~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/sunrpc/xprtsock.c

Version: ~ [ linux-5.8-rc3 ] ~ [ linux-5.7.5 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.48 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.129 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.185 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.228 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.228 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * linux/net/sunrpc/xprtsock.c
  3  *
  4  * Client-side transport implementation for sockets.
  5  *
  6  * TCP callback races fixes (C) 1998 Red Hat
  7  * TCP send fixes (C) 1998 Red Hat
  8  * TCP NFS related read + write fixes
  9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
 10  *
 11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
 12  * Fix behaviour when socket buffer is full.
 13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
 14  *
 15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
 16  *
 17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
 18  *   <gilles.quillard@bull.net>
 19  */
 20 
 21 #include <linux/types.h>
 22 #include <linux/string.h>
 23 #include <linux/slab.h>
 24 #include <linux/module.h>
 25 #include <linux/capability.h>
 26 #include <linux/pagemap.h>
 27 #include <linux/errno.h>
 28 #include <linux/socket.h>
 29 #include <linux/in.h>
 30 #include <linux/net.h>
 31 #include <linux/mm.h>
 32 #include <linux/un.h>
 33 #include <linux/udp.h>
 34 #include <linux/tcp.h>
 35 #include <linux/sunrpc/clnt.h>
 36 #include <linux/sunrpc/addr.h>
 37 #include <linux/sunrpc/sched.h>
 38 #include <linux/sunrpc/svcsock.h>
 39 #include <linux/sunrpc/xprtsock.h>
 40 #include <linux/file.h>
 41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
 42 #include <linux/sunrpc/bc_xprt.h>
 43 #endif
 44 
 45 #include <net/sock.h>
 46 #include <net/checksum.h>
 47 #include <net/udp.h>
 48 #include <net/tcp.h>
 49 
 50 #include <trace/events/sunrpc.h>
 51 
 52 #include "sunrpc.h"
 53 
 54 static void xs_close(struct rpc_xprt *xprt);
 55 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
 56                 struct socket *sock);
 57 
 58 /*
 59  * xprtsock tunables
 60  */
 61 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
 62 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
 63 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
 64 
 65 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
 66 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
 67 
 68 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 69 
 70 #define XS_TCP_LINGER_TO        (15U * HZ)
 71 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
 72 
 73 /*
 74  * We can register our own files under /proc/sys/sunrpc by
 75  * calling register_sysctl_table() again.  The files in that
 76  * directory become the union of all files registered there.
 77  *
 78  * We simply need to make sure that we don't collide with
 79  * someone else's file names!
 80  */
 81 
 82 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
 83 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
 84 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
 85 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
 86 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
 87 
 88 static struct ctl_table_header *sunrpc_table_header;
 89 
 90 /*
 91  * FIXME: changing the UDP slot table size should also resize the UDP
 92  *        socket buffers for existing UDP transports
 93  */
 94 static struct ctl_table xs_tunables_table[] = {
 95         {
 96                 .procname       = "udp_slot_table_entries",
 97                 .data           = &xprt_udp_slot_table_entries,
 98                 .maxlen         = sizeof(unsigned int),
 99                 .mode           = 0644,
100                 .proc_handler   = proc_dointvec_minmax,
101                 .extra1         = &min_slot_table_size,
102                 .extra2         = &max_slot_table_size
103         },
104         {
105                 .procname       = "tcp_slot_table_entries",
106                 .data           = &xprt_tcp_slot_table_entries,
107                 .maxlen         = sizeof(unsigned int),
108                 .mode           = 0644,
109                 .proc_handler   = proc_dointvec_minmax,
110                 .extra1         = &min_slot_table_size,
111                 .extra2         = &max_slot_table_size
112         },
113         {
114                 .procname       = "tcp_max_slot_table_entries",
115                 .data           = &xprt_max_tcp_slot_table_entries,
116                 .maxlen         = sizeof(unsigned int),
117                 .mode           = 0644,
118                 .proc_handler   = proc_dointvec_minmax,
119                 .extra1         = &min_slot_table_size,
120                 .extra2         = &max_tcp_slot_table_limit
121         },
122         {
123                 .procname       = "min_resvport",
124                 .data           = &xprt_min_resvport,
125                 .maxlen         = sizeof(unsigned int),
126                 .mode           = 0644,
127                 .proc_handler   = proc_dointvec_minmax,
128                 .extra1         = &xprt_min_resvport_limit,
129                 .extra2         = &xprt_max_resvport
130         },
131         {
132                 .procname       = "max_resvport",
133                 .data           = &xprt_max_resvport,
134                 .maxlen         = sizeof(unsigned int),
135                 .mode           = 0644,
136                 .proc_handler   = proc_dointvec_minmax,
137                 .extra1         = &xprt_min_resvport,
138                 .extra2         = &xprt_max_resvport_limit
139         },
140         {
141                 .procname       = "tcp_fin_timeout",
142                 .data           = &xs_tcp_fin_timeout,
143                 .maxlen         = sizeof(xs_tcp_fin_timeout),
144                 .mode           = 0644,
145                 .proc_handler   = proc_dointvec_jiffies,
146         },
147         { },
148 };
149 
150 static struct ctl_table sunrpc_table[] = {
151         {
152                 .procname       = "sunrpc",
153                 .mode           = 0555,
154                 .child          = xs_tunables_table
155         },
156         { },
157 };
158 
159 #endif
160 
161 /*
162  * Wait duration for a reply from the RPC portmapper.
163  */
164 #define XS_BIND_TO              (60U * HZ)
165 
166 /*
167  * Delay if a UDP socket connect error occurs.  This is most likely some
168  * kind of resource problem on the local host.
169  */
170 #define XS_UDP_REEST_TO         (2U * HZ)
171 
172 /*
173  * The reestablish timeout allows clients to delay for a bit before attempting
174  * to reconnect to a server that just dropped our connection.
175  *
176  * We implement an exponential backoff when trying to reestablish a TCP
177  * transport connection with the server.  Some servers like to drop a TCP
178  * connection when they are overworked, so we start with a short timeout and
179  * increase over time if the server is down or not responding.
180  */
181 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
182 
183 /*
184  * TCP idle timeout; client drops the transport socket if it is idle
185  * for this long.  Note that we also timeout UDP sockets to prevent
186  * holding port numbers when there is no RPC traffic.
187  */
188 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
189 
190 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
191 # undef  RPC_DEBUG_DATA
192 # define RPCDBG_FACILITY        RPCDBG_TRANS
193 #endif
194 
195 #ifdef RPC_DEBUG_DATA
196 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
197 {
198         u8 *buf = (u8 *) packet;
199         int j;
200 
201         dprintk("RPC:       %s\n", msg);
202         for (j = 0; j < count && j < 128; j += 4) {
203                 if (!(j & 31)) {
204                         if (j)
205                                 dprintk("\n");
206                         dprintk("0x%04x ", j);
207                 }
208                 dprintk("%02x%02x%02x%02x ",
209                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
210         }
211         dprintk("\n");
212 }
213 #else
214 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
215 {
216         /* NOP */
217 }
218 #endif
219 
220 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
221 {
222         return (struct rpc_xprt *) sk->sk_user_data;
223 }
224 
225 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
226 {
227         return (struct sockaddr *) &xprt->addr;
228 }
229 
230 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
231 {
232         return (struct sockaddr_un *) &xprt->addr;
233 }
234 
235 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
236 {
237         return (struct sockaddr_in *) &xprt->addr;
238 }
239 
240 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
241 {
242         return (struct sockaddr_in6 *) &xprt->addr;
243 }
244 
245 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
246 {
247         struct sockaddr *sap = xs_addr(xprt);
248         struct sockaddr_in6 *sin6;
249         struct sockaddr_in *sin;
250         struct sockaddr_un *sun;
251         char buf[128];
252 
253         switch (sap->sa_family) {
254         case AF_LOCAL:
255                 sun = xs_addr_un(xprt);
256                 strlcpy(buf, sun->sun_path, sizeof(buf));
257                 xprt->address_strings[RPC_DISPLAY_ADDR] =
258                                                 kstrdup(buf, GFP_KERNEL);
259                 break;
260         case AF_INET:
261                 (void)rpc_ntop(sap, buf, sizeof(buf));
262                 xprt->address_strings[RPC_DISPLAY_ADDR] =
263                                                 kstrdup(buf, GFP_KERNEL);
264                 sin = xs_addr_in(xprt);
265                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
266                 break;
267         case AF_INET6:
268                 (void)rpc_ntop(sap, buf, sizeof(buf));
269                 xprt->address_strings[RPC_DISPLAY_ADDR] =
270                                                 kstrdup(buf, GFP_KERNEL);
271                 sin6 = xs_addr_in6(xprt);
272                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
273                 break;
274         default:
275                 BUG();
276         }
277 
278         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
279 }
280 
281 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
282 {
283         struct sockaddr *sap = xs_addr(xprt);
284         char buf[128];
285 
286         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
287         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
288 
289         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
290         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
291 }
292 
293 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
294                                      const char *protocol,
295                                      const char *netid)
296 {
297         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
298         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
299         xs_format_common_peer_addresses(xprt);
300         xs_format_common_peer_ports(xprt);
301 }
302 
303 static void xs_update_peer_port(struct rpc_xprt *xprt)
304 {
305         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
306         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
307 
308         xs_format_common_peer_ports(xprt);
309 }
310 
311 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
312 {
313         unsigned int i;
314 
315         for (i = 0; i < RPC_DISPLAY_MAX; i++)
316                 switch (i) {
317                 case RPC_DISPLAY_PROTO:
318                 case RPC_DISPLAY_NETID:
319                         continue;
320                 default:
321                         kfree(xprt->address_strings[i]);
322                 }
323 }
324 
325 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
326 
327 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
328 {
329         struct msghdr msg = {
330                 .msg_name       = addr,
331                 .msg_namelen    = addrlen,
332                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
333         };
334         struct kvec iov = {
335                 .iov_base       = vec->iov_base + base,
336                 .iov_len        = vec->iov_len - base,
337         };
338 
339         if (iov.iov_len != 0)
340                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
341         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
342 }
343 
344 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
345 {
346         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
347                         int offset, size_t size, int flags);
348         struct page **ppage;
349         unsigned int remainder;
350         int err;
351 
352         remainder = xdr->page_len - base;
353         base += xdr->page_base;
354         ppage = xdr->pages + (base >> PAGE_SHIFT);
355         base &= ~PAGE_MASK;
356         do_sendpage = sock->ops->sendpage;
357         if (!zerocopy)
358                 do_sendpage = sock_no_sendpage;
359         for(;;) {
360                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
361                 int flags = XS_SENDMSG_FLAGS;
362 
363                 remainder -= len;
364                 if (more)
365                         flags |= MSG_MORE;
366                 if (remainder != 0)
367                         flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
368                 err = do_sendpage(sock, *ppage, base, len, flags);
369                 if (remainder == 0 || err != len)
370                         break;
371                 *sent_p += err;
372                 ppage++;
373                 base = 0;
374         }
375         if (err > 0) {
376                 *sent_p += err;
377                 err = 0;
378         }
379         return err;
380 }
381 
382 /**
383  * xs_sendpages - write pages directly to a socket
384  * @sock: socket to send on
385  * @addr: UDP only -- address of destination
386  * @addrlen: UDP only -- length of destination address
387  * @xdr: buffer containing this request
388  * @base: starting position in the buffer
389  * @zerocopy: true if it is safe to use sendpage()
390  * @sent_p: return the total number of bytes successfully queued for sending
391  *
392  */
393 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
394 {
395         unsigned int remainder = xdr->len - base;
396         int err = 0;
397         int sent = 0;
398 
399         if (unlikely(!sock))
400                 return -ENOTSOCK;
401 
402         if (base != 0) {
403                 addr = NULL;
404                 addrlen = 0;
405         }
406 
407         if (base < xdr->head[0].iov_len || addr != NULL) {
408                 unsigned int len = xdr->head[0].iov_len - base;
409                 remainder -= len;
410                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
411                 if (remainder == 0 || err != len)
412                         goto out;
413                 *sent_p += err;
414                 base = 0;
415         } else
416                 base -= xdr->head[0].iov_len;
417 
418         if (base < xdr->page_len) {
419                 unsigned int len = xdr->page_len - base;
420                 remainder -= len;
421                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
422                 *sent_p += sent;
423                 if (remainder == 0 || sent != len)
424                         goto out;
425                 base = 0;
426         } else
427                 base -= xdr->page_len;
428 
429         if (base >= xdr->tail[0].iov_len)
430                 return 0;
431         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
432 out:
433         if (err > 0) {
434                 *sent_p += err;
435                 err = 0;
436         }
437         return err;
438 }
439 
440 static void xs_nospace_callback(struct rpc_task *task)
441 {
442         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
443 
444         transport->inet->sk_write_pending--;
445 }
446 
447 /**
448  * xs_nospace - place task on wait queue if transmit was incomplete
449  * @task: task to put to sleep
450  *
451  */
452 static int xs_nospace(struct rpc_task *task)
453 {
454         struct rpc_rqst *req = task->tk_rqstp;
455         struct rpc_xprt *xprt = req->rq_xprt;
456         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
457         struct sock *sk = transport->inet;
458         int ret = -EAGAIN;
459 
460         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
461                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
462                         req->rq_slen);
463 
464         /* Protect against races with write_space */
465         spin_lock_bh(&xprt->transport_lock);
466 
467         /* Don't race with disconnect */
468         if (xprt_connected(xprt)) {
469                 /* wait for more buffer space */
470                 sk->sk_write_pending++;
471                 xprt_wait_for_buffer_space(task, xs_nospace_callback);
472         } else
473                 ret = -ENOTCONN;
474 
475         spin_unlock_bh(&xprt->transport_lock);
476 
477         /* Race breaker in case memory is freed before above code is called */
478         if (ret == -EAGAIN) {
479                 struct socket_wq *wq;
480 
481                 rcu_read_lock();
482                 wq = rcu_dereference(sk->sk_wq);
483                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
484                 rcu_read_unlock();
485 
486                 sk->sk_write_space(sk);
487         }
488         return ret;
489 }
490 
491 /*
492  * Construct a stream transport record marker in @buf.
493  */
494 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
495 {
496         u32 reclen = buf->len - sizeof(rpc_fraghdr);
497         rpc_fraghdr *base = buf->head[0].iov_base;
498         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
499 }
500 
501 /**
502  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
503  * @task: RPC task that manages the state of an RPC request
504  *
505  * Return values:
506  *        0:    The request has been sent
507  *   EAGAIN:    The socket was blocked, please call again later to
508  *              complete the request
509  * ENOTCONN:    Caller needs to invoke connect logic then call again
510  *    other:    Some other error occured, the request was not sent
511  */
512 static int xs_local_send_request(struct rpc_task *task)
513 {
514         struct rpc_rqst *req = task->tk_rqstp;
515         struct rpc_xprt *xprt = req->rq_xprt;
516         struct sock_xprt *transport =
517                                 container_of(xprt, struct sock_xprt, xprt);
518         struct xdr_buf *xdr = &req->rq_snd_buf;
519         int status;
520         int sent = 0;
521 
522         xs_encode_stream_record_marker(&req->rq_snd_buf);
523 
524         xs_pktdump("packet data:",
525                         req->rq_svec->iov_base, req->rq_svec->iov_len);
526 
527         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
528                               true, &sent);
529         dprintk("RPC:       %s(%u) = %d\n",
530                         __func__, xdr->len - req->rq_bytes_sent, status);
531 
532         if (status == -EAGAIN && sock_writeable(transport->inet))
533                 status = -ENOBUFS;
534 
535         if (likely(sent > 0) || status == 0) {
536                 req->rq_bytes_sent += sent;
537                 req->rq_xmit_bytes_sent += sent;
538                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
539                         req->rq_bytes_sent = 0;
540                         return 0;
541                 }
542                 status = -EAGAIN;
543         }
544 
545         switch (status) {
546         case -ENOBUFS:
547                 break;
548         case -EAGAIN:
549                 status = xs_nospace(task);
550                 break;
551         default:
552                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
553                         -status);
554         case -EPIPE:
555                 xs_close(xprt);
556                 status = -ENOTCONN;
557         }
558 
559         return status;
560 }
561 
562 /**
563  * xs_udp_send_request - write an RPC request to a UDP socket
564  * @task: address of RPC task that manages the state of an RPC request
565  *
566  * Return values:
567  *        0:    The request has been sent
568  *   EAGAIN:    The socket was blocked, please call again later to
569  *              complete the request
570  * ENOTCONN:    Caller needs to invoke connect logic then call again
571  *    other:    Some other error occurred, the request was not sent
572  */
573 static int xs_udp_send_request(struct rpc_task *task)
574 {
575         struct rpc_rqst *req = task->tk_rqstp;
576         struct rpc_xprt *xprt = req->rq_xprt;
577         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
578         struct xdr_buf *xdr = &req->rq_snd_buf;
579         int sent = 0;
580         int status;
581 
582         xs_pktdump("packet data:",
583                                 req->rq_svec->iov_base,
584                                 req->rq_svec->iov_len);
585 
586         if (!xprt_bound(xprt))
587                 return -ENOTCONN;
588         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
589                               xdr, req->rq_bytes_sent, true, &sent);
590 
591         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
592                         xdr->len - req->rq_bytes_sent, status);
593 
594         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
595         if (status == -EPERM)
596                 goto process_status;
597 
598         if (status == -EAGAIN && sock_writeable(transport->inet))
599                 status = -ENOBUFS;
600 
601         if (sent > 0 || status == 0) {
602                 req->rq_xmit_bytes_sent += sent;
603                 if (sent >= req->rq_slen)
604                         return 0;
605                 /* Still some bytes left; set up for a retry later. */
606                 status = -EAGAIN;
607         }
608 
609 process_status:
610         switch (status) {
611         case -ENOTSOCK:
612                 status = -ENOTCONN;
613                 /* Should we call xs_close() here? */
614                 break;
615         case -EAGAIN:
616                 status = xs_nospace(task);
617                 break;
618         case -ENETUNREACH:
619         case -ENOBUFS:
620         case -EPIPE:
621         case -ECONNREFUSED:
622         case -EPERM:
623                 /* When the server has died, an ICMP port unreachable message
624                  * prompts ECONNREFUSED. */
625                 break;
626         default:
627                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
628                         -status);
629         }
630 
631         return status;
632 }
633 
634 /**
635  * xs_tcp_send_request - write an RPC request to a TCP socket
636  * @task: address of RPC task that manages the state of an RPC request
637  *
638  * Return values:
639  *        0:    The request has been sent
640  *   EAGAIN:    The socket was blocked, please call again later to
641  *              complete the request
642  * ENOTCONN:    Caller needs to invoke connect logic then call again
643  *    other:    Some other error occurred, the request was not sent
644  *
645  * XXX: In the case of soft timeouts, should we eventually give up
646  *      if sendmsg is not able to make progress?
647  */
648 static int xs_tcp_send_request(struct rpc_task *task)
649 {
650         struct rpc_rqst *req = task->tk_rqstp;
651         struct rpc_xprt *xprt = req->rq_xprt;
652         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
653         struct xdr_buf *xdr = &req->rq_snd_buf;
654         bool zerocopy = true;
655         bool vm_wait = false;
656         int status;
657         int sent;
658 
659         xs_encode_stream_record_marker(&req->rq_snd_buf);
660 
661         xs_pktdump("packet data:",
662                                 req->rq_svec->iov_base,
663                                 req->rq_svec->iov_len);
664         /* Don't use zero copy if this is a resend. If the RPC call
665          * completes while the socket holds a reference to the pages,
666          * then we may end up resending corrupted data.
667          */
668         if (task->tk_flags & RPC_TASK_SENT)
669                 zerocopy = false;
670 
671         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
672                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
673 
674         /* Continue transmitting the packet/record. We must be careful
675          * to cope with writespace callbacks arriving _after_ we have
676          * called sendmsg(). */
677         while (1) {
678                 sent = 0;
679                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
680                                       req->rq_bytes_sent, zerocopy, &sent);
681 
682                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
683                                 xdr->len - req->rq_bytes_sent, status);
684 
685                 /* If we've sent the entire packet, immediately
686                  * reset the count of bytes sent. */
687                 req->rq_bytes_sent += sent;
688                 req->rq_xmit_bytes_sent += sent;
689                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
690                         req->rq_bytes_sent = 0;
691                         return 0;
692                 }
693 
694                 WARN_ON_ONCE(sent == 0 && status == 0);
695 
696                 if (status == -EAGAIN ) {
697                         /*
698                          * Return EAGAIN if we're sure we're hitting the
699                          * socket send buffer limits.
700                          */
701                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
702                                 break;
703                         /*
704                          * Did we hit a memory allocation failure?
705                          */
706                         if (sent == 0) {
707                                 status = -ENOBUFS;
708                                 if (vm_wait)
709                                         break;
710                                 /* Retry, knowing now that we're below the
711                                  * socket send buffer limit
712                                  */
713                                 vm_wait = true;
714                         }
715                         continue;
716                 }
717                 if (status < 0)
718                         break;
719                 vm_wait = false;
720         }
721 
722         switch (status) {
723         case -ENOTSOCK:
724                 status = -ENOTCONN;
725                 /* Should we call xs_close() here? */
726                 break;
727         case -EAGAIN:
728                 status = xs_nospace(task);
729                 break;
730         case -ECONNRESET:
731         case -ECONNREFUSED:
732         case -ENOTCONN:
733         case -EADDRINUSE:
734         case -ENOBUFS:
735         case -EPIPE:
736                 break;
737         default:
738                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
739                         -status);
740         }
741 
742         return status;
743 }
744 
745 /**
746  * xs_tcp_release_xprt - clean up after a tcp transmission
747  * @xprt: transport
748  * @task: rpc task
749  *
750  * This cleans up if an error causes us to abort the transmission of a request.
751  * In this case, the socket may need to be reset in order to avoid confusing
752  * the server.
753  */
754 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
755 {
756         struct rpc_rqst *req;
757 
758         if (task != xprt->snd_task)
759                 return;
760         if (task == NULL)
761                 goto out_release;
762         req = task->tk_rqstp;
763         if (req == NULL)
764                 goto out_release;
765         if (req->rq_bytes_sent == 0)
766                 goto out_release;
767         if (req->rq_bytes_sent == req->rq_snd_buf.len)
768                 goto out_release;
769         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
770 out_release:
771         xprt_release_xprt(xprt, task);
772 }
773 
774 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
775 {
776         transport->old_data_ready = sk->sk_data_ready;
777         transport->old_state_change = sk->sk_state_change;
778         transport->old_write_space = sk->sk_write_space;
779         transport->old_error_report = sk->sk_error_report;
780 }
781 
782 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
783 {
784         sk->sk_data_ready = transport->old_data_ready;
785         sk->sk_state_change = transport->old_state_change;
786         sk->sk_write_space = transport->old_write_space;
787         sk->sk_error_report = transport->old_error_report;
788 }
789 
790 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
791 {
792         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
793 
794         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
795 }
796 
797 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
798 {
799         smp_mb__before_atomic();
800         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
801         clear_bit(XPRT_CLOSING, &xprt->state);
802         xs_sock_reset_state_flags(xprt);
803         smp_mb__after_atomic();
804 }
805 
806 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
807 {
808         xs_sock_reset_connection_flags(xprt);
809         /* Mark transport as closed and wake up all pending tasks */
810         xprt_disconnect_done(xprt);
811 }
812 
813 /**
814  * xs_error_report - callback to handle TCP socket state errors
815  * @sk: socket
816  *
817  * Note: we don't call sock_error() since there may be a rpc_task
818  * using the socket, and so we don't want to clear sk->sk_err.
819  */
820 static void xs_error_report(struct sock *sk)
821 {
822         struct rpc_xprt *xprt;
823         int err;
824 
825         read_lock_bh(&sk->sk_callback_lock);
826         if (!(xprt = xprt_from_sock(sk)))
827                 goto out;
828 
829         err = -sk->sk_err;
830         if (err == 0)
831                 goto out;
832         /* Is this a reset event? */
833         if (sk->sk_state == TCP_CLOSE)
834                 xs_sock_mark_closed(xprt);
835         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
836                         xprt, -err);
837         trace_rpc_socket_error(xprt, sk->sk_socket, err);
838         xprt_wake_pending_tasks(xprt, err);
839  out:
840         read_unlock_bh(&sk->sk_callback_lock);
841 }
842 
843 static void xs_reset_transport(struct sock_xprt *transport)
844 {
845         struct socket *sock = transport->sock;
846         struct sock *sk = transport->inet;
847         struct rpc_xprt *xprt = &transport->xprt;
848 
849         if (sk == NULL)
850                 return;
851 
852         if (atomic_read(&transport->xprt.swapper))
853                 sk_clear_memalloc(sk);
854 
855         kernel_sock_shutdown(sock, SHUT_RDWR);
856 
857         mutex_lock(&transport->recv_mutex);
858         write_lock_bh(&sk->sk_callback_lock);
859         transport->inet = NULL;
860         transport->sock = NULL;
861 
862         sk->sk_user_data = NULL;
863 
864         xs_restore_old_callbacks(transport, sk);
865         xprt_clear_connected(xprt);
866         write_unlock_bh(&sk->sk_callback_lock);
867         xs_sock_reset_connection_flags(xprt);
868         mutex_unlock(&transport->recv_mutex);
869 
870         trace_rpc_socket_close(xprt, sock);
871         sock_release(sock);
872 }
873 
874 /**
875  * xs_close - close a socket
876  * @xprt: transport
877  *
878  * This is used when all requests are complete; ie, no DRC state remains
879  * on the server we want to save.
880  *
881  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
882  * xs_reset_transport() zeroing the socket from underneath a writer.
883  */
884 static void xs_close(struct rpc_xprt *xprt)
885 {
886         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
887 
888         dprintk("RPC:       xs_close xprt %p\n", xprt);
889 
890         xs_reset_transport(transport);
891         xprt->reestablish_timeout = 0;
892 
893         xprt_disconnect_done(xprt);
894 }
895 
896 static void xs_inject_disconnect(struct rpc_xprt *xprt)
897 {
898         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
899                 xprt);
900         xprt_disconnect_done(xprt);
901 }
902 
903 static void xs_xprt_free(struct rpc_xprt *xprt)
904 {
905         xs_free_peer_addresses(xprt);
906         xprt_free(xprt);
907 }
908 
909 /**
910  * xs_destroy - prepare to shutdown a transport
911  * @xprt: doomed transport
912  *
913  */
914 static void xs_destroy(struct rpc_xprt *xprt)
915 {
916         struct sock_xprt *transport = container_of(xprt,
917                         struct sock_xprt, xprt);
918         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
919 
920         cancel_delayed_work_sync(&transport->connect_worker);
921         xs_close(xprt);
922         cancel_work_sync(&transport->recv_worker);
923         xs_xprt_free(xprt);
924         module_put(THIS_MODULE);
925 }
926 
927 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
928 {
929         struct xdr_skb_reader desc = {
930                 .skb            = skb,
931                 .offset         = sizeof(rpc_fraghdr),
932                 .count          = skb->len - sizeof(rpc_fraghdr),
933         };
934 
935         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
936                 return -1;
937         if (desc.count)
938                 return -1;
939         return 0;
940 }
941 
942 /**
943  * xs_local_data_read_skb
944  * @xprt: transport
945  * @sk: socket
946  * @skb: skbuff
947  *
948  * Currently this assumes we can read the whole reply in a single gulp.
949  */
950 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
951                 struct sock *sk,
952                 struct sk_buff *skb)
953 {
954         struct rpc_task *task;
955         struct rpc_rqst *rovr;
956         int repsize, copied;
957         u32 _xid;
958         __be32 *xp;
959 
960         repsize = skb->len - sizeof(rpc_fraghdr);
961         if (repsize < 4) {
962                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
963                 return;
964         }
965 
966         /* Copy the XID from the skb... */
967         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
968         if (xp == NULL)
969                 return;
970 
971         /* Look up and lock the request corresponding to the given XID */
972         spin_lock_bh(&xprt->transport_lock);
973         rovr = xprt_lookup_rqst(xprt, *xp);
974         if (!rovr)
975                 goto out_unlock;
976         task = rovr->rq_task;
977 
978         copied = rovr->rq_private_buf.buflen;
979         if (copied > repsize)
980                 copied = repsize;
981 
982         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
983                 dprintk("RPC:       sk_buff copy failed\n");
984                 goto out_unlock;
985         }
986 
987         xprt_complete_rqst(task, copied);
988 
989  out_unlock:
990         spin_unlock_bh(&xprt->transport_lock);
991 }
992 
993 static void xs_local_data_receive(struct sock_xprt *transport)
994 {
995         struct sk_buff *skb;
996         struct sock *sk;
997         int err;
998 
999         mutex_lock(&transport->recv_mutex);
1000         sk = transport->inet;
1001         if (sk == NULL)
1002                 goto out;
1003         for (;;) {
1004                 skb = skb_recv_datagram(sk, 0, 1, &err);
1005                 if (skb != NULL) {
1006                         xs_local_data_read_skb(&transport->xprt, sk, skb);
1007                         skb_free_datagram(sk, skb);
1008                         continue;
1009                 }
1010                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1011                         break;
1012         }
1013 out:
1014         mutex_unlock(&transport->recv_mutex);
1015 }
1016 
1017 static void xs_local_data_receive_workfn(struct work_struct *work)
1018 {
1019         struct sock_xprt *transport =
1020                 container_of(work, struct sock_xprt, recv_worker);
1021         xs_local_data_receive(transport);
1022 }
1023 
1024 /**
1025  * xs_udp_data_read_skb - receive callback for UDP sockets
1026  * @xprt: transport
1027  * @sk: socket
1028  * @skb: skbuff
1029  *
1030  */
1031 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1032                 struct sock *sk,
1033                 struct sk_buff *skb)
1034 {
1035         struct rpc_task *task;
1036         struct rpc_rqst *rovr;
1037         int repsize, copied;
1038         u32 _xid;
1039         __be32 *xp;
1040 
1041         repsize = skb->len;
1042         if (repsize < 4) {
1043                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1044                 return;
1045         }
1046 
1047         /* Copy the XID from the skb... */
1048         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1049         if (xp == NULL)
1050                 return;
1051 
1052         /* Look up and lock the request corresponding to the given XID */
1053         spin_lock_bh(&xprt->transport_lock);
1054         rovr = xprt_lookup_rqst(xprt, *xp);
1055         if (!rovr)
1056                 goto out_unlock;
1057         task = rovr->rq_task;
1058 
1059         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1060                 copied = repsize;
1061 
1062         /* Suck it into the iovec, verify checksum if not done by hw. */
1063         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1064                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1065                 goto out_unlock;
1066         }
1067 
1068         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1069 
1070         xprt_adjust_cwnd(xprt, task, copied);
1071         xprt_complete_rqst(task, copied);
1072 
1073  out_unlock:
1074         spin_unlock_bh(&xprt->transport_lock);
1075 }
1076 
1077 static void xs_udp_data_receive(struct sock_xprt *transport)
1078 {
1079         struct sk_buff *skb;
1080         struct sock *sk;
1081         int err;
1082 
1083         mutex_lock(&transport->recv_mutex);
1084         sk = transport->inet;
1085         if (sk == NULL)
1086                 goto out;
1087         for (;;) {
1088                 skb = skb_recv_udp(sk, 0, 1, &err);
1089                 if (skb != NULL) {
1090                         xs_udp_data_read_skb(&transport->xprt, sk, skb);
1091                         consume_skb(skb);
1092                         continue;
1093                 }
1094                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1095                         break;
1096         }
1097 out:
1098         mutex_unlock(&transport->recv_mutex);
1099 }
1100 
1101 static void xs_udp_data_receive_workfn(struct work_struct *work)
1102 {
1103         struct sock_xprt *transport =
1104                 container_of(work, struct sock_xprt, recv_worker);
1105         xs_udp_data_receive(transport);
1106 }
1107 
1108 /**
1109  * xs_data_ready - "data ready" callback for UDP sockets
1110  * @sk: socket with data to read
1111  *
1112  */
1113 static void xs_data_ready(struct sock *sk)
1114 {
1115         struct rpc_xprt *xprt;
1116 
1117         read_lock_bh(&sk->sk_callback_lock);
1118         dprintk("RPC:       xs_data_ready...\n");
1119         xprt = xprt_from_sock(sk);
1120         if (xprt != NULL) {
1121                 struct sock_xprt *transport = container_of(xprt,
1122                                 struct sock_xprt, xprt);
1123                 transport->old_data_ready(sk);
1124                 /* Any data means we had a useful conversation, so
1125                  * then we don't need to delay the next reconnect
1126                  */
1127                 if (xprt->reestablish_timeout)
1128                         xprt->reestablish_timeout = 0;
1129                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1130                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1131         }
1132         read_unlock_bh(&sk->sk_callback_lock);
1133 }
1134 
1135 /*
1136  * Helper function to force a TCP close if the server is sending
1137  * junk and/or it has put us in CLOSE_WAIT
1138  */
1139 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1140 {
1141         xprt_force_disconnect(xprt);
1142 }
1143 
1144 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1145 {
1146         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1147         size_t len, used;
1148         char *p;
1149 
1150         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1151         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1152         used = xdr_skb_read_bits(desc, p, len);
1153         transport->tcp_offset += used;
1154         if (used != len)
1155                 return;
1156 
1157         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1158         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1159                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1160         else
1161                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1162         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1163 
1164         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1165         transport->tcp_offset = 0;
1166 
1167         /* Sanity check of the record length */
1168         if (unlikely(transport->tcp_reclen < 8)) {
1169                 dprintk("RPC:       invalid TCP record fragment length\n");
1170                 xs_tcp_force_close(xprt);
1171                 return;
1172         }
1173         dprintk("RPC:       reading TCP record fragment of length %d\n",
1174                         transport->tcp_reclen);
1175 }
1176 
1177 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1178 {
1179         if (transport->tcp_offset == transport->tcp_reclen) {
1180                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1181                 transport->tcp_offset = 0;
1182                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1183                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1184                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1185                         transport->tcp_copied = 0;
1186                 }
1187         }
1188 }
1189 
1190 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1191 {
1192         size_t len, used;
1193         char *p;
1194 
1195         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1196         dprintk("RPC:       reading XID (%zu bytes)\n", len);
1197         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1198         used = xdr_skb_read_bits(desc, p, len);
1199         transport->tcp_offset += used;
1200         if (used != len)
1201                 return;
1202         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1203         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1204         transport->tcp_copied = 4;
1205         dprintk("RPC:       reading %s XID %08x\n",
1206                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1207                                                               : "request with",
1208                         ntohl(transport->tcp_xid));
1209         xs_tcp_check_fraghdr(transport);
1210 }
1211 
1212 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1213                                        struct xdr_skb_reader *desc)
1214 {
1215         size_t len, used;
1216         u32 offset;
1217         char *p;
1218 
1219         /*
1220          * We want transport->tcp_offset to be 8 at the end of this routine
1221          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1222          * When this function is called for the first time,
1223          * transport->tcp_offset is 4 (after having already read the xid).
1224          */
1225         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1226         len = sizeof(transport->tcp_calldir) - offset;
1227         dprintk("RPC:       reading CALL/REPLY flag (%zu bytes)\n", len);
1228         p = ((char *) &transport->tcp_calldir) + offset;
1229         used = xdr_skb_read_bits(desc, p, len);
1230         transport->tcp_offset += used;
1231         if (used != len)
1232                 return;
1233         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1234         /*
1235          * We don't yet have the XDR buffer, so we will write the calldir
1236          * out after we get the buffer from the 'struct rpc_rqst'
1237          */
1238         switch (ntohl(transport->tcp_calldir)) {
1239         case RPC_REPLY:
1240                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1241                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1242                 transport->tcp_flags |= TCP_RPC_REPLY;
1243                 break;
1244         case RPC_CALL:
1245                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1246                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1247                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1248                 break;
1249         default:
1250                 dprintk("RPC:       invalid request message type\n");
1251                 xs_tcp_force_close(&transport->xprt);
1252         }
1253         xs_tcp_check_fraghdr(transport);
1254 }
1255 
1256 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1257                                      struct xdr_skb_reader *desc,
1258                                      struct rpc_rqst *req)
1259 {
1260         struct sock_xprt *transport =
1261                                 container_of(xprt, struct sock_xprt, xprt);
1262         struct xdr_buf *rcvbuf;
1263         size_t len;
1264         ssize_t r;
1265 
1266         rcvbuf = &req->rq_private_buf;
1267 
1268         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1269                 /*
1270                  * Save the RPC direction in the XDR buffer
1271                  */
1272                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1273                         &transport->tcp_calldir,
1274                         sizeof(transport->tcp_calldir));
1275                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1276                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1277         }
1278 
1279         len = desc->count;
1280         if (len > transport->tcp_reclen - transport->tcp_offset) {
1281                 struct xdr_skb_reader my_desc;
1282 
1283                 len = transport->tcp_reclen - transport->tcp_offset;
1284                 memcpy(&my_desc, desc, sizeof(my_desc));
1285                 my_desc.count = len;
1286                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1287                                           &my_desc, xdr_skb_read_bits);
1288                 desc->count -= r;
1289                 desc->offset += r;
1290         } else
1291                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1292                                           desc, xdr_skb_read_bits);
1293 
1294         if (r > 0) {
1295                 transport->tcp_copied += r;
1296                 transport->tcp_offset += r;
1297         }
1298         if (r != len) {
1299                 /* Error when copying to the receive buffer,
1300                  * usually because we weren't able to allocate
1301                  * additional buffer pages. All we can do now
1302                  * is turn off TCP_RCV_COPY_DATA, so the request
1303                  * will not receive any additional updates,
1304                  * and time out.
1305                  * Any remaining data from this record will
1306                  * be discarded.
1307                  */
1308                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1309                 dprintk("RPC:       XID %08x truncated request\n",
1310                                 ntohl(transport->tcp_xid));
1311                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1312                                 "tcp_offset = %u, tcp_reclen = %u\n",
1313                                 xprt, transport->tcp_copied,
1314                                 transport->tcp_offset, transport->tcp_reclen);
1315                 return;
1316         }
1317 
1318         dprintk("RPC:       XID %08x read %zd bytes\n",
1319                         ntohl(transport->tcp_xid), r);
1320         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1321                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1322                         transport->tcp_offset, transport->tcp_reclen);
1323 
1324         if (transport->tcp_copied == req->rq_private_buf.buflen)
1325                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1326         else if (transport->tcp_offset == transport->tcp_reclen) {
1327                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1328                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1329         }
1330 }
1331 
1332 /*
1333  * Finds the request corresponding to the RPC xid and invokes the common
1334  * tcp read code to read the data.
1335  */
1336 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1337                                     struct xdr_skb_reader *desc)
1338 {
1339         struct sock_xprt *transport =
1340                                 container_of(xprt, struct sock_xprt, xprt);
1341         struct rpc_rqst *req;
1342 
1343         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1344 
1345         /* Find and lock the request corresponding to this xid */
1346         spin_lock_bh(&xprt->transport_lock);
1347         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1348         if (!req) {
1349                 dprintk("RPC:       XID %08x request not found!\n",
1350                                 ntohl(transport->tcp_xid));
1351                 spin_unlock_bh(&xprt->transport_lock);
1352                 return -1;
1353         }
1354 
1355         xs_tcp_read_common(xprt, desc, req);
1356 
1357         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1358                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1359 
1360         spin_unlock_bh(&xprt->transport_lock);
1361         return 0;
1362 }
1363 
1364 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1365 /*
1366  * Obtains an rpc_rqst previously allocated and invokes the common
1367  * tcp read code to read the data.  The result is placed in the callback
1368  * queue.
1369  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1370  * connection and return -1.
1371  */
1372 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1373                                        struct xdr_skb_reader *desc)
1374 {
1375         struct sock_xprt *transport =
1376                                 container_of(xprt, struct sock_xprt, xprt);
1377         struct rpc_rqst *req;
1378 
1379         /* Look up and lock the request corresponding to the given XID */
1380         spin_lock_bh(&xprt->transport_lock);
1381         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1382         if (req == NULL) {
1383                 spin_unlock_bh(&xprt->transport_lock);
1384                 printk(KERN_WARNING "Callback slot table overflowed\n");
1385                 xprt_force_disconnect(xprt);
1386                 return -1;
1387         }
1388 
1389         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1390         xs_tcp_read_common(xprt, desc, req);
1391 
1392         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1393                 xprt_complete_bc_request(req, transport->tcp_copied);
1394         spin_unlock_bh(&xprt->transport_lock);
1395 
1396         return 0;
1397 }
1398 
1399 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1400                                         struct xdr_skb_reader *desc)
1401 {
1402         struct sock_xprt *transport =
1403                                 container_of(xprt, struct sock_xprt, xprt);
1404 
1405         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1406                 xs_tcp_read_reply(xprt, desc) :
1407                 xs_tcp_read_callback(xprt, desc);
1408 }
1409 
1410 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1411 {
1412         int ret;
1413 
1414         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1415                               SVC_SOCK_ANONYMOUS);
1416         if (ret < 0)
1417                 return ret;
1418         return 0;
1419 }
1420 
1421 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1422 {
1423         return PAGE_SIZE;
1424 }
1425 #else
1426 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1427                                         struct xdr_skb_reader *desc)
1428 {
1429         return xs_tcp_read_reply(xprt, desc);
1430 }
1431 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1432 
1433 /*
1434  * Read data off the transport.  This can be either an RPC_CALL or an
1435  * RPC_REPLY.  Relay the processing to helper functions.
1436  */
1437 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1438                                     struct xdr_skb_reader *desc)
1439 {
1440         struct sock_xprt *transport =
1441                                 container_of(xprt, struct sock_xprt, xprt);
1442 
1443         if (_xs_tcp_read_data(xprt, desc) == 0)
1444                 xs_tcp_check_fraghdr(transport);
1445         else {
1446                 /*
1447                  * The transport_lock protects the request handling.
1448                  * There's no need to hold it to update the tcp_flags.
1449                  */
1450                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1451         }
1452 }
1453 
1454 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1455 {
1456         size_t len;
1457 
1458         len = transport->tcp_reclen - transport->tcp_offset;
1459         if (len > desc->count)
1460                 len = desc->count;
1461         desc->count -= len;
1462         desc->offset += len;
1463         transport->tcp_offset += len;
1464         dprintk("RPC:       discarded %zu bytes\n", len);
1465         xs_tcp_check_fraghdr(transport);
1466 }
1467 
1468 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1469 {
1470         struct rpc_xprt *xprt = rd_desc->arg.data;
1471         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1472         struct xdr_skb_reader desc = {
1473                 .skb    = skb,
1474                 .offset = offset,
1475                 .count  = len,
1476         };
1477 
1478         dprintk("RPC:       xs_tcp_data_recv started\n");
1479         do {
1480                 trace_xs_tcp_data_recv(transport);
1481                 /* Read in a new fragment marker if necessary */
1482                 /* Can we ever really expect to get completely empty fragments? */
1483                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1484                         xs_tcp_read_fraghdr(xprt, &desc);
1485                         continue;
1486                 }
1487                 /* Read in the xid if necessary */
1488                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1489                         xs_tcp_read_xid(transport, &desc);
1490                         continue;
1491                 }
1492                 /* Read in the call/reply flag */
1493                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1494                         xs_tcp_read_calldir(transport, &desc);
1495                         continue;
1496                 }
1497                 /* Read in the request data */
1498                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1499                         xs_tcp_read_data(xprt, &desc);
1500                         continue;
1501                 }
1502                 /* Skip over any trailing bytes on short reads */
1503                 xs_tcp_read_discard(transport, &desc);
1504         } while (desc.count);
1505         trace_xs_tcp_data_recv(transport);
1506         dprintk("RPC:       xs_tcp_data_recv done\n");
1507         return len - desc.count;
1508 }
1509 
1510 static void xs_tcp_data_receive(struct sock_xprt *transport)
1511 {
1512         struct rpc_xprt *xprt = &transport->xprt;
1513         struct sock *sk;
1514         read_descriptor_t rd_desc = {
1515                 .count = 2*1024*1024,
1516                 .arg.data = xprt,
1517         };
1518         unsigned long total = 0;
1519         int read = 0;
1520 
1521         mutex_lock(&transport->recv_mutex);
1522         sk = transport->inet;
1523         if (sk == NULL)
1524                 goto out;
1525 
1526         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1527         for (;;) {
1528                 lock_sock(sk);
1529                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1530                 if (read <= 0) {
1531                         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1532                         release_sock(sk);
1533                         if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1534                                 break;
1535                 } else {
1536                         release_sock(sk);
1537                         total += read;
1538                 }
1539                 rd_desc.count = 65536;
1540         }
1541 out:
1542         mutex_unlock(&transport->recv_mutex);
1543         trace_xs_tcp_data_ready(xprt, read, total);
1544 }
1545 
1546 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1547 {
1548         struct sock_xprt *transport =
1549                 container_of(work, struct sock_xprt, recv_worker);
1550         xs_tcp_data_receive(transport);
1551 }
1552 
1553 /**
1554  * xs_tcp_state_change - callback to handle TCP socket state changes
1555  * @sk: socket whose state has changed
1556  *
1557  */
1558 static void xs_tcp_state_change(struct sock *sk)
1559 {
1560         struct rpc_xprt *xprt;
1561         struct sock_xprt *transport;
1562 
1563         read_lock_bh(&sk->sk_callback_lock);
1564         if (!(xprt = xprt_from_sock(sk)))
1565                 goto out;
1566         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1567         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1568                         sk->sk_state, xprt_connected(xprt),
1569                         sock_flag(sk, SOCK_DEAD),
1570                         sock_flag(sk, SOCK_ZAPPED),
1571                         sk->sk_shutdown);
1572 
1573         transport = container_of(xprt, struct sock_xprt, xprt);
1574         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1575         switch (sk->sk_state) {
1576         case TCP_ESTABLISHED:
1577                 spin_lock(&xprt->transport_lock);
1578                 if (!xprt_test_and_set_connected(xprt)) {
1579 
1580                         /* Reset TCP record info */
1581                         transport->tcp_offset = 0;
1582                         transport->tcp_reclen = 0;
1583                         transport->tcp_copied = 0;
1584                         transport->tcp_flags =
1585                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1586                         xprt->connect_cookie++;
1587                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1588                         xprt_clear_connecting(xprt);
1589 
1590                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1591                 }
1592                 spin_unlock(&xprt->transport_lock);
1593                 break;
1594         case TCP_FIN_WAIT1:
1595                 /* The client initiated a shutdown of the socket */
1596                 xprt->connect_cookie++;
1597                 xprt->reestablish_timeout = 0;
1598                 set_bit(XPRT_CLOSING, &xprt->state);
1599                 smp_mb__before_atomic();
1600                 clear_bit(XPRT_CONNECTED, &xprt->state);
1601                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1602                 smp_mb__after_atomic();
1603                 break;
1604         case TCP_CLOSE_WAIT:
1605                 /* The server initiated a shutdown of the socket */
1606                 xprt->connect_cookie++;
1607                 clear_bit(XPRT_CONNECTED, &xprt->state);
1608                 xs_tcp_force_close(xprt);
1609         case TCP_CLOSING:
1610                 /*
1611                  * If the server closed down the connection, make sure that
1612                  * we back off before reconnecting
1613                  */
1614                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1615                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1616                 break;
1617         case TCP_LAST_ACK:
1618                 set_bit(XPRT_CLOSING, &xprt->state);
1619                 smp_mb__before_atomic();
1620                 clear_bit(XPRT_CONNECTED, &xprt->state);
1621                 smp_mb__after_atomic();
1622                 break;
1623         case TCP_CLOSE:
1624                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1625                                         &transport->sock_state))
1626                         xprt_clear_connecting(xprt);
1627                 if (sk->sk_err)
1628                         xprt_wake_pending_tasks(xprt, -sk->sk_err);
1629                 xs_sock_mark_closed(xprt);
1630         }
1631  out:
1632         read_unlock_bh(&sk->sk_callback_lock);
1633 }
1634 
1635 static void xs_write_space(struct sock *sk)
1636 {
1637         struct socket_wq *wq;
1638         struct rpc_xprt *xprt;
1639 
1640         if (!sk->sk_socket)
1641                 return;
1642         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1643 
1644         if (unlikely(!(xprt = xprt_from_sock(sk))))
1645                 return;
1646         rcu_read_lock();
1647         wq = rcu_dereference(sk->sk_wq);
1648         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1649                 goto out;
1650 
1651         xprt_write_space(xprt);
1652 out:
1653         rcu_read_unlock();
1654 }
1655 
1656 /**
1657  * xs_udp_write_space - callback invoked when socket buffer space
1658  *                             becomes available
1659  * @sk: socket whose state has changed
1660  *
1661  * Called when more output buffer space is available for this socket.
1662  * We try not to wake our writers until they can make "significant"
1663  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1664  * with a bunch of small requests.
1665  */
1666 static void xs_udp_write_space(struct sock *sk)
1667 {
1668         read_lock_bh(&sk->sk_callback_lock);
1669 
1670         /* from net/core/sock.c:sock_def_write_space */
1671         if (sock_writeable(sk))
1672                 xs_write_space(sk);
1673 
1674         read_unlock_bh(&sk->sk_callback_lock);
1675 }
1676 
1677 /**
1678  * xs_tcp_write_space - callback invoked when socket buffer space
1679  *                             becomes available
1680  * @sk: socket whose state has changed
1681  *
1682  * Called when more output buffer space is available for this socket.
1683  * We try not to wake our writers until they can make "significant"
1684  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1685  * with a bunch of small requests.
1686  */
1687 static void xs_tcp_write_space(struct sock *sk)
1688 {
1689         read_lock_bh(&sk->sk_callback_lock);
1690 
1691         /* from net/core/stream.c:sk_stream_write_space */
1692         if (sk_stream_is_writeable(sk))
1693                 xs_write_space(sk);
1694 
1695         read_unlock_bh(&sk->sk_callback_lock);
1696 }
1697 
1698 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1699 {
1700         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1701         struct sock *sk = transport->inet;
1702 
1703         if (transport->rcvsize) {
1704                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1705                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1706         }
1707         if (transport->sndsize) {
1708                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1709                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1710                 sk->sk_write_space(sk);
1711         }
1712 }
1713 
1714 /**
1715  * xs_udp_set_buffer_size - set send and receive limits
1716  * @xprt: generic transport
1717  * @sndsize: requested size of send buffer, in bytes
1718  * @rcvsize: requested size of receive buffer, in bytes
1719  *
1720  * Set socket send and receive buffer size limits.
1721  */
1722 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1723 {
1724         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1725 
1726         transport->sndsize = 0;
1727         if (sndsize)
1728                 transport->sndsize = sndsize + 1024;
1729         transport->rcvsize = 0;
1730         if (rcvsize)
1731                 transport->rcvsize = rcvsize + 1024;
1732 
1733         xs_udp_do_set_buffer_size(xprt);
1734 }
1735 
1736 /**
1737  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1738  * @task: task that timed out
1739  *
1740  * Adjust the congestion window after a retransmit timeout has occurred.
1741  */
1742 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1743 {
1744         spin_lock_bh(&xprt->transport_lock);
1745         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1746         spin_unlock_bh(&xprt->transport_lock);
1747 }
1748 
1749 static unsigned short xs_get_random_port(void)
1750 {
1751         unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1752         unsigned short rand = (unsigned short) prandom_u32() % range;
1753         return rand + xprt_min_resvport;
1754 }
1755 
1756 /**
1757  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1758  * @sock: socket
1759  *
1760  * Note that this function has to be called on all sockets that share the
1761  * same port, and it must be called before binding.
1762  */
1763 static void xs_sock_set_reuseport(struct socket *sock)
1764 {
1765         int opt = 1;
1766 
1767         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1768                         (char *)&opt, sizeof(opt));
1769 }
1770 
1771 static unsigned short xs_sock_getport(struct socket *sock)
1772 {
1773         struct sockaddr_storage buf;
1774         int buflen;
1775         unsigned short port = 0;
1776 
1777         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1778                 goto out;
1779         switch (buf.ss_family) {
1780         case AF_INET6:
1781                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1782                 break;
1783         case AF_INET:
1784                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1785         }
1786 out:
1787         return port;
1788 }
1789 
1790 /**
1791  * xs_set_port - reset the port number in the remote endpoint address
1792  * @xprt: generic transport
1793  * @port: new port number
1794  *
1795  */
1796 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1797 {
1798         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1799 
1800         rpc_set_port(xs_addr(xprt), port);
1801         xs_update_peer_port(xprt);
1802 }
1803 
1804 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1805 {
1806         if (transport->srcport == 0)
1807                 transport->srcport = xs_sock_getport(sock);
1808 }
1809 
1810 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1811 {
1812         unsigned short port = transport->srcport;
1813 
1814         if (port == 0 && transport->xprt.resvport)
1815                 port = xs_get_random_port();
1816         return port;
1817 }
1818 
1819 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1820 {
1821         if (transport->srcport != 0)
1822                 transport->srcport = 0;
1823         if (!transport->xprt.resvport)
1824                 return 0;
1825         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1826                 return xprt_max_resvport;
1827         return --port;
1828 }
1829 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1830 {
1831         struct sockaddr_storage myaddr;
1832         int err, nloop = 0;
1833         unsigned short port = xs_get_srcport(transport);
1834         unsigned short last;
1835 
1836         /*
1837          * If we are asking for any ephemeral port (i.e. port == 0 &&
1838          * transport->xprt.resvport == 0), don't bind.  Let the local
1839          * port selection happen implicitly when the socket is used
1840          * (for example at connect time).
1841          *
1842          * This ensures that we can continue to establish TCP
1843          * connections even when all local ephemeral ports are already
1844          * a part of some TCP connection.  This makes no difference
1845          * for UDP sockets, but also doens't harm them.
1846          *
1847          * If we're asking for any reserved port (i.e. port == 0 &&
1848          * transport->xprt.resvport == 1) xs_get_srcport above will
1849          * ensure that port is non-zero and we will bind as needed.
1850          */
1851         if (port == 0)
1852                 return 0;
1853 
1854         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1855         do {
1856                 rpc_set_port((struct sockaddr *)&myaddr, port);
1857                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1858                                 transport->xprt.addrlen);
1859                 if (err == 0) {
1860                         transport->srcport = port;
1861                         break;
1862                 }
1863                 last = port;
1864                 port = xs_next_srcport(transport, port);
1865                 if (port > last)
1866                         nloop++;
1867         } while (err == -EADDRINUSE && nloop != 2);
1868 
1869         if (myaddr.ss_family == AF_INET)
1870                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1871                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1872                                 port, err ? "failed" : "ok", err);
1873         else
1874                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1875                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1876                                 port, err ? "failed" : "ok", err);
1877         return err;
1878 }
1879 
1880 /*
1881  * We don't support autobind on AF_LOCAL sockets
1882  */
1883 static void xs_local_rpcbind(struct rpc_task *task)
1884 {
1885         xprt_set_bound(task->tk_xprt);
1886 }
1887 
1888 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1889 {
1890 }
1891 
1892 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1893 static struct lock_class_key xs_key[2];
1894 static struct lock_class_key xs_slock_key[2];
1895 
1896 static inline void xs_reclassify_socketu(struct socket *sock)
1897 {
1898         struct sock *sk = sock->sk;
1899 
1900         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1901                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1902 }
1903 
1904 static inline void xs_reclassify_socket4(struct socket *sock)
1905 {
1906         struct sock *sk = sock->sk;
1907 
1908         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1909                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1910 }
1911 
1912 static inline void xs_reclassify_socket6(struct socket *sock)
1913 {
1914         struct sock *sk = sock->sk;
1915 
1916         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1917                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1918 }
1919 
1920 static inline void xs_reclassify_socket(int family, struct socket *sock)
1921 {
1922         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1923                 return;
1924 
1925         switch (family) {
1926         case AF_LOCAL:
1927                 xs_reclassify_socketu(sock);
1928                 break;
1929         case AF_INET:
1930                 xs_reclassify_socket4(sock);
1931                 break;
1932         case AF_INET6:
1933                 xs_reclassify_socket6(sock);
1934                 break;
1935         }
1936 }
1937 #else
1938 static inline void xs_reclassify_socket(int family, struct socket *sock)
1939 {
1940 }
1941 #endif
1942 
1943 static void xs_dummy_setup_socket(struct work_struct *work)
1944 {
1945 }
1946 
1947 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1948                 struct sock_xprt *transport, int family, int type,
1949                 int protocol, bool reuseport)
1950 {
1951         struct socket *sock;
1952         int err;
1953 
1954         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1955         if (err < 0) {
1956                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1957                                 protocol, -err);
1958                 goto out;
1959         }
1960         xs_reclassify_socket(family, sock);
1961 
1962         if (reuseport)
1963                 xs_sock_set_reuseport(sock);
1964 
1965         err = xs_bind(transport, sock);
1966         if (err) {
1967                 sock_release(sock);
1968                 goto out;
1969         }
1970 
1971         return sock;
1972 out:
1973         return ERR_PTR(err);
1974 }
1975 
1976 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1977                                       struct socket *sock)
1978 {
1979         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1980                                                                         xprt);
1981 
1982         if (!transport->inet) {
1983                 struct sock *sk = sock->sk;
1984 
1985                 write_lock_bh(&sk->sk_callback_lock);
1986 
1987                 xs_save_old_callbacks(transport, sk);
1988 
1989                 sk->sk_user_data = xprt;
1990                 sk->sk_data_ready = xs_data_ready;
1991                 sk->sk_write_space = xs_udp_write_space;
1992                 sock_set_flag(sk, SOCK_FASYNC);
1993                 sk->sk_error_report = xs_error_report;
1994                 sk->sk_allocation = GFP_NOIO;
1995 
1996                 xprt_clear_connected(xprt);
1997 
1998                 /* Reset to new socket */
1999                 transport->sock = sock;
2000                 transport->inet = sk;
2001 
2002                 write_unlock_bh(&sk->sk_callback_lock);
2003         }
2004 
2005         /* Tell the socket layer to start connecting... */
2006         xprt->stat.connect_count++;
2007         xprt->stat.connect_start = jiffies;
2008         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2009 }
2010 
2011 /**
2012  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2013  * @transport: socket transport to connect
2014  */
2015 static int xs_local_setup_socket(struct sock_xprt *transport)
2016 {
2017         struct rpc_xprt *xprt = &transport->xprt;
2018         struct socket *sock;
2019         int status = -EIO;
2020 
2021         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2022                                         SOCK_STREAM, 0, &sock, 1);
2023         if (status < 0) {
2024                 dprintk("RPC:       can't create AF_LOCAL "
2025                         "transport socket (%d).\n", -status);
2026                 goto out;
2027         }
2028         xs_reclassify_socket(AF_LOCAL, sock);
2029 
2030         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2031                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2032 
2033         status = xs_local_finish_connecting(xprt, sock);
2034         trace_rpc_socket_connect(xprt, sock, status);
2035         switch (status) {
2036         case 0:
2037                 dprintk("RPC:       xprt %p connected to %s\n",
2038                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2039                 xprt_set_connected(xprt);
2040         case -ENOBUFS:
2041                 break;
2042         case -ENOENT:
2043                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2044                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2045                 break;
2046         case -ECONNREFUSED:
2047                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2048                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2049                 break;
2050         default:
2051                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2052                                 __func__, -status,
2053                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2054         }
2055 
2056 out:
2057         xprt_clear_connecting(xprt);
2058         xprt_wake_pending_tasks(xprt, status);
2059         return status;
2060 }
2061 
2062 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2063 {
2064         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2065         int ret;
2066 
2067          if (RPC_IS_ASYNC(task)) {
2068                 /*
2069                  * We want the AF_LOCAL connect to be resolved in the
2070                  * filesystem namespace of the process making the rpc
2071                  * call.  Thus we connect synchronously.
2072                  *
2073                  * If we want to support asynchronous AF_LOCAL calls,
2074                  * we'll need to figure out how to pass a namespace to
2075                  * connect.
2076                  */
2077                 rpc_exit(task, -ENOTCONN);
2078                 return;
2079         }
2080         ret = xs_local_setup_socket(transport);
2081         if (ret && !RPC_IS_SOFTCONN(task))
2082                 msleep_interruptible(15000);
2083 }
2084 
2085 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2086 /*
2087  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2088  * know that we have exclusive access to the socket), to guard against
2089  * races with xs_reset_transport.
2090  */
2091 static void xs_set_memalloc(struct rpc_xprt *xprt)
2092 {
2093         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2094                         xprt);
2095 
2096         /*
2097          * If there's no sock, then we have nothing to set. The
2098          * reconnecting process will get it for us.
2099          */
2100         if (!transport->inet)
2101                 return;
2102         if (atomic_read(&xprt->swapper))
2103                 sk_set_memalloc(transport->inet);
2104 }
2105 
2106 /**
2107  * xs_enable_swap - Tag this transport as being used for swap.
2108  * @xprt: transport to tag
2109  *
2110  * Take a reference to this transport on behalf of the rpc_clnt, and
2111  * optionally mark it for swapping if it wasn't already.
2112  */
2113 static int
2114 xs_enable_swap(struct rpc_xprt *xprt)
2115 {
2116         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2117 
2118         if (atomic_inc_return(&xprt->swapper) != 1)
2119                 return 0;
2120         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2121                 return -ERESTARTSYS;
2122         if (xs->inet)
2123                 sk_set_memalloc(xs->inet);
2124         xprt_release_xprt(xprt, NULL);
2125         return 0;
2126 }
2127 
2128 /**
2129  * xs_disable_swap - Untag this transport as being used for swap.
2130  * @xprt: transport to tag
2131  *
2132  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2133  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2134  */
2135 static void
2136 xs_disable_swap(struct rpc_xprt *xprt)
2137 {
2138         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2139 
2140         if (!atomic_dec_and_test(&xprt->swapper))
2141                 return;
2142         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2143                 return;
2144         if (xs->inet)
2145                 sk_clear_memalloc(xs->inet);
2146         xprt_release_xprt(xprt, NULL);
2147 }
2148 #else
2149 static void xs_set_memalloc(struct rpc_xprt *xprt)
2150 {
2151 }
2152 
2153 static int
2154 xs_enable_swap(struct rpc_xprt *xprt)
2155 {
2156         return -EINVAL;
2157 }
2158 
2159 static void
2160 xs_disable_swap(struct rpc_xprt *xprt)
2161 {
2162 }
2163 #endif
2164 
2165 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2166 {
2167         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2168 
2169         if (!transport->inet) {
2170                 struct sock *sk = sock->sk;
2171 
2172                 write_lock_bh(&sk->sk_callback_lock);
2173 
2174                 xs_save_old_callbacks(transport, sk);
2175 
2176                 sk->sk_user_data = xprt;
2177                 sk->sk_data_ready = xs_data_ready;
2178                 sk->sk_write_space = xs_udp_write_space;
2179                 sock_set_flag(sk, SOCK_FASYNC);
2180                 sk->sk_allocation = GFP_NOIO;
2181 
2182                 xprt_set_connected(xprt);
2183 
2184                 /* Reset to new socket */
2185                 transport->sock = sock;
2186                 transport->inet = sk;
2187 
2188                 xs_set_memalloc(xprt);
2189 
2190                 write_unlock_bh(&sk->sk_callback_lock);
2191         }
2192         xs_udp_do_set_buffer_size(xprt);
2193 
2194         xprt->stat.connect_start = jiffies;
2195 }
2196 
2197 static void xs_udp_setup_socket(struct work_struct *work)
2198 {
2199         struct sock_xprt *transport =
2200                 container_of(work, struct sock_xprt, connect_worker.work);
2201         struct rpc_xprt *xprt = &transport->xprt;
2202         struct socket *sock = transport->sock;
2203         int status = -EIO;
2204 
2205         sock = xs_create_sock(xprt, transport,
2206                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2207                         IPPROTO_UDP, false);
2208         if (IS_ERR(sock))
2209                 goto out;
2210 
2211         dprintk("RPC:       worker connecting xprt %p via %s to "
2212                                 "%s (port %s)\n", xprt,
2213                         xprt->address_strings[RPC_DISPLAY_PROTO],
2214                         xprt->address_strings[RPC_DISPLAY_ADDR],
2215                         xprt->address_strings[RPC_DISPLAY_PORT]);
2216 
2217         xs_udp_finish_connecting(xprt, sock);
2218         trace_rpc_socket_connect(xprt, sock, 0);
2219         status = 0;
2220 out:
2221         xprt_unlock_connect(xprt, transport);
2222         xprt_clear_connecting(xprt);
2223         xprt_wake_pending_tasks(xprt, status);
2224 }
2225 
2226 /**
2227  * xs_tcp_shutdown - gracefully shut down a TCP socket
2228  * @xprt: transport
2229  *
2230  * Initiates a graceful shutdown of the TCP socket by calling the
2231  * equivalent of shutdown(SHUT_RDWR);
2232  */
2233 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2234 {
2235         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2236         struct socket *sock = transport->sock;
2237 
2238         if (sock == NULL)
2239                 return;
2240         if (xprt_connected(xprt)) {
2241                 kernel_sock_shutdown(sock, SHUT_RDWR);
2242                 trace_rpc_socket_shutdown(xprt, sock);
2243         } else
2244                 xs_reset_transport(transport);
2245 }
2246 
2247 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2248                 struct socket *sock)
2249 {
2250         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2251         unsigned int keepidle;
2252         unsigned int keepcnt;
2253         unsigned int opt_on = 1;
2254         unsigned int timeo;
2255 
2256         spin_lock_bh(&xprt->transport_lock);
2257         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2258         keepcnt = xprt->timeout->to_retries + 1;
2259         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2260                 (xprt->timeout->to_retries + 1);
2261         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2262         spin_unlock_bh(&xprt->transport_lock);
2263 
2264         /* TCP Keepalive options */
2265         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2266                         (char *)&opt_on, sizeof(opt_on));
2267         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2268                         (char *)&keepidle, sizeof(keepidle));
2269         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2270                         (char *)&keepidle, sizeof(keepidle));
2271         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2272                         (char *)&keepcnt, sizeof(keepcnt));
2273 
2274         /* TCP user timeout (see RFC5482) */
2275         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2276                         (char *)&timeo, sizeof(timeo));
2277 }
2278 
2279 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2280                 unsigned long connect_timeout,
2281                 unsigned long reconnect_timeout)
2282 {
2283         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2284         struct rpc_timeout to;
2285         unsigned long initval;
2286 
2287         spin_lock_bh(&xprt->transport_lock);
2288         if (reconnect_timeout < xprt->max_reconnect_timeout)
2289                 xprt->max_reconnect_timeout = reconnect_timeout;
2290         if (connect_timeout < xprt->connect_timeout) {
2291                 memcpy(&to, xprt->timeout, sizeof(to));
2292                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2293                 /* Arbitrary lower limit */
2294                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2295                         initval = XS_TCP_INIT_REEST_TO << 1;
2296                 to.to_initval = initval;
2297                 to.to_maxval = initval;
2298                 memcpy(&transport->tcp_timeout, &to,
2299                                 sizeof(transport->tcp_timeout));
2300                 xprt->timeout = &transport->tcp_timeout;
2301                 xprt->connect_timeout = connect_timeout;
2302         }
2303         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2304         spin_unlock_bh(&xprt->transport_lock);
2305 }
2306 
2307 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2308 {
2309         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2310         int ret = -ENOTCONN;
2311 
2312         if (!transport->inet) {
2313                 struct sock *sk = sock->sk;
2314                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2315 
2316                 /* Avoid temporary address, they are bad for long-lived
2317                  * connections such as NFS mounts.
2318                  * RFC4941, section 3.6 suggests that:
2319                  *    Individual applications, which have specific
2320                  *    knowledge about the normal duration of connections,
2321                  *    MAY override this as appropriate.
2322                  */
2323                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2324                                 (char *)&addr_pref, sizeof(addr_pref));
2325 
2326                 xs_tcp_set_socket_timeouts(xprt, sock);
2327 
2328                 write_lock_bh(&sk->sk_callback_lock);
2329 
2330                 xs_save_old_callbacks(transport, sk);
2331 
2332                 sk->sk_user_data = xprt;
2333                 sk->sk_data_ready = xs_data_ready;
2334                 sk->sk_state_change = xs_tcp_state_change;
2335                 sk->sk_write_space = xs_tcp_write_space;
2336                 sock_set_flag(sk, SOCK_FASYNC);
2337                 sk->sk_error_report = xs_error_report;
2338                 sk->sk_allocation = GFP_NOIO;
2339 
2340                 /* socket options */
2341                 sock_reset_flag(sk, SOCK_LINGER);
2342                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2343 
2344                 xprt_clear_connected(xprt);
2345 
2346                 /* Reset to new socket */
2347                 transport->sock = sock;
2348                 transport->inet = sk;
2349 
2350                 write_unlock_bh(&sk->sk_callback_lock);
2351         }
2352 
2353         if (!xprt_bound(xprt))
2354                 goto out;
2355 
2356         xs_set_memalloc(xprt);
2357 
2358         /* Tell the socket layer to start connecting... */
2359         xprt->stat.connect_count++;
2360         xprt->stat.connect_start = jiffies;
2361         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2362         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2363         switch (ret) {
2364         case 0:
2365                 xs_set_srcport(transport, sock);
2366         case -EINPROGRESS:
2367                 /* SYN_SENT! */
2368                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2369                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2370                 break;
2371         case -EADDRNOTAVAIL:
2372                 /* Source port number is unavailable. Try a new one! */
2373                 transport->srcport = 0;
2374         }
2375 out:
2376         return ret;
2377 }
2378 
2379 /**
2380  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2381  *
2382  * Invoked by a work queue tasklet.
2383  */
2384 static void xs_tcp_setup_socket(struct work_struct *work)
2385 {
2386         struct sock_xprt *transport =
2387                 container_of(work, struct sock_xprt, connect_worker.work);
2388         struct socket *sock = transport->sock;
2389         struct rpc_xprt *xprt = &transport->xprt;
2390         int status = -EIO;
2391 
2392         if (!sock) {
2393                 sock = xs_create_sock(xprt, transport,
2394                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2395                                 IPPROTO_TCP, true);
2396                 if (IS_ERR(sock)) {
2397                         status = PTR_ERR(sock);
2398                         goto out;
2399                 }
2400         }
2401 
2402         dprintk("RPC:       worker connecting xprt %p via %s to "
2403                                 "%s (port %s)\n", xprt,
2404                         xprt->address_strings[RPC_DISPLAY_PROTO],
2405                         xprt->address_strings[RPC_DISPLAY_ADDR],
2406                         xprt->address_strings[RPC_DISPLAY_PORT]);
2407 
2408         status = xs_tcp_finish_connecting(xprt, sock);
2409         trace_rpc_socket_connect(xprt, sock, status);
2410         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2411                         xprt, -status, xprt_connected(xprt),
2412                         sock->sk->sk_state);
2413         switch (status) {
2414         default:
2415                 printk("%s: connect returned unhandled error %d\n",
2416                         __func__, status);
2417         case -EADDRNOTAVAIL:
2418                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2419                  * and retry
2420                  */
2421                 xs_tcp_force_close(xprt);
2422                 break;
2423         case 0:
2424         case -EINPROGRESS:
2425         case -EALREADY:
2426                 xprt_unlock_connect(xprt, transport);
2427                 return;
2428         case -EINVAL:
2429                 /* Happens, for instance, if the user specified a link
2430                  * local IPv6 address without a scope-id.
2431                  */
2432         case -ECONNREFUSED:
2433         case -ECONNRESET:
2434         case -ENETUNREACH:
2435         case -EADDRINUSE:
2436         case -ENOBUFS:
2437                 /*
2438                  * xs_tcp_force_close() wakes tasks with -EIO.
2439                  * We need to wake them first to ensure the
2440                  * correct error code.
2441                  */
2442                 xprt_wake_pending_tasks(xprt, status);
2443                 xs_tcp_force_close(xprt);
2444                 goto out;
2445         }
2446         status = -EAGAIN;
2447 out:
2448         xprt_unlock_connect(xprt, transport);
2449         xprt_clear_connecting(xprt);
2450         xprt_wake_pending_tasks(xprt, status);
2451 }
2452 
2453 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2454 {
2455         unsigned long start, now = jiffies;
2456 
2457         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2458         if (time_after(start, now))
2459                 return start - now;
2460         return 0;
2461 }
2462 
2463 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2464 {
2465         xprt->reestablish_timeout <<= 1;
2466         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2467                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2468         if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2469                 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2470 }
2471 
2472 /**
2473  * xs_connect - connect a socket to a remote endpoint
2474  * @xprt: pointer to transport structure
2475  * @task: address of RPC task that manages state of connect request
2476  *
2477  * TCP: If the remote end dropped the connection, delay reconnecting.
2478  *
2479  * UDP socket connects are synchronous, but we use a work queue anyway
2480  * to guarantee that even unprivileged user processes can set up a
2481  * socket on a privileged port.
2482  *
2483  * If a UDP socket connect fails, the delay behavior here prevents
2484  * retry floods (hard mounts).
2485  */
2486 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2487 {
2488         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2489         unsigned long delay = 0;
2490 
2491         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2492 
2493         if (transport->sock != NULL) {
2494                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2495                                 "seconds\n",
2496                                 xprt, xprt->reestablish_timeout / HZ);
2497 
2498                 /* Start by resetting any existing state */
2499                 xs_reset_transport(transport);
2500 
2501                 delay = xs_reconnect_delay(xprt);
2502                 xs_reconnect_backoff(xprt);
2503 
2504         } else
2505                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2506 
2507         queue_delayed_work(xprtiod_workqueue,
2508                         &transport->connect_worker,
2509                         delay);
2510 }
2511 
2512 /**
2513  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2514  * @xprt: rpc_xprt struct containing statistics
2515  * @seq: output file
2516  *
2517  */
2518 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2519 {
2520         long idle_time = 0;
2521 
2522         if (xprt_connected(xprt))
2523                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2524 
2525         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2526                         "%llu %llu %lu %llu %llu\n",
2527                         xprt->stat.bind_count,
2528                         xprt->stat.connect_count,
2529                         xprt->stat.connect_time,
2530                         idle_time,
2531                         xprt->stat.sends,
2532                         xprt->stat.recvs,
2533                         xprt->stat.bad_xids,
2534                         xprt->stat.req_u,
2535                         xprt->stat.bklog_u,
2536                         xprt->stat.max_slots,
2537                         xprt->stat.sending_u,
2538                         xprt->stat.pending_u);
2539 }
2540 
2541 /**
2542  * xs_udp_print_stats - display UDP socket-specifc stats
2543  * @xprt: rpc_xprt struct containing statistics
2544  * @seq: output file
2545  *
2546  */
2547 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2548 {
2549         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2550 
2551         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2552                         "%lu %llu %llu\n",
2553                         transport->srcport,
2554                         xprt->stat.bind_count,
2555                         xprt->stat.sends,
2556                         xprt->stat.recvs,
2557                         xprt->stat.bad_xids,
2558                         xprt->stat.req_u,
2559                         xprt->stat.bklog_u,
2560                         xprt->stat.max_slots,
2561                         xprt->stat.sending_u,
2562                         xprt->stat.pending_u);
2563 }
2564 
2565 /**
2566  * xs_tcp_print_stats - display TCP socket-specifc stats
2567  * @xprt: rpc_xprt struct containing statistics
2568  * @seq: output file
2569  *
2570  */
2571 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2572 {
2573         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2574         long idle_time = 0;
2575 
2576         if (xprt_connected(xprt))
2577                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2578 
2579         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2580                         "%llu %llu %lu %llu %llu\n",
2581                         transport->srcport,
2582                         xprt->stat.bind_count,
2583                         xprt->stat.connect_count,
2584                         xprt->stat.connect_time,
2585                         idle_time,
2586                         xprt->stat.sends,
2587                         xprt->stat.recvs,
2588                         xprt->stat.bad_xids,
2589                         xprt->stat.req_u,
2590                         xprt->stat.bklog_u,
2591                         xprt->stat.max_slots,
2592                         xprt->stat.sending_u,
2593                         xprt->stat.pending_u);
2594 }
2595 
2596 /*
2597  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2598  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2599  * to use the server side send routines.
2600  */
2601 static int bc_malloc(struct rpc_task *task)
2602 {
2603         struct rpc_rqst *rqst = task->tk_rqstp;
2604         size_t size = rqst->rq_callsize;
2605         struct page *page;
2606         struct rpc_buffer *buf;
2607 
2608         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2609                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2610                           size);
2611                 return -EINVAL;
2612         }
2613 
2614         page = alloc_page(GFP_KERNEL);
2615         if (!page)
2616                 return -ENOMEM;
2617 
2618         buf = page_address(page);
2619         buf->len = PAGE_SIZE;
2620 
2621         rqst->rq_buffer = buf->data;
2622         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2623         return 0;
2624 }
2625 
2626 /*
2627  * Free the space allocated in the bc_alloc routine
2628  */
2629 static void bc_free(struct rpc_task *task)
2630 {
2631         void *buffer = task->tk_rqstp->rq_buffer;
2632         struct rpc_buffer *buf;
2633 
2634         buf = container_of(buffer, struct rpc_buffer, data);
2635         free_page((unsigned long)buf);
2636 }
2637 
2638 /*
2639  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2640  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2641  */
2642 static int bc_sendto(struct rpc_rqst *req)
2643 {
2644         int len;
2645         struct xdr_buf *xbufp = &req->rq_snd_buf;
2646         struct rpc_xprt *xprt = req->rq_xprt;
2647         struct sock_xprt *transport =
2648                                 container_of(xprt, struct sock_xprt, xprt);
2649         struct socket *sock = transport->sock;
2650         unsigned long headoff;
2651         unsigned long tailoff;
2652 
2653         xs_encode_stream_record_marker(xbufp);
2654 
2655         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2656         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2657         len = svc_send_common(sock, xbufp,
2658                               virt_to_page(xbufp->head[0].iov_base), headoff,
2659                               xbufp->tail[0].iov_base, tailoff);
2660 
2661         if (len != xbufp->len) {
2662                 printk(KERN_NOTICE "Error sending entire callback!\n");
2663                 len = -EAGAIN;
2664         }
2665 
2666         return len;
2667 }
2668 
2669 /*
2670  * The send routine. Borrows from svc_send
2671  */
2672 static int bc_send_request(struct rpc_task *task)
2673 {
2674         struct rpc_rqst *req = task->tk_rqstp;
2675         struct svc_xprt *xprt;
2676         int len;
2677 
2678         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2679         /*
2680          * Get the server socket associated with this callback xprt
2681          */
2682         xprt = req->rq_xprt->bc_xprt;
2683 
2684         /*
2685          * Grab the mutex to serialize data as the connection is shared
2686          * with the fore channel
2687          */
2688         if (!mutex_trylock(&xprt->xpt_mutex)) {
2689                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2690                 if (!mutex_trylock(&xprt->xpt_mutex))
2691                         return -EAGAIN;
2692                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2693         }
2694         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2695                 len = -ENOTCONN;
2696         else
2697                 len = bc_sendto(req);
2698         mutex_unlock(&xprt->xpt_mutex);
2699 
2700         if (len > 0)
2701                 len = 0;
2702 
2703         return len;
2704 }
2705 
2706 /*
2707  * The close routine. Since this is client initiated, we do nothing
2708  */
2709 
2710 static void bc_close(struct rpc_xprt *xprt)
2711 {
2712 }
2713 
2714 /*
2715  * The xprt destroy routine. Again, because this connection is client
2716  * initiated, we do nothing
2717  */
2718 
2719 static void bc_destroy(struct rpc_xprt *xprt)
2720 {
2721         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2722 
2723         xs_xprt_free(xprt);
2724         module_put(THIS_MODULE);
2725 }
2726 
2727 static struct rpc_xprt_ops xs_local_ops = {
2728         .reserve_xprt           = xprt_reserve_xprt,
2729         .release_xprt           = xs_tcp_release_xprt,
2730         .alloc_slot             = xprt_alloc_slot,
2731         .rpcbind                = xs_local_rpcbind,
2732         .set_port               = xs_local_set_port,
2733         .connect                = xs_local_connect,
2734         .buf_alloc              = rpc_malloc,
2735         .buf_free               = rpc_free,
2736         .send_request           = xs_local_send_request,
2737         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2738         .close                  = xs_close,
2739         .destroy                = xs_destroy,
2740         .print_stats            = xs_local_print_stats,
2741         .enable_swap            = xs_enable_swap,
2742         .disable_swap           = xs_disable_swap,
2743 };
2744 
2745 static struct rpc_xprt_ops xs_udp_ops = {
2746         .set_buffer_size        = xs_udp_set_buffer_size,
2747         .reserve_xprt           = xprt_reserve_xprt_cong,
2748         .release_xprt           = xprt_release_xprt_cong,
2749         .alloc_slot             = xprt_alloc_slot,
2750         .rpcbind                = rpcb_getport_async,
2751         .set_port               = xs_set_port,
2752         .connect                = xs_connect,
2753         .buf_alloc              = rpc_malloc,
2754         .buf_free               = rpc_free,
2755         .send_request           = xs_udp_send_request,
2756         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2757         .timer                  = xs_udp_timer,
2758         .release_request        = xprt_release_rqst_cong,
2759         .close                  = xs_close,
2760         .destroy                = xs_destroy,
2761         .print_stats            = xs_udp_print_stats,
2762         .enable_swap            = xs_enable_swap,
2763         .disable_swap           = xs_disable_swap,
2764         .inject_disconnect      = xs_inject_disconnect,
2765 };
2766 
2767 static struct rpc_xprt_ops xs_tcp_ops = {
2768         .reserve_xprt           = xprt_reserve_xprt,
2769         .release_xprt           = xs_tcp_release_xprt,
2770         .alloc_slot             = xprt_lock_and_alloc_slot,
2771         .rpcbind                = rpcb_getport_async,
2772         .set_port               = xs_set_port,
2773         .connect                = xs_connect,
2774         .buf_alloc              = rpc_malloc,
2775         .buf_free               = rpc_free,
2776         .send_request           = xs_tcp_send_request,
2777         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2778         .close                  = xs_tcp_shutdown,
2779         .destroy                = xs_destroy,
2780         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2781         .print_stats            = xs_tcp_print_stats,
2782         .enable_swap            = xs_enable_swap,
2783         .disable_swap           = xs_disable_swap,
2784         .inject_disconnect      = xs_inject_disconnect,
2785 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2786         .bc_setup               = xprt_setup_bc,
2787         .bc_up                  = xs_tcp_bc_up,
2788         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2789         .bc_free_rqst           = xprt_free_bc_rqst,
2790         .bc_destroy             = xprt_destroy_bc,
2791 #endif
2792 };
2793 
2794 /*
2795  * The rpc_xprt_ops for the server backchannel
2796  */
2797 
2798 static struct rpc_xprt_ops bc_tcp_ops = {
2799         .reserve_xprt           = xprt_reserve_xprt,
2800         .release_xprt           = xprt_release_xprt,
2801         .alloc_slot             = xprt_alloc_slot,
2802         .buf_alloc              = bc_malloc,
2803         .buf_free               = bc_free,
2804         .send_request           = bc_send_request,
2805         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2806         .close                  = bc_close,
2807         .destroy                = bc_destroy,
2808         .print_stats            = xs_tcp_print_stats,
2809         .enable_swap            = xs_enable_swap,
2810         .disable_swap           = xs_disable_swap,
2811         .inject_disconnect      = xs_inject_disconnect,
2812 };
2813 
2814 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2815 {
2816         static const struct sockaddr_in sin = {
2817                 .sin_family             = AF_INET,
2818                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2819         };
2820         static const struct sockaddr_in6 sin6 = {
2821                 .sin6_family            = AF_INET6,
2822                 .sin6_addr              = IN6ADDR_ANY_INIT,
2823         };
2824 
2825         switch (family) {
2826         case AF_LOCAL:
2827                 break;
2828         case AF_INET:
2829                 memcpy(sap, &sin, sizeof(sin));
2830                 break;
2831         case AF_INET6:
2832                 memcpy(sap, &sin6, sizeof(sin6));
2833                 break;
2834         default:
2835                 dprintk("RPC:       %s: Bad address family\n", __func__);
2836                 return -EAFNOSUPPORT;
2837         }
2838         return 0;
2839 }
2840 
2841 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2842                                       unsigned int slot_table_size,
2843                                       unsigned int max_slot_table_size)
2844 {
2845         struct rpc_xprt *xprt;
2846         struct sock_xprt *new;
2847 
2848         if (args->addrlen > sizeof(xprt->addr)) {
2849                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2850                 return ERR_PTR(-EBADF);
2851         }
2852 
2853         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2854                         max_slot_table_size);
2855         if (xprt == NULL) {
2856                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2857                                 "rpc_xprt\n");
2858                 return ERR_PTR(-ENOMEM);
2859         }
2860 
2861         new = container_of(xprt, struct sock_xprt, xprt);
2862         mutex_init(&new->recv_mutex);
2863         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2864         xprt->addrlen = args->addrlen;
2865         if (args->srcaddr)
2866                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2867         else {
2868                 int err;
2869                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2870                                         (struct sockaddr *)&new->srcaddr);
2871                 if (err != 0) {
2872                         xprt_free(xprt);
2873                         return ERR_PTR(err);
2874                 }
2875         }
2876 
2877         return xprt;
2878 }
2879 
2880 static const struct rpc_timeout xs_local_default_timeout = {
2881         .to_initval = 10 * HZ,
2882         .to_maxval = 10 * HZ,
2883         .to_retries = 2,
2884 };
2885 
2886 /**
2887  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2888  * @args: rpc transport creation arguments
2889  *
2890  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2891  */
2892 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2893 {
2894         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2895         struct sock_xprt *transport;
2896         struct rpc_xprt *xprt;
2897         struct rpc_xprt *ret;
2898 
2899         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2900                         xprt_max_tcp_slot_table_entries);
2901         if (IS_ERR(xprt))
2902                 return xprt;
2903         transport = container_of(xprt, struct sock_xprt, xprt);
2904 
2905         xprt->prot = 0;
2906         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2907         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2908 
2909         xprt->bind_timeout = XS_BIND_TO;
2910         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2911         xprt->idle_timeout = XS_IDLE_DISC_TO;
2912 
2913         xprt->ops = &xs_local_ops;
2914         xprt->timeout = &xs_local_default_timeout;
2915 
2916         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2917         INIT_DELAYED_WORK(&transport->connect_worker,
2918                         xs_dummy_setup_socket);
2919 
2920         switch (sun->sun_family) {
2921         case AF_LOCAL:
2922                 if (sun->sun_path[0] != '/') {
2923                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2924                                         sun->sun_path);
2925                         ret = ERR_PTR(-EINVAL);
2926                         goto out_err;
2927                 }
2928                 xprt_set_bound(xprt);
2929                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2930                 ret = ERR_PTR(xs_local_setup_socket(transport));
2931                 if (ret)
2932                         goto out_err;
2933                 break;
2934         default:
2935                 ret = ERR_PTR(-EAFNOSUPPORT);
2936                 goto out_err;
2937         }
2938 
2939         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2940                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2941 
2942         if (try_module_get(THIS_MODULE))
2943                 return xprt;
2944         ret = ERR_PTR(-EINVAL);
2945 out_err:
2946         xs_xprt_free(xprt);
2947         return ret;
2948 }
2949 
2950 static const struct rpc_timeout xs_udp_default_timeout = {
2951         .to_initval = 5 * HZ,
2952         .to_maxval = 30 * HZ,
2953         .to_increment = 5 * HZ,
2954         .to_retries = 5,
2955 };
2956 
2957 /**
2958  * xs_setup_udp - Set up transport to use a UDP socket
2959  * @args: rpc transport creation arguments
2960  *
2961  */
2962 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2963 {
2964         struct sockaddr *addr = args->dstaddr;
2965         struct rpc_xprt *xprt;
2966         struct sock_xprt *transport;
2967         struct rpc_xprt *ret;
2968 
2969         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2970                         xprt_udp_slot_table_entries);
2971         if (IS_ERR(xprt))
2972                 return xprt;
2973         transport = container_of(xprt, struct sock_xprt, xprt);
2974 
2975         xprt->prot = IPPROTO_UDP;
2976         xprt->tsh_size = 0;
2977         /* XXX: header size can vary due to auth type, IPv6, etc. */
2978         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2979 
2980         xprt->bind_timeout = XS_BIND_TO;
2981         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2982         xprt->idle_timeout = XS_IDLE_DISC_TO;
2983 
2984         xprt->ops = &xs_udp_ops;
2985 
2986         xprt->timeout = &xs_udp_default_timeout;
2987 
2988         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2989         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2990 
2991         switch (addr->sa_family) {
2992         case AF_INET:
2993                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2994                         xprt_set_bound(xprt);
2995 
2996                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2997                 break;
2998         case AF_INET6:
2999                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3000                         xprt_set_bound(xprt);
3001 
3002                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3003                 break;
3004         default:
3005                 ret = ERR_PTR(-EAFNOSUPPORT);
3006                 goto out_err;
3007         }
3008 
3009         if (xprt_bound(xprt))
3010                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3011                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3012                                 xprt->address_strings[RPC_DISPLAY_PORT],
3013                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3014         else
3015                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3016                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3017                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3018 
3019         if (try_module_get(THIS_MODULE))
3020                 return xprt;
3021         ret = ERR_PTR(-EINVAL);
3022 out_err:
3023         xs_xprt_free(xprt);
3024         return ret;
3025 }
3026 
3027 static const struct rpc_timeout xs_tcp_default_timeout = {
3028         .to_initval = 60 * HZ,
3029         .to_maxval = 60 * HZ,
3030         .to_retries = 2,
3031 };
3032 
3033 /**
3034  * xs_setup_tcp - Set up transport to use a TCP socket
3035  * @args: rpc transport creation arguments
3036  *
3037  */
3038 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3039 {
3040         struct sockaddr *addr = args->dstaddr;
3041         struct rpc_xprt *xprt;
3042         struct sock_xprt *transport;
3043         struct rpc_xprt *ret;
3044         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3045 
3046         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3047                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3048 
3049         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3050                         max_slot_table_size);
3051         if (IS_ERR(xprt))
3052                 return xprt;
3053         transport = container_of(xprt, struct sock_xprt, xprt);
3054 
3055         xprt->prot = IPPROTO_TCP;
3056         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3057         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3058 
3059         xprt->bind_timeout = XS_BIND_TO;
3060         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3061         xprt->idle_timeout = XS_IDLE_DISC_TO;
3062 
3063         xprt->ops = &xs_tcp_ops;
3064         xprt->timeout = &xs_tcp_default_timeout;
3065 
3066         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3067         xprt->connect_timeout = xprt->timeout->to_initval *
3068                 (xprt->timeout->to_retries + 1);
3069 
3070         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3071         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3072 
3073         switch (addr->sa_family) {
3074         case AF_INET:
3075                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3076                         xprt_set_bound(xprt);
3077 
3078                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3079                 break;
3080         case AF_INET6:
3081                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3082                         xprt_set_bound(xprt);
3083 
3084                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3085                 break;
3086         default:
3087                 ret = ERR_PTR(-EAFNOSUPPORT);
3088                 goto out_err;
3089         }
3090 
3091         if (xprt_bound(xprt))
3092                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3093                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3094                                 xprt->address_strings[RPC_DISPLAY_PORT],
3095                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3096         else
3097                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3098                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3099                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3100 
3101         if (try_module_get(THIS_MODULE))
3102                 return xprt;
3103         ret = ERR_PTR(-EINVAL);
3104 out_err:
3105         xs_xprt_free(xprt);
3106         return ret;
3107 }
3108 
3109 /**
3110  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3111  * @args: rpc transport creation arguments
3112  *
3113  */
3114 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3115 {
3116         struct sockaddr *addr = args->dstaddr;
3117         struct rpc_xprt *xprt;
3118         struct sock_xprt *transport;
3119         struct svc_sock *bc_sock;
3120         struct rpc_xprt *ret;
3121 
3122         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3123                         xprt_tcp_slot_table_entries);
3124         if (IS_ERR(xprt))
3125                 return xprt;
3126         transport = container_of(xprt, struct sock_xprt, xprt);
3127 
3128         xprt->prot = IPPROTO_TCP;
3129         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3130         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3131         xprt->timeout = &xs_tcp_default_timeout;
3132 
3133         /* backchannel */
3134         xprt_set_bound(xprt);
3135         xprt->bind_timeout = 0;
3136         xprt->reestablish_timeout = 0;
3137         xprt->idle_timeout = 0;
3138 
3139         xprt->ops = &bc_tcp_ops;
3140 
3141         switch (addr->sa_family) {
3142         case AF_INET:
3143                 xs_format_peer_addresses(xprt, "tcp",
3144                                          RPCBIND_NETID_TCP);
3145                 break;
3146         case AF_INET6:
3147                 xs_format_peer_addresses(xprt, "tcp",
3148                                    RPCBIND_NETID_TCP6);
3149                 break;
3150         default:
3151                 ret = ERR_PTR(-EAFNOSUPPORT);
3152                 goto out_err;
3153         }
3154 
3155         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3156                         xprt->address_strings[RPC_DISPLAY_ADDR],
3157                         xprt->address_strings[RPC_DISPLAY_PORT],
3158                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3159 
3160         /*
3161          * Once we've associated a backchannel xprt with a connection,
3162          * we want to keep it around as long as the connection lasts,
3163          * in case we need to start using it for a backchannel again;
3164          * this reference won't be dropped until bc_xprt is destroyed.
3165          */
3166         xprt_get(xprt);
3167         args->bc_xprt->xpt_bc_xprt = xprt;
3168         xprt->bc_xprt = args->bc_xprt;
3169         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3170         transport->sock = bc_sock->sk_sock;
3171         transport->inet = bc_sock->sk_sk;
3172 
3173         /*
3174          * Since we don't want connections for the backchannel, we set
3175          * the xprt status to connected
3176          */
3177         xprt_set_connected(xprt);
3178 
3179         if (try_module_get(THIS_MODULE))
3180                 return xprt;
3181 
3182         args->bc_xprt->xpt_bc_xprt = NULL;
3183         args->bc_xprt->xpt_bc_xps = NULL;
3184         xprt_put(xprt);
3185         ret = ERR_PTR(-EINVAL);
3186 out_err:
3187         xs_xprt_free(xprt);
3188         return ret;
3189 }
3190 
3191 static struct xprt_class        xs_local_transport = {
3192         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3193         .name           = "named UNIX socket",
3194         .owner          = THIS_MODULE,
3195         .ident          = XPRT_TRANSPORT_LOCAL,
3196         .setup          = xs_setup_local,
3197 };
3198 
3199 static struct xprt_class        xs_udp_transport = {
3200         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3201         .name           = "udp",
3202         .owner          = THIS_MODULE,
3203         .ident          = XPRT_TRANSPORT_UDP,
3204         .setup          = xs_setup_udp,
3205 };
3206 
3207 static struct xprt_class        xs_tcp_transport = {
3208         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3209         .name           = "tcp",
3210         .owner          = THIS_MODULE,
3211         .ident          = XPRT_TRANSPORT_TCP,
3212         .setup          = xs_setup_tcp,
3213 };
3214 
3215 static struct xprt_class        xs_bc_tcp_transport = {
3216         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3217         .name           = "tcp NFSv4.1 backchannel",
3218         .owner          = THIS_MODULE,
3219         .ident          = XPRT_TRANSPORT_BC_TCP,
3220         .setup          = xs_setup_bc_tcp,
3221 };
3222 
3223 /**
3224  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3225  *
3226  */
3227 int init_socket_xprt(void)
3228 {
3229 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3230         if (!sunrpc_table_header)
3231                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3232 #endif
3233 
3234         xprt_register_transport(&xs_local_transport);
3235         xprt_register_transport(&xs_udp_transport);
3236         xprt_register_transport(&xs_tcp_transport);
3237         xprt_register_transport(&xs_bc_tcp_transport);
3238 
3239         return 0;
3240 }
3241 
3242 /**
3243  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3244  *
3245  */
3246 void cleanup_socket_xprt(void)
3247 {
3248 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3249         if (sunrpc_table_header) {
3250                 unregister_sysctl_table(sunrpc_table_header);
3251                 sunrpc_table_header = NULL;
3252         }
3253 #endif
3254 
3255         xprt_unregister_transport(&xs_local_transport);
3256         xprt_unregister_transport(&xs_udp_transport);
3257         xprt_unregister_transport(&xs_tcp_transport);
3258         xprt_unregister_transport(&xs_bc_tcp_transport);
3259 }
3260 
3261 static int param_set_uint_minmax(const char *val,
3262                 const struct kernel_param *kp,
3263                 unsigned int min, unsigned int max)
3264 {
3265         unsigned int num;
3266         int ret;
3267 
3268         if (!val)
3269                 return -EINVAL;
3270         ret = kstrtouint(val, 0, &num);
3271         if (ret)
3272                 return ret;
3273         if (num < min || num > max)
3274                 return -EINVAL;
3275         *((unsigned int *)kp->arg) = num;
3276         return 0;
3277 }
3278 
3279 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3280 {
3281         if (kp->arg == &xprt_min_resvport)
3282                 return param_set_uint_minmax(val, kp,
3283                         RPC_MIN_RESVPORT,
3284                         xprt_max_resvport);
3285         return param_set_uint_minmax(val, kp,
3286                         xprt_min_resvport,
3287                         RPC_MAX_RESVPORT);
3288 }
3289 
3290 static const struct kernel_param_ops param_ops_portnr = {
3291         .set = param_set_portnr,
3292         .get = param_get_uint,
3293 };
3294 
3295 #define param_check_portnr(name, p) \
3296         __param_check(name, p, unsigned int);
3297 
3298 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3299 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3300 
3301 static int param_set_slot_table_size(const char *val,
3302                                      const struct kernel_param *kp)
3303 {
3304         return param_set_uint_minmax(val, kp,
3305                         RPC_MIN_SLOT_TABLE,
3306                         RPC_MAX_SLOT_TABLE);
3307 }
3308 
3309 static const struct kernel_param_ops param_ops_slot_table_size = {
3310         .set = param_set_slot_table_size,
3311         .get = param_get_uint,
3312 };
3313 
3314 #define param_check_slot_table_size(name, p) \
3315         __param_check(name, p, unsigned int);
3316 
3317 static int param_set_max_slot_table_size(const char *val,
3318                                      const struct kernel_param *kp)
3319 {
3320         return param_set_uint_minmax(val, kp,
3321                         RPC_MIN_SLOT_TABLE,
3322                         RPC_MAX_SLOT_TABLE_LIMIT);
3323 }
3324 
3325 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3326         .set = param_set_max_slot_table_size,
3327         .get = param_get_uint,
3328 };
3329 
3330 #define param_check_max_slot_table_size(name, p) \
3331         __param_check(name, p, unsigned int);
3332 
3333 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3334                    slot_table_size, 0644);
3335 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3336                    max_slot_table_size, 0644);
3337 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3338                    slot_table_size, 0644);
3339 
3340 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp