock_xprt, xprt); | |
1374 | ||
1375 | if (_xs_tcp_read_data(xprt, desc) == 0) | |
1376 | xs_tcp_check_fraghdr(transport); | |
1377 | else { | |
1378 | /* | |
1379 | * The transport_lock protects the request handling. | |
1380 | * There's no need to hold it to update the tcp_flags. | |
1381 | */ | |
1382 | transport->tcp_flags &= ~TCP_RCV_COPY_DATA; | |
1383 | } | |
a246b010 CL |
1384 | } |
1385 | ||
dd456471 | 1386 | static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) |
a246b010 CL |
1387 | { |
1388 | size_t len; | |
1389 | ||
51971139 | 1390 | len = transport->tcp_reclen - transport->tcp_offset; |
a246b010 CL |
1391 | if (len > desc->count) |
1392 | len = desc->count; | |
1393 | desc->count -= len; | |
1394 | desc->offset += len; | |
51971139 | 1395 | transport->tcp_offset += len; |
46121cf7 | 1396 | dprintk("RPC: discarded %Zu bytes\n", len); |
51971139 | 1397 | xs_tcp_check_fraghdr(transport); |
a246b010 CL |
1398 | } |
1399 | ||
9903cd1c | 1400 | static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) |
a246b010 CL |
1401 | { |
1402 | struct rpc_xprt *xprt = rd_desc->arg.data; | |
51971139 | 1403 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
dd456471 | 1404 | struct xdr_skb_reader desc = { |
a246b010 CL |
1405 | .skb = skb, |
1406 | .offset = offset, | |
1407 | .count = len, | |
9903cd1c | 1408 | }; |
a246b010 | 1409 | |
46121cf7 | 1410 | dprintk("RPC: xs_tcp_data_recv started\n"); |
a246b010 CL |
1411 | do { |
1412 | /* Read in a new fragment marker if necessary */ | |
1413 | /* Can we ever really expect to get completely empty fragments? */ | |
e136d092 | 1414 | if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { |
9903cd1c | 1415 | xs_tcp_read_fraghdr(xprt, &desc); |
a246b010 CL |
1416 | continue; |
1417 | } | |
1418 | /* Read in the xid if necessary */ | |
e136d092 | 1419 | if (transport->tcp_flags & TCP_RCV_COPY_XID) { |
51971139 | 1420 | xs_tcp_read_xid(transport, &desc); |
a246b010 CL |
1421 | continue; |
1422 | } | |
18dca02a | 1423 | /* Read in the call/reply flag */ |
f4a2e418 | 1424 | if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { |
18dca02a RL |
1425 | xs_tcp_read_calldir(transport, &desc); |
1426 | continue; | |
1427 | } | |
a246b010 | 1428 | /* Read in the request data */ |
e136d092 | 1429 | if (transport->tcp_flags & TCP_RCV_COPY_DATA) { |
44b98efd | 1430 | xs_tcp_read_data(xprt, &desc); |
a246b010 CL |
1431 | continue; |
1432 | } | |
1433 | /* Skip over any trailing bytes on short reads */ | |
51971139 | 1434 | xs_tcp_read_discard(transport, &desc); |
a246b010 | 1435 | } while (desc.count); |
46121cf7 | 1436 | dprintk("RPC: xs_tcp_data_recv done\n"); |
a246b010 CL |
1437 | return len - desc.count; |
1438 | } | |
1439 | ||
9903cd1c CL |
1440 | /** |
1441 | * xs_tcp_data_ready - "data ready" callback for TCP sockets | |
1442 | * @sk: socket with data to read | |
1443 | * @bytes: how much data to read | |
1444 | * | |
1445 | */ | |
676d2369 | 1446 | static void xs_tcp_data_ready(struct sock *sk) |
a246b010 CL |
1447 | { |
1448 | struct rpc_xprt *xprt; | |
1449 | read_descriptor_t rd_desc; | |
ff2d7db8 | 1450 | int read; |
a246b010 | 1451 | |
46121cf7 CL |
1452 | dprintk("RPC: xs_tcp_data_ready...\n"); |
1453 | ||
f064af1e | 1454 | read_lock_bh(&sk->sk_callback_lock); |
9903cd1c | 1455 | if (!(xprt = xprt_from_sock(sk))) |
a246b010 | 1456 | goto out; |
61d0a8e6 NB |
1457 | /* Any data means we had a useful conversation, so |
1458 | * the we don't need to delay the next reconnect | |
1459 | */ | |
1460 | if (xprt->reestablish_timeout) | |
1461 | xprt->reestablish_timeout = 0; | |
1462 | ||
9903cd1c | 1463 | /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ |
a246b010 | 1464 | rd_desc.arg.data = xprt; |
ff2d7db8 TM |
1465 | do { |
1466 | rd_desc.count = 65536; | |
1467 | read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); | |
1468 | } while (read > 0); | |
a246b010 | 1469 | out: |
f064af1e | 1470 | read_unlock_bh(&sk->sk_callback_lock); |
a246b010 CL |
1471 | } |
1472 | ||
7d1e8255 TM |
1473 | /* |
1474 | * Do the equivalent of linger/linger2 handling for dealing with | |
1475 | * broken servers that don't close the socket in a timely | |
1476 | * fashion | |
1477 | */ | |
1478 | static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt, | |
1479 | unsigned long timeout) | |
1480 | { | |
1481 | struct sock_xprt *transport; | |
1482 | ||
1483 | if (xprt_test_and_set_connecting(xprt)) | |
1484 | return; | |
1485 | set_bit(XPRT_CONNECTION_ABORT, &xprt->state); | |
1486 | transport = container_of(xprt, struct sock_xprt, xprt); | |
1487 | queue_delayed_work(rpciod_workqueue, &transport->connect_worker, | |
1488 | timeout); | |
1489 | } | |
1490 | ||
1491 | static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt) | |
1492 | { | |
1493 | struct sock_xprt *transport; | |
1494 | ||
1495 | transport = container_of(xprt, struct sock_xprt, xprt); | |
1496 | ||
1497 | if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) || | |
1498 | !cancel_delayed_work(&transport->connect_worker)) | |
1499 | return; | |
1500 | clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); | |
1501 | xprt_clear_connecting(xprt); | |
1502 | } | |
1503 | ||
4bc1e68e | 1504 | static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) |
7d1e8255 | 1505 | { |
4e857c58 | 1506 | smp_mb__before_atomic(); |
a519fc7a TM |
1507 | clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); |
1508 | clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state); | |
7d1e8255 TM |
1509 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); |
1510 | clear_bit(XPRT_CLOSING, &xprt->state); | |
4e857c58 | 1511 | smp_mb__after_atomic(); |
4bc1e68e TM |
1512 | } |
1513 | ||
1514 | static void xs_sock_mark_closed(struct rpc_xprt *xprt) | |
1515 | { | |
1516 | xs_sock_reset_connection_flags(xprt); | |
7d1e8255 TM |
1517 | /* Mark transport as closed and wake up all pending tasks */ |
1518 | xprt_disconnect_done(xprt); | |
1519 | } | |
1520 | ||
9903cd1c CL |
1521 | /** |
1522 | * xs_tcp_state_change - callback to handle TCP socket state changes | |
1523 | * @sk: socket whose state has changed | |
1524 | * | |
1525 | */ | |
1526 | static void xs_tcp_state_change(struct sock *sk) | |
a246b010 | 1527 | { |
9903cd1c | 1528 | struct rpc_xprt *xprt; |
a246b010 | 1529 | |
f064af1e | 1530 | read_lock_bh(&sk->sk_callback_lock); |
a246b010 CL |
1531 | if (!(xprt = xprt_from_sock(sk))) |
1532 | goto out; | |
46121cf7 | 1533 | dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); |
669502ff | 1534 | dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", |
46121cf7 CL |
1535 | sk->sk_state, xprt_connected(xprt), |
1536 | sock_flag(sk, SOCK_DEAD), | |
669502ff AC |
1537 | sock_flag(sk, SOCK_ZAPPED), |
1538 | sk->sk_shutdown); | |
a246b010 | 1539 | |
40b5ea0c | 1540 | trace_rpc_socket_state_change(xprt, sk->sk_socket); |
a246b010 CL |
1541 | switch (sk->sk_state) { |
1542 | case TCP_ESTABLISHED: | |
f064af1e | 1543 | spin_lock(&xprt->transport_lock); |
a246b010 | 1544 | if (!xprt_test_and_set_connected(xprt)) { |
51971139 CL |
1545 | struct sock_xprt *transport = container_of(xprt, |
1546 | struct sock_xprt, xprt); | |
1547 | ||
a246b010 | 1548 | /* Reset TCP record info */ |
51971139 CL |
1549 | transport->tcp_offset = 0; |
1550 | transport->tcp_reclen = 0; | |
1551 | transport->tcp_copied = 0; | |
e136d092 CL |
1552 | transport->tcp_flags = |
1553 | TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; | |
8b71798c | 1554 | xprt->connect_cookie++; |
51971139 | 1555 | |
2a491991 | 1556 | xprt_wake_pending_tasks(xprt, -EAGAIN); |
a246b010 | 1557 | } |
f064af1e | 1558 | spin_unlock(&xprt->transport_lock); |
a246b010 | 1559 | break; |
3b948ae5 TM |
1560 | case TCP_FIN_WAIT1: |
1561 | /* The client initiated a shutdown of the socket */ | |
7c1d71cf | 1562 | xprt->connect_cookie++; |
663b8858 | 1563 | xprt->reestablish_timeout = 0; |
3b948ae5 | 1564 | set_bit(XPRT_CLOSING, &xprt->state); |
4e857c58 | 1565 | smp_mb__before_atomic(); |
3b948ae5 | 1566 | clear_bit(XPRT_CONNECTED, &xprt->state); |
ef803670 | 1567 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); |
4e857c58 | 1568 | smp_mb__after_atomic(); |
25fe6142 | 1569 | xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); |
a246b010 | 1570 | break; |
632e3bdc | 1571 | case TCP_CLOSE_WAIT: |
3b948ae5 | 1572 | /* The server initiated a shutdown of the socket */ |
7c1d71cf | 1573 | xprt->connect_cookie++; |
d0bea455 | 1574 | clear_bit(XPRT_CONNECTED, &xprt->state); |
a519fc7a | 1575 | xs_tcp_force_close(xprt); |
663b8858 TM |
1576 | case TCP_CLOSING: |
1577 | /* | |
1578 | * If the server closed down the connection, make sure that | |
1579 | * we back off before reconnecting | |
1580 | */ | |
1581 | if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) | |
1582 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
3b948ae5 TM |
1583 | break; |
1584 | case TCP_LAST_ACK: | |
670f9457 | 1585 | set_bit(XPRT_CLOSING, &xprt->state); |
25fe6142 | 1586 | xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); |
4e857c58 | 1587 | smp_mb__before_atomic(); |
3b948ae5 | 1588 | clear_bit(XPRT_CONNECTED, &xprt->state); |
4e857c58 | 1589 | smp_mb__after_atomic(); |
3b948ae5 TM |
1590 | break; |
1591 | case TCP_CLOSE: | |
7d1e8255 TM |
1592 | xs_tcp_cancel_linger_timeout(xprt); |
1593 | xs_sock_mark_closed(xprt); | |
a246b010 CL |
1594 | } |
1595 | out: | |
f064af1e | 1596 | read_unlock_bh(&sk->sk_callback_lock); |
a246b010 CL |
1597 | } |
1598 | ||
1f0fa154 IJ |
1599 | static void xs_write_space(struct sock *sk) |
1600 | { | |
1601 | struct socket *sock; | |
1602 | struct rpc_xprt *xprt; | |
1603 | ||
1604 | if (unlikely(!(sock = sk->sk_socket))) | |
1605 | return; | |
1606 | clear_bit(SOCK_NOSPACE, &sock->flags); | |
1607 | ||
1608 | if (unlikely(!(xprt = xprt_from_sock(sk)))) | |
1609 | return; | |
1610 | if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) | |
1611 | return; | |
1612 | ||
1613 | xprt_write_space(xprt); | |
1614 | } | |
1615 | ||
9903cd1c | 1616 | /** |
c7b2cae8 CL |
1617 | * xs_udp_write_space - callback invoked when socket buffer space |
1618 | * becomes available | |
9903cd1c CL |
1619 | * @sk: socket whose state has changed |
1620 | * | |
a246b010 CL |
1621 | * Called when more output buffer space is available for this socket. |
1622 | * We try not to wake our writers until they can make "significant" | |
c7b2cae8 | 1623 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg |
a246b010 CL |
1624 | * with a bunch of small requests. |
1625 | */ | |
c7b2cae8 | 1626 | static void xs_udp_write_space(struct sock *sk) |
a246b010 | 1627 | { |
f064af1e | 1628 | read_lock_bh(&sk->sk_callback_lock); |
a246b010 | 1629 | |
c7b2cae8 | 1630 | /* from net/core/sock.c:sock_def_write_space */ |
1f0fa154 IJ |
1631 | if (sock_writeable(sk)) |
1632 | xs_write_space(sk); | |
a246b010 | 1633 | |
f064af1e | 1634 | read_unlock_bh(&sk->sk_callback_lock); |
c7b2cae8 | 1635 | } |
a246b010 | 1636 | |
c7b2cae8 CL |
1637 | /** |
1638 | * xs_tcp_write_space - callback invoked when socket buffer space | |
1639 | * becomes available | |
1640 | * @sk: socket whose state has changed | |
1641 | * | |
1642 | * Called when more output buffer space is available for this socket. | |
1643 | * We try not to wake our writers until they can make "significant" | |
1644 | * progress, otherwise we'll waste resources thrashing kernel_sendmsg | |
1645 | * with a bunch of small requests. | |
1646 | */ | |
1647 | static void xs_tcp_write_space(struct sock *sk) | |
1648 | { | |
f064af1e | 1649 | read_lock_bh(&sk->sk_callback_lock); |
c7b2cae8 CL |
1650 | |
1651 | /* from net/core/stream.c:sk_stream_write_space */ | |
64dc6130 | 1652 | if (sk_stream_is_writeable(sk)) |
1f0fa154 | 1653 | xs_write_space(sk); |
b6ddf64f | 1654 | |
f064af1e | 1655 | read_unlock_bh(&sk->sk_callback_lock); |
a246b010 CL |
1656 | } |
1657 | ||
470056c2 | 1658 | static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) |
a246b010 | 1659 | { |
ee0ac0c2 CL |
1660 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
1661 | struct sock *sk = transport->inet; | |
a246b010 | 1662 | |
7c6e066e | 1663 | if (transport->rcvsize) { |
a246b010 | 1664 | sk->sk_userlocks |= SOCK_RCVBUF_LOCK; |
7c6e066e | 1665 | sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; |
a246b010 | 1666 | } |
7c6e066e | 1667 | if (transport->sndsize) { |
a246b010 | 1668 | sk->sk_userlocks |= SOCK_SNDBUF_LOCK; |
7c6e066e | 1669 | sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; |
a246b010 CL |
1670 | sk->sk_write_space(sk); |
1671 | } | |
1672 | } | |
1673 | ||
43118c29 | 1674 | /** |
470056c2 | 1675 | * xs_udp_set_buffer_size - set send and receive limits |
43118c29 | 1676 | * @xprt: generic transport |
470056c2 CL |
1677 | * @sndsize: requested size of send buffer, in bytes |
1678 | * @rcvsize: requested size of receive buffer, in bytes | |
43118c29 | 1679 | * |
470056c2 | 1680 | * Set socket send and receive buffer size limits. |
43118c29 | 1681 | */ |
470056c2 | 1682 | static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) |
43118c29 | 1683 | { |
7c6e066e CL |
1684 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
1685 | ||
1686 | transport->sndsize = 0; | |
470056c2 | 1687 | if (sndsize) |
7c6e066e CL |
1688 | transport->sndsize = sndsize + 1024; |
1689 | transport->rcvsize = 0; | |
470056c2 | 1690 | if (rcvsize) |
7c6e066e | 1691 | transport->rcvsize = rcvsize + 1024; |
470056c2 CL |
1692 | |
1693 | xs_udp_do_set_buffer_size(xprt); | |
43118c29 CL |
1694 | } |
1695 | ||
46c0ee8b CL |
1696 | /** |
1697 | * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport | |
1698 | * @task: task that timed out | |
1699 | * | |
1700 | * Adjust the congestion window after a retransmit timeout has occurred. | |
1701 | */ | |
6a24dfb6 | 1702 | static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) |
46c0ee8b | 1703 | { |
6a24dfb6 | 1704 | xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); |
46c0ee8b CL |
1705 | } |
1706 | ||
b85d8806 CL |
1707 | static unsigned short xs_get_random_port(void) |
1708 | { | |
1709 | unsigned short range = xprt_max_resvport - xprt_min_resvport; | |
63862b5b | 1710 | unsigned short rand = (unsigned short) prandom_u32() % range; |
b85d8806 CL |
1711 | return rand + xprt_min_resvport; |
1712 | } | |
1713 | ||
92200412 CL |
1714 | /** |
1715 | * xs_set_port - reset the port number in the remote endpoint address | |
1716 | * @xprt: generic transport | |
1717 | * @port: new port number | |
1718 | * | |
1719 | */ | |
1720 | static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) | |
1721 | { | |
46121cf7 | 1722 | dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); |
c4efcb1d | 1723 | |
9dc3b095 CL |
1724 | rpc_set_port(xs_addr(xprt), port); |
1725 | xs_update_peer_port(xprt); | |
92200412 CL |
1726 | } |
1727 | ||
5d4ec932 | 1728 | static unsigned short xs_get_srcport(struct sock_xprt *transport) |
67a391d7 | 1729 | { |
fbfffbd5 | 1730 | unsigned short port = transport->srcport; |
67a391d7 TM |
1731 | |
1732 | if (port == 0 && transport->xprt.resvport) | |
1733 | port = xs_get_random_port(); | |
1734 | return port; | |
1735 | } | |
1736 | ||
baaf4e48 | 1737 | static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) |
67a391d7 | 1738 | { |
fbfffbd5 CL |
1739 | if (transport->srcport != 0) |
1740 | transport->srcport = 0; | |
67a391d7 TM |
1741 | if (!transport->xprt.resvport) |
1742 | return 0; | |
1743 | if (port <= xprt_min_resvport || port > xprt_max_resvport) | |
1744 | return xprt_max_resvport; | |
1745 | return --port; | |
1746 | } | |
beb59b68 | 1747 | static int xs_bind(struct sock_xprt *transport, struct socket *sock) |
a246b010 | 1748 | { |
beb59b68 | 1749 | struct sockaddr_storage myaddr; |
67a391d7 | 1750 | int err, nloop = 0; |
5d4ec932 | 1751 | unsigned short port = xs_get_srcport(transport); |
67a391d7 | 1752 | unsigned short last; |
a246b010 | 1753 | |
0f7a622c CP |
1754 | /* |
1755 | * If we are asking for any ephemeral port (i.e. port == 0 && | |
1756 | * transport->xprt.resvport == 0), don't bind. Let the local | |
1757 | * port selection happen implicitly when the socket is used | |
1758 | * (for example at connect time). | |
1759 | * | |
1760 | * This ensures that we can continue to establish TCP | |
1761 | * connections even when all local ephemeral ports are already | |
1762 | * a part of some TCP connection. This makes no difference | |
1763 | * for UDP sockets, but also doens't harm them. | |
1764 | * | |
1765 | * If we're asking for any reserved port (i.e. port == 0 && | |
1766 | * transport->xprt.resvport == 1) xs_get_srcport above will | |
1767 | * ensure that port is non-zero and we will bind as needed. | |
1768 | */ | |
1769 | if (port == 0) | |
1770 | return 0; | |
1771 | ||
beb59b68 | 1772 | memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); |
a246b010 | 1773 | do { |
beb59b68 PE |
1774 | rpc_set_port((struct sockaddr *)&myaddr, port); |
1775 | err = kernel_bind(sock, (struct sockaddr *)&myaddr, | |
1776 | transport->xprt.addrlen); | |
a246b010 | 1777 | if (err == 0) { |
fbfffbd5 | 1778 | transport->srcport = port; |
d3bc9a1d | 1779 | break; |
a246b010 | 1780 | } |
67a391d7 | 1781 | last = port; |
baaf4e48 | 1782 | port = xs_next_srcport(transport, port); |
67a391d7 TM |
1783 | if (port > last) |
1784 | nloop++; | |
1785 | } while (err == -EADDRINUSE && nloop != 2); | |
90058d37 | 1786 | |
4232e863 | 1787 | if (myaddr.ss_family == AF_INET) |
beb59b68 PE |
1788 | dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, |
1789 | &((struct sockaddr_in *)&myaddr)->sin_addr, | |
1790 | port, err ? "failed" : "ok", err); | |
1791 | else | |
1792 | dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, | |
1793 | &((struct sockaddr_in6 *)&myaddr)->sin6_addr, | |
1794 | port, err ? "failed" : "ok", err); | |
a246b010 CL |
1795 | return err; |
1796 | } | |
1797 | ||
176e21ee CL |
1798 | /* |
1799 | * We don't support autobind on AF_LOCAL sockets | |
1800 | */ | |
1801 | static void xs_local_rpcbind(struct rpc_task *task) | |
1802 | { | |
3dc0da27 TM |
1803 | rcu_read_lock(); |
1804 | xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt)); | |
1805 | rcu_read_unlock(); | |
176e21ee CL |
1806 | } |
1807 | ||
1808 | static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) | |
1809 | { | |
1810 | } | |
beb59b68 | 1811 | |
ed07536e PZ |
1812 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1813 | static struct lock_class_key xs_key[2]; | |
1814 | static struct lock_class_key xs_slock_key[2]; | |
1815 | ||
176e21ee CL |
1816 | static inline void xs_reclassify_socketu(struct socket *sock) |
1817 | { | |
1818 | struct sock *sk = sock->sk; | |
1819 | ||
176e21ee CL |
1820 | sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", |
1821 | &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); | |
1822 | } | |
1823 | ||
8945ee5e | 1824 | static inline void xs_reclassify_socket4(struct socket *sock) |
ed07536e PZ |
1825 | { |
1826 | struct sock *sk = sock->sk; | |
8945ee5e | 1827 | |
8945ee5e CL |
1828 | sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", |
1829 | &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); | |
1830 | } | |
ed07536e | 1831 | |
8945ee5e CL |
1832 | static inline void xs_reclassify_socket6(struct socket *sock) |
1833 | { | |
1834 | struct sock *sk = sock->sk; | |
ed07536e | 1835 | |
8945ee5e CL |
1836 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", |
1837 | &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); | |
ed07536e | 1838 | } |
6bc9638a PE |
1839 | |
1840 | static inline void xs_reclassify_socket(int family, struct socket *sock) | |
1841 | { | |
1b7a1819 WAA |
1842 | WARN_ON_ONCE(sock_owned_by_user(sock->sk)); |
1843 | if (sock_owned_by_user(sock->sk)) | |
1844 | return; | |
1845 | ||
4232e863 | 1846 | switch (family) { |
176e21ee CL |
1847 | case AF_LOCAL: |
1848 | xs_reclassify_socketu(sock); | |
1849 | break; | |
4232e863 | 1850 | case AF_INET: |
6bc9638a | 1851 | xs_reclassify_socket4(sock); |
4232e863 CL |
1852 | break; |
1853 | case AF_INET6: | |
6bc9638a | 1854 | xs_reclassify_socket6(sock); |
4232e863 CL |
1855 | break; |
1856 | } | |
6bc9638a | 1857 | } |
ed07536e | 1858 | #else |
176e21ee CL |
1859 | static inline void xs_reclassify_socketu(struct socket *sock) |
1860 | { | |
1861 | } | |
1862 | ||
8945ee5e CL |
1863 | static inline void xs_reclassify_socket4(struct socket *sock) |
1864 | { | |
1865 | } | |
1866 | ||
1867 | static inline void xs_reclassify_socket6(struct socket *sock) | |
ed07536e PZ |
1868 | { |
1869 | } | |
6bc9638a PE |
1870 | |
1871 | static inline void xs_reclassify_socket(int family, struct socket *sock) | |
1872 | { | |
1873 | } | |
ed07536e PZ |
1874 | #endif |
1875 | ||
93dc41bd N |
1876 | static void xs_dummy_setup_socket(struct work_struct *work) |
1877 | { | |
1878 | } | |
1879 | ||
6bc9638a PE |
1880 | static struct socket *xs_create_sock(struct rpc_xprt *xprt, |
1881 | struct sock_xprt *transport, int family, int type, int protocol) | |
22f79326 PE |
1882 | { |
1883 | struct socket *sock; | |
1884 | int err; | |
1885 | ||
6bc9638a | 1886 | err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); |
22f79326 PE |
1887 | if (err < 0) { |
1888 | dprintk("RPC: can't create %d transport socket (%d).\n", | |
1889 | protocol, -err); | |
1890 | goto out; | |
1891 | } | |
6bc9638a | 1892 | xs_reclassify_socket(family, sock); |
22f79326 | 1893 | |
4cea288a BH |
1894 | err = xs_bind(transport, sock); |
1895 | if (err) { | |
22f79326 PE |
1896 | sock_release(sock); |
1897 | goto out; | |
1898 | } | |
1899 | ||
1900 | return sock; | |
1901 | out: | |
1902 | return ERR_PTR(err); | |
1903 | } | |
1904 | ||
176e21ee CL |
1905 | static int xs_local_finish_connecting(struct rpc_xprt *xprt, |
1906 | struct socket *sock) | |
1907 | { | |
1908 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, | |
1909 | xprt); | |
1910 | ||
1911 | if (!transport->inet) { | |
1912 | struct sock *sk = sock->sk; | |
1913 | ||
1914 | write_lock_bh(&sk->sk_callback_lock); | |
1915 | ||
1916 | xs_save_old_callbacks(transport, sk); | |
1917 | ||
1918 | sk->sk_user_data = xprt; | |
1919 | sk->sk_data_ready = xs_local_data_ready; | |
1920 | sk->sk_write_space = xs_udp_write_space; | |
2118071d | 1921 | sk->sk_error_report = xs_error_report; |
176e21ee CL |
1922 | sk->sk_allocation = GFP_ATOMIC; |
1923 | ||
1924 | xprt_clear_connected(xprt); | |
1925 | ||
1926 | /* Reset to new socket */ | |
1927 | transport->sock = sock; | |
1928 | transport->inet = sk; | |
1929 | ||
1930 | write_unlock_bh(&sk->sk_callback_lock); | |
1931 | } | |
1932 | ||
1933 | /* Tell the socket layer to start connecting... */ | |
1934 | xprt->stat.connect_count++; | |
1935 | xprt->stat.connect_start = jiffies; | |
1936 | return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); | |
1937 | } | |
1938 | ||
1939 | /** | |
1940 | * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint | |
1941 | * @xprt: RPC transport to connect | |
1942 | * @transport: socket transport to connect | |
1943 | * @create_sock: function to create a socket of the correct type | |
176e21ee | 1944 | */ |
dc107402 | 1945 | static int xs_local_setup_socket(struct sock_xprt *transport) |
176e21ee | 1946 | { |
176e21ee CL |
1947 | struct rpc_xprt *xprt = &transport->xprt; |
1948 | struct socket *sock; | |
1949 | int status = -EIO; | |
1950 | ||
5cf02d09 JL |
1951 | current->flags |= PF_FSTRANS; |
1952 | ||
176e21ee CL |
1953 | clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); |
1954 | status = __sock_create(xprt->xprt_net, AF_LOCAL, | |
1955 | SOCK_STREAM, 0, &sock, 1); | |
1956 | if (status < 0) { | |
1957 | dprintk("RPC: can't create AF_LOCAL " | |
1958 | "transport socket (%d).\n", -status); | |
1959 | goto out; | |
1960 | } | |
1961 | xs_reclassify_socketu(sock); | |
1962 | ||
1963 | dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", | |
1964 | xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); | |
1965 | ||
1966 | status = xs_local_finish_connecting(xprt, sock); | |
40b5ea0c | 1967 | trace_rpc_socket_connect(xprt, sock, status); |
176e21ee CL |
1968 | switch (status) { |
1969 | case 0: | |
1970 | dprintk("RPC: xprt %p connected to %s\n", | |
1971 | xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); | |
1972 | xprt_set_connected(xprt); | |
3601c4a9 | 1973 | case -ENOBUFS: |
176e21ee CL |
1974 | break; |
1975 | case -ENOENT: | |
1976 | dprintk("RPC: xprt %p: socket %s does not exist\n", | |
1977 | xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); | |
1978 | break; | |
4a20a988 TM |
1979 | case -ECONNREFUSED: |
1980 | dprintk("RPC: xprt %p: connection refused for %s\n", | |
1981 | xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); | |
1982 | break; | |
176e21ee CL |
1983 | default: |
1984 | printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", | |
1985 | __func__, -status, | |
1986 | xprt->address_strings[RPC_DISPLAY_ADDR]); | |
1987 | } | |
1988 | ||
1989 | out: | |
1990 | xprt_clear_connecting(xprt); | |
1991 | xprt_wake_pending_tasks(xprt, status); | |
5cf02d09 | 1992 | current->flags &= ~PF_FSTRANS; |
dc107402 BF |
1993 | return status; |
1994 | } | |
1995 | ||
b6669737 | 1996 | static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) |
dc107402 | 1997 | { |
dc107402 BF |
1998 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
1999 | int ret; | |
2000 | ||
2001 | if (RPC_IS_ASYNC(task)) { | |
2002 | /* | |
2003 | * We want the AF_LOCAL connect to be resolved in the | |
2004 | * filesystem namespace of the process making the rpc | |
2005 | * call. Thus we connect synchronously. | |
2006 | * | |
2007 | * If we want to support asynchronous AF_LOCAL calls, | |
2008 | * we'll need to figure out how to pass a namespace to | |
2009 | * connect. | |
2010 | */ | |
2011 | rpc_exit(task, -ENOTCONN); | |
2012 | return; | |
2013 | } | |
2014 | ret = xs_local_setup_socket(transport); | |
2015 | if (ret && !RPC_IS_SOFTCONN(task)) | |
2016 | msleep_interruptible(15000); | |
176e21ee CL |
2017 | } |
2018 | ||
a564b8f0 MG |
2019 | #ifdef CONFIG_SUNRPC_SWAP |
2020 | static void xs_set_memalloc(struct rpc_xprt *xprt) | |
2021 | { | |
2022 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, | |
2023 | xprt); | |
2024 | ||
2025 | if (xprt->swapper) | |
2026 | sk_set_memalloc(transport->inet); | |
2027 | } | |
2028 | ||
2029 | /** | |
2030 | * xs_swapper - Tag this transport as being used for swap. | |
2031 | * @xprt: transport to tag | |
2032 | * @enable: enable/disable | |
2033 | * | |
2034 | */ | |
2035 | int xs_swapper(struct rpc_xprt *xprt, int enable) | |
2036 | { | |
2037 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, | |
2038 | xprt); | |
2039 | int err = 0; | |
2040 | ||
2041 | if (enable) { | |
2042 | xprt->swapper++; | |
2043 | xs_set_memalloc(xprt); | |
2044 | } else if (xprt->swapper) { | |
2045 | xprt->swapper--; | |
2046 | sk_clear_memalloc(transport->inet); | |
2047 | } | |
2048 | ||
2049 | return err; | |
2050 | } | |
2051 | EXPORT_SYMBOL_GPL(xs_swapper); | |
2052 | #else | |
2053 | static void xs_set_memalloc(struct rpc_xprt *xprt) | |
2054 | { | |
2055 | } | |
2056 | #endif | |
2057 | ||
16be2d20 CL |
2058 | static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) |
2059 | { | |
2060 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); | |
2061 | ||
2062 | if (!transport->inet) { | |
2063 | struct sock *sk = sock->sk; | |
2064 | ||
2065 | write_lock_bh(&sk->sk_callback_lock); | |
2066 | ||
2a9e1cfa TM |
2067 | xs_save_old_callbacks(transport, sk); |
2068 | ||
16be2d20 | 2069 | sk->sk_user_data = xprt; |
16be2d20 CL |
2070 | sk->sk_data_ready = xs_udp_data_ready; |
2071 | sk->sk_write_space = xs_udp_write_space; | |
16be2d20 CL |
2072 | sk->sk_allocation = GFP_ATOMIC; |
2073 | ||
2074 | xprt_set_connected(xprt); | |
2075 | ||
2076 | /* Reset to new socket */ | |
2077 | transport->sock = sock; | |
2078 | transport->inet = sk; | |
2079 | ||
a564b8f0 MG |
2080 | xs_set_memalloc(xprt); |
2081 | ||
16be2d20 CL |
2082 | write_unlock_bh(&sk->sk_callback_lock); |
2083 | } | |
2084 | xs_udp_do_set_buffer_size(xprt); | |
2085 | } | |
2086 | ||
8c14ff2a | 2087 | static void xs_udp_setup_socket(struct work_struct *work) |
a246b010 | 2088 | { |
34161db6 TM |
2089 | struct sock_xprt *transport = |
2090 | container_of(work, struct sock_xprt, connect_worker.work); | |
c8475461 | 2091 | struct rpc_xprt *xprt = &transport->xprt; |
ee0ac0c2 | 2092 | struct socket *sock = transport->sock; |
b65c0310 | 2093 | int status = -EIO; |
9903cd1c | 2094 | |
5cf02d09 JL |
2095 | current->flags |= PF_FSTRANS; |
2096 | ||
b0d93ad5 | 2097 | /* Start by resetting any existing state */ |
fe315e76 | 2098 | xs_reset_transport(transport); |
8c14ff2a PE |
2099 | sock = xs_create_sock(xprt, transport, |
2100 | xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP); | |
b65c0310 | 2101 | if (IS_ERR(sock)) |
b0d93ad5 | 2102 | goto out; |
68e220bd | 2103 | |
c740eff8 CL |
2104 | dprintk("RPC: worker connecting xprt %p via %s to " |
2105 | "%s (port %s)\n", xprt, | |
2106 | xprt->address_strings[RPC_DISPLAY_PROTO], | |
2107 | xprt->address_strings[RPC_DISPLAY_ADDR], | |
2108 | xprt->address_strings[RPC_DISPLAY_PORT]); | |
68e220bd CL |
2109 | |
2110 | xs_udp_finish_connecting(xprt, sock); | |
40b5ea0c | 2111 | trace_rpc_socket_connect(xprt, sock, 0); |
b0d93ad5 CL |
2112 | status = 0; |
2113 | out: | |
b0d93ad5 | 2114 | xprt_clear_connecting(xprt); |
7d1e8255 | 2115 | xprt_wake_pending_tasks(xprt, status); |
5cf02d09 | 2116 | current->flags &= ~PF_FSTRANS; |
a246b010 CL |
2117 | } |
2118 | ||
3167e12c CL |
2119 | /* |
2120 | * We need to preserve the port number so the reply cache on the server can | |
2121 | * find our cached RPC replies when we get around to reconnecting. | |
2122 | */ | |
58dddac9 | 2123 | static void xs_abort_connection(struct sock_xprt *transport) |
3167e12c CL |
2124 | { |
2125 | int result; | |
3167e12c CL |
2126 | struct sockaddr any; |
2127 | ||
58dddac9 | 2128 | dprintk("RPC: disconnecting xprt %p to reuse port\n", transport); |
3167e12c CL |
2129 | |
2130 | /* | |
2131 | * Disconnect the transport socket by doing a connect operation | |
2132 | * with AF_UNSPEC. This should return immediately... | |
2133 | */ | |
2134 | memset(&any, 0, sizeof(any)); | |
2135 | any.sa_family = AF_UNSPEC; | |
ee0ac0c2 | 2136 | result = kernel_connect(transport->sock, &any, sizeof(any), 0); |
40b5ea0c TM |
2137 | trace_rpc_socket_reset_connection(&transport->xprt, |
2138 | transport->sock, result); | |
7d1e8255 | 2139 | if (!result) |
4bc1e68e TM |
2140 | xs_sock_reset_connection_flags(&transport->xprt); |
2141 | dprintk("RPC: AF_UNSPEC connect return code %d\n", result); | |
3167e12c CL |
2142 | } |
2143 | ||
58dddac9 | 2144 | static void xs_tcp_reuse_connection(struct sock_xprt *transport) |
40d2549d TM |
2145 | { |
2146 | unsigned int state = transport->inet->sk_state; | |
2147 | ||
669502ff AC |
2148 | if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) { |
2149 | /* we don't need to abort the connection if the socket | |
2150 | * hasn't undergone a shutdown | |
2151 | */ | |
2152 | if (transport->inet->sk_shutdown == 0) | |
2153 | return; | |
2154 | dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n", | |
2155 | __func__, transport->inet->sk_shutdown); | |
2156 | } | |
2157 | if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) { | |
2158 | /* we don't need to abort the connection if the socket | |
2159 | * hasn't undergone a shutdown | |
2160 | */ | |
2161 | if (transport->inet->sk_shutdown == 0) | |
2162 | return; | |
2163 | dprintk("RPC: %s: ESTABLISHED/SYN_SENT " | |
2164 | "sk_shutdown set to %d\n", | |
2165 | __func__, transport->inet->sk_shutdown); | |
2166 | } | |
58dddac9 | 2167 | xs_abort_connection(transport); |
40d2549d TM |
2168 | } |
2169 | ||
16be2d20 | 2170 | static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) |
a246b010 | 2171 | { |
16be2d20 | 2172 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
fe19a96b | 2173 | int ret = -ENOTCONN; |
edb267a6 | 2174 | |
ee0ac0c2 | 2175 | if (!transport->inet) { |
b0d93ad5 | 2176 | struct sock *sk = sock->sk; |
7f260e85 TM |
2177 | unsigned int keepidle = xprt->timeout->to_initval / HZ; |
2178 | unsigned int keepcnt = xprt->timeout->to_retries + 1; | |
2179 | unsigned int opt_on = 1; | |
2180 | ||
2181 | /* TCP Keepalive options */ | |
2182 | kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, | |
2183 | (char *)&opt_on, sizeof(opt_on)); | |
2184 | kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, | |
2185 | (char *)&keepidle, sizeof(keepidle)); | |
2186 | kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, | |
2187 | (char *)&keepidle, sizeof(keepidle)); | |
2188 | kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, | |
2189 | (char *)&keepcnt, sizeof(keepcnt)); | |
b0d93ad5 CL |
2190 | |
2191 | write_lock_bh(&sk->sk_callback_lock); | |
2192 | ||
2a9e1cfa TM |
2193 | xs_save_old_callbacks(transport, sk); |
2194 | ||
b0d93ad5 | 2195 | sk->sk_user_data = xprt; |
b0d93ad5 CL |
2196 | sk->sk_data_ready = xs_tcp_data_ready; |
2197 | sk->sk_state_change = xs_tcp_state_change; | |
2198 | sk->sk_write_space = xs_tcp_write_space; | |
2118071d | 2199 | sk->sk_error_report = xs_error_report; |
b079fa7b | 2200 | sk->sk_allocation = GFP_ATOMIC; |
3167e12c CL |
2201 | |
2202 | /* socket options */ | |
2203 | sk->sk_userlocks |= SOCK_BINDPORT_LOCK; | |
2204 | sock_reset_flag(sk, SOCK_LINGER); | |
2205 | tcp_sk(sk)->linger2 = 0; | |
2206 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; | |
b0d93ad5 CL |
2207 | |
2208 | xprt_clear_connected(xprt); | |
2209 | ||
2210 | /* Reset to new socket */ | |
ee0ac0c2 CL |
2211 | transport->sock = sock; |
2212 | transport->inet = sk; | |
b0d93ad5 CL |
2213 | |
2214 | write_unlock_bh(&sk->sk_callback_lock); | |
2215 | } | |
2216 | ||
01d37c42 | 2217 | if (!xprt_bound(xprt)) |
fe19a96b | 2218 | goto out; |
01d37c42 | 2219 | |
a564b8f0 MG |
2220 | xs_set_memalloc(xprt); |
2221 | ||
b0d93ad5 | 2222 | /* Tell the socket layer to start connecting... */ |
262ca07d CL |
2223 | xprt->stat.connect_count++; |
2224 | xprt->stat.connect_start = jiffies; | |
fe19a96b TM |
2225 | ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); |
2226 | switch (ret) { | |
2227 | case 0: | |
2228 | case -EINPROGRESS: | |
2229 | /* SYN_SENT! */ | |
fe19a96b TM |
2230 | if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) |
2231 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
2232 | } | |
2233 | out: | |
2234 | return ret; | |
16be2d20 CL |
2235 | } |
2236 | ||
9903cd1c | 2237 | /** |
b61d59ff TM |
2238 | * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint |
2239 | * @xprt: RPC transport to connect | |
2240 | * @transport: socket transport to connect | |
2241 | * @create_sock: function to create a socket of the correct type | |
9903cd1c CL |
2242 | * |
2243 | * Invoked by a work queue tasklet. | |
a246b010 | 2244 | */ |
cdd518d5 | 2245 | static void xs_tcp_setup_socket(struct work_struct *work) |
a246b010 | 2246 | { |
cdd518d5 PE |
2247 | struct sock_xprt *transport = |
2248 | container_of(work, struct sock_xprt, connect_worker.work); | |
ee0ac0c2 | 2249 | struct socket *sock = transport->sock; |
a9f5f0f7 | 2250 | struct rpc_xprt *xprt = &transport->xprt; |
b61d59ff | 2251 | int status = -EIO; |
a246b010 | 2252 | |
5cf02d09 JL |
2253 | current->flags |= PF_FSTRANS; |
2254 | ||
ee0ac0c2 | 2255 | if (!sock) { |
7d1e8255 | 2256 | clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); |
cdd518d5 PE |
2257 | sock = xs_create_sock(xprt, transport, |
2258 | xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP); | |
b61d59ff TM |
2259 | if (IS_ERR(sock)) { |
2260 | status = PTR_ERR(sock); | |
3167e12c CL |
2261 | goto out; |
2262 | } | |
7d1e8255 TM |
2263 | } else { |
2264 | int abort_and_exit; | |
a246b010 | 2265 | |
7d1e8255 TM |
2266 | abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT, |
2267 | &xprt->state); | |
3167e12c | 2268 | /* "close" the socket, preserving the local port */ |
a743419f | 2269 | set_bit(XPRT_CONNECTION_REUSE, &xprt->state); |
58dddac9 | 2270 | xs_tcp_reuse_connection(transport); |
a743419f | 2271 | clear_bit(XPRT_CONNECTION_REUSE, &xprt->state); |
a246b010 | 2272 | |
7d1e8255 TM |
2273 | if (abort_and_exit) |
2274 | goto out_eagain; | |
2275 | } | |
a246b010 | 2276 | |
c740eff8 CL |
2277 | dprintk("RPC: worker connecting xprt %p via %s to " |
2278 | "%s (port %s)\n", xprt, | |
2279 | xprt->address_strings[RPC_DISPLAY_PROTO], | |
2280 | xprt->address_strings[RPC_DISPLAY_ADDR], | |
2281 | xprt->address_strings[RPC_DISPLAY_PORT]); | |
edb267a6 | 2282 | |
16be2d20 | 2283 | status = xs_tcp_finish_connecting(xprt, sock); |
40b5ea0c | 2284 | trace_rpc_socket_connect(xprt, sock, status); |
46121cf7 CL |
2285 | dprintk("RPC: %p connect status %d connected %d sock state %d\n", |
2286 | xprt, -status, xprt_connected(xprt), | |
2287 | sock->sk->sk_state); | |
2a491991 | 2288 | switch (status) { |
f75e6745 TM |
2289 | default: |
2290 | printk("%s: connect returned unhandled error %d\n", | |
2291 | __func__, status); | |
2292 | case -EADDRNOTAVAIL: | |
2293 | /* We're probably in TIME_WAIT. Get rid of existing socket, | |
2294 | * and retry | |
2295 | */ | |
a519fc7a | 2296 | xs_tcp_force_close(xprt); |
88b5ed73 | 2297 | break; |
2a491991 TM |
2298 | case 0: |
2299 | case -EINPROGRESS: | |
2300 | case -EALREADY: | |
7d1e8255 | 2301 | xprt_clear_connecting(xprt); |
5cf02d09 | 2302 | current->flags &= ~PF_FSTRANS; |
7d1e8255 | 2303 | return; |
9fcfe0c8 TM |
2304 | case -EINVAL: |
2305 | /* Happens, for instance, if the user specified a link | |
2306 | * local IPv6 address without a scope-id. | |
2307 | */ | |
3ed5e2a2 TM |
2308 | case -ECONNREFUSED: |
2309 | case -ECONNRESET: | |
2310 | case -ENETUNREACH: | |
3601c4a9 | 2311 | case -ENOBUFS: |
3ed5e2a2 | 2312 | /* retry with existing socket, after a delay */ |
9fcfe0c8 | 2313 | goto out; |
a246b010 | 2314 | } |
7d1e8255 | 2315 | out_eagain: |
2a491991 | 2316 | status = -EAGAIN; |
a246b010 | 2317 | out: |
68e220bd | 2318 | xprt_clear_connecting(xprt); |
7d1e8255 | 2319 | xprt_wake_pending_tasks(xprt, status); |
5cf02d09 | 2320 | current->flags &= ~PF_FSTRANS; |
68e220bd | 2321 | } |
b0d93ad5 | 2322 | |
9903cd1c CL |
2323 | /** |
2324 | * xs_connect - connect a socket to a remote endpoint | |
1b092092 | 2325 | * @xprt: pointer to transport structure |
9903cd1c CL |
2326 | * @task: address of RPC task that manages state of connect request |
2327 | * | |
2328 | * TCP: If the remote end dropped the connection, delay reconnecting. | |
03bf4b70 CL |
2329 | * |
2330 | * UDP socket connects are synchronous, but we use a work queue anyway | |
2331 | * to guarantee that even unprivileged user processes can set up a | |
2332 | * socket on a privileged port. | |
2333 | * | |
2334 | * If a UDP socket connect fails, the delay behavior here prevents | |
2335 | * retry floods (hard mounts). | |
9903cd1c | 2336 | */ |
1b092092 | 2337 | static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) |
a246b010 | 2338 | { |
ee0ac0c2 | 2339 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 | 2340 | |
09a21c41 | 2341 | if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) { |
46121cf7 CL |
2342 | dprintk("RPC: xs_connect delayed xprt %p for %lu " |
2343 | "seconds\n", | |
03bf4b70 | 2344 | xprt, xprt->reestablish_timeout / HZ); |
c1384c9c TM |
2345 | queue_delayed_work(rpciod_workqueue, |
2346 | &transport->connect_worker, | |
2347 | xprt->reestablish_timeout); | |
03bf4b70 | 2348 | xprt->reestablish_timeout <<= 1; |
61d0a8e6 NB |
2349 | if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) |
2350 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
03bf4b70 CL |
2351 | if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) |
2352 | xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; | |
b0d93ad5 | 2353 | } else { |
46121cf7 | 2354 | dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); |
c1384c9c TM |
2355 | queue_delayed_work(rpciod_workqueue, |
2356 | &transport->connect_worker, 0); | |
a246b010 CL |
2357 | } |
2358 | } | |
2359 | ||
176e21ee CL |
2360 | /** |
2361 | * xs_local_print_stats - display AF_LOCAL socket-specifc stats | |
2362 | * @xprt: rpc_xprt struct containing statistics | |
2363 | * @seq: output file | |
2364 | * | |
2365 | */ | |
2366 | static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
2367 | { | |
2368 | long idle_time = 0; | |
2369 | ||
2370 | if (xprt_connected(xprt)) | |
2371 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
2372 | ||
2373 | seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " | |
15a45206 | 2374 | "%llu %llu %lu %llu %llu\n", |
176e21ee CL |
2375 | xprt->stat.bind_count, |
2376 | xprt->stat.connect_count, | |
2377 | xprt->stat.connect_time, | |
2378 | idle_time, | |
2379 | xprt->stat.sends, | |
2380 | xprt->stat.recvs, | |
2381 | xprt->stat.bad_xids, | |
2382 | xprt->stat.req_u, | |
15a45206 AA |
2383 | xprt->stat.bklog_u, |
2384 | xprt->stat.max_slots, | |
2385 | xprt->stat.sending_u, | |
2386 | xprt->stat.pending_u); | |
176e21ee CL |
2387 | } |
2388 | ||
262ca07d CL |
2389 | /** |
2390 | * xs_udp_print_stats - display UDP socket-specifc stats | |
2391 | * @xprt: rpc_xprt struct containing statistics | |
2392 | * @seq: output file | |
2393 | * | |
2394 | */ | |
2395 | static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
2396 | { | |
c8475461 CL |
2397 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
2398 | ||
15a45206 AA |
2399 | seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " |
2400 | "%lu %llu %llu\n", | |
fbfffbd5 | 2401 | transport->srcport, |
262ca07d CL |
2402 | xprt->stat.bind_count, |
2403 | xprt->stat.sends, | |
2404 | xprt->stat.recvs, | |
2405 | xprt->stat.bad_xids, | |
2406 | xprt->stat.req_u, | |
15a45206 AA |
2407 | xprt->stat.bklog_u, |
2408 | xprt->stat.max_slots, | |
2409 | xprt->stat.sending_u, | |
2410 | xprt->stat.pending_u); | |
262ca07d CL |
2411 | } |
2412 | ||
2413 | /** | |
2414 | * xs_tcp_print_stats - display TCP socket-specifc stats | |
2415 | * @xprt: rpc_xprt struct containing statistics | |
2416 | * @seq: output file | |
2417 | * | |
2418 | */ | |
2419 | static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
2420 | { | |
c8475461 | 2421 | struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); |
262ca07d CL |
2422 | long idle_time = 0; |
2423 | ||
2424 | if (xprt_connected(xprt)) | |
2425 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
2426 | ||
15a45206 AA |
2427 | seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " |
2428 | "%llu %llu %lu %llu %llu\n", | |
fbfffbd5 | 2429 | transport->srcport, |
262ca07d CL |
2430 | xprt->stat.bind_count, |
2431 | xprt->stat.connect_count, | |
2432 | xprt->stat.connect_time, | |
2433 | idle_time, | |
2434 | xprt->stat.sends, | |
2435 | xprt->stat.recvs, | |
2436 | xprt->stat.bad_xids, | |
2437 | xprt->stat.req_u, | |
15a45206 AA |
2438 | xprt->stat.bklog_u, |
2439 | xprt->stat.max_slots, | |
2440 | xprt->stat.sending_u, | |
2441 | xprt->stat.pending_u); | |
262ca07d CL |
2442 | } |
2443 | ||
4cfc7e60 RI |
2444 | /* |
2445 | * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason | |
2446 | * we allocate pages instead doing a kmalloc like rpc_malloc is because we want | |
2447 | * to use the server side send routines. | |
2448 | */ | |
5a51f13a | 2449 | static void *bc_malloc(struct rpc_task *task, size_t size) |
4cfc7e60 RI |
2450 | { |
2451 | struct page *page; | |
2452 | struct rpc_buffer *buf; | |
2453 | ||
b8a13d03 WAA |
2454 | WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer)); |
2455 | if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) | |
2456 | return NULL; | |
4cfc7e60 | 2457 | |
b8a13d03 | 2458 | page = alloc_page(GFP_KERNEL); |
4cfc7e60 RI |
2459 | if (!page) |
2460 | return NULL; | |
2461 | ||
2462 | buf = page_address(page); | |
2463 | buf->len = PAGE_SIZE; | |
2464 | ||
2465 | return buf->data; | |
2466 | } | |
2467 | ||
2468 | /* | |
2469 | * Free the space allocated in the bc_alloc routine | |
2470 | */ | |
5a51f13a | 2471 | static void bc_free(void *buffer) |
4cfc7e60 RI |
2472 | { |
2473 | struct rpc_buffer *buf; | |
2474 | ||
2475 | if (!buffer) | |
2476 | return; | |
2477 | ||
2478 | buf = container_of(buffer, struct rpc_buffer, data); | |
2479 | free_page((unsigned long)buf); | |
2480 | } | |
2481 | ||
2482 | /* | |
2483 | * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex | |
2484 | * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. | |
2485 | */ | |
2486 | static int bc_sendto(struct rpc_rqst *req) | |
2487 | { | |
2488 | int len; | |
2489 | struct xdr_buf *xbufp = &req->rq_snd_buf; | |
2490 | struct rpc_xprt *xprt = req->rq_xprt; | |
2491 | struct sock_xprt *transport = | |
2492 | container_of(xprt, struct sock_xprt, xprt); | |
2493 | struct socket *sock = transport->sock; | |
2494 | unsigned long headoff; | |
2495 | unsigned long tailoff; | |
2496 | ||
61677eee | 2497 | xs_encode_stream_record_marker(xbufp); |
4cfc7e60 RI |
2498 | |
2499 | tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; | |
2500 | headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; | |
2501 | len = svc_send_common(sock, xbufp, | |
2502 | virt_to_page(xbufp->head[0].iov_base), headoff, | |
2503 | xbufp->tail[0].iov_base, tailoff); | |
2504 | ||
2505 | if (len != xbufp->len) { | |
2506 | printk(KERN_NOTICE "Error sending entire callback!\n"); | |
2507 | len = -EAGAIN; | |
2508 | } | |
2509 | ||
2510 | return len; | |
2511 | } | |
2512 | ||
2513 | /* | |
2514 | * The send routine. Borrows from svc_send | |
2515 | */ | |
2516 | static int bc_send_request(struct rpc_task *task) | |
2517 | { | |
2518 | struct rpc_rqst *req = task->tk_rqstp; | |
2519 | struct svc_xprt *xprt; | |
4cfc7e60 RI |
2520 | u32 len; |
2521 | ||
2522 | dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); | |
2523 | /* | |
2524 | * Get the server socket associated with this callback xprt | |
2525 | */ | |
2526 | xprt = req->rq_xprt->bc_xprt; | |
4cfc7e60 RI |
2527 | |
2528 | /* | |
2529 | * Grab the mutex to serialize data as the connection is shared | |
2530 | * with the fore channel | |
2531 | */ | |
2532 | if (!mutex_trylock(&xprt->xpt_mutex)) { | |
2533 | rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL); | |
2534 | if (!mutex_trylock(&xprt->xpt_mutex)) | |
2535 | return -EAGAIN; | |
2536 | rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task); | |
2537 | } | |
2538 | if (test_bit(XPT_DEAD, &xprt->xpt_flags)) | |
2539 | len = -ENOTCONN; | |
2540 | else | |
2541 | len = bc_sendto(req); | |
2542 | mutex_unlock(&xprt->xpt_mutex); | |
2543 | ||
2544 | if (len > 0) | |
2545 | len = 0; | |
2546 | ||
2547 | return len; | |
2548 | } | |
2549 | ||
2550 | /* | |
2551 | * The close routine. Since this is client initiated, we do nothing | |
2552 | */ | |
2553 | ||
2554 | static void bc_close(struct rpc_xprt *xprt) | |
2555 | { | |
4cfc7e60 RI |
2556 | } |
2557 | ||
2558 | /* | |
2559 | * The xprt destroy routine. Again, because this connection is client | |
2560 | * initiated, we do nothing | |
2561 | */ | |
2562 | ||
2563 | static void bc_destroy(struct rpc_xprt *xprt) | |
2564 | { | |
47f72efa KM |
2565 | dprintk("RPC: bc_destroy xprt %p\n", xprt); |
2566 | ||
2567 | xs_xprt_free(xprt); | |
2568 | module_put(THIS_MODULE); | |
4cfc7e60 RI |
2569 | } |
2570 | ||
176e21ee CL |
2571 | static struct rpc_xprt_ops xs_local_ops = { |
2572 | .reserve_xprt = xprt_reserve_xprt, | |
2573 | .release_xprt = xs_tcp_release_xprt, | |
f39c1bfb | 2574 | .alloc_slot = xprt_alloc_slot, |
176e21ee CL |
2575 | .rpcbind = xs_local_rpcbind, |
2576 | .set_port = xs_local_set_port, | |
dc107402 | 2577 | .connect = xs_local_connect, |
176e21ee CL |
2578 | .buf_alloc = rpc_malloc, |
2579 | .buf_free = rpc_free, | |
2580 | .send_request = xs_local_send_request, | |
2581 | .set_retrans_timeout = xprt_set_retrans_timeout_def, | |
2582 | .close = xs_close, | |
a1311d87 | 2583 | .destroy = xs_destroy, |
176e21ee CL |
2584 | .print_stats = xs_local_print_stats, |
2585 | }; | |
2586 | ||
262965f5 | 2587 | static struct rpc_xprt_ops xs_udp_ops = { |
43118c29 | 2588 | .set_buffer_size = xs_udp_set_buffer_size, |
12a80469 | 2589 | .reserve_xprt = xprt_reserve_xprt_cong, |
49e9a890 | 2590 | .release_xprt = xprt_release_xprt_cong, |
f39c1bfb | 2591 | .alloc_slot = xprt_alloc_slot, |
45160d62 | 2592 | .rpcbind = rpcb_getport_async, |
92200412 | 2593 | .set_port = xs_set_port, |
262965f5 | 2594 | .connect = xs_connect, |
02107148 CL |
2595 | .buf_alloc = rpc_malloc, |
2596 | .buf_free = rpc_free, | |
262965f5 | 2597 | .send_request = xs_udp_send_request, |
fe3aca29 | 2598 | .set_retrans_timeout = xprt_set_retrans_timeout_rtt, |
46c0ee8b | 2599 | .timer = xs_udp_timer, |
a58dd398 | 2600 | .release_request = xprt_release_rqst_cong, |
262965f5 CL |
2601 | .close = xs_close, |
2602 | .destroy = xs_destroy, | |
262ca07d | 2603 | .print_stats = xs_udp_print_stats, |
262965f5 CL |
2604 | }; |
2605 | ||
2606 | static struct rpc_xprt_ops xs_tcp_ops = { | |
12a80469 | 2607 | .reserve_xprt = xprt_reserve_xprt, |
e0ab53de | 2608 | .release_xprt = xs_tcp_release_xprt, |
f39c1bfb | 2609 | .alloc_slot = xprt_lock_and_alloc_slot, |
45160d62 | 2610 | .rpcbind = rpcb_getport_async, |
92200412 | 2611 | .set_port = xs_set_port, |
0b9e7943 | 2612 | .connect = xs_connect, |
02107148 CL |
2613 | .buf_alloc = rpc_malloc, |
2614 | .buf_free = rpc_free, | |
262965f5 | 2615 | .send_request = xs_tcp_send_request, |
fe3aca29 | 2616 | .set_retrans_timeout = xprt_set_retrans_timeout_def, |
f75e6745 | 2617 | .close = xs_tcp_close, |
9903cd1c | 2618 | .destroy = xs_destroy, |
262ca07d | 2619 | .print_stats = xs_tcp_print_stats, |
a246b010 CL |
2620 | }; |
2621 | ||
4cfc7e60 RI |
2622 | /* |
2623 | * The rpc_xprt_ops for the server backchannel | |
2624 | */ | |
2625 | ||
2626 | static struct rpc_xprt_ops bc_tcp_ops = { | |
2627 | .reserve_xprt = xprt_reserve_xprt, | |
2628 | .release_xprt = xprt_release_xprt, | |
84e28a30 | 2629 | .alloc_slot = xprt_alloc_slot, |
4cfc7e60 RI |
2630 | .buf_alloc = bc_malloc, |
2631 | .buf_free = bc_free, | |
2632 | .send_request = bc_send_request, | |
2633 | .set_retrans_timeout = xprt_set_retrans_timeout_def, | |
2634 | .close = bc_close, | |
2635 | .destroy = bc_destroy, | |
2636 | .print_stats = xs_tcp_print_stats, | |
2637 | }; | |
2638 | ||
92476850 CL |
2639 | static int xs_init_anyaddr(const int family, struct sockaddr *sap) |
2640 | { | |
2641 | static const struct sockaddr_in sin = { | |
2642 | .sin_family = AF_INET, | |
2643 | .sin_addr.s_addr = htonl(INADDR_ANY), | |
2644 | }; | |
2645 | static const struct sockaddr_in6 sin6 = { | |
2646 | .sin6_family = AF_INET6, | |
2647 | .sin6_addr = IN6ADDR_ANY_INIT, | |
2648 | }; | |
2649 | ||
2650 | switch (family) { | |
176e21ee CL |
2651 | case AF_LOCAL: |
2652 | break; | |
92476850 CL |
2653 | case AF_INET: |
2654 | memcpy(sap, &sin, sizeof(sin)); | |
2655 | break; | |
2656 | case AF_INET6: | |
2657 | memcpy(sap, &sin6, sizeof(sin6)); | |
2658 | break; | |
2659 | default: | |
2660 | dprintk("RPC: %s: Bad address family\n", __func__); | |
2661 | return -EAFNOSUPPORT; | |
2662 | } | |
2663 | return 0; | |
2664 | } | |
2665 | ||
3c341b0b | 2666 | static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, |
d9ba131d TM |
2667 | unsigned int slot_table_size, |
2668 | unsigned int max_slot_table_size) | |
c8541ecd CL |
2669 | { |
2670 | struct rpc_xprt *xprt; | |
ffc2e518 | 2671 | struct sock_xprt *new; |
c8541ecd | 2672 | |
96802a09 | 2673 | if (args->addrlen > sizeof(xprt->addr)) { |
46121cf7 | 2674 | dprintk("RPC: xs_setup_xprt: address too large\n"); |
c8541ecd CL |
2675 | return ERR_PTR(-EBADF); |
2676 | } | |
2677 | ||
d9ba131d TM |
2678 | xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, |
2679 | max_slot_table_size); | |
bd1722d4 | 2680 | if (xprt == NULL) { |
46121cf7 CL |
2681 | dprintk("RPC: xs_setup_xprt: couldn't allocate " |
2682 | "rpc_xprt\n"); | |
c8541ecd CL |
2683 | return ERR_PTR(-ENOMEM); |
2684 | } | |
c8541ecd | 2685 | |
bd1722d4 | 2686 | new = container_of(xprt, struct sock_xprt, xprt); |
96802a09 FM |
2687 | memcpy(&xprt->addr, args->dstaddr, args->addrlen); |
2688 | xprt->addrlen = args->addrlen; | |
d3bc9a1d | 2689 | if (args->srcaddr) |
fbfffbd5 | 2690 | memcpy(&new->srcaddr, args->srcaddr, args->addrlen); |
92476850 CL |
2691 | else { |
2692 | int err; | |
2693 | err = xs_init_anyaddr(args->dstaddr->sa_family, | |
2694 | (struct sockaddr *)&new->srcaddr); | |
2aa13531 SK |
2695 | if (err != 0) { |
2696 | xprt_free(xprt); | |
92476850 | 2697 | return ERR_PTR(err); |
2aa13531 | 2698 | } |
92476850 | 2699 | } |
c8541ecd CL |
2700 | |
2701 | return xprt; | |
2702 | } | |
2703 | ||
176e21ee CL |
2704 | static const struct rpc_timeout xs_local_default_timeout = { |
2705 | .to_initval = 10 * HZ, | |
2706 | .to_maxval = 10 * HZ, | |
2707 | .to_retries = 2, | |
2708 | }; | |
2709 | ||
2710 | /** | |
2711 | * xs_setup_local - Set up transport to use an AF_LOCAL socket | |
2712 | * @args: rpc transport creation arguments | |
2713 | * | |
2714 | * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP | |
2715 | */ | |
2716 | static struct rpc_xprt *xs_setup_local(struct xprt_create *args) | |
2717 | { | |
2718 | struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; | |
2719 | struct sock_xprt *transport; | |
2720 | struct rpc_xprt *xprt; | |
2721 | struct rpc_xprt *ret; | |
2722 | ||
d9ba131d TM |
2723 | xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, |
2724 | xprt_max_tcp_slot_table_entries); | |
176e21ee CL |
2725 | if (IS_ERR(xprt)) |
2726 | return xprt; | |
2727 | transport = container_of(xprt, struct sock_xprt, xprt); | |
2728 | ||
2729 | xprt->prot = 0; | |
2730 | xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); | |
2731 | xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; | |
2732 | ||
2733 | xprt->bind_timeout = XS_BIND_TO; | |
2734 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; | |
2735 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
2736 | ||
2737 | xprt->ops = &xs_local_ops; | |
2738 | xprt->timeout = &xs_local_default_timeout; | |
2739 | ||
93dc41bd N |
2740 | INIT_DELAYED_WORK(&transport->connect_worker, |
2741 | xs_dummy_setup_socket); | |
2742 | ||
176e21ee CL |
2743 | switch (sun->sun_family) { |
2744 | case AF_LOCAL: | |
2745 | if (sun->sun_path[0] != '/') { | |
2746 | dprintk("RPC: bad AF_LOCAL address: %s\n", | |
2747 | sun->sun_path); | |
2748 | ret = ERR_PTR(-EINVAL); | |
2749 | goto out_err; | |
2750 | } | |
2751 | xprt_set_bound(xprt); | |
176e21ee | 2752 | xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); |
7073ea87 BF |
2753 | ret = ERR_PTR(xs_local_setup_socket(transport)); |
2754 | if (ret) | |
2755 | goto out_err; | |
176e21ee CL |
2756 | break; |
2757 | default: | |
2758 | ret = ERR_PTR(-EAFNOSUPPORT); | |
2759 | goto out_err; | |
2760 | } | |
2761 | ||
2762 | dprintk("RPC: set up xprt to %s via AF_LOCAL\n", | |
2763 | xprt->address_strings[RPC_DISPLAY_ADDR]); | |
2764 | ||
2765 | if (try_module_get(THIS_MODULE)) | |
2766 | return xprt; | |
2767 | ret = ERR_PTR(-EINVAL); | |
2768 | out_err: | |
315f3812 | 2769 | xs_xprt_free(xprt); |
176e21ee CL |
2770 | return ret; |
2771 | } | |
2772 | ||
2881ae74 TM |
2773 | static const struct rpc_timeout xs_udp_default_timeout = { |
2774 | .to_initval = 5 * HZ, | |
2775 | .to_maxval = 30 * HZ, | |
2776 | .to_increment = 5 * HZ, | |
2777 | .to_retries = 5, | |
2778 | }; | |
2779 | ||
9903cd1c CL |
2780 | /** |
2781 | * xs_setup_udp - Set up transport to use a UDP socket | |
96802a09 | 2782 | * @args: rpc transport creation arguments |
9903cd1c CL |
2783 | * |
2784 | */ | |
483066d6 | 2785 | static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) |
a246b010 | 2786 | { |
8f9d5b1a | 2787 | struct sockaddr *addr = args->dstaddr; |
c8541ecd | 2788 | struct rpc_xprt *xprt; |
c8475461 | 2789 | struct sock_xprt *transport; |
0a68b0be | 2790 | struct rpc_xprt *ret; |
a246b010 | 2791 | |
d9ba131d TM |
2792 | xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, |
2793 | xprt_udp_slot_table_entries); | |
c8541ecd CL |
2794 | if (IS_ERR(xprt)) |
2795 | return xprt; | |
c8475461 | 2796 | transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 | 2797 | |
ec739ef0 | 2798 | xprt->prot = IPPROTO_UDP; |
808012fb | 2799 | xprt->tsh_size = 0; |
a246b010 CL |
2800 | /* XXX: header size can vary due to auth type, IPv6, etc. */ |
2801 | xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); | |
2802 | ||
03bf4b70 | 2803 | xprt->bind_timeout = XS_BIND_TO; |
03bf4b70 CL |
2804 | xprt->reestablish_timeout = XS_UDP_REEST_TO; |
2805 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 2806 | |
262965f5 | 2807 | xprt->ops = &xs_udp_ops; |
a246b010 | 2808 | |
ba7392bb | 2809 | xprt->timeout = &xs_udp_default_timeout; |
a246b010 | 2810 | |
8f9d5b1a CL |
2811 | switch (addr->sa_family) { |
2812 | case AF_INET: | |
2813 | if (((struct sockaddr_in *)addr)->sin_port != htons(0)) | |
2814 | xprt_set_bound(xprt); | |
2815 | ||
2816 | INIT_DELAYED_WORK(&transport->connect_worker, | |
8c14ff2a | 2817 | xs_udp_setup_socket); |
9dc3b095 | 2818 | xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); |
8f9d5b1a CL |
2819 | break; |
2820 | case AF_INET6: | |
2821 | if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) | |
2822 | xprt_set_bound(xprt); | |
2823 | ||
2824 | INIT_DELAYED_WORK(&transport->connect_worker, | |
8c14ff2a | 2825 | xs_udp_setup_socket); |
9dc3b095 | 2826 | xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); |
8f9d5b1a CL |
2827 | break; |
2828 | default: | |
0a68b0be BF |
2829 | ret = ERR_PTR(-EAFNOSUPPORT); |
2830 | goto out_err; | |
8f9d5b1a CL |
2831 | } |
2832 | ||
c740eff8 CL |
2833 | if (xprt_bound(xprt)) |
2834 | dprintk("RPC: set up xprt to %s (port %s) via %s\n", | |
2835 | xprt->address_strings[RPC_DISPLAY_ADDR], | |
2836 | xprt->address_strings[RPC_DISPLAY_PORT], | |
2837 | xprt->address_strings[RPC_DISPLAY_PROTO]); | |
2838 | else | |
2839 | dprintk("RPC: set up xprt to %s (autobind) via %s\n", | |
2840 | xprt->address_strings[RPC_DISPLAY_ADDR], | |
2841 | xprt->address_strings[RPC_DISPLAY_PROTO]); | |
edb267a6 | 2842 | |
bc25571e TT |
2843 | if (try_module_get(THIS_MODULE)) |
2844 | return xprt; | |
0a68b0be BF |
2845 | ret = ERR_PTR(-EINVAL); |
2846 | out_err: | |
315f3812 | 2847 | xs_xprt_free(xprt); |
0a68b0be | 2848 | return ret; |
a246b010 CL |
2849 | } |
2850 | ||
2881ae74 TM |
2851 | static const struct rpc_timeout xs_tcp_default_timeout = { |
2852 | .to_initval = 60 * HZ, | |
2853 | .to_maxval = 60 * HZ, | |
2854 | .to_retries = 2, | |
2855 | }; | |
2856 | ||
9903cd1c CL |
2857 | /** |
2858 | * xs_setup_tcp - Set up transport to use a TCP socket | |
96802a09 | 2859 | * @args: rpc transport creation arguments |
9903cd1c CL |
2860 | * |
2861 | */ | |
483066d6 | 2862 | static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) |
a246b010 | 2863 | { |
8f9d5b1a | 2864 | struct sockaddr *addr = args->dstaddr; |
c8541ecd | 2865 | struct rpc_xprt *xprt; |
c8475461 | 2866 | struct sock_xprt *transport; |
0a68b0be | 2867 | struct rpc_xprt *ret; |
b7993ceb TM |
2868 | unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; |
2869 | ||
2870 | if (args->flags & XPRT_CREATE_INFINITE_SLOTS) | |
2871 | max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; | |
a246b010 | 2872 | |
d9ba131d | 2873 | xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, |
b7993ceb | 2874 | max_slot_table_size); |
c8541ecd CL |
2875 | if (IS_ERR(xprt)) |
2876 | return xprt; | |
c8475461 | 2877 | transport = container_of(xprt, struct sock_xprt, xprt); |
a246b010 | 2878 | |
ec739ef0 | 2879 | xprt->prot = IPPROTO_TCP; |
808012fb | 2880 | xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); |
808012fb | 2881 | xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; |
a246b010 | 2882 | |
03bf4b70 | 2883 | xprt->bind_timeout = XS_BIND_TO; |
03bf4b70 CL |
2884 | xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; |
2885 | xprt->idle_timeout = XS_IDLE_DISC_TO; | |
a246b010 | 2886 | |
262965f5 | 2887 | xprt->ops = &xs_tcp_ops; |
ba7392bb | 2888 | xprt->timeout = &xs_tcp_default_timeout; |
a246b010 | 2889 | |
8f9d5b1a CL |
2890 | switch (addr->sa_family) { |
2891 | case AF_INET: | |
2892 | if (((struct sockaddr_in *)addr)->sin_port != htons(0)) | |
2893 | xprt_set_bound(xprt); | |
2894 | ||
9dc3b095 | 2895 | INIT_DELAYED_WORK(&transport->connect_worker, |
cdd518d5 | 2896 | xs_tcp_setup_socket); |
9dc3b095 | 2897 | xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); |
8f9d5b1a CL |
2898 | break; |
2899 | case AF_INET6: | |
2900 | if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) | |
2901 | xprt_set_bound(xprt); | |
2902 | ||
9dc3b095 | 2903 | INIT_DELAYED_WORK(&transport->connect_worker, |
cdd518d5 | 2904 | xs_tcp_setup_socket); |
9dc3b095 | 2905 | xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); |
8f9d5b1a CL |
2906 | break; |
2907 | default: | |
0a68b0be BF |
2908 | ret = ERR_PTR(-EAFNOSUPPORT); |
2909 | goto out_err; | |
8f9d5b1a CL |
2910 | } |
2911 | ||
c740eff8 CL |
2912 | if (xprt_bound(xprt)) |
2913 | dprintk("RPC: set up xprt to %s (port %s) via %s\n", | |
2914 | xprt->address_strings[RPC_DISPLAY_ADDR], | |
2915 | xprt->address_strings[RPC_DISPLAY_PORT], | |
2916 | xprt->address_strings[RPC_DISPLAY_PROTO]); | |
2917 | else | |
2918 | dprintk("RPC: set up xprt to %s (autobind) via %s\n", | |
2919 | xprt->address_strings[RPC_DISPLAY_ADDR], | |
2920 | xprt->address_strings[RPC_DISPLAY_PROTO]); | |
2921 | ||
bc25571e TT |
2922 | if (try_module_get(THIS_MODULE)) |
2923 | return xprt; | |
0a68b0be BF |
2924 | ret = ERR_PTR(-EINVAL); |
2925 | out_err: | |
315f3812 | 2926 | xs_xprt_free(xprt); |
0a68b0be | 2927 | return ret; |
a246b010 | 2928 | } |
282b32e1 | 2929 | |
f300baba AB |
2930 | /** |
2931 | * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket | |
2932 | * @args: rpc transport creation arguments | |
2933 | * | |
2934 | */ | |
2935 | static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) | |
2936 | { | |
2937 | struct sockaddr *addr = args->dstaddr; | |
2938 | struct rpc_xprt *xprt; | |
2939 | struct sock_xprt *transport; | |
2940 | struct svc_sock *bc_sock; | |
0a68b0be | 2941 | struct rpc_xprt *ret; |
f300baba | 2942 | |
d9ba131d TM |
2943 | xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, |
2944 | xprt_tcp_slot_table_entries); | |
f300baba AB |
2945 | if (IS_ERR(xprt)) |
2946 | return xprt; | |
2947 | transport = container_of(xprt, struct sock_xprt, xprt); | |
2948 | ||
2949 | xprt->prot = IPPROTO_TCP; | |
2950 | xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); | |
2951 | xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; | |
2952 | xprt->timeout = &xs_tcp_default_timeout; | |
2953 | ||
2954 | /* backchannel */ | |
2955 | xprt_set_bound(xprt); | |
2956 | xprt->bind_timeout = 0; | |
f300baba AB |
2957 | xprt->reestablish_timeout = 0; |
2958 | xprt->idle_timeout = 0; | |
2959 | ||
f300baba AB |
2960 | xprt->ops = &bc_tcp_ops; |
2961 | ||
2962 | switch (addr->sa_family) { | |
2963 | case AF_INET: | |
2964 | xs_format_peer_addresses(xprt, "tcp", | |
2965 | RPCBIND_NETID_TCP); | |
2966 | break; | |
2967 | case AF_INET6: | |
2968 | xs_format_peer_addresses(xprt, "tcp", | |
2969 | RPCBIND_NETID_TCP6); | |
2970 | break; | |
2971 | default: | |
0a68b0be BF |
2972 | ret = ERR_PTR(-EAFNOSUPPORT); |
2973 | goto out_err; | |
f300baba AB |
2974 | } |
2975 | ||
50fa0d40 PE |
2976 | dprintk("RPC: set up xprt to %s (port %s) via %s\n", |
2977 | xprt->address_strings[RPC_DISPLAY_ADDR], | |
2978 | xprt->address_strings[RPC_DISPLAY_PORT], | |
2979 | xprt->address_strings[RPC_DISPLAY_PROTO]); | |
f300baba | 2980 | |
99de8ea9 BF |
2981 | /* |
2982 | * Once we've associated a backchannel xprt with a connection, | |
28303ca3 WM |
2983 | * we want to keep it around as long as the connection lasts, |
2984 | * in case we need to start using it for a backchannel again; | |
2985 | * this reference won't be dropped until bc_xprt is destroyed. | |
99de8ea9 BF |
2986 | */ |
2987 | xprt_get(xprt); | |
2988 | args->bc_xprt->xpt_bc_xprt = xprt; | |
2989 | xprt->bc_xprt = args->bc_xprt; | |
2990 | bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); | |
2991 | transport->sock = bc_sock->sk_sock; | |
2992 | transport->inet = bc_sock->sk_sk; | |
2993 | ||
f300baba AB |
2994 | /* |
2995 | * Since we don't want connections for the backchannel, we set | |
2996 | * the xprt status to connected | |
2997 | */ | |
2998 | xprt_set_connected(xprt); | |
2999 | ||
f300baba AB |
3000 | if (try_module_get(THIS_MODULE)) |
3001 | return xprt; | |
642aab58 KM |
3002 | |
3003 | args->bc_xprt->xpt_bc_xprt = NULL; | |
99de8ea9 | 3004 | xprt_put(xprt); |
0a68b0be BF |
3005 | ret = ERR_PTR(-EINVAL); |
3006 | out_err: | |
315f3812 | 3007 | xs_xprt_free(xprt); |
0a68b0be | 3008 | return ret; |
f300baba AB |
3009 | } |
3010 | ||
176e21ee CL |
3011 | static struct xprt_class xs_local_transport = { |
3012 | .list = LIST_HEAD_INIT(xs_local_transport.list), | |
3013 | .name = "named UNIX socket", | |
3014 | .owner = THIS_MODULE, | |
3015 | .ident = XPRT_TRANSPORT_LOCAL, | |
3016 | .setup = xs_setup_local, | |
3017 | }; | |
3018 | ||
bc25571e TT |
3019 | static struct xprt_class xs_udp_transport = { |
3020 | .list = LIST_HEAD_INIT(xs_udp_transport.list), | |
3021 | .name = "udp", | |
3022 | .owner = THIS_MODULE, | |
f300baba | 3023 | .ident = XPRT_TRANSPORT_UDP, |
bc25571e TT |
3024 | .setup = xs_setup_udp, |
3025 | }; | |
3026 | ||
3027 | static struct xprt_class xs_tcp_transport = { | |
3028 | .list = LIST_HEAD_INIT(xs_tcp_transport.list), | |
3029 | .name = "tcp", | |
3030 | .owner = THIS_MODULE, | |
f300baba | 3031 | .ident = XPRT_TRANSPORT_TCP, |
bc25571e TT |
3032 | .setup = xs_setup_tcp, |
3033 | }; | |
3034 | ||
f300baba AB |
3035 | static struct xprt_class xs_bc_tcp_transport = { |
3036 | .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), | |
3037 | .name = "tcp NFSv4.1 backchannel", | |
3038 | .owner = THIS_MODULE, | |
3039 | .ident = XPRT_TRANSPORT_BC_TCP, | |
3040 | .setup = xs_setup_bc_tcp, | |
3041 | }; | |
3042 | ||
282b32e1 | 3043 | /** |
bc25571e | 3044 | * init_socket_xprt - set up xprtsock's sysctls, register with RPC client |
282b32e1 CL |
3045 | * |
3046 | */ | |
3047 | int init_socket_xprt(void) | |
3048 | { | |
fbf76683 | 3049 | #ifdef RPC_DEBUG |
2b1bec5f | 3050 | if (!sunrpc_table_header) |
0b4d4147 | 3051 | sunrpc_table_header = register_sysctl_table(sunrpc_table); |
fbf76683 CL |
3052 | #endif |
3053 | ||
176e21ee | 3054 | xprt_register_transport(&xs_local_transport); |
bc25571e TT |
3055 | xprt_register_transport(&xs_udp_transport); |
3056 | xprt_register_transport(&xs_tcp_transport); | |
f300baba | 3057 | xprt_register_transport(&xs_bc_tcp_transport); |
bc25571e | 3058 | |
282b32e1 CL |
3059 | return 0; |
3060 | } | |
3061 | ||
3062 | /** | |
bc25571e | 3063 | * cleanup_socket_xprt - remove xprtsock's sysctls, unregister |
282b32e1 CL |
3064 | * |
3065 | */ | |
3066 | void cleanup_socket_xprt(void) | |
3067 | { | |
fbf76683 CL |
3068 | #ifdef RPC_DEBUG |
3069 | if (sunrpc_table_header) { | |
3070 | unregister_sysctl_table(sunrpc_table_header); | |
3071 | sunrpc_table_header = NULL; | |
3072 | } | |
3073 | #endif | |
bc25571e | 3074 | |
176e21ee | 3075 | xprt_unregister_transport(&xs_local_transport); |
bc25571e TT |
3076 | xprt_unregister_transport(&xs_udp_transport); |
3077 | xprt_unregister_transport(&xs_tcp_transport); | |
f300baba | 3078 | xprt_unregister_transport(&xs_bc_tcp_transport); |
282b32e1 | 3079 | } |
cbf11071 | 3080 | |
9bbb9e5a RR |
3081 | static int param_set_uint_minmax(const char *val, |
3082 | const struct kernel_param *kp, | |
cbf11071 TM |
3083 | unsigned int min, unsigned int max) |
3084 | { | |
00cfaa94 | 3085 | unsigned int num; |
cbf11071 TM |
3086 | int ret; |
3087 | ||
3088 | if (!val) | |
3089 | return -EINVAL; | |
00cfaa94 | 3090 | ret = kstrtouint(val, 0, &num); |
cbf11071 TM |
3091 | if (ret == -EINVAL || num < min || num > max) |
3092 | return -EINVAL; | |
3093 | *((unsigned int *)kp->arg) = num; | |
3094 | return 0; | |
3095 | } | |
3096 | ||
9bbb9e5a | 3097 | static int param_set_portnr(const char *val, const struct kernel_param *kp) |
cbf11071 TM |
3098 | { |
3099 | return param_set_uint_minmax(val, kp, | |
3100 | RPC_MIN_RESVPORT, | |
3101 | RPC_MAX_RESVPORT); | |
3102 | } | |
3103 | ||
9bbb9e5a RR |
3104 | static struct kernel_param_ops param_ops_portnr = { |
3105 | .set = param_set_portnr, | |
3106 | .get = param_get_uint, | |
3107 | }; | |
3108 | ||
cbf11071 TM |
3109 | #define param_check_portnr(name, p) \ |
3110 | __param_check(name, p, unsigned int); | |
3111 | ||
3112 | module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); | |
3113 | module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); | |
3114 | ||
9bbb9e5a RR |
3115 | static int param_set_slot_table_size(const char *val, |
3116 | const struct kernel_param *kp) | |
cbf11071 TM |
3117 | { |
3118 | return param_set_uint_minmax(val, kp, | |
3119 | RPC_MIN_SLOT_TABLE, | |
3120 | RPC_MAX_SLOT_TABLE); | |
3121 | } | |
3122 | ||
9bbb9e5a RR |
3123 | static struct kernel_param_ops param_ops_slot_table_size = { |
3124 | .set = param_set_slot_table_size, | |
3125 | .get = param_get_uint, | |
3126 | }; | |
3127 | ||
cbf11071 TM |
3128 | #define param_check_slot_table_size(name, p) \ |
3129 | __param_check(name, p, unsigned int); | |
3130 | ||
d9ba131d TM |
3131 | static int param_set_max_slot_table_size(const char *val, |
3132 | const struct kernel_param *kp) | |
3133 | { | |
3134 | return param_set_uint_minmax(val, kp, | |
3135 | RPC_MIN_SLOT_TABLE, | |
3136 | RPC_MAX_SLOT_TABLE_LIMIT); | |
3137 | } | |
3138 | ||
3139 | static struct kernel_param_ops param_ops_max_slot_table_size = { | |
3140 | .set = param_set_max_slot_table_size, | |
3141 | .get = param_get_uint, | |
3142 | }; | |
3143 | ||
3144 | #define param_check_max_slot_table_size(name, p) \ | |
3145 | __param_check(name, p, unsigned int); | |
3146 | ||
cbf11071 TM |
3147 | module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, |
3148 | slot_table_size, 0644); | |
d9ba131d TM |
3149 | module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, |
3150 | max_slot_table_size, 0644); | |
cbf11071 TM |
3151 | module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, |
3152 | slot_table_size, 0644); | |
3153 |