Merge tag 'nfs-for-4.21-1' of git://git.linux-nfs.org/projects/anna/linux-nfs
[sfrench/cifs-2.6.git] / net / sunrpc / xprtsock.c
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/uio.h>
52
53 #include <trace/events/sunrpc.h>
54
55 #include "sunrpc.h"
56
57 static void xs_close(struct rpc_xprt *xprt);
58 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
59                 struct socket *sock);
60
61 /*
62  * xprtsock tunables
63  */
64 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
67
68 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
70
71 #define XS_TCP_LINGER_TO        (15U * HZ)
72 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
73
74 /*
75  * We can register our own files under /proc/sys/sunrpc by
76  * calling register_sysctl_table() again.  The files in that
77  * directory become the union of all files registered there.
78  *
79  * We simply need to make sure that we don't collide with
80  * someone else's file names!
81  */
82
83 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
84 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
85 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
86 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
87 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
88
89 static struct ctl_table_header *sunrpc_table_header;
90
91 /*
92  * FIXME: changing the UDP slot table size should also resize the UDP
93  *        socket buffers for existing UDP transports
94  */
95 static struct ctl_table xs_tunables_table[] = {
96         {
97                 .procname       = "udp_slot_table_entries",
98                 .data           = &xprt_udp_slot_table_entries,
99                 .maxlen         = sizeof(unsigned int),
100                 .mode           = 0644,
101                 .proc_handler   = proc_dointvec_minmax,
102                 .extra1         = &min_slot_table_size,
103                 .extra2         = &max_slot_table_size
104         },
105         {
106                 .procname       = "tcp_slot_table_entries",
107                 .data           = &xprt_tcp_slot_table_entries,
108                 .maxlen         = sizeof(unsigned int),
109                 .mode           = 0644,
110                 .proc_handler   = proc_dointvec_minmax,
111                 .extra1         = &min_slot_table_size,
112                 .extra2         = &max_slot_table_size
113         },
114         {
115                 .procname       = "tcp_max_slot_table_entries",
116                 .data           = &xprt_max_tcp_slot_table_entries,
117                 .maxlen         = sizeof(unsigned int),
118                 .mode           = 0644,
119                 .proc_handler   = proc_dointvec_minmax,
120                 .extra1         = &min_slot_table_size,
121                 .extra2         = &max_tcp_slot_table_limit
122         },
123         {
124                 .procname       = "min_resvport",
125                 .data           = &xprt_min_resvport,
126                 .maxlen         = sizeof(unsigned int),
127                 .mode           = 0644,
128                 .proc_handler   = proc_dointvec_minmax,
129                 .extra1         = &xprt_min_resvport_limit,
130                 .extra2         = &xprt_max_resvport_limit
131         },
132         {
133                 .procname       = "max_resvport",
134                 .data           = &xprt_max_resvport,
135                 .maxlen         = sizeof(unsigned int),
136                 .mode           = 0644,
137                 .proc_handler   = proc_dointvec_minmax,
138                 .extra1         = &xprt_min_resvport_limit,
139                 .extra2         = &xprt_max_resvport_limit
140         },
141         {
142                 .procname       = "tcp_fin_timeout",
143                 .data           = &xs_tcp_fin_timeout,
144                 .maxlen         = sizeof(xs_tcp_fin_timeout),
145                 .mode           = 0644,
146                 .proc_handler   = proc_dointvec_jiffies,
147         },
148         { },
149 };
150
151 static struct ctl_table sunrpc_table[] = {
152         {
153                 .procname       = "sunrpc",
154                 .mode           = 0555,
155                 .child          = xs_tunables_table
156         },
157         { },
158 };
159
160 /*
161  * Wait duration for a reply from the RPC portmapper.
162  */
163 #define XS_BIND_TO              (60U * HZ)
164
165 /*
166  * Delay if a UDP socket connect error occurs.  This is most likely some
167  * kind of resource problem on the local host.
168  */
169 #define XS_UDP_REEST_TO         (2U * HZ)
170
171 /*
172  * The reestablish timeout allows clients to delay for a bit before attempting
173  * to reconnect to a server that just dropped our connection.
174  *
175  * We implement an exponential backoff when trying to reestablish a TCP
176  * transport connection with the server.  Some servers like to drop a TCP
177  * connection when they are overworked, so we start with a short timeout and
178  * increase over time if the server is down or not responding.
179  */
180 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
181
182 /*
183  * TCP idle timeout; client drops the transport socket if it is idle
184  * for this long.  Note that we also timeout UDP sockets to prevent
185  * holding port numbers when there is no RPC traffic.
186  */
187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
188
189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
190 # undef  RPC_DEBUG_DATA
191 # define RPCDBG_FACILITY        RPCDBG_TRANS
192 #endif
193
194 #ifdef RPC_DEBUG_DATA
195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196 {
197         u8 *buf = (u8 *) packet;
198         int j;
199
200         dprintk("RPC:       %s\n", msg);
201         for (j = 0; j < count && j < 128; j += 4) {
202                 if (!(j & 31)) {
203                         if (j)
204                                 dprintk("\n");
205                         dprintk("0x%04x ", j);
206                 }
207                 dprintk("%02x%02x%02x%02x ",
208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
209         }
210         dprintk("\n");
211 }
212 #else
213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
214 {
215         /* NOP */
216 }
217 #endif
218
219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220 {
221         return (struct rpc_xprt *) sk->sk_user_data;
222 }
223
224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225 {
226         return (struct sockaddr *) &xprt->addr;
227 }
228
229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230 {
231         return (struct sockaddr_un *) &xprt->addr;
232 }
233
234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
235 {
236         return (struct sockaddr_in *) &xprt->addr;
237 }
238
239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240 {
241         return (struct sockaddr_in6 *) &xprt->addr;
242 }
243
244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
245 {
246         struct sockaddr *sap = xs_addr(xprt);
247         struct sockaddr_in6 *sin6;
248         struct sockaddr_in *sin;
249         struct sockaddr_un *sun;
250         char buf[128];
251
252         switch (sap->sa_family) {
253         case AF_LOCAL:
254                 sun = xs_addr_un(xprt);
255                 strlcpy(buf, sun->sun_path, sizeof(buf));
256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
257                                                 kstrdup(buf, GFP_KERNEL);
258                 break;
259         case AF_INET:
260                 (void)rpc_ntop(sap, buf, sizeof(buf));
261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
262                                                 kstrdup(buf, GFP_KERNEL);
263                 sin = xs_addr_in(xprt);
264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
265                 break;
266         case AF_INET6:
267                 (void)rpc_ntop(sap, buf, sizeof(buf));
268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
269                                                 kstrdup(buf, GFP_KERNEL);
270                 sin6 = xs_addr_in6(xprt);
271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
272                 break;
273         default:
274                 BUG();
275         }
276
277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
278 }
279
280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
281 {
282         struct sockaddr *sap = xs_addr(xprt);
283         char buf[128];
284
285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
287
288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
290 }
291
292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293                                      const char *protocol,
294                                      const char *netid)
295 {
296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
298         xs_format_common_peer_addresses(xprt);
299         xs_format_common_peer_ports(xprt);
300 }
301
302 static void xs_update_peer_port(struct rpc_xprt *xprt)
303 {
304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
306
307         xs_format_common_peer_ports(xprt);
308 }
309
310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
311 {
312         unsigned int i;
313
314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
315                 switch (i) {
316                 case RPC_DISPLAY_PROTO:
317                 case RPC_DISPLAY_NETID:
318                         continue;
319                 default:
320                         kfree(xprt->address_strings[i]);
321                 }
322 }
323
324 static size_t
325 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
326 {
327         size_t i,n;
328
329         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
330                 return want;
331         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
332         for (i = 0; i < n; i++) {
333                 if (buf->pages[i])
334                         continue;
335                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
336                 if (!buf->pages[i]) {
337                         i *= PAGE_SIZE;
338                         return i > buf->page_base ? i - buf->page_base : 0;
339                 }
340         }
341         return want;
342 }
343
344 static ssize_t
345 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
346 {
347         ssize_t ret;
348         if (seek != 0)
349                 iov_iter_advance(&msg->msg_iter, seek);
350         ret = sock_recvmsg(sock, msg, flags);
351         return ret > 0 ? ret + seek : ret;
352 }
353
354 static ssize_t
355 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
356                 struct kvec *kvec, size_t count, size_t seek)
357 {
358         iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
359         return xs_sock_recvmsg(sock, msg, flags, seek);
360 }
361
362 static ssize_t
363 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
364                 struct bio_vec *bvec, unsigned long nr, size_t count,
365                 size_t seek)
366 {
367         iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
368         return xs_sock_recvmsg(sock, msg, flags, seek);
369 }
370
371 static ssize_t
372 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
373                 size_t count)
374 {
375         iov_iter_discard(&msg->msg_iter, READ, count);
376         return sock_recvmsg(sock, msg, flags);
377 }
378
379 static ssize_t
380 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
381                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
382 {
383         size_t want, seek_init = seek, offset = 0;
384         ssize_t ret;
385
386         if (seek < buf->head[0].iov_len) {
387                 want = min_t(size_t, count, buf->head[0].iov_len);
388                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
389                 if (ret <= 0)
390                         goto sock_err;
391                 offset += ret;
392                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
393                         goto out;
394                 if (ret != want)
395                         goto out;
396                 seek = 0;
397         } else {
398                 seek -= buf->head[0].iov_len;
399                 offset += buf->head[0].iov_len;
400         }
401
402         want = xs_alloc_sparse_pages(buf,
403                         min_t(size_t, count - offset, buf->page_len),
404                         GFP_NOWAIT);
405         if (seek < want) {
406                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
407                                 xdr_buf_pagecount(buf),
408                                 want + buf->page_base,
409                                 seek + buf->page_base);
410                 if (ret <= 0)
411                         goto sock_err;
412                 offset += ret - buf->page_base;
413                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
414                         goto out;
415                 if (ret != want)
416                         goto out;
417                 seek = 0;
418         } else {
419                 seek -= want;
420                 offset += want;
421         }
422
423         if (seek < buf->tail[0].iov_len) {
424                 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
425                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
426                 if (ret <= 0)
427                         goto sock_err;
428                 offset += ret;
429                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
430                         goto out;
431                 if (ret != want)
432                         goto out;
433         } else
434                 offset += buf->tail[0].iov_len;
435         ret = -EMSGSIZE;
436 out:
437         *read = offset - seek_init;
438         return ret;
439 sock_err:
440         offset += seek;
441         goto out;
442 }
443
444 static void
445 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
446 {
447         if (!transport->recv.copied) {
448                 if (buf->head[0].iov_len >= transport->recv.offset)
449                         memcpy(buf->head[0].iov_base,
450                                         &transport->recv.xid,
451                                         transport->recv.offset);
452                 transport->recv.copied = transport->recv.offset;
453         }
454 }
455
456 static bool
457 xs_read_stream_request_done(struct sock_xprt *transport)
458 {
459         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
460 }
461
462 static ssize_t
463 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
464                 int flags, struct rpc_rqst *req)
465 {
466         struct xdr_buf *buf = &req->rq_private_buf;
467         size_t want, read;
468         ssize_t ret;
469
470         xs_read_header(transport, buf);
471
472         want = transport->recv.len - transport->recv.offset;
473         ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
474                         transport->recv.copied + want, transport->recv.copied,
475                         &read);
476         transport->recv.offset += read;
477         transport->recv.copied += read;
478         if (transport->recv.offset == transport->recv.len) {
479                 if (xs_read_stream_request_done(transport))
480                         msg->msg_flags |= MSG_EOR;
481                 return read;
482         }
483
484         switch (ret) {
485         default:
486                 break;
487         case -EFAULT:
488         case -EMSGSIZE:
489                 msg->msg_flags |= MSG_TRUNC;
490                 return read;
491         case 0:
492                 return -ESHUTDOWN;
493         }
494         return ret < 0 ? ret : read;
495 }
496
497 static size_t
498 xs_read_stream_headersize(bool isfrag)
499 {
500         if (isfrag)
501                 return sizeof(__be32);
502         return 3 * sizeof(__be32);
503 }
504
505 static ssize_t
506 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
507                 int flags, size_t want, size_t seek)
508 {
509         struct kvec kvec = {
510                 .iov_base = &transport->recv.fraghdr,
511                 .iov_len = want,
512         };
513         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
514 }
515
516 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
517 static ssize_t
518 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
519 {
520         struct rpc_xprt *xprt = &transport->xprt;
521         struct rpc_rqst *req;
522         ssize_t ret;
523
524         /* Look up and lock the request corresponding to the given XID */
525         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
526         if (!req) {
527                 printk(KERN_WARNING "Callback slot table overflowed\n");
528                 return -ESHUTDOWN;
529         }
530
531         ret = xs_read_stream_request(transport, msg, flags, req);
532         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
533                 xprt_complete_bc_request(req, transport->recv.copied);
534
535         return ret;
536 }
537 #else /* CONFIG_SUNRPC_BACKCHANNEL */
538 static ssize_t
539 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
540 {
541         return -ESHUTDOWN;
542 }
543 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
544
545 static ssize_t
546 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
547 {
548         struct rpc_xprt *xprt = &transport->xprt;
549         struct rpc_rqst *req;
550         ssize_t ret = 0;
551
552         /* Look up and lock the request corresponding to the given XID */
553         spin_lock(&xprt->queue_lock);
554         req = xprt_lookup_rqst(xprt, transport->recv.xid);
555         if (!req) {
556                 msg->msg_flags |= MSG_TRUNC;
557                 goto out;
558         }
559         xprt_pin_rqst(req);
560         spin_unlock(&xprt->queue_lock);
561
562         ret = xs_read_stream_request(transport, msg, flags, req);
563
564         spin_lock(&xprt->queue_lock);
565         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
566                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
567         xprt_unpin_rqst(req);
568 out:
569         spin_unlock(&xprt->queue_lock);
570         return ret;
571 }
572
573 static ssize_t
574 xs_read_stream(struct sock_xprt *transport, int flags)
575 {
576         struct msghdr msg = { 0 };
577         size_t want, read = 0;
578         ssize_t ret = 0;
579
580         if (transport->recv.len == 0) {
581                 want = xs_read_stream_headersize(transport->recv.copied != 0);
582                 ret = xs_read_stream_header(transport, &msg, flags, want,
583                                 transport->recv.offset);
584                 if (ret <= 0)
585                         goto out_err;
586                 transport->recv.offset = ret;
587                 if (transport->recv.offset != want)
588                         return transport->recv.offset;
589                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
590                         RPC_FRAGMENT_SIZE_MASK;
591                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
592                 read = ret;
593         }
594
595         switch (be32_to_cpu(transport->recv.calldir)) {
596         default:
597                 msg.msg_flags |= MSG_TRUNC;
598                 break;
599         case RPC_CALL:
600                 ret = xs_read_stream_call(transport, &msg, flags);
601                 break;
602         case RPC_REPLY:
603                 ret = xs_read_stream_reply(transport, &msg, flags);
604         }
605         if (msg.msg_flags & MSG_TRUNC) {
606                 transport->recv.calldir = cpu_to_be32(-1);
607                 transport->recv.copied = -1;
608         }
609         if (ret < 0)
610                 goto out_err;
611         read += ret;
612         if (transport->recv.offset < transport->recv.len) {
613                 if (!(msg.msg_flags & MSG_TRUNC))
614                         return read;
615                 msg.msg_flags = 0;
616                 ret = xs_read_discard(transport->sock, &msg, flags,
617                                 transport->recv.len - transport->recv.offset);
618                 if (ret <= 0)
619                         goto out_err;
620                 transport->recv.offset += ret;
621                 read += ret;
622                 if (transport->recv.offset != transport->recv.len)
623                         return read;
624         }
625         if (xs_read_stream_request_done(transport)) {
626                 trace_xs_stream_read_request(transport);
627                 transport->recv.copied = 0;
628         }
629         transport->recv.offset = 0;
630         transport->recv.len = 0;
631         return read;
632 out_err:
633         return ret != 0 ? ret : -ESHUTDOWN;
634 }
635
636 static void xs_stream_data_receive(struct sock_xprt *transport)
637 {
638         size_t read = 0;
639         ssize_t ret = 0;
640
641         mutex_lock(&transport->recv_mutex);
642         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
643         if (transport->sock == NULL)
644                 goto out;
645         for (;;) {
646                 ret = xs_read_stream(transport, MSG_DONTWAIT);
647                 if (ret < 0)
648                         break;
649                 read += ret;
650                 cond_resched();
651         }
652 out:
653         mutex_unlock(&transport->recv_mutex);
654         trace_xs_stream_read_data(&transport->xprt, ret, read);
655 }
656
657 static void xs_stream_data_receive_workfn(struct work_struct *work)
658 {
659         struct sock_xprt *transport =
660                 container_of(work, struct sock_xprt, recv_worker);
661         xs_stream_data_receive(transport);
662 }
663
664 static void
665 xs_stream_reset_connect(struct sock_xprt *transport)
666 {
667         transport->recv.offset = 0;
668         transport->recv.len = 0;
669         transport->recv.copied = 0;
670         transport->xmit.offset = 0;
671         transport->xprt.stat.connect_count++;
672         transport->xprt.stat.connect_start = jiffies;
673 }
674
675 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
676
677 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
678 {
679         struct msghdr msg = {
680                 .msg_name       = addr,
681                 .msg_namelen    = addrlen,
682                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
683         };
684         struct kvec iov = {
685                 .iov_base       = vec->iov_base + base,
686                 .iov_len        = vec->iov_len - base,
687         };
688
689         if (iov.iov_len != 0)
690                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
691         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
692 }
693
694 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
695 {
696         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
697                         int offset, size_t size, int flags);
698         struct page **ppage;
699         unsigned int remainder;
700         int err;
701
702         remainder = xdr->page_len - base;
703         base += xdr->page_base;
704         ppage = xdr->pages + (base >> PAGE_SHIFT);
705         base &= ~PAGE_MASK;
706         do_sendpage = sock->ops->sendpage;
707         if (!zerocopy)
708                 do_sendpage = sock_no_sendpage;
709         for(;;) {
710                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
711                 int flags = XS_SENDMSG_FLAGS;
712
713                 remainder -= len;
714                 if (more)
715                         flags |= MSG_MORE;
716                 if (remainder != 0)
717                         flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
718                 err = do_sendpage(sock, *ppage, base, len, flags);
719                 if (remainder == 0 || err != len)
720                         break;
721                 *sent_p += err;
722                 ppage++;
723                 base = 0;
724         }
725         if (err > 0) {
726                 *sent_p += err;
727                 err = 0;
728         }
729         return err;
730 }
731
732 /**
733  * xs_sendpages - write pages directly to a socket
734  * @sock: socket to send on
735  * @addr: UDP only -- address of destination
736  * @addrlen: UDP only -- length of destination address
737  * @xdr: buffer containing this request
738  * @base: starting position in the buffer
739  * @zerocopy: true if it is safe to use sendpage()
740  * @sent_p: return the total number of bytes successfully queued for sending
741  *
742  */
743 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
744 {
745         unsigned int remainder = xdr->len - base;
746         int err = 0;
747         int sent = 0;
748
749         if (unlikely(!sock))
750                 return -ENOTSOCK;
751
752         if (base != 0) {
753                 addr = NULL;
754                 addrlen = 0;
755         }
756
757         if (base < xdr->head[0].iov_len || addr != NULL) {
758                 unsigned int len = xdr->head[0].iov_len - base;
759                 remainder -= len;
760                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
761                 if (remainder == 0 || err != len)
762                         goto out;
763                 *sent_p += err;
764                 base = 0;
765         } else
766                 base -= xdr->head[0].iov_len;
767
768         if (base < xdr->page_len) {
769                 unsigned int len = xdr->page_len - base;
770                 remainder -= len;
771                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
772                 *sent_p += sent;
773                 if (remainder == 0 || sent != len)
774                         goto out;
775                 base = 0;
776         } else
777                 base -= xdr->page_len;
778
779         if (base >= xdr->tail[0].iov_len)
780                 return 0;
781         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
782 out:
783         if (err > 0) {
784                 *sent_p += err;
785                 err = 0;
786         }
787         return err;
788 }
789
790 /**
791  * xs_nospace - handle transmit was incomplete
792  * @req: pointer to RPC request
793  *
794  */
795 static int xs_nospace(struct rpc_rqst *req)
796 {
797         struct rpc_xprt *xprt = req->rq_xprt;
798         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
799         struct sock *sk = transport->inet;
800         int ret = -EAGAIN;
801
802         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
803                         req->rq_task->tk_pid,
804                         req->rq_slen - transport->xmit.offset,
805                         req->rq_slen);
806
807         /* Protect against races with write_space */
808         spin_lock_bh(&xprt->transport_lock);
809
810         /* Don't race with disconnect */
811         if (xprt_connected(xprt)) {
812                 /* wait for more buffer space */
813                 sk->sk_write_pending++;
814                 xprt_wait_for_buffer_space(xprt);
815         } else
816                 ret = -ENOTCONN;
817
818         spin_unlock_bh(&xprt->transport_lock);
819
820         /* Race breaker in case memory is freed before above code is called */
821         if (ret == -EAGAIN) {
822                 struct socket_wq *wq;
823
824                 rcu_read_lock();
825                 wq = rcu_dereference(sk->sk_wq);
826                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
827                 rcu_read_unlock();
828
829                 sk->sk_write_space(sk);
830         }
831         return ret;
832 }
833
834 static void
835 xs_stream_prepare_request(struct rpc_rqst *req)
836 {
837         req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_NOIO);
838 }
839
840 /*
841  * Determine if the previous message in the stream was aborted before it
842  * could complete transmission.
843  */
844 static bool
845 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
846 {
847         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
848 }
849
850 /*
851  * Construct a stream transport record marker in @buf.
852  */
853 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
854 {
855         u32 reclen = buf->len - sizeof(rpc_fraghdr);
856         rpc_fraghdr *base = buf->head[0].iov_base;
857         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
858 }
859
860 /**
861  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
862  * @req: pointer to RPC request
863  *
864  * Return values:
865  *        0:    The request has been sent
866  *   EAGAIN:    The socket was blocked, please call again later to
867  *              complete the request
868  * ENOTCONN:    Caller needs to invoke connect logic then call again
869  *    other:    Some other error occured, the request was not sent
870  */
871 static int xs_local_send_request(struct rpc_rqst *req)
872 {
873         struct rpc_xprt *xprt = req->rq_xprt;
874         struct sock_xprt *transport =
875                                 container_of(xprt, struct sock_xprt, xprt);
876         struct xdr_buf *xdr = &req->rq_snd_buf;
877         int status;
878         int sent = 0;
879
880         /* Close the stream if the previous transmission was incomplete */
881         if (xs_send_request_was_aborted(transport, req)) {
882                 xs_close(xprt);
883                 return -ENOTCONN;
884         }
885
886         xs_encode_stream_record_marker(&req->rq_snd_buf);
887
888         xs_pktdump("packet data:",
889                         req->rq_svec->iov_base, req->rq_svec->iov_len);
890
891         req->rq_xtime = ktime_get();
892         status = xs_sendpages(transport->sock, NULL, 0, xdr,
893                               transport->xmit.offset,
894                               true, &sent);
895         dprintk("RPC:       %s(%u) = %d\n",
896                         __func__, xdr->len - transport->xmit.offset, status);
897
898         if (status == -EAGAIN && sock_writeable(transport->inet))
899                 status = -ENOBUFS;
900
901         if (likely(sent > 0) || status == 0) {
902                 transport->xmit.offset += sent;
903                 req->rq_bytes_sent = transport->xmit.offset;
904                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
905                         req->rq_xmit_bytes_sent += transport->xmit.offset;
906                         req->rq_bytes_sent = 0;
907                         transport->xmit.offset = 0;
908                         return 0;
909                 }
910                 status = -EAGAIN;
911         }
912
913         switch (status) {
914         case -ENOBUFS:
915                 break;
916         case -EAGAIN:
917                 status = xs_nospace(req);
918                 break;
919         default:
920                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
921                         -status);
922                 /* fall through */
923         case -EPIPE:
924                 xs_close(xprt);
925                 status = -ENOTCONN;
926         }
927
928         return status;
929 }
930
931 /**
932  * xs_udp_send_request - write an RPC request to a UDP socket
933  * @req: pointer to RPC request
934  *
935  * Return values:
936  *        0:    The request has been sent
937  *   EAGAIN:    The socket was blocked, please call again later to
938  *              complete the request
939  * ENOTCONN:    Caller needs to invoke connect logic then call again
940  *    other:    Some other error occurred, the request was not sent
941  */
942 static int xs_udp_send_request(struct rpc_rqst *req)
943 {
944         struct rpc_xprt *xprt = req->rq_xprt;
945         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
946         struct xdr_buf *xdr = &req->rq_snd_buf;
947         int sent = 0;
948         int status;
949
950         xs_pktdump("packet data:",
951                                 req->rq_svec->iov_base,
952                                 req->rq_svec->iov_len);
953
954         if (!xprt_bound(xprt))
955                 return -ENOTCONN;
956
957         if (!xprt_request_get_cong(xprt, req))
958                 return -EBADSLT;
959
960         req->rq_xtime = ktime_get();
961         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
962                               xdr, 0, true, &sent);
963
964         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
965                         xdr->len, status);
966
967         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
968         if (status == -EPERM)
969                 goto process_status;
970
971         if (status == -EAGAIN && sock_writeable(transport->inet))
972                 status = -ENOBUFS;
973
974         if (sent > 0 || status == 0) {
975                 req->rq_xmit_bytes_sent += sent;
976                 if (sent >= req->rq_slen)
977                         return 0;
978                 /* Still some bytes left; set up for a retry later. */
979                 status = -EAGAIN;
980         }
981
982 process_status:
983         switch (status) {
984         case -ENOTSOCK:
985                 status = -ENOTCONN;
986                 /* Should we call xs_close() here? */
987                 break;
988         case -EAGAIN:
989                 status = xs_nospace(req);
990                 break;
991         case -ENETUNREACH:
992         case -ENOBUFS:
993         case -EPIPE:
994         case -ECONNREFUSED:
995         case -EPERM:
996                 /* When the server has died, an ICMP port unreachable message
997                  * prompts ECONNREFUSED. */
998                 break;
999         default:
1000                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1001                         -status);
1002         }
1003
1004         return status;
1005 }
1006
1007 /**
1008  * xs_tcp_send_request - write an RPC request to a TCP socket
1009  * @req: pointer to RPC request
1010  *
1011  * Return values:
1012  *        0:    The request has been sent
1013  *   EAGAIN:    The socket was blocked, please call again later to
1014  *              complete the request
1015  * ENOTCONN:    Caller needs to invoke connect logic then call again
1016  *    other:    Some other error occurred, the request was not sent
1017  *
1018  * XXX: In the case of soft timeouts, should we eventually give up
1019  *      if sendmsg is not able to make progress?
1020  */
1021 static int xs_tcp_send_request(struct rpc_rqst *req)
1022 {
1023         struct rpc_xprt *xprt = req->rq_xprt;
1024         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1025         struct xdr_buf *xdr = &req->rq_snd_buf;
1026         bool zerocopy = true;
1027         bool vm_wait = false;
1028         int status;
1029         int sent;
1030
1031         /* Close the stream if the previous transmission was incomplete */
1032         if (xs_send_request_was_aborted(transport, req)) {
1033                 if (transport->sock != NULL)
1034                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1035                 return -ENOTCONN;
1036         }
1037
1038         xs_encode_stream_record_marker(&req->rq_snd_buf);
1039
1040         xs_pktdump("packet data:",
1041                                 req->rq_svec->iov_base,
1042                                 req->rq_svec->iov_len);
1043         /* Don't use zero copy if this is a resend. If the RPC call
1044          * completes while the socket holds a reference to the pages,
1045          * then we may end up resending corrupted data.
1046          */
1047         if (req->rq_task->tk_flags & RPC_TASK_SENT)
1048                 zerocopy = false;
1049
1050         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1051                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1052
1053         /* Continue transmitting the packet/record. We must be careful
1054          * to cope with writespace callbacks arriving _after_ we have
1055          * called sendmsg(). */
1056         req->rq_xtime = ktime_get();
1057         while (1) {
1058                 sent = 0;
1059                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1060                                       transport->xmit.offset,
1061                                       zerocopy, &sent);
1062
1063                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1064                                 xdr->len - transport->xmit.offset, status);
1065
1066                 /* If we've sent the entire packet, immediately
1067                  * reset the count of bytes sent. */
1068                 transport->xmit.offset += sent;
1069                 req->rq_bytes_sent = transport->xmit.offset;
1070                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
1071                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1072                         req->rq_bytes_sent = 0;
1073                         transport->xmit.offset = 0;
1074                         return 0;
1075                 }
1076
1077                 WARN_ON_ONCE(sent == 0 && status == 0);
1078
1079                 if (status == -EAGAIN ) {
1080                         /*
1081                          * Return EAGAIN if we're sure we're hitting the
1082                          * socket send buffer limits.
1083                          */
1084                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1085                                 break;
1086                         /*
1087                          * Did we hit a memory allocation failure?
1088                          */
1089                         if (sent == 0) {
1090                                 status = -ENOBUFS;
1091                                 if (vm_wait)
1092                                         break;
1093                                 /* Retry, knowing now that we're below the
1094                                  * socket send buffer limit
1095                                  */
1096                                 vm_wait = true;
1097                         }
1098                         continue;
1099                 }
1100                 if (status < 0)
1101                         break;
1102                 vm_wait = false;
1103         }
1104
1105         switch (status) {
1106         case -ENOTSOCK:
1107                 status = -ENOTCONN;
1108                 /* Should we call xs_close() here? */
1109                 break;
1110         case -EAGAIN:
1111                 status = xs_nospace(req);
1112                 break;
1113         case -ECONNRESET:
1114         case -ECONNREFUSED:
1115         case -ENOTCONN:
1116         case -EADDRINUSE:
1117         case -ENOBUFS:
1118         case -EPIPE:
1119                 break;
1120         default:
1121                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1122                         -status);
1123         }
1124
1125         return status;
1126 }
1127
1128 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1129 {
1130         transport->old_data_ready = sk->sk_data_ready;
1131         transport->old_state_change = sk->sk_state_change;
1132         transport->old_write_space = sk->sk_write_space;
1133         transport->old_error_report = sk->sk_error_report;
1134 }
1135
1136 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1137 {
1138         sk->sk_data_ready = transport->old_data_ready;
1139         sk->sk_state_change = transport->old_state_change;
1140         sk->sk_write_space = transport->old_write_space;
1141         sk->sk_error_report = transport->old_error_report;
1142 }
1143
1144 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1145 {
1146         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1147
1148         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1149 }
1150
1151 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1152 {
1153         smp_mb__before_atomic();
1154         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1155         clear_bit(XPRT_CLOSING, &xprt->state);
1156         xs_sock_reset_state_flags(xprt);
1157         smp_mb__after_atomic();
1158 }
1159
1160 /**
1161  * xs_error_report - callback to handle TCP socket state errors
1162  * @sk: socket
1163  *
1164  * Note: we don't call sock_error() since there may be a rpc_task
1165  * using the socket, and so we don't want to clear sk->sk_err.
1166  */
1167 static void xs_error_report(struct sock *sk)
1168 {
1169         struct rpc_xprt *xprt;
1170         int err;
1171
1172         read_lock_bh(&sk->sk_callback_lock);
1173         if (!(xprt = xprt_from_sock(sk)))
1174                 goto out;
1175
1176         err = -sk->sk_err;
1177         if (err == 0)
1178                 goto out;
1179         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1180                         xprt, -err);
1181         trace_rpc_socket_error(xprt, sk->sk_socket, err);
1182         xprt_wake_pending_tasks(xprt, err);
1183  out:
1184         read_unlock_bh(&sk->sk_callback_lock);
1185 }
1186
1187 static void xs_reset_transport(struct sock_xprt *transport)
1188 {
1189         struct socket *sock = transport->sock;
1190         struct sock *sk = transport->inet;
1191         struct rpc_xprt *xprt = &transport->xprt;
1192
1193         if (sk == NULL)
1194                 return;
1195
1196         if (atomic_read(&transport->xprt.swapper))
1197                 sk_clear_memalloc(sk);
1198
1199         kernel_sock_shutdown(sock, SHUT_RDWR);
1200
1201         mutex_lock(&transport->recv_mutex);
1202         write_lock_bh(&sk->sk_callback_lock);
1203         transport->inet = NULL;
1204         transport->sock = NULL;
1205
1206         sk->sk_user_data = NULL;
1207
1208         xs_restore_old_callbacks(transport, sk);
1209         xprt_clear_connected(xprt);
1210         write_unlock_bh(&sk->sk_callback_lock);
1211         xs_sock_reset_connection_flags(xprt);
1212         mutex_unlock(&transport->recv_mutex);
1213
1214         trace_rpc_socket_close(xprt, sock);
1215         sock_release(sock);
1216
1217         xprt_disconnect_done(xprt);
1218 }
1219
1220 /**
1221  * xs_close - close a socket
1222  * @xprt: transport
1223  *
1224  * This is used when all requests are complete; ie, no DRC state remains
1225  * on the server we want to save.
1226  *
1227  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1228  * xs_reset_transport() zeroing the socket from underneath a writer.
1229  */
1230 static void xs_close(struct rpc_xprt *xprt)
1231 {
1232         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1233
1234         dprintk("RPC:       xs_close xprt %p\n", xprt);
1235
1236         xs_reset_transport(transport);
1237         xprt->reestablish_timeout = 0;
1238 }
1239
1240 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1241 {
1242         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1243                 xprt);
1244         xprt_disconnect_done(xprt);
1245 }
1246
1247 static void xs_xprt_free(struct rpc_xprt *xprt)
1248 {
1249         xs_free_peer_addresses(xprt);
1250         xprt_free(xprt);
1251 }
1252
1253 /**
1254  * xs_destroy - prepare to shutdown a transport
1255  * @xprt: doomed transport
1256  *
1257  */
1258 static void xs_destroy(struct rpc_xprt *xprt)
1259 {
1260         struct sock_xprt *transport = container_of(xprt,
1261                         struct sock_xprt, xprt);
1262         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1263
1264         cancel_delayed_work_sync(&transport->connect_worker);
1265         xs_close(xprt);
1266         cancel_work_sync(&transport->recv_worker);
1267         xs_xprt_free(xprt);
1268         module_put(THIS_MODULE);
1269 }
1270
1271 /**
1272  * xs_udp_data_read_skb - receive callback for UDP sockets
1273  * @xprt: transport
1274  * @sk: socket
1275  * @skb: skbuff
1276  *
1277  */
1278 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1279                 struct sock *sk,
1280                 struct sk_buff *skb)
1281 {
1282         struct rpc_task *task;
1283         struct rpc_rqst *rovr;
1284         int repsize, copied;
1285         u32 _xid;
1286         __be32 *xp;
1287
1288         repsize = skb->len;
1289         if (repsize < 4) {
1290                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1291                 return;
1292         }
1293
1294         /* Copy the XID from the skb... */
1295         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1296         if (xp == NULL)
1297                 return;
1298
1299         /* Look up and lock the request corresponding to the given XID */
1300         spin_lock(&xprt->queue_lock);
1301         rovr = xprt_lookup_rqst(xprt, *xp);
1302         if (!rovr)
1303                 goto out_unlock;
1304         xprt_pin_rqst(rovr);
1305         xprt_update_rtt(rovr->rq_task);
1306         spin_unlock(&xprt->queue_lock);
1307         task = rovr->rq_task;
1308
1309         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1310                 copied = repsize;
1311
1312         /* Suck it into the iovec, verify checksum if not done by hw. */
1313         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1314                 spin_lock(&xprt->queue_lock);
1315                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1316                 goto out_unpin;
1317         }
1318
1319
1320         spin_lock_bh(&xprt->transport_lock);
1321         xprt_adjust_cwnd(xprt, task, copied);
1322         spin_unlock_bh(&xprt->transport_lock);
1323         spin_lock(&xprt->queue_lock);
1324         xprt_complete_rqst(task, copied);
1325         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1326 out_unpin:
1327         xprt_unpin_rqst(rovr);
1328  out_unlock:
1329         spin_unlock(&xprt->queue_lock);
1330 }
1331
1332 static void xs_udp_data_receive(struct sock_xprt *transport)
1333 {
1334         struct sk_buff *skb;
1335         struct sock *sk;
1336         int err;
1337
1338         mutex_lock(&transport->recv_mutex);
1339         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1340         sk = transport->inet;
1341         if (sk == NULL)
1342                 goto out;
1343         for (;;) {
1344                 skb = skb_recv_udp(sk, 0, 1, &err);
1345                 if (skb == NULL)
1346                         break;
1347                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1348                 consume_skb(skb);
1349                 cond_resched();
1350         }
1351 out:
1352         mutex_unlock(&transport->recv_mutex);
1353 }
1354
1355 static void xs_udp_data_receive_workfn(struct work_struct *work)
1356 {
1357         struct sock_xprt *transport =
1358                 container_of(work, struct sock_xprt, recv_worker);
1359         xs_udp_data_receive(transport);
1360 }
1361
1362 /**
1363  * xs_data_ready - "data ready" callback for UDP sockets
1364  * @sk: socket with data to read
1365  *
1366  */
1367 static void xs_data_ready(struct sock *sk)
1368 {
1369         struct rpc_xprt *xprt;
1370
1371         read_lock_bh(&sk->sk_callback_lock);
1372         dprintk("RPC:       xs_data_ready...\n");
1373         xprt = xprt_from_sock(sk);
1374         if (xprt != NULL) {
1375                 struct sock_xprt *transport = container_of(xprt,
1376                                 struct sock_xprt, xprt);
1377                 transport->old_data_ready(sk);
1378                 /* Any data means we had a useful conversation, so
1379                  * then we don't need to delay the next reconnect
1380                  */
1381                 if (xprt->reestablish_timeout)
1382                         xprt->reestablish_timeout = 0;
1383                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1384                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1385         }
1386         read_unlock_bh(&sk->sk_callback_lock);
1387 }
1388
1389 /*
1390  * Helper function to force a TCP close if the server is sending
1391  * junk and/or it has put us in CLOSE_WAIT
1392  */
1393 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1394 {
1395         xprt_force_disconnect(xprt);
1396 }
1397
1398 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1399 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1400 {
1401         return PAGE_SIZE;
1402 }
1403 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1404
1405 /**
1406  * xs_tcp_state_change - callback to handle TCP socket state changes
1407  * @sk: socket whose state has changed
1408  *
1409  */
1410 static void xs_tcp_state_change(struct sock *sk)
1411 {
1412         struct rpc_xprt *xprt;
1413         struct sock_xprt *transport;
1414
1415         read_lock_bh(&sk->sk_callback_lock);
1416         if (!(xprt = xprt_from_sock(sk)))
1417                 goto out;
1418         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1419         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1420                         sk->sk_state, xprt_connected(xprt),
1421                         sock_flag(sk, SOCK_DEAD),
1422                         sock_flag(sk, SOCK_ZAPPED),
1423                         sk->sk_shutdown);
1424
1425         transport = container_of(xprt, struct sock_xprt, xprt);
1426         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1427         switch (sk->sk_state) {
1428         case TCP_ESTABLISHED:
1429                 spin_lock(&xprt->transport_lock);
1430                 if (!xprt_test_and_set_connected(xprt)) {
1431                         xprt->connect_cookie++;
1432                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1433                         xprt_clear_connecting(xprt);
1434
1435                         xprt->stat.connect_count++;
1436                         xprt->stat.connect_time += (long)jiffies -
1437                                                    xprt->stat.connect_start;
1438                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1439                 }
1440                 spin_unlock(&xprt->transport_lock);
1441                 break;
1442         case TCP_FIN_WAIT1:
1443                 /* The client initiated a shutdown of the socket */
1444                 xprt->connect_cookie++;
1445                 xprt->reestablish_timeout = 0;
1446                 set_bit(XPRT_CLOSING, &xprt->state);
1447                 smp_mb__before_atomic();
1448                 clear_bit(XPRT_CONNECTED, &xprt->state);
1449                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1450                 smp_mb__after_atomic();
1451                 break;
1452         case TCP_CLOSE_WAIT:
1453                 /* The server initiated a shutdown of the socket */
1454                 xprt->connect_cookie++;
1455                 clear_bit(XPRT_CONNECTED, &xprt->state);
1456                 xs_tcp_force_close(xprt);
1457                 /* fall through */
1458         case TCP_CLOSING:
1459                 /*
1460                  * If the server closed down the connection, make sure that
1461                  * we back off before reconnecting
1462                  */
1463                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1464                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1465                 break;
1466         case TCP_LAST_ACK:
1467                 set_bit(XPRT_CLOSING, &xprt->state);
1468                 smp_mb__before_atomic();
1469                 clear_bit(XPRT_CONNECTED, &xprt->state);
1470                 smp_mb__after_atomic();
1471                 break;
1472         case TCP_CLOSE:
1473                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1474                                         &transport->sock_state))
1475                         xprt_clear_connecting(xprt);
1476                 clear_bit(XPRT_CLOSING, &xprt->state);
1477                 /* Trigger the socket release */
1478                 xs_tcp_force_close(xprt);
1479         }
1480  out:
1481         read_unlock_bh(&sk->sk_callback_lock);
1482 }
1483
1484 static void xs_write_space(struct sock *sk)
1485 {
1486         struct socket_wq *wq;
1487         struct rpc_xprt *xprt;
1488
1489         if (!sk->sk_socket)
1490                 return;
1491         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1492
1493         if (unlikely(!(xprt = xprt_from_sock(sk))))
1494                 return;
1495         rcu_read_lock();
1496         wq = rcu_dereference(sk->sk_wq);
1497         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1498                 goto out;
1499
1500         if (xprt_write_space(xprt))
1501                 sk->sk_write_pending--;
1502 out:
1503         rcu_read_unlock();
1504 }
1505
1506 /**
1507  * xs_udp_write_space - callback invoked when socket buffer space
1508  *                             becomes available
1509  * @sk: socket whose state has changed
1510  *
1511  * Called when more output buffer space is available for this socket.
1512  * We try not to wake our writers until they can make "significant"
1513  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1514  * with a bunch of small requests.
1515  */
1516 static void xs_udp_write_space(struct sock *sk)
1517 {
1518         read_lock_bh(&sk->sk_callback_lock);
1519
1520         /* from net/core/sock.c:sock_def_write_space */
1521         if (sock_writeable(sk))
1522                 xs_write_space(sk);
1523
1524         read_unlock_bh(&sk->sk_callback_lock);
1525 }
1526
1527 /**
1528  * xs_tcp_write_space - callback invoked when socket buffer space
1529  *                             becomes available
1530  * @sk: socket whose state has changed
1531  *
1532  * Called when more output buffer space is available for this socket.
1533  * We try not to wake our writers until they can make "significant"
1534  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1535  * with a bunch of small requests.
1536  */
1537 static void xs_tcp_write_space(struct sock *sk)
1538 {
1539         read_lock_bh(&sk->sk_callback_lock);
1540
1541         /* from net/core/stream.c:sk_stream_write_space */
1542         if (sk_stream_is_writeable(sk))
1543                 xs_write_space(sk);
1544
1545         read_unlock_bh(&sk->sk_callback_lock);
1546 }
1547
1548 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1549 {
1550         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1551         struct sock *sk = transport->inet;
1552
1553         if (transport->rcvsize) {
1554                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1555                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1556         }
1557         if (transport->sndsize) {
1558                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1559                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1560                 sk->sk_write_space(sk);
1561         }
1562 }
1563
1564 /**
1565  * xs_udp_set_buffer_size - set send and receive limits
1566  * @xprt: generic transport
1567  * @sndsize: requested size of send buffer, in bytes
1568  * @rcvsize: requested size of receive buffer, in bytes
1569  *
1570  * Set socket send and receive buffer size limits.
1571  */
1572 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1573 {
1574         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1575
1576         transport->sndsize = 0;
1577         if (sndsize)
1578                 transport->sndsize = sndsize + 1024;
1579         transport->rcvsize = 0;
1580         if (rcvsize)
1581                 transport->rcvsize = rcvsize + 1024;
1582
1583         xs_udp_do_set_buffer_size(xprt);
1584 }
1585
1586 /**
1587  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1588  * @xprt: controlling transport
1589  * @task: task that timed out
1590  *
1591  * Adjust the congestion window after a retransmit timeout has occurred.
1592  */
1593 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1594 {
1595         spin_lock_bh(&xprt->transport_lock);
1596         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1597         spin_unlock_bh(&xprt->transport_lock);
1598 }
1599
1600 static int xs_get_random_port(void)
1601 {
1602         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1603         unsigned short range;
1604         unsigned short rand;
1605
1606         if (max < min)
1607                 return -EADDRINUSE;
1608         range = max - min + 1;
1609         rand = (unsigned short) prandom_u32() % range;
1610         return rand + min;
1611 }
1612
1613 /**
1614  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1615  * @sock: socket
1616  *
1617  * Note that this function has to be called on all sockets that share the
1618  * same port, and it must be called before binding.
1619  */
1620 static void xs_sock_set_reuseport(struct socket *sock)
1621 {
1622         int opt = 1;
1623
1624         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1625                         (char *)&opt, sizeof(opt));
1626 }
1627
1628 static unsigned short xs_sock_getport(struct socket *sock)
1629 {
1630         struct sockaddr_storage buf;
1631         unsigned short port = 0;
1632
1633         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1634                 goto out;
1635         switch (buf.ss_family) {
1636         case AF_INET6:
1637                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1638                 break;
1639         case AF_INET:
1640                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1641         }
1642 out:
1643         return port;
1644 }
1645
1646 /**
1647  * xs_set_port - reset the port number in the remote endpoint address
1648  * @xprt: generic transport
1649  * @port: new port number
1650  *
1651  */
1652 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1653 {
1654         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1655
1656         rpc_set_port(xs_addr(xprt), port);
1657         xs_update_peer_port(xprt);
1658 }
1659
1660 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1661 {
1662         if (transport->srcport == 0)
1663                 transport->srcport = xs_sock_getport(sock);
1664 }
1665
1666 static int xs_get_srcport(struct sock_xprt *transport)
1667 {
1668         int port = transport->srcport;
1669
1670         if (port == 0 && transport->xprt.resvport)
1671                 port = xs_get_random_port();
1672         return port;
1673 }
1674
1675 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1676 {
1677         if (transport->srcport != 0)
1678                 transport->srcport = 0;
1679         if (!transport->xprt.resvport)
1680                 return 0;
1681         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1682                 return xprt_max_resvport;
1683         return --port;
1684 }
1685 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1686 {
1687         struct sockaddr_storage myaddr;
1688         int err, nloop = 0;
1689         int port = xs_get_srcport(transport);
1690         unsigned short last;
1691
1692         /*
1693          * If we are asking for any ephemeral port (i.e. port == 0 &&
1694          * transport->xprt.resvport == 0), don't bind.  Let the local
1695          * port selection happen implicitly when the socket is used
1696          * (for example at connect time).
1697          *
1698          * This ensures that we can continue to establish TCP
1699          * connections even when all local ephemeral ports are already
1700          * a part of some TCP connection.  This makes no difference
1701          * for UDP sockets, but also doens't harm them.
1702          *
1703          * If we're asking for any reserved port (i.e. port == 0 &&
1704          * transport->xprt.resvport == 1) xs_get_srcport above will
1705          * ensure that port is non-zero and we will bind as needed.
1706          */
1707         if (port <= 0)
1708                 return port;
1709
1710         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1711         do {
1712                 rpc_set_port((struct sockaddr *)&myaddr, port);
1713                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1714                                 transport->xprt.addrlen);
1715                 if (err == 0) {
1716                         transport->srcport = port;
1717                         break;
1718                 }
1719                 last = port;
1720                 port = xs_next_srcport(transport, port);
1721                 if (port > last)
1722                         nloop++;
1723         } while (err == -EADDRINUSE && nloop != 2);
1724
1725         if (myaddr.ss_family == AF_INET)
1726                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1727                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1728                                 port, err ? "failed" : "ok", err);
1729         else
1730                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1731                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1732                                 port, err ? "failed" : "ok", err);
1733         return err;
1734 }
1735
1736 /*
1737  * We don't support autobind on AF_LOCAL sockets
1738  */
1739 static void xs_local_rpcbind(struct rpc_task *task)
1740 {
1741         xprt_set_bound(task->tk_xprt);
1742 }
1743
1744 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1745 {
1746 }
1747
1748 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1749 static struct lock_class_key xs_key[2];
1750 static struct lock_class_key xs_slock_key[2];
1751
1752 static inline void xs_reclassify_socketu(struct socket *sock)
1753 {
1754         struct sock *sk = sock->sk;
1755
1756         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1757                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1758 }
1759
1760 static inline void xs_reclassify_socket4(struct socket *sock)
1761 {
1762         struct sock *sk = sock->sk;
1763
1764         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1765                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1766 }
1767
1768 static inline void xs_reclassify_socket6(struct socket *sock)
1769 {
1770         struct sock *sk = sock->sk;
1771
1772         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1773                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1774 }
1775
1776 static inline void xs_reclassify_socket(int family, struct socket *sock)
1777 {
1778         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1779                 return;
1780
1781         switch (family) {
1782         case AF_LOCAL:
1783                 xs_reclassify_socketu(sock);
1784                 break;
1785         case AF_INET:
1786                 xs_reclassify_socket4(sock);
1787                 break;
1788         case AF_INET6:
1789                 xs_reclassify_socket6(sock);
1790                 break;
1791         }
1792 }
1793 #else
1794 static inline void xs_reclassify_socket(int family, struct socket *sock)
1795 {
1796 }
1797 #endif
1798
1799 static void xs_dummy_setup_socket(struct work_struct *work)
1800 {
1801 }
1802
1803 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1804                 struct sock_xprt *transport, int family, int type,
1805                 int protocol, bool reuseport)
1806 {
1807         struct socket *sock;
1808         int err;
1809
1810         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1811         if (err < 0) {
1812                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1813                                 protocol, -err);
1814                 goto out;
1815         }
1816         xs_reclassify_socket(family, sock);
1817
1818         if (reuseport)
1819                 xs_sock_set_reuseport(sock);
1820
1821         err = xs_bind(transport, sock);
1822         if (err) {
1823                 sock_release(sock);
1824                 goto out;
1825         }
1826
1827         return sock;
1828 out:
1829         return ERR_PTR(err);
1830 }
1831
1832 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1833                                       struct socket *sock)
1834 {
1835         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1836                                                                         xprt);
1837
1838         if (!transport->inet) {
1839                 struct sock *sk = sock->sk;
1840
1841                 write_lock_bh(&sk->sk_callback_lock);
1842
1843                 xs_save_old_callbacks(transport, sk);
1844
1845                 sk->sk_user_data = xprt;
1846                 sk->sk_data_ready = xs_data_ready;
1847                 sk->sk_write_space = xs_udp_write_space;
1848                 sock_set_flag(sk, SOCK_FASYNC);
1849                 sk->sk_error_report = xs_error_report;
1850                 sk->sk_allocation = GFP_NOIO;
1851
1852                 xprt_clear_connected(xprt);
1853
1854                 /* Reset to new socket */
1855                 transport->sock = sock;
1856                 transport->inet = sk;
1857
1858                 write_unlock_bh(&sk->sk_callback_lock);
1859         }
1860
1861         xs_stream_reset_connect(transport);
1862
1863         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1864 }
1865
1866 /**
1867  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1868  * @transport: socket transport to connect
1869  */
1870 static int xs_local_setup_socket(struct sock_xprt *transport)
1871 {
1872         struct rpc_xprt *xprt = &transport->xprt;
1873         struct socket *sock;
1874         int status = -EIO;
1875
1876         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1877                                         SOCK_STREAM, 0, &sock, 1);
1878         if (status < 0) {
1879                 dprintk("RPC:       can't create AF_LOCAL "
1880                         "transport socket (%d).\n", -status);
1881                 goto out;
1882         }
1883         xs_reclassify_socket(AF_LOCAL, sock);
1884
1885         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1886                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1887
1888         status = xs_local_finish_connecting(xprt, sock);
1889         trace_rpc_socket_connect(xprt, sock, status);
1890         switch (status) {
1891         case 0:
1892                 dprintk("RPC:       xprt %p connected to %s\n",
1893                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1894                 xprt->stat.connect_count++;
1895                 xprt->stat.connect_time += (long)jiffies -
1896                                            xprt->stat.connect_start;
1897                 xprt_set_connected(xprt);
1898         case -ENOBUFS:
1899                 break;
1900         case -ENOENT:
1901                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1902                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1903                 break;
1904         case -ECONNREFUSED:
1905                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1906                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1907                 break;
1908         default:
1909                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1910                                 __func__, -status,
1911                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1912         }
1913
1914 out:
1915         xprt_clear_connecting(xprt);
1916         xprt_wake_pending_tasks(xprt, status);
1917         return status;
1918 }
1919
1920 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1921 {
1922         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1923         int ret;
1924
1925          if (RPC_IS_ASYNC(task)) {
1926                 /*
1927                  * We want the AF_LOCAL connect to be resolved in the
1928                  * filesystem namespace of the process making the rpc
1929                  * call.  Thus we connect synchronously.
1930                  *
1931                  * If we want to support asynchronous AF_LOCAL calls,
1932                  * we'll need to figure out how to pass a namespace to
1933                  * connect.
1934                  */
1935                 rpc_exit(task, -ENOTCONN);
1936                 return;
1937         }
1938         ret = xs_local_setup_socket(transport);
1939         if (ret && !RPC_IS_SOFTCONN(task))
1940                 msleep_interruptible(15000);
1941 }
1942
1943 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1944 /*
1945  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1946  * know that we have exclusive access to the socket), to guard against
1947  * races with xs_reset_transport.
1948  */
1949 static void xs_set_memalloc(struct rpc_xprt *xprt)
1950 {
1951         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1952                         xprt);
1953
1954         /*
1955          * If there's no sock, then we have nothing to set. The
1956          * reconnecting process will get it for us.
1957          */
1958         if (!transport->inet)
1959                 return;
1960         if (atomic_read(&xprt->swapper))
1961                 sk_set_memalloc(transport->inet);
1962 }
1963
1964 /**
1965  * xs_enable_swap - Tag this transport as being used for swap.
1966  * @xprt: transport to tag
1967  *
1968  * Take a reference to this transport on behalf of the rpc_clnt, and
1969  * optionally mark it for swapping if it wasn't already.
1970  */
1971 static int
1972 xs_enable_swap(struct rpc_xprt *xprt)
1973 {
1974         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1975
1976         if (atomic_inc_return(&xprt->swapper) != 1)
1977                 return 0;
1978         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1979                 return -ERESTARTSYS;
1980         if (xs->inet)
1981                 sk_set_memalloc(xs->inet);
1982         xprt_release_xprt(xprt, NULL);
1983         return 0;
1984 }
1985
1986 /**
1987  * xs_disable_swap - Untag this transport as being used for swap.
1988  * @xprt: transport to tag
1989  *
1990  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1991  * swapper refcount goes to 0, untag the socket as a memalloc socket.
1992  */
1993 static void
1994 xs_disable_swap(struct rpc_xprt *xprt)
1995 {
1996         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1997
1998         if (!atomic_dec_and_test(&xprt->swapper))
1999                 return;
2000         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2001                 return;
2002         if (xs->inet)
2003                 sk_clear_memalloc(xs->inet);
2004         xprt_release_xprt(xprt, NULL);
2005 }
2006 #else
2007 static void xs_set_memalloc(struct rpc_xprt *xprt)
2008 {
2009 }
2010
2011 static int
2012 xs_enable_swap(struct rpc_xprt *xprt)
2013 {
2014         return -EINVAL;
2015 }
2016
2017 static void
2018 xs_disable_swap(struct rpc_xprt *xprt)
2019 {
2020 }
2021 #endif
2022
2023 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2024 {
2025         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2026
2027         if (!transport->inet) {
2028                 struct sock *sk = sock->sk;
2029
2030                 write_lock_bh(&sk->sk_callback_lock);
2031
2032                 xs_save_old_callbacks(transport, sk);
2033
2034                 sk->sk_user_data = xprt;
2035                 sk->sk_data_ready = xs_data_ready;
2036                 sk->sk_write_space = xs_udp_write_space;
2037                 sock_set_flag(sk, SOCK_FASYNC);
2038                 sk->sk_allocation = GFP_NOIO;
2039
2040                 xprt_set_connected(xprt);
2041
2042                 /* Reset to new socket */
2043                 transport->sock = sock;
2044                 transport->inet = sk;
2045
2046                 xs_set_memalloc(xprt);
2047
2048                 write_unlock_bh(&sk->sk_callback_lock);
2049         }
2050         xs_udp_do_set_buffer_size(xprt);
2051
2052         xprt->stat.connect_start = jiffies;
2053 }
2054
2055 static void xs_udp_setup_socket(struct work_struct *work)
2056 {
2057         struct sock_xprt *transport =
2058                 container_of(work, struct sock_xprt, connect_worker.work);
2059         struct rpc_xprt *xprt = &transport->xprt;
2060         struct socket *sock;
2061         int status = -EIO;
2062
2063         sock = xs_create_sock(xprt, transport,
2064                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2065                         IPPROTO_UDP, false);
2066         if (IS_ERR(sock))
2067                 goto out;
2068
2069         dprintk("RPC:       worker connecting xprt %p via %s to "
2070                                 "%s (port %s)\n", xprt,
2071                         xprt->address_strings[RPC_DISPLAY_PROTO],
2072                         xprt->address_strings[RPC_DISPLAY_ADDR],
2073                         xprt->address_strings[RPC_DISPLAY_PORT]);
2074
2075         xs_udp_finish_connecting(xprt, sock);
2076         trace_rpc_socket_connect(xprt, sock, 0);
2077         status = 0;
2078 out:
2079         xprt_clear_connecting(xprt);
2080         xprt_unlock_connect(xprt, transport);
2081         xprt_wake_pending_tasks(xprt, status);
2082 }
2083
2084 /**
2085  * xs_tcp_shutdown - gracefully shut down a TCP socket
2086  * @xprt: transport
2087  *
2088  * Initiates a graceful shutdown of the TCP socket by calling the
2089  * equivalent of shutdown(SHUT_RDWR);
2090  */
2091 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2092 {
2093         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2094         struct socket *sock = transport->sock;
2095         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2096
2097         if (sock == NULL)
2098                 return;
2099         switch (skst) {
2100         default:
2101                 kernel_sock_shutdown(sock, SHUT_RDWR);
2102                 trace_rpc_socket_shutdown(xprt, sock);
2103                 break;
2104         case TCP_CLOSE:
2105         case TCP_TIME_WAIT:
2106                 xs_reset_transport(transport);
2107         }
2108 }
2109
2110 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2111                 struct socket *sock)
2112 {
2113         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2114         unsigned int keepidle;
2115         unsigned int keepcnt;
2116         unsigned int opt_on = 1;
2117         unsigned int timeo;
2118
2119         spin_lock_bh(&xprt->transport_lock);
2120         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2121         keepcnt = xprt->timeout->to_retries + 1;
2122         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2123                 (xprt->timeout->to_retries + 1);
2124         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2125         spin_unlock_bh(&xprt->transport_lock);
2126
2127         /* TCP Keepalive options */
2128         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2129                         (char *)&opt_on, sizeof(opt_on));
2130         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2131                         (char *)&keepidle, sizeof(keepidle));
2132         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2133                         (char *)&keepidle, sizeof(keepidle));
2134         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2135                         (char *)&keepcnt, sizeof(keepcnt));
2136
2137         /* TCP user timeout (see RFC5482) */
2138         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2139                         (char *)&timeo, sizeof(timeo));
2140 }
2141
2142 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2143                 unsigned long connect_timeout,
2144                 unsigned long reconnect_timeout)
2145 {
2146         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2147         struct rpc_timeout to;
2148         unsigned long initval;
2149
2150         spin_lock_bh(&xprt->transport_lock);
2151         if (reconnect_timeout < xprt->max_reconnect_timeout)
2152                 xprt->max_reconnect_timeout = reconnect_timeout;
2153         if (connect_timeout < xprt->connect_timeout) {
2154                 memcpy(&to, xprt->timeout, sizeof(to));
2155                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2156                 /* Arbitrary lower limit */
2157                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2158                         initval = XS_TCP_INIT_REEST_TO << 1;
2159                 to.to_initval = initval;
2160                 to.to_maxval = initval;
2161                 memcpy(&transport->tcp_timeout, &to,
2162                                 sizeof(transport->tcp_timeout));
2163                 xprt->timeout = &transport->tcp_timeout;
2164                 xprt->connect_timeout = connect_timeout;
2165         }
2166         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2167         spin_unlock_bh(&xprt->transport_lock);
2168 }
2169
2170 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2171 {
2172         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2173         int ret = -ENOTCONN;
2174
2175         if (!transport->inet) {
2176                 struct sock *sk = sock->sk;
2177                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2178
2179                 /* Avoid temporary address, they are bad for long-lived
2180                  * connections such as NFS mounts.
2181                  * RFC4941, section 3.6 suggests that:
2182                  *    Individual applications, which have specific
2183                  *    knowledge about the normal duration of connections,
2184                  *    MAY override this as appropriate.
2185                  */
2186                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2187                                 (char *)&addr_pref, sizeof(addr_pref));
2188
2189                 xs_tcp_set_socket_timeouts(xprt, sock);
2190
2191                 write_lock_bh(&sk->sk_callback_lock);
2192
2193                 xs_save_old_callbacks(transport, sk);
2194
2195                 sk->sk_user_data = xprt;
2196                 sk->sk_data_ready = xs_data_ready;
2197                 sk->sk_state_change = xs_tcp_state_change;
2198                 sk->sk_write_space = xs_tcp_write_space;
2199                 sock_set_flag(sk, SOCK_FASYNC);
2200                 sk->sk_error_report = xs_error_report;
2201                 sk->sk_allocation = GFP_NOIO;
2202
2203                 /* socket options */
2204                 sock_reset_flag(sk, SOCK_LINGER);
2205                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2206
2207                 xprt_clear_connected(xprt);
2208
2209                 /* Reset to new socket */
2210                 transport->sock = sock;
2211                 transport->inet = sk;
2212
2213                 write_unlock_bh(&sk->sk_callback_lock);
2214         }
2215
2216         if (!xprt_bound(xprt))
2217                 goto out;
2218
2219         xs_set_memalloc(xprt);
2220
2221         /* Reset TCP record info */
2222         xs_stream_reset_connect(transport);
2223
2224         /* Tell the socket layer to start connecting... */
2225         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2226         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2227         switch (ret) {
2228         case 0:
2229                 xs_set_srcport(transport, sock);
2230                 /* fall through */
2231         case -EINPROGRESS:
2232                 /* SYN_SENT! */
2233                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2234                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2235                 break;
2236         case -EADDRNOTAVAIL:
2237                 /* Source port number is unavailable. Try a new one! */
2238                 transport->srcport = 0;
2239         }
2240 out:
2241         return ret;
2242 }
2243
2244 /**
2245  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2246  * @work: queued work item
2247  *
2248  * Invoked by a work queue tasklet.
2249  */
2250 static void xs_tcp_setup_socket(struct work_struct *work)
2251 {
2252         struct sock_xprt *transport =
2253                 container_of(work, struct sock_xprt, connect_worker.work);
2254         struct socket *sock = transport->sock;
2255         struct rpc_xprt *xprt = &transport->xprt;
2256         int status = -EIO;
2257
2258         if (!sock) {
2259                 sock = xs_create_sock(xprt, transport,
2260                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2261                                 IPPROTO_TCP, true);
2262                 if (IS_ERR(sock)) {
2263                         status = PTR_ERR(sock);
2264                         goto out;
2265                 }
2266         }
2267
2268         dprintk("RPC:       worker connecting xprt %p via %s to "
2269                                 "%s (port %s)\n", xprt,
2270                         xprt->address_strings[RPC_DISPLAY_PROTO],
2271                         xprt->address_strings[RPC_DISPLAY_ADDR],
2272                         xprt->address_strings[RPC_DISPLAY_PORT]);
2273
2274         status = xs_tcp_finish_connecting(xprt, sock);
2275         trace_rpc_socket_connect(xprt, sock, status);
2276         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2277                         xprt, -status, xprt_connected(xprt),
2278                         sock->sk->sk_state);
2279         switch (status) {
2280         default:
2281                 printk("%s: connect returned unhandled error %d\n",
2282                         __func__, status);
2283                 /* fall through */
2284         case -EADDRNOTAVAIL:
2285                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2286                  * and retry
2287                  */
2288                 xs_tcp_force_close(xprt);
2289                 break;
2290         case 0:
2291         case -EINPROGRESS:
2292         case -EALREADY:
2293                 xprt_unlock_connect(xprt, transport);
2294                 return;
2295         case -EINVAL:
2296                 /* Happens, for instance, if the user specified a link
2297                  * local IPv6 address without a scope-id.
2298                  */
2299         case -ECONNREFUSED:
2300         case -ECONNRESET:
2301         case -ENETDOWN:
2302         case -ENETUNREACH:
2303         case -EHOSTUNREACH:
2304         case -EADDRINUSE:
2305         case -ENOBUFS:
2306                 /*
2307                  * xs_tcp_force_close() wakes tasks with -EIO.
2308                  * We need to wake them first to ensure the
2309                  * correct error code.
2310                  */
2311                 xprt_wake_pending_tasks(xprt, status);
2312                 xs_tcp_force_close(xprt);
2313                 goto out;
2314         }
2315         status = -EAGAIN;
2316 out:
2317         xprt_clear_connecting(xprt);
2318         xprt_unlock_connect(xprt, transport);
2319         xprt_wake_pending_tasks(xprt, status);
2320 }
2321
2322 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2323 {
2324         unsigned long start, now = jiffies;
2325
2326         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2327         if (time_after(start, now))
2328                 return start - now;
2329         return 0;
2330 }
2331
2332 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2333 {
2334         xprt->reestablish_timeout <<= 1;
2335         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2336                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2337         if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2338                 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2339 }
2340
2341 /**
2342  * xs_connect - connect a socket to a remote endpoint
2343  * @xprt: pointer to transport structure
2344  * @task: address of RPC task that manages state of connect request
2345  *
2346  * TCP: If the remote end dropped the connection, delay reconnecting.
2347  *
2348  * UDP socket connects are synchronous, but we use a work queue anyway
2349  * to guarantee that even unprivileged user processes can set up a
2350  * socket on a privileged port.
2351  *
2352  * If a UDP socket connect fails, the delay behavior here prevents
2353  * retry floods (hard mounts).
2354  */
2355 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2356 {
2357         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2358         unsigned long delay = 0;
2359
2360         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2361
2362         if (transport->sock != NULL) {
2363                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2364                                 "seconds\n",
2365                                 xprt, xprt->reestablish_timeout / HZ);
2366
2367                 /* Start by resetting any existing state */
2368                 xs_reset_transport(transport);
2369
2370                 delay = xs_reconnect_delay(xprt);
2371                 xs_reconnect_backoff(xprt);
2372
2373         } else
2374                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2375
2376         queue_delayed_work(xprtiod_workqueue,
2377                         &transport->connect_worker,
2378                         delay);
2379 }
2380
2381 /**
2382  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2383  * @xprt: rpc_xprt struct containing statistics
2384  * @seq: output file
2385  *
2386  */
2387 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2388 {
2389         long idle_time = 0;
2390
2391         if (xprt_connected(xprt))
2392                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2393
2394         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2395                         "%llu %llu %lu %llu %llu\n",
2396                         xprt->stat.bind_count,
2397                         xprt->stat.connect_count,
2398                         xprt->stat.connect_time / HZ,
2399                         idle_time,
2400                         xprt->stat.sends,
2401                         xprt->stat.recvs,
2402                         xprt->stat.bad_xids,
2403                         xprt->stat.req_u,
2404                         xprt->stat.bklog_u,
2405                         xprt->stat.max_slots,
2406                         xprt->stat.sending_u,
2407                         xprt->stat.pending_u);
2408 }
2409
2410 /**
2411  * xs_udp_print_stats - display UDP socket-specifc stats
2412  * @xprt: rpc_xprt struct containing statistics
2413  * @seq: output file
2414  *
2415  */
2416 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2417 {
2418         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2419
2420         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2421                         "%lu %llu %llu\n",
2422                         transport->srcport,
2423                         xprt->stat.bind_count,
2424                         xprt->stat.sends,
2425                         xprt->stat.recvs,
2426                         xprt->stat.bad_xids,
2427                         xprt->stat.req_u,
2428                         xprt->stat.bklog_u,
2429                         xprt->stat.max_slots,
2430                         xprt->stat.sending_u,
2431                         xprt->stat.pending_u);
2432 }
2433
2434 /**
2435  * xs_tcp_print_stats - display TCP socket-specifc stats
2436  * @xprt: rpc_xprt struct containing statistics
2437  * @seq: output file
2438  *
2439  */
2440 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2441 {
2442         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2443         long idle_time = 0;
2444
2445         if (xprt_connected(xprt))
2446                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2447
2448         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2449                         "%llu %llu %lu %llu %llu\n",
2450                         transport->srcport,
2451                         xprt->stat.bind_count,
2452                         xprt->stat.connect_count,
2453                         xprt->stat.connect_time / HZ,
2454                         idle_time,
2455                         xprt->stat.sends,
2456                         xprt->stat.recvs,
2457                         xprt->stat.bad_xids,
2458                         xprt->stat.req_u,
2459                         xprt->stat.bklog_u,
2460                         xprt->stat.max_slots,
2461                         xprt->stat.sending_u,
2462                         xprt->stat.pending_u);
2463 }
2464
2465 /*
2466  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2467  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2468  * to use the server side send routines.
2469  */
2470 static int bc_malloc(struct rpc_task *task)
2471 {
2472         struct rpc_rqst *rqst = task->tk_rqstp;
2473         size_t size = rqst->rq_callsize;
2474         struct page *page;
2475         struct rpc_buffer *buf;
2476
2477         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2478                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2479                           size);
2480                 return -EINVAL;
2481         }
2482
2483         page = alloc_page(GFP_KERNEL);
2484         if (!page)
2485                 return -ENOMEM;
2486
2487         buf = page_address(page);
2488         buf->len = PAGE_SIZE;
2489
2490         rqst->rq_buffer = buf->data;
2491         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2492         return 0;
2493 }
2494
2495 /*
2496  * Free the space allocated in the bc_alloc routine
2497  */
2498 static void bc_free(struct rpc_task *task)
2499 {
2500         void *buffer = task->tk_rqstp->rq_buffer;
2501         struct rpc_buffer *buf;
2502
2503         buf = container_of(buffer, struct rpc_buffer, data);
2504         free_page((unsigned long)buf);
2505 }
2506
2507 /*
2508  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2509  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2510  */
2511 static int bc_sendto(struct rpc_rqst *req)
2512 {
2513         int len;
2514         struct xdr_buf *xbufp = &req->rq_snd_buf;
2515         struct rpc_xprt *xprt = req->rq_xprt;
2516         struct sock_xprt *transport =
2517                                 container_of(xprt, struct sock_xprt, xprt);
2518         struct socket *sock = transport->sock;
2519         unsigned long headoff;
2520         unsigned long tailoff;
2521
2522         xs_encode_stream_record_marker(xbufp);
2523
2524         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2525         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2526         len = svc_send_common(sock, xbufp,
2527                               virt_to_page(xbufp->head[0].iov_base), headoff,
2528                               xbufp->tail[0].iov_base, tailoff);
2529
2530         if (len != xbufp->len) {
2531                 printk(KERN_NOTICE "Error sending entire callback!\n");
2532                 len = -EAGAIN;
2533         }
2534
2535         return len;
2536 }
2537
2538 /*
2539  * The send routine. Borrows from svc_send
2540  */
2541 static int bc_send_request(struct rpc_rqst *req)
2542 {
2543         struct svc_xprt *xprt;
2544         int len;
2545
2546         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2547         /*
2548          * Get the server socket associated with this callback xprt
2549          */
2550         xprt = req->rq_xprt->bc_xprt;
2551
2552         /*
2553          * Grab the mutex to serialize data as the connection is shared
2554          * with the fore channel
2555          */
2556         mutex_lock(&xprt->xpt_mutex);
2557         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2558                 len = -ENOTCONN;
2559         else
2560                 len = bc_sendto(req);
2561         mutex_unlock(&xprt->xpt_mutex);
2562
2563         if (len > 0)
2564                 len = 0;
2565
2566         return len;
2567 }
2568
2569 /*
2570  * The close routine. Since this is client initiated, we do nothing
2571  */
2572
2573 static void bc_close(struct rpc_xprt *xprt)
2574 {
2575 }
2576
2577 /*
2578  * The xprt destroy routine. Again, because this connection is client
2579  * initiated, we do nothing
2580  */
2581
2582 static void bc_destroy(struct rpc_xprt *xprt)
2583 {
2584         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2585
2586         xs_xprt_free(xprt);
2587         module_put(THIS_MODULE);
2588 }
2589
2590 static const struct rpc_xprt_ops xs_local_ops = {
2591         .reserve_xprt           = xprt_reserve_xprt,
2592         .release_xprt           = xprt_release_xprt,
2593         .alloc_slot             = xprt_alloc_slot,
2594         .free_slot              = xprt_free_slot,
2595         .rpcbind                = xs_local_rpcbind,
2596         .set_port               = xs_local_set_port,
2597         .connect                = xs_local_connect,
2598         .buf_alloc              = rpc_malloc,
2599         .buf_free               = rpc_free,
2600         .prepare_request        = xs_stream_prepare_request,
2601         .send_request           = xs_local_send_request,
2602         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2603         .close                  = xs_close,
2604         .destroy                = xs_destroy,
2605         .print_stats            = xs_local_print_stats,
2606         .enable_swap            = xs_enable_swap,
2607         .disable_swap           = xs_disable_swap,
2608 };
2609
2610 static const struct rpc_xprt_ops xs_udp_ops = {
2611         .set_buffer_size        = xs_udp_set_buffer_size,
2612         .reserve_xprt           = xprt_reserve_xprt_cong,
2613         .release_xprt           = xprt_release_xprt_cong,
2614         .alloc_slot             = xprt_alloc_slot,
2615         .free_slot              = xprt_free_slot,
2616         .rpcbind                = rpcb_getport_async,
2617         .set_port               = xs_set_port,
2618         .connect                = xs_connect,
2619         .buf_alloc              = rpc_malloc,
2620         .buf_free               = rpc_free,
2621         .send_request           = xs_udp_send_request,
2622         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2623         .timer                  = xs_udp_timer,
2624         .release_request        = xprt_release_rqst_cong,
2625         .close                  = xs_close,
2626         .destroy                = xs_destroy,
2627         .print_stats            = xs_udp_print_stats,
2628         .enable_swap            = xs_enable_swap,
2629         .disable_swap           = xs_disable_swap,
2630         .inject_disconnect      = xs_inject_disconnect,
2631 };
2632
2633 static const struct rpc_xprt_ops xs_tcp_ops = {
2634         .reserve_xprt           = xprt_reserve_xprt,
2635         .release_xprt           = xprt_release_xprt,
2636         .alloc_slot             = xprt_alloc_slot,
2637         .free_slot              = xprt_free_slot,
2638         .rpcbind                = rpcb_getport_async,
2639         .set_port               = xs_set_port,
2640         .connect                = xs_connect,
2641         .buf_alloc              = rpc_malloc,
2642         .buf_free               = rpc_free,
2643         .prepare_request        = xs_stream_prepare_request,
2644         .send_request           = xs_tcp_send_request,
2645         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2646         .close                  = xs_tcp_shutdown,
2647         .destroy                = xs_destroy,
2648         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2649         .print_stats            = xs_tcp_print_stats,
2650         .enable_swap            = xs_enable_swap,
2651         .disable_swap           = xs_disable_swap,
2652         .inject_disconnect      = xs_inject_disconnect,
2653 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2654         .bc_setup               = xprt_setup_bc,
2655         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2656         .bc_free_rqst           = xprt_free_bc_rqst,
2657         .bc_destroy             = xprt_destroy_bc,
2658 #endif
2659 };
2660
2661 /*
2662  * The rpc_xprt_ops for the server backchannel
2663  */
2664
2665 static const struct rpc_xprt_ops bc_tcp_ops = {
2666         .reserve_xprt           = xprt_reserve_xprt,
2667         .release_xprt           = xprt_release_xprt,
2668         .alloc_slot             = xprt_alloc_slot,
2669         .free_slot              = xprt_free_slot,
2670         .buf_alloc              = bc_malloc,
2671         .buf_free               = bc_free,
2672         .send_request           = bc_send_request,
2673         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2674         .close                  = bc_close,
2675         .destroy                = bc_destroy,
2676         .print_stats            = xs_tcp_print_stats,
2677         .enable_swap            = xs_enable_swap,
2678         .disable_swap           = xs_disable_swap,
2679         .inject_disconnect      = xs_inject_disconnect,
2680 };
2681
2682 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2683 {
2684         static const struct sockaddr_in sin = {
2685                 .sin_family             = AF_INET,
2686                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2687         };
2688         static const struct sockaddr_in6 sin6 = {
2689                 .sin6_family            = AF_INET6,
2690                 .sin6_addr              = IN6ADDR_ANY_INIT,
2691         };
2692
2693         switch (family) {
2694         case AF_LOCAL:
2695                 break;
2696         case AF_INET:
2697                 memcpy(sap, &sin, sizeof(sin));
2698                 break;
2699         case AF_INET6:
2700                 memcpy(sap, &sin6, sizeof(sin6));
2701                 break;
2702         default:
2703                 dprintk("RPC:       %s: Bad address family\n", __func__);
2704                 return -EAFNOSUPPORT;
2705         }
2706         return 0;
2707 }
2708
2709 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2710                                       unsigned int slot_table_size,
2711                                       unsigned int max_slot_table_size)
2712 {
2713         struct rpc_xprt *xprt;
2714         struct sock_xprt *new;
2715
2716         if (args->addrlen > sizeof(xprt->addr)) {
2717                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2718                 return ERR_PTR(-EBADF);
2719         }
2720
2721         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2722                         max_slot_table_size);
2723         if (xprt == NULL) {
2724                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2725                                 "rpc_xprt\n");
2726                 return ERR_PTR(-ENOMEM);
2727         }
2728
2729         new = container_of(xprt, struct sock_xprt, xprt);
2730         mutex_init(&new->recv_mutex);
2731         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2732         xprt->addrlen = args->addrlen;
2733         if (args->srcaddr)
2734                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2735         else {
2736                 int err;
2737                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2738                                         (struct sockaddr *)&new->srcaddr);
2739                 if (err != 0) {
2740                         xprt_free(xprt);
2741                         return ERR_PTR(err);
2742                 }
2743         }
2744
2745         return xprt;
2746 }
2747
2748 static const struct rpc_timeout xs_local_default_timeout = {
2749         .to_initval = 10 * HZ,
2750         .to_maxval = 10 * HZ,
2751         .to_retries = 2,
2752 };
2753
2754 /**
2755  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2756  * @args: rpc transport creation arguments
2757  *
2758  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2759  */
2760 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2761 {
2762         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2763         struct sock_xprt *transport;
2764         struct rpc_xprt *xprt;
2765         struct rpc_xprt *ret;
2766
2767         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2768                         xprt_max_tcp_slot_table_entries);
2769         if (IS_ERR(xprt))
2770                 return xprt;
2771         transport = container_of(xprt, struct sock_xprt, xprt);
2772
2773         xprt->prot = 0;
2774         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2775         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2776
2777         xprt->bind_timeout = XS_BIND_TO;
2778         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2779         xprt->idle_timeout = XS_IDLE_DISC_TO;
2780
2781         xprt->ops = &xs_local_ops;
2782         xprt->timeout = &xs_local_default_timeout;
2783
2784         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2785         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2786
2787         switch (sun->sun_family) {
2788         case AF_LOCAL:
2789                 if (sun->sun_path[0] != '/') {
2790                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2791                                         sun->sun_path);
2792                         ret = ERR_PTR(-EINVAL);
2793                         goto out_err;
2794                 }
2795                 xprt_set_bound(xprt);
2796                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2797                 ret = ERR_PTR(xs_local_setup_socket(transport));
2798                 if (ret)
2799                         goto out_err;
2800                 break;
2801         default:
2802                 ret = ERR_PTR(-EAFNOSUPPORT);
2803                 goto out_err;
2804         }
2805
2806         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2807                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2808
2809         if (try_module_get(THIS_MODULE))
2810                 return xprt;
2811         ret = ERR_PTR(-EINVAL);
2812 out_err:
2813         xs_xprt_free(xprt);
2814         return ret;
2815 }
2816
2817 static const struct rpc_timeout xs_udp_default_timeout = {
2818         .to_initval = 5 * HZ,
2819         .to_maxval = 30 * HZ,
2820         .to_increment = 5 * HZ,
2821         .to_retries = 5,
2822 };
2823
2824 /**
2825  * xs_setup_udp - Set up transport to use a UDP socket
2826  * @args: rpc transport creation arguments
2827  *
2828  */
2829 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2830 {
2831         struct sockaddr *addr = args->dstaddr;
2832         struct rpc_xprt *xprt;
2833         struct sock_xprt *transport;
2834         struct rpc_xprt *ret;
2835
2836         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2837                         xprt_udp_slot_table_entries);
2838         if (IS_ERR(xprt))
2839                 return xprt;
2840         transport = container_of(xprt, struct sock_xprt, xprt);
2841
2842         xprt->prot = IPPROTO_UDP;
2843         xprt->tsh_size = 0;
2844         /* XXX: header size can vary due to auth type, IPv6, etc. */
2845         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2846
2847         xprt->bind_timeout = XS_BIND_TO;
2848         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2849         xprt->idle_timeout = XS_IDLE_DISC_TO;
2850
2851         xprt->ops = &xs_udp_ops;
2852
2853         xprt->timeout = &xs_udp_default_timeout;
2854
2855         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2856         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2857
2858         switch (addr->sa_family) {
2859         case AF_INET:
2860                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2861                         xprt_set_bound(xprt);
2862
2863                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2864                 break;
2865         case AF_INET6:
2866                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2867                         xprt_set_bound(xprt);
2868
2869                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2870                 break;
2871         default:
2872                 ret = ERR_PTR(-EAFNOSUPPORT);
2873                 goto out_err;
2874         }
2875
2876         if (xprt_bound(xprt))
2877                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2878                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2879                                 xprt->address_strings[RPC_DISPLAY_PORT],
2880                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2881         else
2882                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2883                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2884                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2885
2886         if (try_module_get(THIS_MODULE))
2887                 return xprt;
2888         ret = ERR_PTR(-EINVAL);
2889 out_err:
2890         xs_xprt_free(xprt);
2891         return ret;
2892 }
2893
2894 static const struct rpc_timeout xs_tcp_default_timeout = {
2895         .to_initval = 60 * HZ,
2896         .to_maxval = 60 * HZ,
2897         .to_retries = 2,
2898 };
2899
2900 /**
2901  * xs_setup_tcp - Set up transport to use a TCP socket
2902  * @args: rpc transport creation arguments
2903  *
2904  */
2905 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2906 {
2907         struct sockaddr *addr = args->dstaddr;
2908         struct rpc_xprt *xprt;
2909         struct sock_xprt *transport;
2910         struct rpc_xprt *ret;
2911         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2912
2913         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2914                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2915
2916         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2917                         max_slot_table_size);
2918         if (IS_ERR(xprt))
2919                 return xprt;
2920         transport = container_of(xprt, struct sock_xprt, xprt);
2921
2922         xprt->prot = IPPROTO_TCP;
2923         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2924         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2925
2926         xprt->bind_timeout = XS_BIND_TO;
2927         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2928         xprt->idle_timeout = XS_IDLE_DISC_TO;
2929
2930         xprt->ops = &xs_tcp_ops;
2931         xprt->timeout = &xs_tcp_default_timeout;
2932
2933         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2934         xprt->connect_timeout = xprt->timeout->to_initval *
2935                 (xprt->timeout->to_retries + 1);
2936
2937         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2938         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2939
2940         switch (addr->sa_family) {
2941         case AF_INET:
2942                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2943                         xprt_set_bound(xprt);
2944
2945                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2946                 break;
2947         case AF_INET6:
2948                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2949                         xprt_set_bound(xprt);
2950
2951                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2952                 break;
2953         default:
2954                 ret = ERR_PTR(-EAFNOSUPPORT);
2955                 goto out_err;
2956         }
2957
2958         if (xprt_bound(xprt))
2959                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2960                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2961                                 xprt->address_strings[RPC_DISPLAY_PORT],
2962                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2963         else
2964                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2965                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2966                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2967
2968         if (try_module_get(THIS_MODULE))
2969                 return xprt;
2970         ret = ERR_PTR(-EINVAL);
2971 out_err:
2972         xs_xprt_free(xprt);
2973         return ret;
2974 }
2975
2976 /**
2977  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2978  * @args: rpc transport creation arguments
2979  *
2980  */
2981 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2982 {
2983         struct sockaddr *addr = args->dstaddr;
2984         struct rpc_xprt *xprt;
2985         struct sock_xprt *transport;
2986         struct svc_sock *bc_sock;
2987         struct rpc_xprt *ret;
2988
2989         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2990                         xprt_tcp_slot_table_entries);
2991         if (IS_ERR(xprt))
2992                 return xprt;
2993         transport = container_of(xprt, struct sock_xprt, xprt);
2994
2995         xprt->prot = IPPROTO_TCP;
2996         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2997         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2998         xprt->timeout = &xs_tcp_default_timeout;
2999
3000         /* backchannel */
3001         xprt_set_bound(xprt);
3002         xprt->bind_timeout = 0;
3003         xprt->reestablish_timeout = 0;
3004         xprt->idle_timeout = 0;
3005
3006         xprt->ops = &bc_tcp_ops;
3007
3008         switch (addr->sa_family) {
3009         case AF_INET:
3010                 xs_format_peer_addresses(xprt, "tcp",
3011                                          RPCBIND_NETID_TCP);
3012                 break;
3013         case AF_INET6:
3014                 xs_format_peer_addresses(xprt, "tcp",
3015                                    RPCBIND_NETID_TCP6);
3016                 break;
3017         default:
3018                 ret = ERR_PTR(-EAFNOSUPPORT);
3019                 goto out_err;
3020         }
3021
3022         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3023                         xprt->address_strings[RPC_DISPLAY_ADDR],
3024                         xprt->address_strings[RPC_DISPLAY_PORT],
3025                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3026
3027         /*
3028          * Once we've associated a backchannel xprt with a connection,
3029          * we want to keep it around as long as the connection lasts,
3030          * in case we need to start using it for a backchannel again;
3031          * this reference won't be dropped until bc_xprt is destroyed.
3032          */
3033         xprt_get(xprt);
3034         args->bc_xprt->xpt_bc_xprt = xprt;
3035         xprt->bc_xprt = args->bc_xprt;
3036         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3037         transport->sock = bc_sock->sk_sock;
3038         transport->inet = bc_sock->sk_sk;
3039
3040         /*
3041          * Since we don't want connections for the backchannel, we set
3042          * the xprt status to connected
3043          */
3044         xprt_set_connected(xprt);
3045
3046         if (try_module_get(THIS_MODULE))
3047                 return xprt;
3048
3049         args->bc_xprt->xpt_bc_xprt = NULL;
3050         args->bc_xprt->xpt_bc_xps = NULL;
3051         xprt_put(xprt);
3052         ret = ERR_PTR(-EINVAL);
3053 out_err:
3054         xs_xprt_free(xprt);
3055         return ret;
3056 }
3057
3058 static struct xprt_class        xs_local_transport = {
3059         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3060         .name           = "named UNIX socket",
3061         .owner          = THIS_MODULE,
3062         .ident          = XPRT_TRANSPORT_LOCAL,
3063         .setup          = xs_setup_local,
3064 };
3065
3066 static struct xprt_class        xs_udp_transport = {
3067         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3068         .name           = "udp",
3069         .owner          = THIS_MODULE,
3070         .ident          = XPRT_TRANSPORT_UDP,
3071         .setup          = xs_setup_udp,
3072 };
3073
3074 static struct xprt_class        xs_tcp_transport = {
3075         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3076         .name           = "tcp",
3077         .owner          = THIS_MODULE,
3078         .ident          = XPRT_TRANSPORT_TCP,
3079         .setup          = xs_setup_tcp,
3080 };
3081
3082 static struct xprt_class        xs_bc_tcp_transport = {
3083         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3084         .name           = "tcp NFSv4.1 backchannel",
3085         .owner          = THIS_MODULE,
3086         .ident          = XPRT_TRANSPORT_BC_TCP,
3087         .setup          = xs_setup_bc_tcp,
3088 };
3089
3090 /**
3091  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3092  *
3093  */
3094 int init_socket_xprt(void)
3095 {
3096         if (!sunrpc_table_header)
3097                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3098
3099         xprt_register_transport(&xs_local_transport);
3100         xprt_register_transport(&xs_udp_transport);
3101         xprt_register_transport(&xs_tcp_transport);
3102         xprt_register_transport(&xs_bc_tcp_transport);
3103
3104         return 0;
3105 }
3106
3107 /**
3108  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3109  *
3110  */
3111 void cleanup_socket_xprt(void)
3112 {
3113         if (sunrpc_table_header) {
3114                 unregister_sysctl_table(sunrpc_table_header);
3115                 sunrpc_table_header = NULL;
3116         }
3117
3118         xprt_unregister_transport(&xs_local_transport);
3119         xprt_unregister_transport(&xs_udp_transport);
3120         xprt_unregister_transport(&xs_tcp_transport);
3121         xprt_unregister_transport(&xs_bc_tcp_transport);
3122 }
3123
3124 static int param_set_uint_minmax(const char *val,
3125                 const struct kernel_param *kp,
3126                 unsigned int min, unsigned int max)
3127 {
3128         unsigned int num;
3129         int ret;
3130
3131         if (!val)
3132                 return -EINVAL;
3133         ret = kstrtouint(val, 0, &num);
3134         if (ret)
3135                 return ret;
3136         if (num < min || num > max)
3137                 return -EINVAL;
3138         *((unsigned int *)kp->arg) = num;
3139         return 0;
3140 }
3141
3142 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3143 {
3144         return param_set_uint_minmax(val, kp,
3145                         RPC_MIN_RESVPORT,
3146                         RPC_MAX_RESVPORT);
3147 }
3148
3149 static const struct kernel_param_ops param_ops_portnr = {
3150         .set = param_set_portnr,
3151         .get = param_get_uint,
3152 };
3153
3154 #define param_check_portnr(name, p) \
3155         __param_check(name, p, unsigned int);
3156
3157 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3158 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3159
3160 static int param_set_slot_table_size(const char *val,
3161                                      const struct kernel_param *kp)
3162 {
3163         return param_set_uint_minmax(val, kp,
3164                         RPC_MIN_SLOT_TABLE,
3165                         RPC_MAX_SLOT_TABLE);
3166 }
3167
3168 static const struct kernel_param_ops param_ops_slot_table_size = {
3169         .set = param_set_slot_table_size,
3170         .get = param_get_uint,
3171 };
3172
3173 #define param_check_slot_table_size(name, p) \
3174         __param_check(name, p, unsigned int);
3175
3176 static int param_set_max_slot_table_size(const char *val,
3177                                      const struct kernel_param *kp)
3178 {
3179         return param_set_uint_minmax(val, kp,
3180                         RPC_MIN_SLOT_TABLE,
3181                         RPC_MAX_SLOT_TABLE_LIMIT);
3182 }
3183
3184 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3185         .set = param_set_max_slot_table_size,
3186         .get = param_get_uint,
3187 };
3188
3189 #define param_check_max_slot_table_size(name, p) \
3190         __param_check(name, p, unsigned int);
3191
3192 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3193                    slot_table_size, 0644);
3194 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3195                    max_slot_table_size, 0644);
3196 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3197                    slot_table_size, 0644);