1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
30 #include <linux/in.h>
31 #include <linux/net.h>
32 #include <linux/mm.h>
33 #include <linux/un.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
44 #endif
45
46 #include <net/sock.h>
47 #include <net/checksum.h>
48 #include <net/udp.h>
49 #include <net/tcp.h>
50 #include <linux/bvec.h>
51 #include <linux/highmem.h>
52 #include <linux/uio.h>
53 #include <linux/sched/mm.h>
54
55 #include <trace/events/sunrpc.h>
56
57 #include "socklib.h"
58 #include "sunrpc.h"
59
60 static void xs_close(struct rpc_xprt *xprt);
61 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
62 struct socket *sock);
63
64 /*
65 * xprtsock tunables
66 */
67 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
70
71 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
73
74 #define XS_TCP_LINGER_TO (15U * HZ)
75 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
76
77 /*
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
81 *
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
84 */
85
86 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91
92 static struct ctl_table_header *sunrpc_table_header;
93
94 /*
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
97 */
98 static struct ctl_table xs_tunables_table[] = {
99 {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
108 {
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
116 },
117 {
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
125 },
126 {
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
135 {
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
143 },
144 {
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_jiffies,
150 },
151 { },
152 };
153
154 static struct ctl_table sunrpc_table[] = {
155 {
156 .procname = "sunrpc",
157 .mode = 0555,
158 .child = xs_tunables_table
159 },
160 { },
161 };
162
163 /*
164 * Wait duration for a reply from the RPC portmapper.
165 */
166 #define XS_BIND_TO (60U * HZ)
167
168 /*
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
171 */
172 #define XS_UDP_REEST_TO (2U * HZ)
173
174 /*
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
177 *
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
182 */
183 #define XS_TCP_INIT_REEST_TO (3U * HZ)
184
185 /*
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
189 */
190 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
191
192 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193 # undef RPC_DEBUG_DATA
194 # define RPCDBG_FACILITY RPCDBG_TRANS
195 #endif
196
197 #ifdef RPC_DEBUG_DATA
xs_pktdump(char * msg,u32 * packet,unsigned int count)198 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
199 {
200 u8 *buf = (u8 *) packet;
201 int j;
202
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
205 if (!(j & 31)) {
206 if (j)
207 dprintk("\n");
208 dprintk("0x%04x ", j);
209 }
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 }
213 dprintk("\n");
214 }
215 #else
xs_pktdump(char * msg,u32 * packet,unsigned int count)216 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217 {
218 /* NOP */
219 }
220 #endif
221
xprt_from_sock(struct sock * sk)222 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
223 {
224 return (struct rpc_xprt *) sk->sk_user_data;
225 }
226
xs_addr(struct rpc_xprt * xprt)227 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
228 {
229 return (struct sockaddr *) &xprt->addr;
230 }
231
xs_addr_un(struct rpc_xprt * xprt)232 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
233 {
234 return (struct sockaddr_un *) &xprt->addr;
235 }
236
xs_addr_in(struct rpc_xprt * xprt)237 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
238 {
239 return (struct sockaddr_in *) &xprt->addr;
240 }
241
xs_addr_in6(struct rpc_xprt * xprt)242 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
243 {
244 return (struct sockaddr_in6 *) &xprt->addr;
245 }
246
xs_format_common_peer_addresses(struct rpc_xprt * xprt)247 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
248 {
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
253 char buf[128];
254
255 switch (sap->sa_family) {
256 case AF_LOCAL:
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
261 break;
262 case AF_INET:
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
268 break;
269 case AF_INET6:
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
275 break;
276 default:
277 BUG();
278 }
279
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
281 }
282
xs_format_common_peer_ports(struct rpc_xprt * xprt)283 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
284 {
285 struct sockaddr *sap = xs_addr(xprt);
286 char buf[128];
287
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
290
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
293 }
294
xs_format_peer_addresses(struct rpc_xprt * xprt,const char * protocol,const char * netid)295 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
297 const char *netid)
298 {
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
303 }
304
xs_update_peer_port(struct rpc_xprt * xprt)305 static void xs_update_peer_port(struct rpc_xprt *xprt)
306 {
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
309
310 xs_format_common_peer_ports(xprt);
311 }
312
xs_free_peer_addresses(struct rpc_xprt * xprt)313 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314 {
315 unsigned int i;
316
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 switch (i) {
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
321 continue;
322 default:
323 kfree(xprt->address_strings[i]);
324 }
325 }
326
327 static size_t
xs_alloc_sparse_pages(struct xdr_buf * buf,size_t want,gfp_t gfp)328 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
329 {
330 size_t i,n;
331
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 return want;
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
336 if (buf->pages[i])
337 continue;
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
340 i *= PAGE_SIZE;
341 return i > buf->page_base ? i - buf->page_base : 0;
342 }
343 }
344 return want;
345 }
346
347 static ssize_t
xs_sock_recvmsg(struct socket * sock,struct msghdr * msg,int flags,size_t seek)348 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
349 {
350 ssize_t ret;
351 if (seek != 0)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
355 }
356
357 static ssize_t
xs_read_kvec(struct socket * sock,struct msghdr * msg,int flags,struct kvec * kvec,size_t count,size_t seek)358 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
360 {
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
363 }
364
365 static ssize_t
xs_read_bvec(struct socket * sock,struct msghdr * msg,int flags,struct bio_vec * bvec,unsigned long nr,size_t count,size_t seek)366 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
368 size_t seek)
369 {
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
372 }
373
374 static ssize_t
xs_read_discard(struct socket * sock,struct msghdr * msg,int flags,size_t count)375 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
376 size_t count)
377 {
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
380 }
381
382 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383 static void
xs_flush_bvec(const struct bio_vec * bvec,size_t count,size_t seek)384 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
385 {
386 struct bvec_iter bi = {
387 .bi_size = count,
388 };
389 struct bio_vec bv;
390
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
394 }
395 #else
396 static inline void
xs_flush_bvec(const struct bio_vec * bvec,size_t count,size_t seek)397 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
398 {
399 }
400 #endif
401
402 static ssize_t
xs_read_xdr_buf(struct socket * sock,struct msghdr * msg,int flags,struct xdr_buf * buf,size_t count,size_t seek,size_t * read)403 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
405 {
406 size_t want, seek_init = seek, offset = 0;
407 ssize_t ret;
408
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 if (seek < want) {
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
412 if (ret <= 0)
413 goto sock_err;
414 offset += ret;
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
416 goto out;
417 if (ret != want)
418 goto out;
419 seek = 0;
420 } else {
421 seek -= want;
422 offset += want;
423 }
424
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
427 GFP_KERNEL);
428 if (seek < want) {
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
433 if (ret <= 0)
434 goto sock_err;
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 ret -= buf->page_base;
437 offset += ret;
438 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
439 goto out;
440 if (ret != want)
441 goto out;
442 seek = 0;
443 } else {
444 seek -= want;
445 offset += want;
446 }
447
448 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
449 if (seek < want) {
450 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
451 if (ret <= 0)
452 goto sock_err;
453 offset += ret;
454 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
455 goto out;
456 if (ret != want)
457 goto out;
458 } else if (offset < seek_init)
459 offset = seek_init;
460 ret = -EMSGSIZE;
461 out:
462 *read = offset - seek_init;
463 return ret;
464 sock_err:
465 offset += seek;
466 goto out;
467 }
468
469 static void
xs_read_header(struct sock_xprt * transport,struct xdr_buf * buf)470 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
471 {
472 if (!transport->recv.copied) {
473 if (buf->head[0].iov_len >= transport->recv.offset)
474 memcpy(buf->head[0].iov_base,
475 &transport->recv.xid,
476 transport->recv.offset);
477 transport->recv.copied = transport->recv.offset;
478 }
479 }
480
481 static bool
xs_read_stream_request_done(struct sock_xprt * transport)482 xs_read_stream_request_done(struct sock_xprt *transport)
483 {
484 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
485 }
486
487 static void
xs_read_stream_check_eor(struct sock_xprt * transport,struct msghdr * msg)488 xs_read_stream_check_eor(struct sock_xprt *transport,
489 struct msghdr *msg)
490 {
491 if (xs_read_stream_request_done(transport))
492 msg->msg_flags |= MSG_EOR;
493 }
494
495 static ssize_t
xs_read_stream_request(struct sock_xprt * transport,struct msghdr * msg,int flags,struct rpc_rqst * req)496 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
497 int flags, struct rpc_rqst *req)
498 {
499 struct xdr_buf *buf = &req->rq_private_buf;
500 size_t want, read;
501 ssize_t ret;
502
503 xs_read_header(transport, buf);
504
505 want = transport->recv.len - transport->recv.offset;
506 if (want != 0) {
507 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
508 transport->recv.copied + want,
509 transport->recv.copied,
510 &read);
511 transport->recv.offset += read;
512 transport->recv.copied += read;
513 }
514
515 if (transport->recv.offset == transport->recv.len)
516 xs_read_stream_check_eor(transport, msg);
517
518 if (want == 0)
519 return 0;
520
521 switch (ret) {
522 default:
523 break;
524 case -EFAULT:
525 case -EMSGSIZE:
526 msg->msg_flags |= MSG_TRUNC;
527 return read;
528 case 0:
529 return -ESHUTDOWN;
530 }
531 return ret < 0 ? ret : read;
532 }
533
534 static size_t
xs_read_stream_headersize(bool isfrag)535 xs_read_stream_headersize(bool isfrag)
536 {
537 if (isfrag)
538 return sizeof(__be32);
539 return 3 * sizeof(__be32);
540 }
541
542 static ssize_t
xs_read_stream_header(struct sock_xprt * transport,struct msghdr * msg,int flags,size_t want,size_t seek)543 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
544 int flags, size_t want, size_t seek)
545 {
546 struct kvec kvec = {
547 .iov_base = &transport->recv.fraghdr,
548 .iov_len = want,
549 };
550 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
551 }
552
553 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
554 static ssize_t
xs_read_stream_call(struct sock_xprt * transport,struct msghdr * msg,int flags)555 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
556 {
557 struct rpc_xprt *xprt = &transport->xprt;
558 struct rpc_rqst *req;
559 ssize_t ret;
560
561 /* Look up and lock the request corresponding to the given XID */
562 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
563 if (!req) {
564 printk(KERN_WARNING "Callback slot table overflowed\n");
565 return -ESHUTDOWN;
566 }
567 if (transport->recv.copied && !req->rq_private_buf.len)
568 return -ESHUTDOWN;
569
570 ret = xs_read_stream_request(transport, msg, flags, req);
571 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
572 xprt_complete_bc_request(req, transport->recv.copied);
573 else
574 req->rq_private_buf.len = transport->recv.copied;
575
576 return ret;
577 }
578 #else /* CONFIG_SUNRPC_BACKCHANNEL */
579 static ssize_t
xs_read_stream_call(struct sock_xprt * transport,struct msghdr * msg,int flags)580 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
581 {
582 return -ESHUTDOWN;
583 }
584 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
585
586 static ssize_t
xs_read_stream_reply(struct sock_xprt * transport,struct msghdr * msg,int flags)587 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
588 {
589 struct rpc_xprt *xprt = &transport->xprt;
590 struct rpc_rqst *req;
591 ssize_t ret = 0;
592
593 /* Look up and lock the request corresponding to the given XID */
594 spin_lock(&xprt->queue_lock);
595 req = xprt_lookup_rqst(xprt, transport->recv.xid);
596 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
597 msg->msg_flags |= MSG_TRUNC;
598 goto out;
599 }
600 xprt_pin_rqst(req);
601 spin_unlock(&xprt->queue_lock);
602
603 ret = xs_read_stream_request(transport, msg, flags, req);
604
605 spin_lock(&xprt->queue_lock);
606 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
607 xprt_complete_rqst(req->rq_task, transport->recv.copied);
608 else
609 req->rq_private_buf.len = transport->recv.copied;
610 xprt_unpin_rqst(req);
611 out:
612 spin_unlock(&xprt->queue_lock);
613 return ret;
614 }
615
616 static ssize_t
xs_read_stream(struct sock_xprt * transport,int flags)617 xs_read_stream(struct sock_xprt *transport, int flags)
618 {
619 struct msghdr msg = { 0 };
620 size_t want, read = 0;
621 ssize_t ret = 0;
622
623 if (transport->recv.len == 0) {
624 want = xs_read_stream_headersize(transport->recv.copied != 0);
625 ret = xs_read_stream_header(transport, &msg, flags, want,
626 transport->recv.offset);
627 if (ret <= 0)
628 goto out_err;
629 transport->recv.offset = ret;
630 if (transport->recv.offset != want)
631 return transport->recv.offset;
632 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
633 RPC_FRAGMENT_SIZE_MASK;
634 transport->recv.offset -= sizeof(transport->recv.fraghdr);
635 read = ret;
636 }
637
638 switch (be32_to_cpu(transport->recv.calldir)) {
639 default:
640 msg.msg_flags |= MSG_TRUNC;
641 break;
642 case RPC_CALL:
643 ret = xs_read_stream_call(transport, &msg, flags);
644 break;
645 case RPC_REPLY:
646 ret = xs_read_stream_reply(transport, &msg, flags);
647 }
648 if (msg.msg_flags & MSG_TRUNC) {
649 transport->recv.calldir = cpu_to_be32(-1);
650 transport->recv.copied = -1;
651 }
652 if (ret < 0)
653 goto out_err;
654 read += ret;
655 if (transport->recv.offset < transport->recv.len) {
656 if (!(msg.msg_flags & MSG_TRUNC))
657 return read;
658 msg.msg_flags = 0;
659 ret = xs_read_discard(transport->sock, &msg, flags,
660 transport->recv.len - transport->recv.offset);
661 if (ret <= 0)
662 goto out_err;
663 transport->recv.offset += ret;
664 read += ret;
665 if (transport->recv.offset != transport->recv.len)
666 return read;
667 }
668 if (xs_read_stream_request_done(transport)) {
669 trace_xs_stream_read_request(transport);
670 transport->recv.copied = 0;
671 }
672 transport->recv.offset = 0;
673 transport->recv.len = 0;
674 return read;
675 out_err:
676 return ret != 0 ? ret : -ESHUTDOWN;
677 }
678
xs_poll_socket(struct sock_xprt * transport)679 static __poll_t xs_poll_socket(struct sock_xprt *transport)
680 {
681 return transport->sock->ops->poll(transport->file, transport->sock,
682 NULL);
683 }
684
xs_poll_socket_readable(struct sock_xprt * transport)685 static bool xs_poll_socket_readable(struct sock_xprt *transport)
686 {
687 __poll_t events = xs_poll_socket(transport);
688
689 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
690 }
691
xs_poll_check_readable(struct sock_xprt * transport)692 static void xs_poll_check_readable(struct sock_xprt *transport)
693 {
694
695 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
696 if (!xs_poll_socket_readable(transport))
697 return;
698 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
699 queue_work(xprtiod_workqueue, &transport->recv_worker);
700 }
701
xs_stream_data_receive(struct sock_xprt * transport)702 static void xs_stream_data_receive(struct sock_xprt *transport)
703 {
704 size_t read = 0;
705 ssize_t ret = 0;
706
707 mutex_lock(&transport->recv_mutex);
708 if (transport->sock == NULL)
709 goto out;
710 for (;;) {
711 ret = xs_read_stream(transport, MSG_DONTWAIT);
712 if (ret < 0)
713 break;
714 read += ret;
715 cond_resched();
716 }
717 if (ret == -ESHUTDOWN)
718 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
719 else
720 xs_poll_check_readable(transport);
721 out:
722 mutex_unlock(&transport->recv_mutex);
723 trace_xs_stream_read_data(&transport->xprt, ret, read);
724 }
725
xs_stream_data_receive_workfn(struct work_struct * work)726 static void xs_stream_data_receive_workfn(struct work_struct *work)
727 {
728 struct sock_xprt *transport =
729 container_of(work, struct sock_xprt, recv_worker);
730 unsigned int pflags = memalloc_nofs_save();
731
732 xs_stream_data_receive(transport);
733 memalloc_nofs_restore(pflags);
734 }
735
736 static void
xs_stream_reset_connect(struct sock_xprt * transport)737 xs_stream_reset_connect(struct sock_xprt *transport)
738 {
739 transport->recv.offset = 0;
740 transport->recv.len = 0;
741 transport->recv.copied = 0;
742 transport->xmit.offset = 0;
743 }
744
745 static void
xs_stream_start_connect(struct sock_xprt * transport)746 xs_stream_start_connect(struct sock_xprt *transport)
747 {
748 transport->xprt.stat.connect_count++;
749 transport->xprt.stat.connect_start = jiffies;
750 }
751
752 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
753
754 /**
755 * xs_nospace - handle transmit was incomplete
756 * @req: pointer to RPC request
757 *
758 */
xs_nospace(struct rpc_rqst * req)759 static int xs_nospace(struct rpc_rqst *req)
760 {
761 struct rpc_xprt *xprt = req->rq_xprt;
762 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
763 struct sock *sk = transport->inet;
764 int ret = -EAGAIN;
765
766 trace_rpc_socket_nospace(req, transport);
767
768 /* Protect against races with write_space */
769 spin_lock(&xprt->transport_lock);
770
771 /* Don't race with disconnect */
772 if (xprt_connected(xprt)) {
773 /* wait for more buffer space */
774 sk->sk_write_pending++;
775 xprt_wait_for_buffer_space(xprt);
776 } else
777 ret = -ENOTCONN;
778
779 spin_unlock(&xprt->transport_lock);
780
781 /* Race breaker in case memory is freed before above code is called */
782 if (ret == -EAGAIN) {
783 struct socket_wq *wq;
784
785 rcu_read_lock();
786 wq = rcu_dereference(sk->sk_wq);
787 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
788 rcu_read_unlock();
789
790 sk->sk_write_space(sk);
791 }
792 return ret;
793 }
794
795 static void
xs_stream_prepare_request(struct rpc_rqst * req)796 xs_stream_prepare_request(struct rpc_rqst *req)
797 {
798 xdr_free_bvec(&req->rq_rcv_buf);
799 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
800 }
801
802 /*
803 * Determine if the previous message in the stream was aborted before it
804 * could complete transmission.
805 */
806 static bool
xs_send_request_was_aborted(struct sock_xprt * transport,struct rpc_rqst * req)807 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
808 {
809 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
810 }
811
812 /*
813 * Return the stream record marker field for a record of length < 2^31-1
814 */
815 static rpc_fraghdr
xs_stream_record_marker(struct xdr_buf * xdr)816 xs_stream_record_marker(struct xdr_buf *xdr)
817 {
818 if (!xdr->len)
819 return 0;
820 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
821 }
822
823 /**
824 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
825 * @req: pointer to RPC request
826 *
827 * Return values:
828 * 0: The request has been sent
829 * EAGAIN: The socket was blocked, please call again later to
830 * complete the request
831 * ENOTCONN: Caller needs to invoke connect logic then call again
832 * other: Some other error occured, the request was not sent
833 */
xs_local_send_request(struct rpc_rqst * req)834 static int xs_local_send_request(struct rpc_rqst *req)
835 {
836 struct rpc_xprt *xprt = req->rq_xprt;
837 struct sock_xprt *transport =
838 container_of(xprt, struct sock_xprt, xprt);
839 struct xdr_buf *xdr = &req->rq_snd_buf;
840 rpc_fraghdr rm = xs_stream_record_marker(xdr);
841 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
842 struct msghdr msg = {
843 .msg_flags = XS_SENDMSG_FLAGS,
844 };
845 unsigned int sent;
846 int status;
847
848 /* Close the stream if the previous transmission was incomplete */
849 if (xs_send_request_was_aborted(transport, req)) {
850 xprt_force_disconnect(xprt);
851 return -ENOTCONN;
852 }
853
854 xs_pktdump("packet data:",
855 req->rq_svec->iov_base, req->rq_svec->iov_len);
856
857 req->rq_xtime = ktime_get();
858 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
859 transport->xmit.offset, rm, &sent);
860 dprintk("RPC: %s(%u) = %d\n",
861 __func__, xdr->len - transport->xmit.offset, status);
862
863 if (status == -EAGAIN && sock_writeable(transport->inet))
864 status = -ENOBUFS;
865
866 if (likely(sent > 0) || status == 0) {
867 transport->xmit.offset += sent;
868 req->rq_bytes_sent = transport->xmit.offset;
869 if (likely(req->rq_bytes_sent >= msglen)) {
870 req->rq_xmit_bytes_sent += transport->xmit.offset;
871 transport->xmit.offset = 0;
872 return 0;
873 }
874 status = -EAGAIN;
875 }
876
877 switch (status) {
878 case -ENOBUFS:
879 break;
880 case -EAGAIN:
881 status = xs_nospace(req);
882 break;
883 default:
884 dprintk("RPC: sendmsg returned unrecognized error %d\n",
885 -status);
886 fallthrough;
887 case -EPIPE:
888 xprt_force_disconnect(xprt);
889 status = -ENOTCONN;
890 }
891
892 return status;
893 }
894
895 /**
896 * xs_udp_send_request - write an RPC request to a UDP socket
897 * @req: pointer to RPC request
898 *
899 * Return values:
900 * 0: The request has been sent
901 * EAGAIN: The socket was blocked, please call again later to
902 * complete the request
903 * ENOTCONN: Caller needs to invoke connect logic then call again
904 * other: Some other error occurred, the request was not sent
905 */
xs_udp_send_request(struct rpc_rqst * req)906 static int xs_udp_send_request(struct rpc_rqst *req)
907 {
908 struct rpc_xprt *xprt = req->rq_xprt;
909 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
910 struct xdr_buf *xdr = &req->rq_snd_buf;
911 struct msghdr msg = {
912 .msg_name = xs_addr(xprt),
913 .msg_namelen = xprt->addrlen,
914 .msg_flags = XS_SENDMSG_FLAGS,
915 };
916 unsigned int sent;
917 int status;
918
919 xs_pktdump("packet data:",
920 req->rq_svec->iov_base,
921 req->rq_svec->iov_len);
922
923 if (!xprt_bound(xprt))
924 return -ENOTCONN;
925
926 if (!xprt_request_get_cong(xprt, req))
927 return -EBADSLT;
928
929 req->rq_xtime = ktime_get();
930 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
931
932 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
933 xdr->len, status);
934
935 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
936 if (status == -EPERM)
937 goto process_status;
938
939 if (status == -EAGAIN && sock_writeable(transport->inet))
940 status = -ENOBUFS;
941
942 if (sent > 0 || status == 0) {
943 req->rq_xmit_bytes_sent += sent;
944 if (sent >= req->rq_slen)
945 return 0;
946 /* Still some bytes left; set up for a retry later. */
947 status = -EAGAIN;
948 }
949
950 process_status:
951 switch (status) {
952 case -ENOTSOCK:
953 status = -ENOTCONN;
954 /* Should we call xs_close() here? */
955 break;
956 case -EAGAIN:
957 status = xs_nospace(req);
958 break;
959 case -ENETUNREACH:
960 case -ENOBUFS:
961 case -EPIPE:
962 case -ECONNREFUSED:
963 case -EPERM:
964 /* When the server has died, an ICMP port unreachable message
965 * prompts ECONNREFUSED. */
966 break;
967 default:
968 dprintk("RPC: sendmsg returned unrecognized error %d\n",
969 -status);
970 }
971
972 return status;
973 }
974
975 /**
976 * xs_tcp_send_request - write an RPC request to a TCP socket
977 * @req: pointer to RPC request
978 *
979 * Return values:
980 * 0: The request has been sent
981 * EAGAIN: The socket was blocked, please call again later to
982 * complete the request
983 * ENOTCONN: Caller needs to invoke connect logic then call again
984 * other: Some other error occurred, the request was not sent
985 *
986 * XXX: In the case of soft timeouts, should we eventually give up
987 * if sendmsg is not able to make progress?
988 */
xs_tcp_send_request(struct rpc_rqst * req)989 static int xs_tcp_send_request(struct rpc_rqst *req)
990 {
991 struct rpc_xprt *xprt = req->rq_xprt;
992 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
993 struct xdr_buf *xdr = &req->rq_snd_buf;
994 rpc_fraghdr rm = xs_stream_record_marker(xdr);
995 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
996 struct msghdr msg = {
997 .msg_flags = XS_SENDMSG_FLAGS,
998 };
999 bool vm_wait = false;
1000 unsigned int sent;
1001 int status;
1002
1003 /* Close the stream if the previous transmission was incomplete */
1004 if (xs_send_request_was_aborted(transport, req)) {
1005 if (transport->sock != NULL)
1006 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1007 return -ENOTCONN;
1008 }
1009
1010 xs_pktdump("packet data:",
1011 req->rq_svec->iov_base,
1012 req->rq_svec->iov_len);
1013
1014 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1015 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1016
1017 /* Continue transmitting the packet/record. We must be careful
1018 * to cope with writespace callbacks arriving _after_ we have
1019 * called sendmsg(). */
1020 req->rq_xtime = ktime_get();
1021 while (1) {
1022 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1023 transport->xmit.offset, rm, &sent);
1024
1025 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1026 xdr->len - transport->xmit.offset, status);
1027
1028 /* If we've sent the entire packet, immediately
1029 * reset the count of bytes sent. */
1030 transport->xmit.offset += sent;
1031 req->rq_bytes_sent = transport->xmit.offset;
1032 if (likely(req->rq_bytes_sent >= msglen)) {
1033 req->rq_xmit_bytes_sent += transport->xmit.offset;
1034 transport->xmit.offset = 0;
1035 return 0;
1036 }
1037
1038 WARN_ON_ONCE(sent == 0 && status == 0);
1039
1040 if (status == -EAGAIN ) {
1041 /*
1042 * Return EAGAIN if we're sure we're hitting the
1043 * socket send buffer limits.
1044 */
1045 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1046 break;
1047 /*
1048 * Did we hit a memory allocation failure?
1049 */
1050 if (sent == 0) {
1051 status = -ENOBUFS;
1052 if (vm_wait)
1053 break;
1054 /* Retry, knowing now that we're below the
1055 * socket send buffer limit
1056 */
1057 vm_wait = true;
1058 }
1059 continue;
1060 }
1061 if (status < 0)
1062 break;
1063 vm_wait = false;
1064 }
1065
1066 switch (status) {
1067 case -ENOTSOCK:
1068 status = -ENOTCONN;
1069 /* Should we call xs_close() here? */
1070 break;
1071 case -EAGAIN:
1072 status = xs_nospace(req);
1073 break;
1074 case -ECONNRESET:
1075 case -ECONNREFUSED:
1076 case -ENOTCONN:
1077 case -EADDRINUSE:
1078 case -ENOBUFS:
1079 case -EPIPE:
1080 break;
1081 default:
1082 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1083 -status);
1084 }
1085
1086 return status;
1087 }
1088
xs_save_old_callbacks(struct sock_xprt * transport,struct sock * sk)1089 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1090 {
1091 transport->old_data_ready = sk->sk_data_ready;
1092 transport->old_state_change = sk->sk_state_change;
1093 transport->old_write_space = sk->sk_write_space;
1094 transport->old_error_report = sk->sk_error_report;
1095 }
1096
xs_restore_old_callbacks(struct sock_xprt * transport,struct sock * sk)1097 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1098 {
1099 sk->sk_data_ready = transport->old_data_ready;
1100 sk->sk_state_change = transport->old_state_change;
1101 sk->sk_write_space = transport->old_write_space;
1102 sk->sk_error_report = transport->old_error_report;
1103 }
1104
xs_sock_reset_state_flags(struct rpc_xprt * xprt)1105 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1106 {
1107 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1108
1109 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1110 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1111 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1112 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1113 }
1114
xs_run_error_worker(struct sock_xprt * transport,unsigned int nr)1115 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1116 {
1117 set_bit(nr, &transport->sock_state);
1118 queue_work(xprtiod_workqueue, &transport->error_worker);
1119 }
1120
xs_sock_reset_connection_flags(struct rpc_xprt * xprt)1121 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1122 {
1123 smp_mb__before_atomic();
1124 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1125 clear_bit(XPRT_CLOSING, &xprt->state);
1126 xs_sock_reset_state_flags(xprt);
1127 smp_mb__after_atomic();
1128 }
1129
1130 /**
1131 * xs_error_report - callback to handle TCP socket state errors
1132 * @sk: socket
1133 *
1134 * Note: we don't call sock_error() since there may be a rpc_task
1135 * using the socket, and so we don't want to clear sk->sk_err.
1136 */
xs_error_report(struct sock * sk)1137 static void xs_error_report(struct sock *sk)
1138 {
1139 struct sock_xprt *transport;
1140 struct rpc_xprt *xprt;
1141
1142 read_lock_bh(&sk->sk_callback_lock);
1143 if (!(xprt = xprt_from_sock(sk)))
1144 goto out;
1145
1146 transport = container_of(xprt, struct sock_xprt, xprt);
1147 transport->xprt_err = -sk->sk_err;
1148 if (transport->xprt_err == 0)
1149 goto out;
1150 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1151 xprt, -transport->xprt_err);
1152 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1153
1154 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1155 smp_mb__before_atomic();
1156 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1157 out:
1158 read_unlock_bh(&sk->sk_callback_lock);
1159 }
1160
xs_reset_transport(struct sock_xprt * transport)1161 static void xs_reset_transport(struct sock_xprt *transport)
1162 {
1163 struct socket *sock = transport->sock;
1164 struct sock *sk = transport->inet;
1165 struct rpc_xprt *xprt = &transport->xprt;
1166 struct file *filp = transport->file;
1167
1168 if (sk == NULL)
1169 return;
1170 /*
1171 * Make sure we're calling this in a context from which it is safe
1172 * to call __fput_sync(). In practice that means rpciod and the
1173 * system workqueue.
1174 */
1175 if (!(current->flags & PF_WQ_WORKER)) {
1176 WARN_ON_ONCE(1);
1177 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1178 return;
1179 }
1180
1181 if (atomic_read(&transport->xprt.swapper))
1182 sk_clear_memalloc(sk);
1183
1184 kernel_sock_shutdown(sock, SHUT_RDWR);
1185
1186 mutex_lock(&transport->recv_mutex);
1187 write_lock_bh(&sk->sk_callback_lock);
1188 transport->inet = NULL;
1189 transport->sock = NULL;
1190 transport->file = NULL;
1191
1192 sk->sk_user_data = NULL;
1193
1194 xs_restore_old_callbacks(transport, sk);
1195 xprt_clear_connected(xprt);
1196 write_unlock_bh(&sk->sk_callback_lock);
1197 xs_sock_reset_connection_flags(xprt);
1198 /* Reset stream record info */
1199 xs_stream_reset_connect(transport);
1200 mutex_unlock(&transport->recv_mutex);
1201
1202 trace_rpc_socket_close(xprt, sock);
1203 __fput_sync(filp);
1204
1205 xprt_disconnect_done(xprt);
1206 }
1207
1208 /**
1209 * xs_close - close a socket
1210 * @xprt: transport
1211 *
1212 * This is used when all requests are complete; ie, no DRC state remains
1213 * on the server we want to save.
1214 *
1215 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1216 * xs_reset_transport() zeroing the socket from underneath a writer.
1217 */
xs_close(struct rpc_xprt * xprt)1218 static void xs_close(struct rpc_xprt *xprt)
1219 {
1220 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1221
1222 dprintk("RPC: xs_close xprt %p\n", xprt);
1223
1224 xs_reset_transport(transport);
1225 xprt->reestablish_timeout = 0;
1226 }
1227
xs_inject_disconnect(struct rpc_xprt * xprt)1228 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1229 {
1230 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1231 xprt);
1232 xprt_disconnect_done(xprt);
1233 }
1234
xs_xprt_free(struct rpc_xprt * xprt)1235 static void xs_xprt_free(struct rpc_xprt *xprt)
1236 {
1237 xs_free_peer_addresses(xprt);
1238 xprt_free(xprt);
1239 }
1240
1241 /**
1242 * xs_destroy - prepare to shutdown a transport
1243 * @xprt: doomed transport
1244 *
1245 */
xs_destroy(struct rpc_xprt * xprt)1246 static void xs_destroy(struct rpc_xprt *xprt)
1247 {
1248 struct sock_xprt *transport = container_of(xprt,
1249 struct sock_xprt, xprt);
1250 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1251
1252 cancel_delayed_work_sync(&transport->connect_worker);
1253 xs_close(xprt);
1254 cancel_work_sync(&transport->recv_worker);
1255 cancel_work_sync(&transport->error_worker);
1256 xs_xprt_free(xprt);
1257 module_put(THIS_MODULE);
1258 }
1259
1260 /**
1261 * xs_udp_data_read_skb - receive callback for UDP sockets
1262 * @xprt: transport
1263 * @sk: socket
1264 * @skb: skbuff
1265 *
1266 */
xs_udp_data_read_skb(struct rpc_xprt * xprt,struct sock * sk,struct sk_buff * skb)1267 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1268 struct sock *sk,
1269 struct sk_buff *skb)
1270 {
1271 struct rpc_task *task;
1272 struct rpc_rqst *rovr;
1273 int repsize, copied;
1274 u32 _xid;
1275 __be32 *xp;
1276
1277 repsize = skb->len;
1278 if (repsize < 4) {
1279 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1280 return;
1281 }
1282
1283 /* Copy the XID from the skb... */
1284 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1285 if (xp == NULL)
1286 return;
1287
1288 /* Look up and lock the request corresponding to the given XID */
1289 spin_lock(&xprt->queue_lock);
1290 rovr = xprt_lookup_rqst(xprt, *xp);
1291 if (!rovr)
1292 goto out_unlock;
1293 xprt_pin_rqst(rovr);
1294 xprt_update_rtt(rovr->rq_task);
1295 spin_unlock(&xprt->queue_lock);
1296 task = rovr->rq_task;
1297
1298 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1299 copied = repsize;
1300
1301 /* Suck it into the iovec, verify checksum if not done by hw. */
1302 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1303 spin_lock(&xprt->queue_lock);
1304 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1305 goto out_unpin;
1306 }
1307
1308
1309 spin_lock(&xprt->transport_lock);
1310 xprt_adjust_cwnd(xprt, task, copied);
1311 spin_unlock(&xprt->transport_lock);
1312 spin_lock(&xprt->queue_lock);
1313 xprt_complete_rqst(task, copied);
1314 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1315 out_unpin:
1316 xprt_unpin_rqst(rovr);
1317 out_unlock:
1318 spin_unlock(&xprt->queue_lock);
1319 }
1320
xs_udp_data_receive(struct sock_xprt * transport)1321 static void xs_udp_data_receive(struct sock_xprt *transport)
1322 {
1323 struct sk_buff *skb;
1324 struct sock *sk;
1325 int err;
1326
1327 mutex_lock(&transport->recv_mutex);
1328 sk = transport->inet;
1329 if (sk == NULL)
1330 goto out;
1331 for (;;) {
1332 skb = skb_recv_udp(sk, 0, 1, &err);
1333 if (skb == NULL)
1334 break;
1335 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1336 consume_skb(skb);
1337 cond_resched();
1338 }
1339 xs_poll_check_readable(transport);
1340 out:
1341 mutex_unlock(&transport->recv_mutex);
1342 }
1343
xs_udp_data_receive_workfn(struct work_struct * work)1344 static void xs_udp_data_receive_workfn(struct work_struct *work)
1345 {
1346 struct sock_xprt *transport =
1347 container_of(work, struct sock_xprt, recv_worker);
1348 unsigned int pflags = memalloc_nofs_save();
1349
1350 xs_udp_data_receive(transport);
1351 memalloc_nofs_restore(pflags);
1352 }
1353
1354 /**
1355 * xs_data_ready - "data ready" callback for UDP sockets
1356 * @sk: socket with data to read
1357 *
1358 */
xs_data_ready(struct sock * sk)1359 static void xs_data_ready(struct sock *sk)
1360 {
1361 struct rpc_xprt *xprt;
1362
1363 read_lock_bh(&sk->sk_callback_lock);
1364 dprintk("RPC: xs_data_ready...\n");
1365 xprt = xprt_from_sock(sk);
1366 if (xprt != NULL) {
1367 struct sock_xprt *transport = container_of(xprt,
1368 struct sock_xprt, xprt);
1369 transport->old_data_ready(sk);
1370 /* Any data means we had a useful conversation, so
1371 * then we don't need to delay the next reconnect
1372 */
1373 if (xprt->reestablish_timeout)
1374 xprt->reestablish_timeout = 0;
1375 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1376 queue_work(xprtiod_workqueue, &transport->recv_worker);
1377 }
1378 read_unlock_bh(&sk->sk_callback_lock);
1379 }
1380
1381 /*
1382 * Helper function to force a TCP close if the server is sending
1383 * junk and/or it has put us in CLOSE_WAIT
1384 */
xs_tcp_force_close(struct rpc_xprt * xprt)1385 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1386 {
1387 xprt_force_disconnect(xprt);
1388 }
1389
1390 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
xs_tcp_bc_maxpayload(struct rpc_xprt * xprt)1391 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1392 {
1393 return PAGE_SIZE;
1394 }
1395 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1396
1397 /**
1398 * xs_tcp_state_change - callback to handle TCP socket state changes
1399 * @sk: socket whose state has changed
1400 *
1401 */
xs_tcp_state_change(struct sock * sk)1402 static void xs_tcp_state_change(struct sock *sk)
1403 {
1404 struct rpc_xprt *xprt;
1405 struct sock_xprt *transport;
1406
1407 read_lock_bh(&sk->sk_callback_lock);
1408 if (!(xprt = xprt_from_sock(sk)))
1409 goto out;
1410 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1411 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1412 sk->sk_state, xprt_connected(xprt),
1413 sock_flag(sk, SOCK_DEAD),
1414 sock_flag(sk, SOCK_ZAPPED),
1415 sk->sk_shutdown);
1416
1417 transport = container_of(xprt, struct sock_xprt, xprt);
1418 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1419 switch (sk->sk_state) {
1420 case TCP_ESTABLISHED:
1421 if (!xprt_test_and_set_connected(xprt)) {
1422 xprt->connect_cookie++;
1423 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1424 xprt_clear_connecting(xprt);
1425
1426 xprt->stat.connect_count++;
1427 xprt->stat.connect_time += (long)jiffies -
1428 xprt->stat.connect_start;
1429 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1430 }
1431 break;
1432 case TCP_FIN_WAIT1:
1433 /* The client initiated a shutdown of the socket */
1434 xprt->connect_cookie++;
1435 xprt->reestablish_timeout = 0;
1436 set_bit(XPRT_CLOSING, &xprt->state);
1437 smp_mb__before_atomic();
1438 clear_bit(XPRT_CONNECTED, &xprt->state);
1439 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1440 smp_mb__after_atomic();
1441 break;
1442 case TCP_CLOSE_WAIT:
1443 /* The server initiated a shutdown of the socket */
1444 xprt->connect_cookie++;
1445 clear_bit(XPRT_CONNECTED, &xprt->state);
1446 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1447 fallthrough;
1448 case TCP_CLOSING:
1449 /*
1450 * If the server closed down the connection, make sure that
1451 * we back off before reconnecting
1452 */
1453 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1454 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1455 break;
1456 case TCP_LAST_ACK:
1457 set_bit(XPRT_CLOSING, &xprt->state);
1458 smp_mb__before_atomic();
1459 clear_bit(XPRT_CONNECTED, &xprt->state);
1460 smp_mb__after_atomic();
1461 break;
1462 case TCP_CLOSE:
1463 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1464 &transport->sock_state))
1465 xprt_clear_connecting(xprt);
1466 clear_bit(XPRT_CLOSING, &xprt->state);
1467 /* Trigger the socket release */
1468 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1469 }
1470 out:
1471 read_unlock_bh(&sk->sk_callback_lock);
1472 }
1473
xs_write_space(struct sock * sk)1474 static void xs_write_space(struct sock *sk)
1475 {
1476 struct socket_wq *wq;
1477 struct sock_xprt *transport;
1478 struct rpc_xprt *xprt;
1479
1480 if (!sk->sk_socket)
1481 return;
1482 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1483
1484 if (unlikely(!(xprt = xprt_from_sock(sk))))
1485 return;
1486 transport = container_of(xprt, struct sock_xprt, xprt);
1487 rcu_read_lock();
1488 wq = rcu_dereference(sk->sk_wq);
1489 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1490 goto out;
1491
1492 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1493 sk->sk_write_pending--;
1494 out:
1495 rcu_read_unlock();
1496 }
1497
1498 /**
1499 * xs_udp_write_space - callback invoked when socket buffer space
1500 * becomes available
1501 * @sk: socket whose state has changed
1502 *
1503 * Called when more output buffer space is available for this socket.
1504 * We try not to wake our writers until they can make "significant"
1505 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1506 * with a bunch of small requests.
1507 */
xs_udp_write_space(struct sock * sk)1508 static void xs_udp_write_space(struct sock *sk)
1509 {
1510 read_lock_bh(&sk->sk_callback_lock);
1511
1512 /* from net/core/sock.c:sock_def_write_space */
1513 if (sock_writeable(sk))
1514 xs_write_space(sk);
1515
1516 read_unlock_bh(&sk->sk_callback_lock);
1517 }
1518
1519 /**
1520 * xs_tcp_write_space - callback invoked when socket buffer space
1521 * becomes available
1522 * @sk: socket whose state has changed
1523 *
1524 * Called when more output buffer space is available for this socket.
1525 * We try not to wake our writers until they can make "significant"
1526 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1527 * with a bunch of small requests.
1528 */
xs_tcp_write_space(struct sock * sk)1529 static void xs_tcp_write_space(struct sock *sk)
1530 {
1531 read_lock_bh(&sk->sk_callback_lock);
1532
1533 /* from net/core/stream.c:sk_stream_write_space */
1534 if (sk_stream_is_writeable(sk))
1535 xs_write_space(sk);
1536
1537 read_unlock_bh(&sk->sk_callback_lock);
1538 }
1539
xs_udp_do_set_buffer_size(struct rpc_xprt * xprt)1540 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1541 {
1542 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1543 struct sock *sk = transport->inet;
1544
1545 if (transport->rcvsize) {
1546 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1547 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1548 }
1549 if (transport->sndsize) {
1550 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1551 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1552 sk->sk_write_space(sk);
1553 }
1554 }
1555
1556 /**
1557 * xs_udp_set_buffer_size - set send and receive limits
1558 * @xprt: generic transport
1559 * @sndsize: requested size of send buffer, in bytes
1560 * @rcvsize: requested size of receive buffer, in bytes
1561 *
1562 * Set socket send and receive buffer size limits.
1563 */
xs_udp_set_buffer_size(struct rpc_xprt * xprt,size_t sndsize,size_t rcvsize)1564 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1565 {
1566 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1567
1568 transport->sndsize = 0;
1569 if (sndsize)
1570 transport->sndsize = sndsize + 1024;
1571 transport->rcvsize = 0;
1572 if (rcvsize)
1573 transport->rcvsize = rcvsize + 1024;
1574
1575 xs_udp_do_set_buffer_size(xprt);
1576 }
1577
1578 /**
1579 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1580 * @xprt: controlling transport
1581 * @task: task that timed out
1582 *
1583 * Adjust the congestion window after a retransmit timeout has occurred.
1584 */
xs_udp_timer(struct rpc_xprt * xprt,struct rpc_task * task)1585 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1586 {
1587 spin_lock(&xprt->transport_lock);
1588 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1589 spin_unlock(&xprt->transport_lock);
1590 }
1591
xs_get_random_port(void)1592 static int xs_get_random_port(void)
1593 {
1594 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1595 unsigned short range;
1596 unsigned short rand;
1597
1598 if (max < min)
1599 return -EADDRINUSE;
1600 range = max - min + 1;
1601 rand = (unsigned short) prandom_u32() % range;
1602 return rand + min;
1603 }
1604
xs_sock_getport(struct socket * sock)1605 static unsigned short xs_sock_getport(struct socket *sock)
1606 {
1607 struct sockaddr_storage buf;
1608 unsigned short port = 0;
1609
1610 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1611 goto out;
1612 switch (buf.ss_family) {
1613 case AF_INET6:
1614 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1615 break;
1616 case AF_INET:
1617 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1618 }
1619 out:
1620 return port;
1621 }
1622
1623 /**
1624 * xs_set_port - reset the port number in the remote endpoint address
1625 * @xprt: generic transport
1626 * @port: new port number
1627 *
1628 */
xs_set_port(struct rpc_xprt * xprt,unsigned short port)1629 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1630 {
1631 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1632
1633 rpc_set_port(xs_addr(xprt), port);
1634 xs_update_peer_port(xprt);
1635 }
1636
xs_set_srcport(struct sock_xprt * transport,struct socket * sock)1637 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1638 {
1639 if (transport->srcport == 0 && transport->xprt.reuseport)
1640 transport->srcport = xs_sock_getport(sock);
1641 }
1642
xs_get_srcport(struct sock_xprt * transport)1643 static int xs_get_srcport(struct sock_xprt *transport)
1644 {
1645 int port = transport->srcport;
1646
1647 if (port == 0 && transport->xprt.resvport)
1648 port = xs_get_random_port();
1649 return port;
1650 }
1651
get_srcport(struct rpc_xprt * xprt)1652 unsigned short get_srcport(struct rpc_xprt *xprt)
1653 {
1654 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1655 return xs_sock_getport(sock->sock);
1656 }
1657 EXPORT_SYMBOL(get_srcport);
1658
xs_next_srcport(struct sock_xprt * transport,unsigned short port)1659 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1660 {
1661 if (transport->srcport != 0)
1662 transport->srcport = 0;
1663 if (!transport->xprt.resvport)
1664 return 0;
1665 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1666 return xprt_max_resvport;
1667 return --port;
1668 }
xs_bind(struct sock_xprt * transport,struct socket * sock)1669 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1670 {
1671 struct sockaddr_storage myaddr;
1672 int err, nloop = 0;
1673 int port = xs_get_srcport(transport);
1674 unsigned short last;
1675
1676 /*
1677 * If we are asking for any ephemeral port (i.e. port == 0 &&
1678 * transport->xprt.resvport == 0), don't bind. Let the local
1679 * port selection happen implicitly when the socket is used
1680 * (for example at connect time).
1681 *
1682 * This ensures that we can continue to establish TCP
1683 * connections even when all local ephemeral ports are already
1684 * a part of some TCP connection. This makes no difference
1685 * for UDP sockets, but also doens't harm them.
1686 *
1687 * If we're asking for any reserved port (i.e. port == 0 &&
1688 * transport->xprt.resvport == 1) xs_get_srcport above will
1689 * ensure that port is non-zero and we will bind as needed.
1690 */
1691 if (port <= 0)
1692 return port;
1693
1694 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1695 do {
1696 rpc_set_port((struct sockaddr *)&myaddr, port);
1697 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1698 transport->xprt.addrlen);
1699 if (err == 0) {
1700 if (transport->xprt.reuseport)
1701 transport->srcport = port;
1702 break;
1703 }
1704 last = port;
1705 port = xs_next_srcport(transport, port);
1706 if (port > last)
1707 nloop++;
1708 } while (err == -EADDRINUSE && nloop != 2);
1709
1710 if (myaddr.ss_family == AF_INET)
1711 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1712 &((struct sockaddr_in *)&myaddr)->sin_addr,
1713 port, err ? "failed" : "ok", err);
1714 else
1715 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1716 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1717 port, err ? "failed" : "ok", err);
1718 return err;
1719 }
1720
1721 /*
1722 * We don't support autobind on AF_LOCAL sockets
1723 */
xs_local_rpcbind(struct rpc_task * task)1724 static void xs_local_rpcbind(struct rpc_task *task)
1725 {
1726 xprt_set_bound(task->tk_xprt);
1727 }
1728
xs_local_set_port(struct rpc_xprt * xprt,unsigned short port)1729 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1730 {
1731 }
1732
1733 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1734 static struct lock_class_key xs_key[2];
1735 static struct lock_class_key xs_slock_key[2];
1736
xs_reclassify_socketu(struct socket * sock)1737 static inline void xs_reclassify_socketu(struct socket *sock)
1738 {
1739 struct sock *sk = sock->sk;
1740
1741 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1742 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1743 }
1744
xs_reclassify_socket4(struct socket * sock)1745 static inline void xs_reclassify_socket4(struct socket *sock)
1746 {
1747 struct sock *sk = sock->sk;
1748
1749 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1750 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1751 }
1752
xs_reclassify_socket6(struct socket * sock)1753 static inline void xs_reclassify_socket6(struct socket *sock)
1754 {
1755 struct sock *sk = sock->sk;
1756
1757 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1758 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1759 }
1760
xs_reclassify_socket(int family,struct socket * sock)1761 static inline void xs_reclassify_socket(int family, struct socket *sock)
1762 {
1763 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1764 return;
1765
1766 switch (family) {
1767 case AF_LOCAL:
1768 xs_reclassify_socketu(sock);
1769 break;
1770 case AF_INET:
1771 xs_reclassify_socket4(sock);
1772 break;
1773 case AF_INET6:
1774 xs_reclassify_socket6(sock);
1775 break;
1776 }
1777 }
1778 #else
xs_reclassify_socket(int family,struct socket * sock)1779 static inline void xs_reclassify_socket(int family, struct socket *sock)
1780 {
1781 }
1782 #endif
1783
xs_dummy_setup_socket(struct work_struct * work)1784 static void xs_dummy_setup_socket(struct work_struct *work)
1785 {
1786 }
1787
xs_create_sock(struct rpc_xprt * xprt,struct sock_xprt * transport,int family,int type,int protocol,bool reuseport)1788 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1789 struct sock_xprt *transport, int family, int type,
1790 int protocol, bool reuseport)
1791 {
1792 struct file *filp;
1793 struct socket *sock;
1794 int err;
1795
1796 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1797 if (err < 0) {
1798 dprintk("RPC: can't create %d transport socket (%d).\n",
1799 protocol, -err);
1800 goto out;
1801 }
1802 xs_reclassify_socket(family, sock);
1803
1804 if (reuseport)
1805 sock_set_reuseport(sock->sk);
1806
1807 err = xs_bind(transport, sock);
1808 if (err) {
1809 sock_release(sock);
1810 goto out;
1811 }
1812
1813 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1814 if (IS_ERR(filp))
1815 return ERR_CAST(filp);
1816 transport->file = filp;
1817
1818 return sock;
1819 out:
1820 return ERR_PTR(err);
1821 }
1822
xs_local_finish_connecting(struct rpc_xprt * xprt,struct socket * sock)1823 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1824 struct socket *sock)
1825 {
1826 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1827 xprt);
1828
1829 if (!transport->inet) {
1830 struct sock *sk = sock->sk;
1831
1832 write_lock_bh(&sk->sk_callback_lock);
1833
1834 xs_save_old_callbacks(transport, sk);
1835
1836 sk->sk_user_data = xprt;
1837 sk->sk_data_ready = xs_data_ready;
1838 sk->sk_write_space = xs_udp_write_space;
1839 sock_set_flag(sk, SOCK_FASYNC);
1840 sk->sk_error_report = xs_error_report;
1841
1842 xprt_clear_connected(xprt);
1843
1844 /* Reset to new socket */
1845 transport->sock = sock;
1846 transport->inet = sk;
1847
1848 write_unlock_bh(&sk->sk_callback_lock);
1849 }
1850
1851 xs_stream_start_connect(transport);
1852
1853 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1854 }
1855
1856 /**
1857 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1858 * @transport: socket transport to connect
1859 */
xs_local_setup_socket(struct sock_xprt * transport)1860 static int xs_local_setup_socket(struct sock_xprt *transport)
1861 {
1862 struct rpc_xprt *xprt = &transport->xprt;
1863 struct file *filp;
1864 struct socket *sock;
1865 int status;
1866
1867 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1868 SOCK_STREAM, 0, &sock, 1);
1869 if (status < 0) {
1870 dprintk("RPC: can't create AF_LOCAL "
1871 "transport socket (%d).\n", -status);
1872 goto out;
1873 }
1874 xs_reclassify_socket(AF_LOCAL, sock);
1875
1876 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1877 if (IS_ERR(filp)) {
1878 status = PTR_ERR(filp);
1879 goto out;
1880 }
1881 transport->file = filp;
1882
1883 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1884 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1885
1886 status = xs_local_finish_connecting(xprt, sock);
1887 trace_rpc_socket_connect(xprt, sock, status);
1888 switch (status) {
1889 case 0:
1890 dprintk("RPC: xprt %p connected to %s\n",
1891 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1892 xprt->stat.connect_count++;
1893 xprt->stat.connect_time += (long)jiffies -
1894 xprt->stat.connect_start;
1895 xprt_set_connected(xprt);
1896 case -ENOBUFS:
1897 break;
1898 case -ENOENT:
1899 dprintk("RPC: xprt %p: socket %s does not exist\n",
1900 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1901 break;
1902 case -ECONNREFUSED:
1903 dprintk("RPC: xprt %p: connection refused for %s\n",
1904 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1905 break;
1906 default:
1907 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1908 __func__, -status,
1909 xprt->address_strings[RPC_DISPLAY_ADDR]);
1910 }
1911
1912 out:
1913 xprt_clear_connecting(xprt);
1914 xprt_wake_pending_tasks(xprt, status);
1915 return status;
1916 }
1917
xs_local_connect(struct rpc_xprt * xprt,struct rpc_task * task)1918 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1919 {
1920 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1921 int ret;
1922
1923 if (RPC_IS_ASYNC(task)) {
1924 /*
1925 * We want the AF_LOCAL connect to be resolved in the
1926 * filesystem namespace of the process making the rpc
1927 * call. Thus we connect synchronously.
1928 *
1929 * If we want to support asynchronous AF_LOCAL calls,
1930 * we'll need to figure out how to pass a namespace to
1931 * connect.
1932 */
1933 task->tk_rpc_status = -ENOTCONN;
1934 rpc_exit(task, -ENOTCONN);
1935 return;
1936 }
1937 ret = xs_local_setup_socket(transport);
1938 if (ret && !RPC_IS_SOFTCONN(task))
1939 msleep_interruptible(15000);
1940 }
1941
1942 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1943 /*
1944 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1945 * know that we have exclusive access to the socket), to guard against
1946 * races with xs_reset_transport.
1947 */
xs_set_memalloc(struct rpc_xprt * xprt)1948 static void xs_set_memalloc(struct rpc_xprt *xprt)
1949 {
1950 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1951 xprt);
1952
1953 /*
1954 * If there's no sock, then we have nothing to set. The
1955 * reconnecting process will get it for us.
1956 */
1957 if (!transport->inet)
1958 return;
1959 if (atomic_read(&xprt->swapper))
1960 sk_set_memalloc(transport->inet);
1961 }
1962
1963 /**
1964 * xs_enable_swap - Tag this transport as being used for swap.
1965 * @xprt: transport to tag
1966 *
1967 * Take a reference to this transport on behalf of the rpc_clnt, and
1968 * optionally mark it for swapping if it wasn't already.
1969 */
1970 static int
xs_enable_swap(struct rpc_xprt * xprt)1971 xs_enable_swap(struct rpc_xprt *xprt)
1972 {
1973 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1974
1975 if (atomic_inc_return(&xprt->swapper) != 1)
1976 return 0;
1977 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1978 return -ERESTARTSYS;
1979 if (xs->inet)
1980 sk_set_memalloc(xs->inet);
1981 xprt_release_xprt(xprt, NULL);
1982 return 0;
1983 }
1984
1985 /**
1986 * xs_disable_swap - Untag this transport as being used for swap.
1987 * @xprt: transport to tag
1988 *
1989 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1990 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1991 */
1992 static void
xs_disable_swap(struct rpc_xprt * xprt)1993 xs_disable_swap(struct rpc_xprt *xprt)
1994 {
1995 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1996
1997 if (!atomic_dec_and_test(&xprt->swapper))
1998 return;
1999 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2000 return;
2001 if (xs->inet)
2002 sk_clear_memalloc(xs->inet);
2003 xprt_release_xprt(xprt, NULL);
2004 }
2005 #else
xs_set_memalloc(struct rpc_xprt * xprt)2006 static void xs_set_memalloc(struct rpc_xprt *xprt)
2007 {
2008 }
2009
2010 static int
xs_enable_swap(struct rpc_xprt * xprt)2011 xs_enable_swap(struct rpc_xprt *xprt)
2012 {
2013 return -EINVAL;
2014 }
2015
2016 static void
xs_disable_swap(struct rpc_xprt * xprt)2017 xs_disable_swap(struct rpc_xprt *xprt)
2018 {
2019 }
2020 #endif
2021
xs_udp_finish_connecting(struct rpc_xprt * xprt,struct socket * sock)2022 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2023 {
2024 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2025
2026 if (!transport->inet) {
2027 struct sock *sk = sock->sk;
2028
2029 write_lock_bh(&sk->sk_callback_lock);
2030
2031 xs_save_old_callbacks(transport, sk);
2032
2033 sk->sk_user_data = xprt;
2034 sk->sk_data_ready = xs_data_ready;
2035 sk->sk_write_space = xs_udp_write_space;
2036 sock_set_flag(sk, SOCK_FASYNC);
2037
2038 xprt_set_connected(xprt);
2039
2040 /* Reset to new socket */
2041 transport->sock = sock;
2042 transport->inet = sk;
2043
2044 xs_set_memalloc(xprt);
2045
2046 write_unlock_bh(&sk->sk_callback_lock);
2047 }
2048 xs_udp_do_set_buffer_size(xprt);
2049
2050 xprt->stat.connect_start = jiffies;
2051 }
2052
xs_udp_setup_socket(struct work_struct * work)2053 static void xs_udp_setup_socket(struct work_struct *work)
2054 {
2055 struct sock_xprt *transport =
2056 container_of(work, struct sock_xprt, connect_worker.work);
2057 struct rpc_xprt *xprt = &transport->xprt;
2058 struct socket *sock;
2059 int status = -EIO;
2060
2061 sock = xs_create_sock(xprt, transport,
2062 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2063 IPPROTO_UDP, false);
2064 if (IS_ERR(sock))
2065 goto out;
2066
2067 dprintk("RPC: worker connecting xprt %p via %s to "
2068 "%s (port %s)\n", xprt,
2069 xprt->address_strings[RPC_DISPLAY_PROTO],
2070 xprt->address_strings[RPC_DISPLAY_ADDR],
2071 xprt->address_strings[RPC_DISPLAY_PORT]);
2072
2073 xs_udp_finish_connecting(xprt, sock);
2074 trace_rpc_socket_connect(xprt, sock, 0);
2075 status = 0;
2076 out:
2077 xprt_clear_connecting(xprt);
2078 xprt_unlock_connect(xprt, transport);
2079 xprt_wake_pending_tasks(xprt, status);
2080 }
2081
2082 /**
2083 * xs_tcp_shutdown - gracefully shut down a TCP socket
2084 * @xprt: transport
2085 *
2086 * Initiates a graceful shutdown of the TCP socket by calling the
2087 * equivalent of shutdown(SHUT_RDWR);
2088 */
xs_tcp_shutdown(struct rpc_xprt * xprt)2089 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2090 {
2091 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2092 struct socket *sock = transport->sock;
2093 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2094
2095 if (sock == NULL)
2096 return;
2097 switch (skst) {
2098 default:
2099 kernel_sock_shutdown(sock, SHUT_RDWR);
2100 trace_rpc_socket_shutdown(xprt, sock);
2101 break;
2102 case TCP_CLOSE:
2103 case TCP_TIME_WAIT:
2104 xs_reset_transport(transport);
2105 }
2106 }
2107
xs_tcp_set_socket_timeouts(struct rpc_xprt * xprt,struct socket * sock)2108 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2109 struct socket *sock)
2110 {
2111 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2112 unsigned int keepidle;
2113 unsigned int keepcnt;
2114 unsigned int timeo;
2115
2116 spin_lock(&xprt->transport_lock);
2117 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2118 keepcnt = xprt->timeout->to_retries + 1;
2119 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2120 (xprt->timeout->to_retries + 1);
2121 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2122 spin_unlock(&xprt->transport_lock);
2123
2124 /* TCP Keepalive options */
2125 sock_set_keepalive(sock->sk);
2126 tcp_sock_set_keepidle(sock->sk, keepidle);
2127 tcp_sock_set_keepintvl(sock->sk, keepidle);
2128 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2129
2130 /* TCP user timeout (see RFC5482) */
2131 tcp_sock_set_user_timeout(sock->sk, timeo);
2132 }
2133
xs_tcp_set_connect_timeout(struct rpc_xprt * xprt,unsigned long connect_timeout,unsigned long reconnect_timeout)2134 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2135 unsigned long connect_timeout,
2136 unsigned long reconnect_timeout)
2137 {
2138 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2139 struct rpc_timeout to;
2140 unsigned long initval;
2141
2142 spin_lock(&xprt->transport_lock);
2143 if (reconnect_timeout < xprt->max_reconnect_timeout)
2144 xprt->max_reconnect_timeout = reconnect_timeout;
2145 if (connect_timeout < xprt->connect_timeout) {
2146 memcpy(&to, xprt->timeout, sizeof(to));
2147 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2148 /* Arbitrary lower limit */
2149 if (initval < XS_TCP_INIT_REEST_TO << 1)
2150 initval = XS_TCP_INIT_REEST_TO << 1;
2151 to.to_initval = initval;
2152 to.to_maxval = initval;
2153 memcpy(&transport->tcp_timeout, &to,
2154 sizeof(transport->tcp_timeout));
2155 xprt->timeout = &transport->tcp_timeout;
2156 xprt->connect_timeout = connect_timeout;
2157 }
2158 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2159 spin_unlock(&xprt->transport_lock);
2160 }
2161
xs_tcp_finish_connecting(struct rpc_xprt * xprt,struct socket * sock)2162 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2163 {
2164 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2165 int ret = -ENOTCONN;
2166
2167 if (!transport->inet) {
2168 struct sock *sk = sock->sk;
2169
2170 /* Avoid temporary address, they are bad for long-lived
2171 * connections such as NFS mounts.
2172 * RFC4941, section 3.6 suggests that:
2173 * Individual applications, which have specific
2174 * knowledge about the normal duration of connections,
2175 * MAY override this as appropriate.
2176 */
2177 if (xs_addr(xprt)->sa_family == PF_INET6) {
2178 ip6_sock_set_addr_preferences(sk,
2179 IPV6_PREFER_SRC_PUBLIC);
2180 }
2181
2182 xs_tcp_set_socket_timeouts(xprt, sock);
2183
2184 write_lock_bh(&sk->sk_callback_lock);
2185
2186 xs_save_old_callbacks(transport, sk);
2187
2188 sk->sk_user_data = xprt;
2189 sk->sk_data_ready = xs_data_ready;
2190 sk->sk_state_change = xs_tcp_state_change;
2191 sk->sk_write_space = xs_tcp_write_space;
2192 sock_set_flag(sk, SOCK_FASYNC);
2193 sk->sk_error_report = xs_error_report;
2194
2195 /* socket options */
2196 sock_reset_flag(sk, SOCK_LINGER);
2197 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2198
2199 xprt_clear_connected(xprt);
2200
2201 /* Reset to new socket */
2202 transport->sock = sock;
2203 transport->inet = sk;
2204
2205 write_unlock_bh(&sk->sk_callback_lock);
2206 }
2207
2208 if (!xprt_bound(xprt))
2209 goto out;
2210
2211 xs_set_memalloc(xprt);
2212
2213 xs_stream_start_connect(transport);
2214
2215 /* Tell the socket layer to start connecting... */
2216 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2217 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2218 switch (ret) {
2219 case 0:
2220 xs_set_srcport(transport, sock);
2221 fallthrough;
2222 case -EINPROGRESS:
2223 /* SYN_SENT! */
2224 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2225 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2226 break;
2227 case -EADDRNOTAVAIL:
2228 /* Source port number is unavailable. Try a new one! */
2229 transport->srcport = 0;
2230 }
2231 out:
2232 return ret;
2233 }
2234
2235 /**
2236 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2237 * @work: queued work item
2238 *
2239 * Invoked by a work queue tasklet.
2240 */
xs_tcp_setup_socket(struct work_struct * work)2241 static void xs_tcp_setup_socket(struct work_struct *work)
2242 {
2243 struct sock_xprt *transport =
2244 container_of(work, struct sock_xprt, connect_worker.work);
2245 struct socket *sock = transport->sock;
2246 struct rpc_xprt *xprt = &transport->xprt;
2247 int status = -EIO;
2248
2249 if (!sock) {
2250 sock = xs_create_sock(xprt, transport,
2251 xs_addr(xprt)->sa_family, SOCK_STREAM,
2252 IPPROTO_TCP, true);
2253 if (IS_ERR(sock)) {
2254 status = PTR_ERR(sock);
2255 goto out;
2256 }
2257 }
2258
2259 dprintk("RPC: worker connecting xprt %p via %s to "
2260 "%s (port %s)\n", xprt,
2261 xprt->address_strings[RPC_DISPLAY_PROTO],
2262 xprt->address_strings[RPC_DISPLAY_ADDR],
2263 xprt->address_strings[RPC_DISPLAY_PORT]);
2264
2265 status = xs_tcp_finish_connecting(xprt, sock);
2266 trace_rpc_socket_connect(xprt, sock, status);
2267 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2268 xprt, -status, xprt_connected(xprt),
2269 sock->sk->sk_state);
2270 switch (status) {
2271 default:
2272 printk("%s: connect returned unhandled error %d\n",
2273 __func__, status);
2274 fallthrough;
2275 case -EADDRNOTAVAIL:
2276 /* We're probably in TIME_WAIT. Get rid of existing socket,
2277 * and retry
2278 */
2279 xs_tcp_force_close(xprt);
2280 break;
2281 case 0:
2282 case -EINPROGRESS:
2283 case -EALREADY:
2284 xprt_unlock_connect(xprt, transport);
2285 return;
2286 case -EINVAL:
2287 /* Happens, for instance, if the user specified a link
2288 * local IPv6 address without a scope-id.
2289 */
2290 case -ECONNREFUSED:
2291 case -ECONNRESET:
2292 case -ENETDOWN:
2293 case -ENETUNREACH:
2294 case -EHOSTUNREACH:
2295 case -EADDRINUSE:
2296 case -ENOBUFS:
2297 /*
2298 * xs_tcp_force_close() wakes tasks with -EIO.
2299 * We need to wake them first to ensure the
2300 * correct error code.
2301 */
2302 xprt_wake_pending_tasks(xprt, status);
2303 xs_tcp_force_close(xprt);
2304 goto out;
2305 }
2306 status = -EAGAIN;
2307 out:
2308 xprt_clear_connecting(xprt);
2309 xprt_unlock_connect(xprt, transport);
2310 xprt_wake_pending_tasks(xprt, status);
2311 }
2312
2313 /**
2314 * xs_connect - connect a socket to a remote endpoint
2315 * @xprt: pointer to transport structure
2316 * @task: address of RPC task that manages state of connect request
2317 *
2318 * TCP: If the remote end dropped the connection, delay reconnecting.
2319 *
2320 * UDP socket connects are synchronous, but we use a work queue anyway
2321 * to guarantee that even unprivileged user processes can set up a
2322 * socket on a privileged port.
2323 *
2324 * If a UDP socket connect fails, the delay behavior here prevents
2325 * retry floods (hard mounts).
2326 */
xs_connect(struct rpc_xprt * xprt,struct rpc_task * task)2327 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2328 {
2329 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2330 unsigned long delay = 0;
2331
2332 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2333
2334 if (transport->sock != NULL) {
2335 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2336 "seconds\n",
2337 xprt, xprt->reestablish_timeout / HZ);
2338
2339 /* Start by resetting any existing state */
2340 xs_reset_transport(transport);
2341
2342 delay = xprt_reconnect_delay(xprt);
2343 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2344
2345 } else
2346 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2347
2348 queue_delayed_work(xprtiod_workqueue,
2349 &transport->connect_worker,
2350 delay);
2351 }
2352
xs_wake_disconnect(struct sock_xprt * transport)2353 static void xs_wake_disconnect(struct sock_xprt *transport)
2354 {
2355 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2356 xs_tcp_force_close(&transport->xprt);
2357 }
2358
xs_wake_write(struct sock_xprt * transport)2359 static void xs_wake_write(struct sock_xprt *transport)
2360 {
2361 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2362 xprt_write_space(&transport->xprt);
2363 }
2364
xs_wake_error(struct sock_xprt * transport)2365 static void xs_wake_error(struct sock_xprt *transport)
2366 {
2367 int sockerr;
2368
2369 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2370 return;
2371 mutex_lock(&transport->recv_mutex);
2372 if (transport->sock == NULL)
2373 goto out;
2374 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2375 goto out;
2376 sockerr = xchg(&transport->xprt_err, 0);
2377 if (sockerr < 0)
2378 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2379 out:
2380 mutex_unlock(&transport->recv_mutex);
2381 }
2382
xs_wake_pending(struct sock_xprt * transport)2383 static void xs_wake_pending(struct sock_xprt *transport)
2384 {
2385 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2386 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2387 }
2388
xs_error_handle(struct work_struct * work)2389 static void xs_error_handle(struct work_struct *work)
2390 {
2391 struct sock_xprt *transport = container_of(work,
2392 struct sock_xprt, error_worker);
2393
2394 xs_wake_disconnect(transport);
2395 xs_wake_write(transport);
2396 xs_wake_error(transport);
2397 xs_wake_pending(transport);
2398 }
2399
2400 /**
2401 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2402 * @xprt: rpc_xprt struct containing statistics
2403 * @seq: output file
2404 *
2405 */
xs_local_print_stats(struct rpc_xprt * xprt,struct seq_file * seq)2406 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2407 {
2408 long idle_time = 0;
2409
2410 if (xprt_connected(xprt))
2411 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2412
2413 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2414 "%llu %llu %lu %llu %llu\n",
2415 xprt->stat.bind_count,
2416 xprt->stat.connect_count,
2417 xprt->stat.connect_time / HZ,
2418 idle_time,
2419 xprt->stat.sends,
2420 xprt->stat.recvs,
2421 xprt->stat.bad_xids,
2422 xprt->stat.req_u,
2423 xprt->stat.bklog_u,
2424 xprt->stat.max_slots,
2425 xprt->stat.sending_u,
2426 xprt->stat.pending_u);
2427 }
2428
2429 /**
2430 * xs_udp_print_stats - display UDP socket-specifc stats
2431 * @xprt: rpc_xprt struct containing statistics
2432 * @seq: output file
2433 *
2434 */
xs_udp_print_stats(struct rpc_xprt * xprt,struct seq_file * seq)2435 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2436 {
2437 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2438
2439 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2440 "%lu %llu %llu\n",
2441 transport->srcport,
2442 xprt->stat.bind_count,
2443 xprt->stat.sends,
2444 xprt->stat.recvs,
2445 xprt->stat.bad_xids,
2446 xprt->stat.req_u,
2447 xprt->stat.bklog_u,
2448 xprt->stat.max_slots,
2449 xprt->stat.sending_u,
2450 xprt->stat.pending_u);
2451 }
2452
2453 /**
2454 * xs_tcp_print_stats - display TCP socket-specifc stats
2455 * @xprt: rpc_xprt struct containing statistics
2456 * @seq: output file
2457 *
2458 */
xs_tcp_print_stats(struct rpc_xprt * xprt,struct seq_file * seq)2459 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2460 {
2461 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2462 long idle_time = 0;
2463
2464 if (xprt_connected(xprt))
2465 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2466
2467 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2468 "%llu %llu %lu %llu %llu\n",
2469 transport->srcport,
2470 xprt->stat.bind_count,
2471 xprt->stat.connect_count,
2472 xprt->stat.connect_time / HZ,
2473 idle_time,
2474 xprt->stat.sends,
2475 xprt->stat.recvs,
2476 xprt->stat.bad_xids,
2477 xprt->stat.req_u,
2478 xprt->stat.bklog_u,
2479 xprt->stat.max_slots,
2480 xprt->stat.sending_u,
2481 xprt->stat.pending_u);
2482 }
2483
2484 /*
2485 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2486 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2487 * to use the server side send routines.
2488 */
bc_malloc(struct rpc_task * task)2489 static int bc_malloc(struct rpc_task *task)
2490 {
2491 struct rpc_rqst *rqst = task->tk_rqstp;
2492 size_t size = rqst->rq_callsize;
2493 struct page *page;
2494 struct rpc_buffer *buf;
2495
2496 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2497 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2498 size);
2499 return -EINVAL;
2500 }
2501
2502 page = alloc_page(GFP_KERNEL);
2503 if (!page)
2504 return -ENOMEM;
2505
2506 buf = page_address(page);
2507 buf->len = PAGE_SIZE;
2508
2509 rqst->rq_buffer = buf->data;
2510 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2511 return 0;
2512 }
2513
2514 /*
2515 * Free the space allocated in the bc_alloc routine
2516 */
bc_free(struct rpc_task * task)2517 static void bc_free(struct rpc_task *task)
2518 {
2519 void *buffer = task->tk_rqstp->rq_buffer;
2520 struct rpc_buffer *buf;
2521
2522 buf = container_of(buffer, struct rpc_buffer, data);
2523 free_page((unsigned long)buf);
2524 }
2525
bc_sendto(struct rpc_rqst * req)2526 static int bc_sendto(struct rpc_rqst *req)
2527 {
2528 struct xdr_buf *xdr = &req->rq_snd_buf;
2529 struct sock_xprt *transport =
2530 container_of(req->rq_xprt, struct sock_xprt, xprt);
2531 struct msghdr msg = {
2532 .msg_flags = 0,
2533 };
2534 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2535 (u32)xdr->len);
2536 unsigned int sent = 0;
2537 int err;
2538
2539 req->rq_xtime = ktime_get();
2540 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2541 xdr_free_bvec(xdr);
2542 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2543 return -EAGAIN;
2544 return sent;
2545 }
2546
2547 /**
2548 * bc_send_request - Send a backchannel Call on a TCP socket
2549 * @req: rpc_rqst containing Call message to be sent
2550 *
2551 * xpt_mutex ensures @rqstp's whole message is written to the socket
2552 * without interruption.
2553 *
2554 * Return values:
2555 * %0 if the message was sent successfully
2556 * %ENOTCONN if the message was not sent
2557 */
bc_send_request(struct rpc_rqst * req)2558 static int bc_send_request(struct rpc_rqst *req)
2559 {
2560 struct svc_xprt *xprt;
2561 int len;
2562
2563 /*
2564 * Get the server socket associated with this callback xprt
2565 */
2566 xprt = req->rq_xprt->bc_xprt;
2567
2568 /*
2569 * Grab the mutex to serialize data as the connection is shared
2570 * with the fore channel
2571 */
2572 mutex_lock(&xprt->xpt_mutex);
2573 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2574 len = -ENOTCONN;
2575 else
2576 len = bc_sendto(req);
2577 mutex_unlock(&xprt->xpt_mutex);
2578
2579 if (len > 0)
2580 len = 0;
2581
2582 return len;
2583 }
2584
2585 /*
2586 * The close routine. Since this is client initiated, we do nothing
2587 */
2588
bc_close(struct rpc_xprt * xprt)2589 static void bc_close(struct rpc_xprt *xprt)
2590 {
2591 xprt_disconnect_done(xprt);
2592 }
2593
2594 /*
2595 * The xprt destroy routine. Again, because this connection is client
2596 * initiated, we do nothing
2597 */
2598
bc_destroy(struct rpc_xprt * xprt)2599 static void bc_destroy(struct rpc_xprt *xprt)
2600 {
2601 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2602
2603 xs_xprt_free(xprt);
2604 module_put(THIS_MODULE);
2605 }
2606
2607 static const struct rpc_xprt_ops xs_local_ops = {
2608 .reserve_xprt = xprt_reserve_xprt,
2609 .release_xprt = xprt_release_xprt,
2610 .alloc_slot = xprt_alloc_slot,
2611 .free_slot = xprt_free_slot,
2612 .rpcbind = xs_local_rpcbind,
2613 .set_port = xs_local_set_port,
2614 .connect = xs_local_connect,
2615 .buf_alloc = rpc_malloc,
2616 .buf_free = rpc_free,
2617 .prepare_request = xs_stream_prepare_request,
2618 .send_request = xs_local_send_request,
2619 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2620 .close = xs_close,
2621 .destroy = xs_destroy,
2622 .print_stats = xs_local_print_stats,
2623 .enable_swap = xs_enable_swap,
2624 .disable_swap = xs_disable_swap,
2625 };
2626
2627 static const struct rpc_xprt_ops xs_udp_ops = {
2628 .set_buffer_size = xs_udp_set_buffer_size,
2629 .reserve_xprt = xprt_reserve_xprt_cong,
2630 .release_xprt = xprt_release_xprt_cong,
2631 .alloc_slot = xprt_alloc_slot,
2632 .free_slot = xprt_free_slot,
2633 .rpcbind = rpcb_getport_async,
2634 .set_port = xs_set_port,
2635 .connect = xs_connect,
2636 .buf_alloc = rpc_malloc,
2637 .buf_free = rpc_free,
2638 .send_request = xs_udp_send_request,
2639 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2640 .timer = xs_udp_timer,
2641 .release_request = xprt_release_rqst_cong,
2642 .close = xs_close,
2643 .destroy = xs_destroy,
2644 .print_stats = xs_udp_print_stats,
2645 .enable_swap = xs_enable_swap,
2646 .disable_swap = xs_disable_swap,
2647 .inject_disconnect = xs_inject_disconnect,
2648 };
2649
2650 static const struct rpc_xprt_ops xs_tcp_ops = {
2651 .reserve_xprt = xprt_reserve_xprt,
2652 .release_xprt = xprt_release_xprt,
2653 .alloc_slot = xprt_alloc_slot,
2654 .free_slot = xprt_free_slot,
2655 .rpcbind = rpcb_getport_async,
2656 .set_port = xs_set_port,
2657 .connect = xs_connect,
2658 .buf_alloc = rpc_malloc,
2659 .buf_free = rpc_free,
2660 .prepare_request = xs_stream_prepare_request,
2661 .send_request = xs_tcp_send_request,
2662 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2663 .close = xs_tcp_shutdown,
2664 .destroy = xs_destroy,
2665 .set_connect_timeout = xs_tcp_set_connect_timeout,
2666 .print_stats = xs_tcp_print_stats,
2667 .enable_swap = xs_enable_swap,
2668 .disable_swap = xs_disable_swap,
2669 .inject_disconnect = xs_inject_disconnect,
2670 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2671 .bc_setup = xprt_setup_bc,
2672 .bc_maxpayload = xs_tcp_bc_maxpayload,
2673 .bc_num_slots = xprt_bc_max_slots,
2674 .bc_free_rqst = xprt_free_bc_rqst,
2675 .bc_destroy = xprt_destroy_bc,
2676 #endif
2677 };
2678
2679 /*
2680 * The rpc_xprt_ops for the server backchannel
2681 */
2682
2683 static const struct rpc_xprt_ops bc_tcp_ops = {
2684 .reserve_xprt = xprt_reserve_xprt,
2685 .release_xprt = xprt_release_xprt,
2686 .alloc_slot = xprt_alloc_slot,
2687 .free_slot = xprt_free_slot,
2688 .buf_alloc = bc_malloc,
2689 .buf_free = bc_free,
2690 .send_request = bc_send_request,
2691 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2692 .close = bc_close,
2693 .destroy = bc_destroy,
2694 .print_stats = xs_tcp_print_stats,
2695 .enable_swap = xs_enable_swap,
2696 .disable_swap = xs_disable_swap,
2697 .inject_disconnect = xs_inject_disconnect,
2698 };
2699
xs_init_anyaddr(const int family,struct sockaddr * sap)2700 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2701 {
2702 static const struct sockaddr_in sin = {
2703 .sin_family = AF_INET,
2704 .sin_addr.s_addr = htonl(INADDR_ANY),
2705 };
2706 static const struct sockaddr_in6 sin6 = {
2707 .sin6_family = AF_INET6,
2708 .sin6_addr = IN6ADDR_ANY_INIT,
2709 };
2710
2711 switch (family) {
2712 case AF_LOCAL:
2713 break;
2714 case AF_INET:
2715 memcpy(sap, &sin, sizeof(sin));
2716 break;
2717 case AF_INET6:
2718 memcpy(sap, &sin6, sizeof(sin6));
2719 break;
2720 default:
2721 dprintk("RPC: %s: Bad address family\n", __func__);
2722 return -EAFNOSUPPORT;
2723 }
2724 return 0;
2725 }
2726
xs_setup_xprt(struct xprt_create * args,unsigned int slot_table_size,unsigned int max_slot_table_size)2727 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2728 unsigned int slot_table_size,
2729 unsigned int max_slot_table_size)
2730 {
2731 struct rpc_xprt *xprt;
2732 struct sock_xprt *new;
2733
2734 if (args->addrlen > sizeof(xprt->addr)) {
2735 dprintk("RPC: xs_setup_xprt: address too large\n");
2736 return ERR_PTR(-EBADF);
2737 }
2738
2739 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2740 max_slot_table_size);
2741 if (xprt == NULL) {
2742 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2743 "rpc_xprt\n");
2744 return ERR_PTR(-ENOMEM);
2745 }
2746
2747 new = container_of(xprt, struct sock_xprt, xprt);
2748 mutex_init(&new->recv_mutex);
2749 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2750 xprt->addrlen = args->addrlen;
2751 if (args->srcaddr)
2752 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2753 else {
2754 int err;
2755 err = xs_init_anyaddr(args->dstaddr->sa_family,
2756 (struct sockaddr *)&new->srcaddr);
2757 if (err != 0) {
2758 xprt_free(xprt);
2759 return ERR_PTR(err);
2760 }
2761 }
2762
2763 return xprt;
2764 }
2765
2766 static const struct rpc_timeout xs_local_default_timeout = {
2767 .to_initval = 10 * HZ,
2768 .to_maxval = 10 * HZ,
2769 .to_retries = 2,
2770 };
2771
2772 /**
2773 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2774 * @args: rpc transport creation arguments
2775 *
2776 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2777 */
xs_setup_local(struct xprt_create * args)2778 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2779 {
2780 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2781 struct sock_xprt *transport;
2782 struct rpc_xprt *xprt;
2783 struct rpc_xprt *ret;
2784
2785 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2786 xprt_max_tcp_slot_table_entries);
2787 if (IS_ERR(xprt))
2788 return xprt;
2789 transport = container_of(xprt, struct sock_xprt, xprt);
2790
2791 xprt->prot = 0;
2792 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2793
2794 xprt->bind_timeout = XS_BIND_TO;
2795 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2796 xprt->idle_timeout = XS_IDLE_DISC_TO;
2797
2798 xprt->ops = &xs_local_ops;
2799 xprt->timeout = &xs_local_default_timeout;
2800
2801 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2802 INIT_WORK(&transport->error_worker, xs_error_handle);
2803 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2804
2805 switch (sun->sun_family) {
2806 case AF_LOCAL:
2807 if (sun->sun_path[0] != '/') {
2808 dprintk("RPC: bad AF_LOCAL address: %s\n",
2809 sun->sun_path);
2810 ret = ERR_PTR(-EINVAL);
2811 goto out_err;
2812 }
2813 xprt_set_bound(xprt);
2814 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2815 ret = ERR_PTR(xs_local_setup_socket(transport));
2816 if (ret)
2817 goto out_err;
2818 break;
2819 default:
2820 ret = ERR_PTR(-EAFNOSUPPORT);
2821 goto out_err;
2822 }
2823
2824 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2825 xprt->address_strings[RPC_DISPLAY_ADDR]);
2826
2827 if (try_module_get(THIS_MODULE))
2828 return xprt;
2829 ret = ERR_PTR(-EINVAL);
2830 out_err:
2831 xs_xprt_free(xprt);
2832 return ret;
2833 }
2834
2835 static const struct rpc_timeout xs_udp_default_timeout = {
2836 .to_initval = 5 * HZ,
2837 .to_maxval = 30 * HZ,
2838 .to_increment = 5 * HZ,
2839 .to_retries = 5,
2840 };
2841
2842 /**
2843 * xs_setup_udp - Set up transport to use a UDP socket
2844 * @args: rpc transport creation arguments
2845 *
2846 */
xs_setup_udp(struct xprt_create * args)2847 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2848 {
2849 struct sockaddr *addr = args->dstaddr;
2850 struct rpc_xprt *xprt;
2851 struct sock_xprt *transport;
2852 struct rpc_xprt *ret;
2853
2854 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2855 xprt_udp_slot_table_entries);
2856 if (IS_ERR(xprt))
2857 return xprt;
2858 transport = container_of(xprt, struct sock_xprt, xprt);
2859
2860 xprt->prot = IPPROTO_UDP;
2861 /* XXX: header size can vary due to auth type, IPv6, etc. */
2862 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2863
2864 xprt->bind_timeout = XS_BIND_TO;
2865 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2866 xprt->idle_timeout = XS_IDLE_DISC_TO;
2867
2868 xprt->ops = &xs_udp_ops;
2869
2870 xprt->timeout = &xs_udp_default_timeout;
2871
2872 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2873 INIT_WORK(&transport->error_worker, xs_error_handle);
2874 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2875
2876 switch (addr->sa_family) {
2877 case AF_INET:
2878 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2879 xprt_set_bound(xprt);
2880
2881 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2882 break;
2883 case AF_INET6:
2884 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2885 xprt_set_bound(xprt);
2886
2887 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2888 break;
2889 default:
2890 ret = ERR_PTR(-EAFNOSUPPORT);
2891 goto out_err;
2892 }
2893
2894 if (xprt_bound(xprt))
2895 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2896 xprt->address_strings[RPC_DISPLAY_ADDR],
2897 xprt->address_strings[RPC_DISPLAY_PORT],
2898 xprt->address_strings[RPC_DISPLAY_PROTO]);
2899 else
2900 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2901 xprt->address_strings[RPC_DISPLAY_ADDR],
2902 xprt->address_strings[RPC_DISPLAY_PROTO]);
2903
2904 if (try_module_get(THIS_MODULE))
2905 return xprt;
2906 ret = ERR_PTR(-EINVAL);
2907 out_err:
2908 xs_xprt_free(xprt);
2909 return ret;
2910 }
2911
2912 static const struct rpc_timeout xs_tcp_default_timeout = {
2913 .to_initval = 60 * HZ,
2914 .to_maxval = 60 * HZ,
2915 .to_retries = 2,
2916 };
2917
2918 /**
2919 * xs_setup_tcp - Set up transport to use a TCP socket
2920 * @args: rpc transport creation arguments
2921 *
2922 */
xs_setup_tcp(struct xprt_create * args)2923 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2924 {
2925 struct sockaddr *addr = args->dstaddr;
2926 struct rpc_xprt *xprt;
2927 struct sock_xprt *transport;
2928 struct rpc_xprt *ret;
2929 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2930
2931 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2932 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2933
2934 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2935 max_slot_table_size);
2936 if (IS_ERR(xprt))
2937 return xprt;
2938 transport = container_of(xprt, struct sock_xprt, xprt);
2939
2940 xprt->prot = IPPROTO_TCP;
2941 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2942
2943 xprt->bind_timeout = XS_BIND_TO;
2944 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2945 xprt->idle_timeout = XS_IDLE_DISC_TO;
2946
2947 xprt->ops = &xs_tcp_ops;
2948 xprt->timeout = &xs_tcp_default_timeout;
2949
2950 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2951 xprt->connect_timeout = xprt->timeout->to_initval *
2952 (xprt->timeout->to_retries + 1);
2953
2954 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2955 INIT_WORK(&transport->error_worker, xs_error_handle);
2956 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2957
2958 switch (addr->sa_family) {
2959 case AF_INET:
2960 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2961 xprt_set_bound(xprt);
2962
2963 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2964 break;
2965 case AF_INET6:
2966 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2967 xprt_set_bound(xprt);
2968
2969 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2970 break;
2971 default:
2972 ret = ERR_PTR(-EAFNOSUPPORT);
2973 goto out_err;
2974 }
2975
2976 if (xprt_bound(xprt))
2977 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2978 xprt->address_strings[RPC_DISPLAY_ADDR],
2979 xprt->address_strings[RPC_DISPLAY_PORT],
2980 xprt->address_strings[RPC_DISPLAY_PROTO]);
2981 else
2982 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2983 xprt->address_strings[RPC_DISPLAY_ADDR],
2984 xprt->address_strings[RPC_DISPLAY_PROTO]);
2985
2986 if (try_module_get(THIS_MODULE))
2987 return xprt;
2988 ret = ERR_PTR(-EINVAL);
2989 out_err:
2990 xs_xprt_free(xprt);
2991 return ret;
2992 }
2993
2994 /**
2995 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2996 * @args: rpc transport creation arguments
2997 *
2998 */
xs_setup_bc_tcp(struct xprt_create * args)2999 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3000 {
3001 struct sockaddr *addr = args->dstaddr;
3002 struct rpc_xprt *xprt;
3003 struct sock_xprt *transport;
3004 struct svc_sock *bc_sock;
3005 struct rpc_xprt *ret;
3006
3007 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3008 xprt_tcp_slot_table_entries);
3009 if (IS_ERR(xprt))
3010 return xprt;
3011 transport = container_of(xprt, struct sock_xprt, xprt);
3012
3013 xprt->prot = IPPROTO_TCP;
3014 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3015 xprt->timeout = &xs_tcp_default_timeout;
3016
3017 /* backchannel */
3018 xprt_set_bound(xprt);
3019 xprt->bind_timeout = 0;
3020 xprt->reestablish_timeout = 0;
3021 xprt->idle_timeout = 0;
3022
3023 xprt->ops = &bc_tcp_ops;
3024
3025 switch (addr->sa_family) {
3026 case AF_INET:
3027 xs_format_peer_addresses(xprt, "tcp",
3028 RPCBIND_NETID_TCP);
3029 break;
3030 case AF_INET6:
3031 xs_format_peer_addresses(xprt, "tcp",
3032 RPCBIND_NETID_TCP6);
3033 break;
3034 default:
3035 ret = ERR_PTR(-EAFNOSUPPORT);
3036 goto out_err;
3037 }
3038
3039 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3040 xprt->address_strings[RPC_DISPLAY_ADDR],
3041 xprt->address_strings[RPC_DISPLAY_PORT],
3042 xprt->address_strings[RPC_DISPLAY_PROTO]);
3043
3044 /*
3045 * Once we've associated a backchannel xprt with a connection,
3046 * we want to keep it around as long as the connection lasts,
3047 * in case we need to start using it for a backchannel again;
3048 * this reference won't be dropped until bc_xprt is destroyed.
3049 */
3050 xprt_get(xprt);
3051 args->bc_xprt->xpt_bc_xprt = xprt;
3052 xprt->bc_xprt = args->bc_xprt;
3053 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3054 transport->sock = bc_sock->sk_sock;
3055 transport->inet = bc_sock->sk_sk;
3056
3057 /*
3058 * Since we don't want connections for the backchannel, we set
3059 * the xprt status to connected
3060 */
3061 xprt_set_connected(xprt);
3062
3063 if (try_module_get(THIS_MODULE))
3064 return xprt;
3065
3066 args->bc_xprt->xpt_bc_xprt = NULL;
3067 args->bc_xprt->xpt_bc_xps = NULL;
3068 xprt_put(xprt);
3069 ret = ERR_PTR(-EINVAL);
3070 out_err:
3071 xs_xprt_free(xprt);
3072 return ret;
3073 }
3074
3075 static struct xprt_class xs_local_transport = {
3076 .list = LIST_HEAD_INIT(xs_local_transport.list),
3077 .name = "named UNIX socket",
3078 .owner = THIS_MODULE,
3079 .ident = XPRT_TRANSPORT_LOCAL,
3080 .setup = xs_setup_local,
3081 .netid = { "" },
3082 };
3083
3084 static struct xprt_class xs_udp_transport = {
3085 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3086 .name = "udp",
3087 .owner = THIS_MODULE,
3088 .ident = XPRT_TRANSPORT_UDP,
3089 .setup = xs_setup_udp,
3090 .netid = { "udp", "udp6", "" },
3091 };
3092
3093 static struct xprt_class xs_tcp_transport = {
3094 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3095 .name = "tcp",
3096 .owner = THIS_MODULE,
3097 .ident = XPRT_TRANSPORT_TCP,
3098 .setup = xs_setup_tcp,
3099 .netid = { "tcp", "tcp6", "" },
3100 };
3101
3102 static struct xprt_class xs_bc_tcp_transport = {
3103 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3104 .name = "tcp NFSv4.1 backchannel",
3105 .owner = THIS_MODULE,
3106 .ident = XPRT_TRANSPORT_BC_TCP,
3107 .setup = xs_setup_bc_tcp,
3108 .netid = { "" },
3109 };
3110
3111 /**
3112 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3113 *
3114 */
init_socket_xprt(void)3115 int init_socket_xprt(void)
3116 {
3117 if (!sunrpc_table_header)
3118 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3119
3120 xprt_register_transport(&xs_local_transport);
3121 xprt_register_transport(&xs_udp_transport);
3122 xprt_register_transport(&xs_tcp_transport);
3123 xprt_register_transport(&xs_bc_tcp_transport);
3124
3125 return 0;
3126 }
3127
3128 /**
3129 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3130 *
3131 */
cleanup_socket_xprt(void)3132 void cleanup_socket_xprt(void)
3133 {
3134 if (sunrpc_table_header) {
3135 unregister_sysctl_table(sunrpc_table_header);
3136 sunrpc_table_header = NULL;
3137 }
3138
3139 xprt_unregister_transport(&xs_local_transport);
3140 xprt_unregister_transport(&xs_udp_transport);
3141 xprt_unregister_transport(&xs_tcp_transport);
3142 xprt_unregister_transport(&xs_bc_tcp_transport);
3143 }
3144
param_set_uint_minmax(const char * val,const struct kernel_param * kp,unsigned int min,unsigned int max)3145 static int param_set_uint_minmax(const char *val,
3146 const struct kernel_param *kp,
3147 unsigned int min, unsigned int max)
3148 {
3149 unsigned int num;
3150 int ret;
3151
3152 if (!val)
3153 return -EINVAL;
3154 ret = kstrtouint(val, 0, &num);
3155 if (ret)
3156 return ret;
3157 if (num < min || num > max)
3158 return -EINVAL;
3159 *((unsigned int *)kp->arg) = num;
3160 return 0;
3161 }
3162
param_set_portnr(const char * val,const struct kernel_param * kp)3163 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3164 {
3165 return param_set_uint_minmax(val, kp,
3166 RPC_MIN_RESVPORT,
3167 RPC_MAX_RESVPORT);
3168 }
3169
3170 static const struct kernel_param_ops param_ops_portnr = {
3171 .set = param_set_portnr,
3172 .get = param_get_uint,
3173 };
3174
3175 #define param_check_portnr(name, p) \
3176 __param_check(name, p, unsigned int);
3177
3178 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3179 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3180
param_set_slot_table_size(const char * val,const struct kernel_param * kp)3181 static int param_set_slot_table_size(const char *val,
3182 const struct kernel_param *kp)
3183 {
3184 return param_set_uint_minmax(val, kp,
3185 RPC_MIN_SLOT_TABLE,
3186 RPC_MAX_SLOT_TABLE);
3187 }
3188
3189 static const struct kernel_param_ops param_ops_slot_table_size = {
3190 .set = param_set_slot_table_size,
3191 .get = param_get_uint,
3192 };
3193
3194 #define param_check_slot_table_size(name, p) \
3195 __param_check(name, p, unsigned int);
3196
param_set_max_slot_table_size(const char * val,const struct kernel_param * kp)3197 static int param_set_max_slot_table_size(const char *val,
3198 const struct kernel_param *kp)
3199 {
3200 return param_set_uint_minmax(val, kp,
3201 RPC_MIN_SLOT_TABLE,
3202 RPC_MAX_SLOT_TABLE_LIMIT);
3203 }
3204
3205 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3206 .set = param_set_max_slot_table_size,
3207 .get = param_get_uint,
3208 };
3209
3210 #define param_check_max_slot_table_size(name, p) \
3211 __param_check(name, p, unsigned int);
3212
3213 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3214 slot_table_size, 0644);
3215 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3216 max_slot_table_size, 0644);
3217 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3218 slot_table_size, 0644);
3219